Chapter  153:  Breastfeeding and Maternal and Infant Health Outcomes in Developed Countries

Approach to Evaluating the Literature

Inclusion Criteria. As it was not feasible to review the large number of primary studies that are relevant to all the outcomes of interest, we consulted the Office on Women's Health and the technical expert panel (TEP) and developed an approach that capitalized on the existing large number of systematic reviews/meta-analyses. For outcomes of interest that have previously been reviewed systematically, we have summarized the findings from those reviews. For acute otitis media, childhood asthma, cognitive development, SIDS, infant mortality, NEC, maternal breast cancer, return to pre-pregnancy weight, and maternal type 2 diabetes, we have also updated those systematic reviews with data from primary studies published subsequent to those reviews. For outcomes that have not been previously evaluated systematically (osteoporosis, ovarian cancer, postpartum depression, infant mortality), we have reviewed those primary studies that met our inclusion criteria. Studies that examined only formula-fed infants were excluded.

Definitions of Breastfeeding. The majority of the studies did not distinguish between exclusive and partially breastfed infants, or explain the difference between “breastfeeding” and “feeding of expressed breast milk.” We elected to use the term “breastfeeding” for studies in full-term infants and the term “human milk feeding” for studies in preterm infants. We elected to accept all definitions of “exclusive breastfeeding” as provided by the different study authors but qualified our conclusions with respect to those specific definitions.

Literature Search Strategy. Comprehensive literature searches of MEDLINE^®, CINAHL, and the Cochrane Database of Systemic Reviews took place in November of 2005. Search terms included subject headings and text words relevant to breastfeeding and the different outcomes. Supplemental searches on selected outcomes were conducted through May of 2006. Other relevant studies were identified by technical experts or in bibliographies of selected reviews.

Specific Inclusion Criteria for Health Conditions Evaluated. We included systematic reviews, meta-analyses, observational studies, randomized controlled trials, and comparative studies that evaluated the effects or associations of breastfeeding on outcomes of interest. All studies must have either a comparator arm that evaluated formula feeding or a comparator arm that evaluated different durations of breastfeeding. Only studies conducted in developed countries were used in updates of systematic reviews/meta-analyses and de novo reviews of primary studies.

Reporting of Evidence

Methodological Quality Grade of Individual Studies. We used a three category grading system (A, B, C) to denote methodological quality of each primary study. We did not evaluate the methodological quality of the individual studies in the systematic reviews/meta-analyses.

A (good): Least bias and results are valid; a primary study that adheres mostly to the commonly held concepts of high quality

B (fair/moderate): Susceptible to some bias, but not sufficient to invalidate the results; a primary study that does not meet all the criteria in category A

C (poor): Significant biases that may invalidate the results; a primary study with serious errors in design, analysis or reporting

Methodological Quality Grade of Systematic reviews/Meta-analyses. We used a similar scheme as above for grading systematic reviews/meta-analyses. But we supplemented the scheme with the MOOSE guideline (standards for reporting for meta-analysis in observational studies in epidemiology) and an additional checklist of items that we devised to evaluate the quality of the systematic review of observational studies. Items in this checklist included questions on the following: appropriate search strategy; justification for inclusion/exclusion criteria for studies; description of well-defined population, intervention/exposure, comparator, outcomes and study designs; effort to minimize errors in data extraction; assessment of quality of individual studies; consideration on the effect of confounders; combinability of the data for meta-analysis; assessment of statistical and clinical heterogeneity; reporting accuracies; and appropriateness of the conclusions based on the reported data.

Full term Infant Outcomes

Acute Otitis Media. Our meta-analysis of five cohort studies of good and moderate methodological quality showed that breastfeeding was associated with a significant reduction in the risk of acute otitis media. Comparing ever breastfeeding with exclusive formula feeding, the risk reduction of acute otitis media was 23 percent (95% CI 9% to 36%). When comparing exclusive breastfeeding with exclusive formula feeding, either for more than 3 or 6 months duration, the reduction was 50 percent (95% CI 30% to 64%). These results were adjusted for potential confounders.

Atopic Dermatitis. One good quality meta-analysis of 18 prospective cohort studies on full term infants reported a reduction in the risk of atopic dermatitis by 42 percent (95% CI 8% to 59%) in children with a family history of atopy and exclusively breastfed for at least 3 months compared with those who were breastfed for less than 3 months. The meta-analysis did not distinguish between atopic dermatitis of infancy (under 2 years of age) and persistent or new atopic dermatitis at older ages. It has been postulated that the diagnosis of atopic dermatitis in patients younger than 2 years of age could be attributed to infectious etiologies, which may be prevented by breastfeeding. However, a stratified analysis by duration of followup found the risk reduction from breastfeeding was similar in subjects with less than 2 years compared with more than 2 years of followup.

Gastrointestinal Infections. For non-specific gastroenteritis, one systematic review identified three primary studies that controlled for potential confounders. These studies reported that there was a reduction in the risk of non-specific gastrointestinal infections during the first year of life in breastfed infants from developed countries. But a summary adjusted estimate taking into account potential confounders could not be determined because the studies did not provide usable quantitative data. However, a recent case-control study from England that took into account the role of potential confounders reported that infants who were breastfeeding had a 64 percent (95% CI 26% to 82%) reduction in the risk of non-specific gastroenteritis compared with infants who were not breastfeeding.

Lower Respiratory Tract Diseases. The summary estimate from a good quality meta-analysis of seven studies reported an overall 72 percent (95% CI 46% to 86%) reduction in the risk of hospitalization due to lower respiratory tract diseases in infants less than 1 year of age who were exclusively breastfed for 4 months or more. The results remained consistent after adjustment for potential confounders.

Asthma. The studies on asthma were equivocal. A previously published good quality meta-analysis reported a moderate protective effect and four recent primary studies reaching mixed conclusions, including two studies finding an increased risk of asthma associated with breastfeeding. We updated the meta-analysis with the new studies. Our analysis showed that breastfeeding for at least 3 months was associated with a 27 percent (95% CI 8% to 41%) reduction in the risk of asthma in those subjects without a family history of asthma compared with those who were not breastfed. For those with a family history of asthma, there was a 40 percent (95% CI 18% to 57%) reduction in the risk of asthma in children less than 10 years of age who were breastfed for at least 3 months compared with those who were not breastfed. However, the relationship between breastfeeding and the risk of asthma in older children and adolescents remains unclear and will need further investigation.

Cognitive Development. One well-performed sibling analysis and three prospective cohort studies of full-term infants, all conducted in developed countries, adjusted their analyses specifically for maternal intelligence. The studies found little or no evidence for an association between breastfeeding in infancy and cognitive performance in childhood. Most of the published studies adjusted their analyses for socioeconomic status and maternal education but not specifically for maternal intelligence. For those studies that reported a significant effect after specific adjustment for maternal intelligence, residual confounding from other factors such as different home environments cannot be ruled out.

Obesity. Three meta-analyses of good and moderate methodological quality reported an association of breastfeeding and a reduction in the risk of obesity in adolescence and adult life compared with those who were not breastfed. One study reported the reduction in the risk of overweight/obesity in breastfeeders compared with non-breastfeeders was 24 percent (95% CI 14% to 33%); another study reported 7 percent (95% CI 1% to 12%). Both of these estimates took into account the role of potential confounders. Furthermore, they also showed that the magnitude of association decreased when more confounders were entered into the analyses. The third study used meta-regression and found a 4 percent reduction in the risk of being overweight in adult life for each additional month of breastfeeding in infancy. Overall, there is an association between a history of breastfeeding and a reduction in the risk of being overweight or obese in adolescence and adult life. One should be cautious in interpreting all these associations because of the possibility of residual confounding.

Risk of Cardiovascular Diseases. Results from two moderate quality meta-analyses concluded that there was a small reduction of less than 1.5 mm Hg in systolic blood pressures and no more than 0.5 mm Hg in diastolic blood pressures among adults who were breastfed in their infancy compared with those who were formula-fed. The association weakened after stratification by study size, suggesting the possibility of bias in the smaller studies.

One meta-analysis of cohort and case-control studies reported that there was a reduction in total and LDL cholesterol levels by 7.0 mg/dL and 7.7 mg/dL, respectively, in adults who were breastfed during infancy compared with those who were not. However, these findings were based on data from adults with a wide age range. The analysis did not segregate the data according to gender and potential confounders were not explicitly analyzed. Detailed information (e.g., fasting or non-fasting) on the collection of specimen for cholesterol testing was not included. Because of these deficiencies, the correct characterization of a relationship between breastfeeding and adult cholesterol levels cannot be determined at this time.

One meta-analysis found little or no difference in all-cause and cardiovascular mortality between adults who were breastfed during infancy and those who were not. There were possible biases and limitations in the studies reviewed, however. Presence of statistical heterogeneity across studies suggests that it may not have been appropriate to combine estimates from individual studies into one summary estimate. Because of these reasons, no definitive conclusion could be drawn regarding the relationship between a history of breastfeeding and cardiovascular mortality.

In summary, the relationship between breastfeeding in infancy and the risk of cardiovascular diseases cannot be confidently characterized at this time and will need further investigation.

Type 1 Diabetes. Two moderate quality meta-analyses suggest that breastfeeding for at least 3 months reduced the risk of childhood type 1 diabetes compared with breastfeeding for less than 3 months. One reported a 19 percent (95% CI 11% to 26%) reduction; the other reported a 27 percent (95% CI 18% to 35%) reduction. In addition, findings from five of six studies published since the meta-analyses reported similar results. However, these results must be interpreted with caution because of the possibility of recall biases and suboptimal adjustments for potential confounders in the studies.

Type 2 Diabetes. In one well-performed meta-analysis of seven studies of various designs, breastfeeding in infancy was associated with a 39 percent (95% CI 15% to 56%) reduced risk of type 2 diabetes in later life compared with those who were not. However, only three of seven studies adjusted for all the important confounders such as birth weight, parental diabetes, socioeconomic status, and individual or maternal body size. Though the crude and adjusted estimates did not differ in these three studies, the lack of adjustments for potential confounders such as birth weight and maternal factors by all studies could exaggerate the magnitude of an association.

Childhood Leukemia. The published studies on childhood acute lymphocytic leukemia (ALL) were equivocal; a good quality meta-analysis reported a moderate protective effect from breastfeeding and the other good quality systematic review reached the opposite conclusion. We conducted a meta-analysis including only good and fair quality case-control studies identified in the systematic review, since the meta-analysis did not provide methodological quality grading of primary studies. We found breastfeeding of at least 6 months duration was associated with a 19 percent (95% CI 9% to 29%) reduction in the risk of childhood ALL. The previous meta-analysis also reported an association between breastfeeding of at least 6 months duration and a 15 percent reduction (95% CI 2% to 27%) in the risk of acute myelogenous leukemia (AML). Overall there is an association between a history of breastfeeding for at least 6 months duration and a reduction in the risk of both leukemias (ALL and AML).

Infant Mortality. One study of moderate methodological quality evaluated the relationship between breastfeeding and infant mortality. The study reported a protective effect of breastfeeding in reducing infant mortality after controlling for some of the potential confounders. However, in subgroup analyses of the study, the only statistically significant association reported was between “never breastfed” and Sudden Infant Death Syndrome (SIDS) or the risk of injury-related deaths. Because of the limited data in this area, the relationship between breastfeeding and infant mortality in developed countries remains unclear. Further investigation is needed.

Sudden Infant Death Syndrome (SIDS). We conducted a meta-analysis by including only studies that reported clear definitions of exposure, outcomes, and results adjusted for well-known confounders or risk factors for SIDS. Our meta-analysis of seven case-control studies found that a history of breastfeeding was associated with a 36 percent (95% CI 19% to 49%) reduction in the risk of SIDS compared to those without a history of breastfeeding.

Preterm Infant Outcomes

Cognitive Development. No definitive conclusion can be made regarding the relationship between breast milk feeding and cognitive development in preterm infants. One meta-analysis reported a five points advantage in standardized mean score and one systematic review identified one primary study that reported an eight points advantage in IQ in preterm or low birth weight infants who received breast milk feeding. In three of four primary studies of moderate quality that controlled for either maternal education or maternal intelligence, the advantage from breastfeeding was reduced to a statistically non-significant level after adjustment. The roles of maternal intelligence and home environment should be accounted for in future studies on breastfeeding and cognitive development. Keeping in mind that cognitive function measured at an early age is not necessarily predictive of later cognitive ability, one should also consider carefully the timing and the selection of appropriate testing instrument in future studies.

Necrotizing Enterocolitis (NEC). Our meta-analysis of four randomized controlled trials of breast milk versus formula in comparing the outcome of NEC demonstrated that there was a marginally statistically significant association between a history of breast milk feeding and a reduction in the risk of NEC (P = 0.04). The estimate of the reduction in relative risk ranged from 4 percent to 82 percent. The absolute risk difference between the two groups was 5 percent. Because of the high case-fatality rate of NEC, this difference is a meaningful clinical outcome. The wide range of the estimate reflects the relatively small number of total subjects in the studies and the small number of events. One must also be cognizant of the heterogeneity underlying these trials in interpreting the findings of the meta-analysis. Examples of which included gestational age that ranged from 23 to more than 33 weeks; birth weight ranged from less than 1,000 g to more than 1,600 g; and some trials included only “healthy” infants, while others included both “healthy” and “ill” infants.

Maternal Outcomes

Return to Pre-pregnancy Weight. Three moderate quality prospective cohort studies reported less than 1 kg weight change from pre-pregnancy or first trimester to 1 to 2 year postpartum period in mothers who breastfed. Results from four moderate quality prospective cohort studies showed that the effects of breastfeeding on postpartum weight loss were unclear. Results from all seven studies consistently showed that many factors other than breastfeeding had larger effects on weight retention or postpartum weight loss. Methodological challenges in these studies included the accurate measurement of weight change, adequate control for numerous covariables including the amount of pregnancy weight gain, and quantifying accurately the exclusivity and the duration of breastfeeding.

Maternal Type 2 Diabetes. Two large cohorts from a high quality longitudinal study of 150,000 parous women in the United States examined the relationship between breastfeeding and the risk of maternal type 2 diabetes. In parous women without a history of gestational diabetes, each additional year of breastfeeding was associated with a 4 percent (95% CI 1% to 9%) reduced risk of developing type 2 diabetes in the first cohort and a 12 percent (95% CI 6% to 18%) reduced risk in the second cohort. In women with a history of gestational diabetes, breastfeeding had no significant effect on the already increased risk of diabetes. Because only nurses were included in the cohorts, generalization of findings to the rest of the population must be done with care.

Osteoporosis. There is little or no evidence from six moderate quality case-control studies for an association between lifetime breastfeeding duration and the risk of fractures due to osteoporosis. In two of three moderate or good quality prospective cohort studies using bone mineral density as a surrogate for osteoporosis, lactation does not appear to have an effect on long-term changes in bone mineral densities. The third study found a small decrease in the bone mineral contents in the distal radius with increased duration of breastfeeding, but no significant changes in bone mineral contents in the femoral neck or the trochanter.

Postpartum Depression. Four prospective cohort studies of moderate methodological quality reported on the relationship between a history of breastfeeding and postpartum depression. None of the studies explicitly screened for depression at baseline before the initiation of breastfeeding and none of them provided detailed data on breastfeeding. Three of the four studies found an association between a history of short duration of breastfeeding or not breastfeeding with postpartum depression. The results were adjusted for socio-demographic and obstetric variables. More investigation will be needed to determine the nature of this association. It is plausible that postpartum depression led to early cessation of breastfeeding, as opposed to breastfeeding altering the risk of depression. Both effects might occur concurrently.

Breast Cancer. Two meta-analyses of moderate methodological quality concluded that there was a reduction of breast cancer risk in women who breastfed their infants. The reduction in breast cancer risk was 4.3 percent for each year of breastfeeding in one meta-analysis and 28 percent for 12 or more months of breastfeeding in the other. In addition, one of the two meta-analyses and another systematic review reported decreased risk of breast cancer primarily in premenopausal women. Findings from primary studies published after the meta-analyses concurred with the findings from the earlier meta-analyses. In summary, consistent evidence from these studies suggests that there is an association between breastfeeding and a reduced risk of breast cancer.

Ovarian Cancer. We reviewed 15 case-control studies that examined the relationship between breastfeeding and the risk of ovarian cancer, and performed meta-analyses in nine studies that adjusted for potential confounders. The overall result from the nine studies showed an association between breastfeeding and a 21 percent (95% CI 9% to 32%) reduction in the risk of ovarian cancer, compared to never breastfeeding. Because not all the studies reported similar comparisons of breastfeeding durations, we had to estimate the comparable risks in five studies. Excluding these five studies from the meta-analysis results in loss of statistical significance for this association.

There was indirect evidence for a dose-response relationship between breastfeeding and a reduced risk of ovarian cancer. Breastfeeding of more than 12 months (cumulative duration) was associated with a reduced risk of ovarian cancer, compared to never breastfeeding. The 12-month cutoff was arbitrary, and the odds ratios were estimated in half of these studies.

Overall, there is evidence to suggest an association between breastfeeding and a reduction in the risk of maternal ovarian cancer. Because of the aforementioned limitations, one must be cautious in interpreting this association.

Limitations

With the availability of many published systematic reviews on breastfeeding, we used this literature as the evidence for a large number of outcomes, supplemented by updates of these systematic reviews with new primary studies. Even though we have assessed the reporting quality of these systematic reviews (using standards of reporting of systematic reviews of observational studies (MOOSE statement), and additional parameters that we devised), we cannot reliably know the validity of the reported summary data without knowing the details of the primary studies. It should also be stressed that a well-performed systematic review does not necessarily imply that the body of evidence for a particular outcome of interest is of high quality. Any systematic review is limited by the quality of the primary studies included in the review. Unless the method used to assess the quality of the primary studies is transparent and the details made available for examination, it would be difficult to reliably determine the validity of the conclusions.

The breastfeeding literature is primarily comprised of observational studies, either cohort or case-control studies. There are a number of potential deficiencies related to the observational study designs that could limit the internal validity and the generalizability of the findings. Some of these potential deficiencies include (1) misclassification of exposure; (2) confounding from the process of self-selection; and (3) residual confounding.

We have summarized the effects of breastfeeding (or breast milk feeding) on a large number of infant and maternal outcomes. Some of the outcomes are well defined and specific (e.g., childhood acute lymphocytic leukemia, breast cancer); and some are not so well defined and non-specific (e.g., asthma, non-specific gastrointestinal infections). When the reported outcome is well defined and specific, it lends confidence that the effect reported is valid for that outcome. When the reported outcome is not well defined, one might have some reservation regarding the validity of the measured effect for that outcome. For all the above reasons, we find that there is a wide range of quality of evidence for the different outcomes examined in this review.

An important area of research that is not systematically reviewed in this report is the use of breastfeeding promotion intervention trial to measure health effects (this topic is not part of the scope of this report and it will be covered in a separate report). The best known of these types of studies is the Promotion of Breastfeeding Intervention Trial (PROBIT) conducted in the Republic of Belarus. Data from this study provided good evidence that breastfeeding is associated with a reduction in the risk of gastrointestinal infection and atopic dermatitis.

Lastly, the outcomes analyzed in this review represent only a portion of all possible health outcomes related to breastfeeding reported by investigators worldwide. To work within the constraints of resources, we relied on the advice from our panel of technical experts in finalizing the list of outcomes included in this review. Thus, some important outcomes (e.g., growth and nutrition) have, by necessity, not been included in this review.

Future Research

Observational studies will remain the major source of information in this field. Clear subject selection criteria, adopting a common definition of “exclusive breastfeeding”, reliable collection of feeding data, specific and properly quantifiable outcomes of interest, controlling for important potential confounders including child-specific factors, and blinded assessment of the outcome measures will help immeasurably to improve the quality of these studies.

Sibling analysis provides a method to control for hereditary and household factors that are important in certain outcomes, provided that those factors are similar for the siblings of interest. Although such analysis may be less susceptible to confounders and effect modifiers that are shared by siblings, one must remember that it is not immune to biases. This method should be used when the appropriate data are available.

Cluster randomized controlled studies similar to the Belarus trial will provide understanding of the effectiveness of various breastfeeding promotion interventions. Any substantial differences in the degree of breastfeeding between the two groups as a result of the intervention will provide further opportunity to investigate any disparity in health outcomes between the two groups.

Selection of Outcomes for This Review

The TEP offered advice on selection of outcomes for review. Final selection of the list of outcomes for review took into account the following factors: the importance of the outcome, whether a systematic review of the outcome has previously been reported from a developed country, whether the existing systematic review of the outcome is outdated, whether the relationship between breastfeeding and the outcome is thought to be equivocal, whether a large number of primary studies has been published recently on the outcome, and the total number of outcomes that could be adequately reviewed for this report given the time constraint.

We included the following outcomes in this report:

Term infant outcomes: acute otitis media, hospitalization for lower respiratory tract infection, gastrointestinal infection, hypertension, cardiovascular diseases, hyperlipidemia, asthma, atopic dermatitis, type 1 and 2 diabetes, obesity, sudden infant death syndrome (SIDS), infant mortality, cognitive development, and childhood cancer (including leukemia)

Preterm infant outcomes: necrotizing enterocolitis (NEC) and cognitive development

Maternal outcomes: maternal weight changes, breast cancer, ovarian cancer, post-partum depression, osteoporosis, and type 2 diabetes

Abstract Screening

All abstracts identified through the literature search were screened. At this stage, eligible studies included all English language primary experimental or observational studies that reported any health outcomes in human subjects in relation to a history of breast milk feeding. As the inclusion criteria were broad at this stage, the abstracts rejected at this stage did not undergo a second rescreening process. Abstracts that were accepted at this stage were examined a second time by different reviewers and categorized into the different outcomes of interest.

Full Article Inclusion/Exclusion Criteria

Articles that passed the abstract screening process were retrieved and the full articles were reviewed for eligibility. Full articles were examined only once unless the articles were equivocal for inclusion or exclusion. In that event, the article in question was screened again by a different reviewer and a consensus was reached after discussion with the first reviewer.

Because the outcomes selected ranged from very broad topic with common occurrence (e.g., non-specific gastrointestinal infection) to a narrowly focused topic with relatively few occurrences (e.g., SIDS), the types of studies available for each outcome varied widely in the distribution of study designs, sample sizes, and quality of breastfeeding data, it was not possible (nor desirable) to design a strict set of inclusion and exclusion criteria that would be applicable to all outcomes. Therefore, additional inclusion/exclusion criteria germane to the specific outcome were also described in the Results section under each outcome.

General inclusion criteria for the studies are as follow:

Study Design. Systematic reviews, experimental (randomized controlled trials) and observational studies (prospective cohort and case-control studies only)

Population. Healthy term infants in developed countries; preterm infants in developed countries (for NEC and cognitive development); healthy mothers in developed countries

Intervention/Exposure. Breastfeeding, breast milk feeding (maternal term and preterm milk, banked term and preterm milk, fortified or unfortified), exclusive or mixed feeding

Comparator. Formula feeding (preterm or term formula, fortified or unfortified)

Data Extraction and Analysis

For those outcomes that have been subjected to a systematic review/meta-analysis, we summarized their results into our report. In addition to the results from the systematic reviews, we have also extracted and summarized the relevant data from primary studies that were published after the latest search dates of the reports for the following infant and maternal outcomes: acute otitis media, childhood asthma, cognitive development, SIDS, infant mortality, NEC, and maternal breast cancer.

For systematic reviews/meta-analyses that reported data from both developed and developing countries, we reported only those results pertaining to developed countries, if the reported data permitted us to do so. In those instances where that were not possible, we noted that fact as a limitation of our findings.

For outcomes that have multiple systematic reviews, we noted the overlapping studies and examined whether their findings were interpreted similarly or differently across reviews and reported our analyses.

For NEC, maternal weight changes, and acute otitis media, in order to better clarify the overall findings, we also extracted relevant data from the primary studies cited in the systematic reviews and combined them with data from the primary studies that were published after the latest search dates of the reports and reanalyzed the data.

For the remaining included outcomes, we extracted and summarized the relevant data from the primary studies.

Data forms were developed separately for extraction of systematic reviews/meta-analysis and primary studies. For systematic reviews/meta-analysis, items extracted were: databases searched, study design, population characteristics, descriptions of intervention/exposure, models used for meta-analysis, results, and authors' conclusions. We reported the estimates in the meta-analyses. We also reported any attempt by the authors of the meta-analyses to explore heterogeneity using sub-group analyses or meta-regression. For primary studies, items extracted were: study design, population characteristics, eligibility criteria, descriptions of intervention/exposure, any adjustments for confounders, and results.

Background

Acute otitis media (AOM) is a common childhood infection. It often begins with an upper respiratory tract infection. The viral infection predisposes the child to the development of AOM by causing eustachian tube dysfunction. The eustachian tube dysfunction enhances nasopharyngeal colonization with middle ear pathogens. The prevalence of a first attack of AOM in children under l year of age was estimated to be 44 percent.²⁵ Almost 70 percent of children under 6 years of age have had an episode of AOM.²⁵ Several risk factors have been identified for increasing the occurrence and recurrence of AOM.²⁶ There is a general consensus that breastfeeding protects against many infections, including AOM. Breast milk contains immunoglobulins with antibody activity against common bacteria such as Haemophilus influenzae and Streptococcus pneumoniae. It also contains components that interfere with the attachment of Haemophilus influenzae and Streptococcus pneumoniae to nasopharyngeal epithelial cells. The intermittent administration of milk with anti-adhesive substances into the nasopharynx of the nursing child may reduce the extent of colonization and protect against infection.²⁷

Commonly considered confounders in the studies of the relationship between breastfeeding and AOM were parental history of allergy, number of siblings, use of day care, maternal smoking, gender, ethnicity, and socioeconomic status.

We identified one meta-analysis of risk factors for AOM that aimed to further clarify possible means of preventing AOM in childhood.²⁸ Duration of breastfeeding was one of the factors examined. Since this meta-analysis, we identified six studies in seven publications that examined the relationship between breastfeeding and AOM.

Additional Methodological Comments

Table 1

Summary of systematic review/meta-analysis on the relationship (more...)

Table 1

Summary of systematic review/meta-analysis on the relationship between breastfeeding and acute otitis media (AOM)

Author Year Population	Study description	Intervention /Comparator	Confounders considered	Results	Quality of SR/MA and limitations
Uhari 1996	MA of 2 case-control and 8 cohort studies in developed countries	Any measures of breastfeeding and comparators	ND	10 studies evaluated the risk of AOM and recurrence of AOM associated with breastfeeding. The diagnosis of AOM varied, but pneumatic otoscopy was used in the diagnosis. There was no restriction on study inclusion according to the diagnostic criteria used. 22 studies evaluating an array of risk factors for development and recurrence of AOM were included in the meta-analysis. Seven of the twenty-two studies specifically evaluated the association between breastfeeding and AOM, while another 3 studies included breastfeeding among other risk factors that were evaluated.	C
AOM in children.				Breastfeeding for at least 3 months reduced the risk of AOM (RR=0.87; 95%CI: 0.79–0.95)	No consideration for potential confounding, no restriction on inclusion according to the diagnostic criteria used for AOM; unclear how breastfeeding data was collected
Specific age groups not specified in the cohorts				Children who were breastfed for 3 or more months had a non-statistically significant reduced risk of recurrent AOM compared with those who were breastfed less than 3 months (RR 0.69, 95%CI 0.46 – 1.03).
Case- control: 671				When children who were breastfed for 6 months or more were compared to those who were breastfed for less than 6 months, a statistically significant reduced risk of recurrent AOM was found (RR=0.69; 95%CI: 0.49–0.97).
Cohort: 4,455

AOM, Acute Otitis Media; SR, systematic review; MA, meta-analysis

Table 2

Summary table for cohort studies on the relationship (more...)

Table 2

Summary table for cohort studies on the relationship between breastfeeding and acute otitis media

Author Year Country Design	Number of subjects (Baseline /Followup)	Definition of AOM	Mean duration Followup	Breastfeeding group (n)	Comparator group (n)	OR (95% CI)		Confounders / risk factors adjusted	Quality and limitations
						Crude	Adjusted
Studies in update search
Vernacchio 2004	15,113 /11,349	Mother selects ear infection from a list of diagnoses by physician in the past month	6 mo	BF at 6 mo (2,771)	No BF at 6 mo (8,558)	0.66 (0.59–0.74)	0.69 (0.61–0.78)	Race/ethnicity, gender, daycare attendance, cigarette smoking, income, access to care	B
USA
Prospective cohort
Scariati 1997	1,743 / 1410	Mother reported symptoms	7 mo	Exclusive BF at 6 mo (299)	Exclusive formula at 6 mo (891)	0.45	0.56, p<0.01	Gender, maternal education, occupation, smoking, households size and income, day care use	B
US					1–57% mixed feeding (81)	0.56	0.63, p=0.03
Prospective cohort					58–88% mixed feeding (80)	0.67	0.71, p<0.05
					89–99% mixed feeding (59)	0.83	0.83, p=NS
Sassen 1994	289 /232	Clinical diagnosis by physician.	23.6 mo	Overall effect before stopping BF (OR per month)	Up to 4 mo after stopping BF		0.92 (0.76–1.07)	Number of siblings, SES	B
Netherlands		Tympanometry results not used.
Prospective cohort
Duffy 1997	Exclusive breast: 178 / 53 @ 3 mo / 28 @ 6 mo	Exclusive formula: 110 @ baseline	Exclusive formula: 110 @ baseline			RR (95% CI)		Age at colonization, day care, maternal smoking	B
Finlandc	Combined BF and formula: 18 @ baseline			Exclusive 3-mo BF	Exclusive 3-mo formula	0.72 (0.52–1.00)	0.72 (0.52–1.00)
Prospective cohort	Exclusive formula: 110 @ baseline			Exclusive 6-mo BF	Exclusive 6-mo formula	0.55 0.34–0.87)	0.55 0.34–0.87)
				6-mo mixed feeding	Exclusive 6-mo formula	0.71 (0.53–0.96)	0.30 (0.19–0.48)
Alho 1996 (2 papers)	2,512 / 825	>1 acute symptoms and one pneumatoscopic finding up to age 2 yrs	22 mo	Ever BF (nd)	Never BF (nd)		0.8 (0.7–0.9) e	Gender, number of siblings, atopy, age, season of birth, day care, parental smoking	C
Finland				BF > 3 mo (735)	BF < 3 mo (90)	0.9 (0.8–1.0)			High dropout rate
Retrospective cohort
Stenstrom, 1997	AOM cases: 179	> 5 episodes of AOM (Otoscopic diagnosis) before age of 30 mo	N/A	Ever BF	Never BF	NS		None	C
Sweden	Control d: 305								Unadjusted estimate only
Case-control study
Studies in MA that reported adjusted OR / RR
Duncan 1993	1220 / 1013	Diagnosed by experienced clinicians with an average of > 9 yrs in pediatric practice	12 mo	Exclusive BF ≥ 6 mo (154)	No BF and BF < 4 mo (465)		0.61 (0.40–0.92)	Parental history of allergy, number of siblings, use of day care, maternal smoking, gender, ethnicity, and maternal education	A
US				BF ≥ 4 mo, suppl 4–6 mo (199)			0.72 (0.54–0.95)
Prospective cohort				BF ≥ 4 mo, suppl <4 mo (200)			0.85 (0.74–0.97)
Teele 1989	1067 / 877	Effusion in one or both middle ears accompanied by ≥ 1 signs of acute illnessb	12 mo	Ever BF (292)	Never BF (585)	≥ 1 episode of AOM: 0.83	≥ 1 episode of AOM: 0.64 (0.44–0.91)	Site of health care, season of birth, birth weight, gender, SES, number of siblings, sibling or parental history of infection, parental smoking	B
US							≥ 3 episode of AOM: 0.51 (0.30–0.89)
Prospective cohort
Howie 1990	750 / 617	Mother reported that infant experienced painful or discharging ear lasting for 48 hrs or more	12 mo	Exclusive BF ≥ 13 wk (89) or partial BF ≥ 13 wk (121)	Bottle feeders (246) and BF < 13 wk(161)	0.7 (0.5–1.0)	NS in the % infants experienced ear infection	Father's social class, maternal age, and parental smoking	B
Scotland
Prospective cohort

BF, breastfeeding; excl, exclusive; ROM, recurrent otitis media; SES, socioeconomic status

a

Mean duration of exclusive BF (95% CI)

b

Signs of acute illness including earache, otorrhea, ear tugging, fever, irritability, lethargy, anorexia, vomiting or diarrhea

c

Proportion of infants in the individual categories of feeding method at different time cut-offs is not available

d

Matched to the case by age and gender

e

Values were estimated from figure. Statistically modeled to fit the whole sample.

Studies Identified after the Systematic Review/Meta-analysis (Table 2)

A total of five cohort studies (in 6 publications)^{3, 32}–³⁶ and one case-control study³⁷ that evaluated the relationship of breastfeeding and AOM published after the search dates of Uhari's meta-analysis met our inclusion criteria. We included only studies conducted in developed countries among children without co-morbidity. In cohort studies, subjects in the studies were followed from birth to a mean of 6 to 24 months. The number of subjects evaluated ranged from 289 to 15,113 at baseline. Four of the five cohort studies were of methodological quality grade B, while the other one was of grade C. In the case-control study, the children were between the ages of 3 and 7 years at the time of examination. A total of 179 AOM cases and 305 controls were analyzed. The methodological quality of the case-control study was rated C because the analysis did not control for potential confounding.

Studies varied in the definitions of breastfeeding and comparison groups. All studies were designed to evaluate a broad range of potential risk factors in AOM, except for Scariati 1997 in which the study specifically aimed to examine the relationship between breastfeeding and infections. We focused only on the relationship between breastfeeding and AOM. The studies also varied in their definitions of the disease conditions. In most studies, definitions of AOM were based on clinical features combined with otoscopic findings. The data were collected from medical records and confirmed by a physician in all studies except for Vernacchio 2004 and Scariati 1997 where a mother was asked to report ear infection in a list of diagnoses or symptoms.

All cohort studies adjusted for potential confounders and the case-control study did not. Confounding factors considered in the studies included gender, number of siblings, family day care, nursery day care, number of children in the home, maternal age, parental race or ethnicity, parity, maternal marital status, and parental smoking.

Breastfeeding was associated with a reduced risk of AOM compared to bottle-feeding after controlling for potential confounders across the five cohort studies, although some studies did not report a statistically significant result.

When exclusive breastfeeding for more than 3 or 6 months was compared with exclusive bottle- or formula- feeding, one study reported 28 percent³³ and another study reported 45 percent³ relative risk reduction in AOM. The case-control study did not show a significant difference in the risk of AOM between the children who were ever breastfed and those who did not. However, the analysis did not control for any potential confounding factors.

Meta-analysis. In addition to the five cohort studies in the update search, three additional cohort studies that reported adjusted odds ratios or risk ratios of AOM from Uhari 1996 were reviewed for possible inclusion into our meta-analyses. Studies were heterogeneous in the definitions of breastfeeding and comparison groups. In order to minimize the heterogeneity, we restricted our analyses to studies that reported an adjusted odds ratio or risk ratio of AOM comparing any definition of breastfeeding duration to exclusive bottle-feeding. Five studies that reported data from a total of six comparisons met the inclusion criteria for our meta-analyses.^{3, 29, 31, 33, 35} Using a random effects model, pooled data from five cohort studies of good and moderate methodological quality showed an adjusted odds ratio of AOM of 0.60 (95%CI 0.46–0.78) comparing breastfeeding to exclusive bottle-feeding. There is a significant difference in the risk reduction of AOM between the studies comparing exclusive breastfeeding with exclusive bottle-feeding and the studies comparing ever breastfeeding with exclusive bottle-feeding (P<0.01). Specifically, the pooled adjusted odds ratio of AOM was 0.77 (95%CI 0.64–0.91), when comparing children who were ever breastfed with children who were exclusively bottle-fed in two studies. The pooled adjusted odds ratio was 0.50 (95%CI 0.36–0.70), when comparing children who were exclusively breastfed for at least 3 or 6 months with those who were exclusively bottle-fed for at least 3 or 6 months in three studies (providing 4 estimates).

The three studies that were excluded from our meta-analyses compared breastfeeding for more than 13 weeks, 4 months, and 6 months with breastfeeding for less than 13 weeks, 4 months, and less than 6 months, respectively.^{29, 30, 32} There was no significant association between the risk of AOM and breastfeeding for more than 13 weeks, while a significant risk reduction in AOM was found when comparing children who were breastfed for more than 4 months or 6 months with those who were breastfed for less than 4 months or 6 months, respectively.

Conclusion

The results from our meta-analyses of cohort studies of good and moderate methodological quality showed that breastfeeding was associated with a significant reduction in the risk of AOM. Comparing ever breastfeeding with exclusive bottle-feeding, the pooled adjusted odds ratio of AOM was 0.77 (95%CI 0.64 – 0.91). When comparing exclusive breastfeeding with exclusive bottle-feeding, either for more than 3 or 6 months duration, the pooled odds ratio was 0.50 (95%CI 0.36 – 0.70).

Background

Atopic dermatitis is a common problem with an estimated lifetime prevalence in children of 10–20 percent.³⁸ Many studies have investigated the possible protective effect of breastfeeding on the development of atopic dermatitis. The results have been conflicting.^{16, 38, 39} Potential confounders considered in the studies included gender, socioeconomic status, family history of atopy, parental smoking, and presence of furry animals in the home.⁴⁰

Published Systematic Review/Meta-Analysis (Table 3)

We identified one systematic review/meta-analysis that examined the relationship between breastfeeding and the development of atopic dermatitis.⁴⁰ The methodological quality of this systematic review/meta-analysis was rated grade A.

Using the MEDLINE database from 1966 to 2000, Gdalevich 2001 identified 18 prospective studies from developed countries that qualified for inclusion in the review. Sample size of the studies ranged from 17 to 991. A total of 4,158 participants were included. Mean followup duration was 4.5 years. Study selection criteria included term infant, restriction of maternal recall of the child's feeding history limited to the first 12 months of the infants' life, the breastfeeding group in the study were exclusively breastfed for at least 3 months, blinding of the feeding history during outcome assessment, strict diagnostic criteria of atopic dermatitis provided by the authors, and control for confounding variables like socioeconomic status and family history of atopy.

Using a fixed effect model, the overall summary odds ratio for the development of atopic dermatitis was 0.68 (95% CI 0.52 – 0.88) in those subjects with at least 3 months of exclusive breastfeeding versus subjects without 3 months of exclusive breastfeeding. When the analysis was restricted to those studies with positive family history of atopy, the odds ratio was 0.58 (95%CI 0.41 – 0.92). When the analysis was restricted to those studies without a family history of atopy, the odds ratio was 0.84 (95%CI 0.59 – 1.19).

The systematic review did not make a distinction between atopic dermatitis of infancy (under 2 years of age) and persistent or new atopic dermatitis at older ages. However, in a primary analysis, the data were stratified according to different durations of followup (because the diagnosis of atopic dermatitis in patients younger than 2 years of age are sometimes attributed to symptoms of infectious origin and breastfeeding may have a protective effect against infections). The summary odds ratio in the group with less than 2 years of followup was 0.74 (95%CI 0.61 – 0.90), whereas the summary odds ratio in the group with 2 or more years of followup was 0.78 (95%CI 0.62 – 0.99).

The authors of the review concluded that there was a substantial protective effect of breastfeeding against atopic dermatitis in children with a family history of atopy.

Conclusion

Available evidence from one well-performed systematic review/meta-analysis on full term infants in developed countries suggests that exclusive breastfeeding for at least 3 months was associated with a reduction in the risk of atopic dermatitis in those subjects with a family history of atopy.

Background

Gastrointestinal infections are common in infants and children. Rate of diarrheal disease in US was estimated to be 1.1 episodes per person-year in children less than 5 years old.⁴¹ Many studies have investigated the possible protective effect of breastfeeding on the development of gastrointestinal infections. A previous review of diarrhea morbidity in both developed and developing countries reported that the risk of diarrhea in infants who did not receive breast milk were 3.5 to 4.9 times higher than infants who had exclusive breastfeeding in the first 6 months of life.⁴² Factors like secretory IgA, oligosaccharides, lactoferrin and others available in breast milk may protect the infant from various infections through passive immunity.¹ In vitro and in vivo binding studies have demonstrated that fucosylated glycans in breast milk inhibit binding by campylobacter jejuni, stable toxin of enterotoxigenic Escherichia coli, and major strains of calciviruses (e.g., noroviruses (also known as Norwalk-like viruses)) to their target host cell receptors.⁴³ One report suggests that glycoprotein lactadherin found in breast milk protects against rotavirus infection.⁴⁴ Socioeconomic status and child care variables (e.g., home versus day care, degree of crowding at home) are thought to be important confounding factors in observational studies on gastrointestinal infection and breastfeeding. As the aforementioned review also included studies before 1950, it would be instructive to review the more recent literature and assess the relationship between breastfeeding and gastrointestinal infections in infants from developed countries.

Published Systematic Review/Meta-Analysis (Table 4)

We identified one systematic review/meta-analysis that examined the relationship between breastfeeding and the development of gastrointestinal infections in children less than 1 year of age from developed countries.⁴⁵ The methodological quality of this systematic review/meta-analysis was rated grade B.

Using the MEDLINE database from 1966 to 1998, and supplementing the results with searches of the bibliographies of the primary and review articles, Chien 2001 identified 16 studies from developed countries that qualified for inclusion in the review. There were 12 prospective cohort studies totaling 5,473 subjects, 2 retrospective cohort studies totaling 504 subjects, and 2 case-control studies with 331 pairs of subjects. For the review, gastrointestinal infection was defined to be “any illness associated with vomiting, change in consistency or frequency of stools, or isolation of a known enteropathogenic bacterial or viral agent.” For the final data analysis, infant feeding practices were dichotomized into two groups: exclusive breastfeeding and partial/mixed feeding, or exclusive artificial feeding.

Results were conflicting. Nine of 16 studies (56 percent) yielded a statistically significant protective effect of breastfeeding on gastrointestinal infections. Majority of the studies suffered from methodological deficiencies. Four studies fulfilled criteria of controlling for detection bias, analyses of confounders, having a clear definition of infant feeding practices and infectious outcomes. Three of the studies reported breastfeeding was protective against non-specific gastrointestinal infection. The fourth study reported that differences in feeding practice did not affect the attack rates of rotavirus gastroenteritis. The potential confounders examined in these studies included infant sex, race, maternal education, family living standards, marital status, paternal social class, and/or parental smoking. Even though these studies adjusted for potential confounders, the actual quantitative adjusted odds ratio or risk ratio were not reported.

The authors of the systematic review stated that “it was not possible to pool the adjusted relative measures of association” in the cohort studies reviewed. Using a fixed effect model, the summary crude odds ratio of the 14 cohort studies for the development of gastrointestinal infection was 0.36 (95% CI 0.32, 0.41; heterogeneity P<0.01); the summary odds ratio of the two case-control studies was 0.54 (95% CI 0.36 – 0.80; heterogeneity, P=0.35).

Conclusion

Available evidence from three primary studies that controlled for potential confounders suggests that breastfeeding is associated with a reduction in the risk of non-specific gastrointestinal infection during the first year of life in infants from developed countries. However, a summary adjusted effect estimate taking into account potential confounders could not be determined because not all the studies that adjusted for potential confounders provided usable quantitative data for meta-analysis.

Addendum

During the final phase for the preparation of this report, we were alerted to a recent primary study on diarrheal disease in infants in 1990s England that provided relevant quantitative data. This case-control study of 304 infants (167 cases and 137 controls) showed that the infants who were breastfeeding had a reduced risk of diarrhea compared to infants who were not breastfeeding (OR 0.36, 95% CI 0.18 to 0.74, P=0.005). The result was adjusted for age, sex, social class, contact with person in and outside household, and other factors. This study also reported that the protective effect of breastfeeding did not persist beyond 2 months after cessation of breastfeeding.⁴⁶

Background

Respiratory infection is the most common medical problem among infants and children. Each year in the United States, three percent of all infants are hospitalized with moderate to severe respiratory infection.⁴⁷ Severe lower respiratory tract diseases may increase the risk of childhood asthma. Viral infections, especially respiratory syncytial virus (RSV) infection, are the most common cause of lower respiratory tract disease in developed countries. RSV infection occurs most frequently between 2 and 8 months of age. The risk factors associated with rates of respiratory illness include age, smoke exposure, day care, race/ethnicity, family size, education, and socioeconomic status. We identified one meta-analysis published in 2003 that compared the risk of hospitalization for respiratory diseases in healthy full term infants who were breastfed with those who were not.⁴⁸

Published Systematic Review/Meta-Analysis (Table 5)

Bachrach 2003's meta-analysis included seven cohort (five prospective and two retrospective) studies that evaluated the relationship between breastfeeding and hospitalization risk secondary to respiratory diseases.⁴⁸ Primary studies published before April 2002 and evaluated healthy full-term infants less than 1 year of age from developed countries were included. Only data from exclusive breastfeeding of at least 2 months or any breastfeeding totaling 9 months or more were included. Exclusive breastfeeding was defined as little or no formula feeding. Studies on cystic fibrosis and allergic conditions were excluded. Unpublished data were also examined. The primary outcome variable was hospitalization for lower respiratory tract disease secondary to bronchiolitis, asthma, pneumonia, empyema, and infections due to specific agents (e.g., RSV). The meta-analysis was restricted to first hospitalization. The analysis evaluated the risk of hospitalization in 3,201 breastfed subjects and 1,324 non-breastfed subjects.

The meta-analysis used a random effects model. There was an overall 72 percent reduction in the risk of hospitalization in infants who were exclusively breastfed for 4 or more months compared with those who were formula-fed (summary relative risk 0.28, 95% CI 0.14 – 0.54). There was no change in summary relative risk in the subgroup analyses of studies that also reported relative risk adjusted for the effects of smoking or socioeconomic status. There was no statistical heterogeneity across the studies. The authors performed sensitivity analyses to assess the appropriateness of combining studies and found no changes in summary risk. In addition, the number needed to treat (NNT) was calculated to be 26. This implies that at least 26 infants will have to breastfeed exclusively for 4 or more months to prevent one infant from hospitalization secondary to respiratory diseases. The methodological quality of the meta-analysis was rated grade A.

Conclusion

The meta-analysis showed an overall 72 percent reduction in the risk of hospitalization secondary to respiratory diseases in infants who were exclusively breastfed for 4 or more months compared with those who were formula-fed. This finding remained statistically significant after adjustment for potential confounders. There could be differences across studies with regard to duration of followup, diagnosis of respiratory disease, and other factors. However, there was no statistically significant heterogeneity detected across the studies. Taking into account the methodological quality of the meta-analysis and the consistent findings reported, we conclude that breastfeeding for 4 or more months is associated with a reduction in the risk of hospitalization secondary to lower respiratory tract diseases.

Background

Asthma is a bronchial disorder accompanied by breathing difficulties, wheezing, coughing, and production of thick mucus. Triggering factors have been attributed to foreign substances, tobacco smoke or pollutants, exercise, infection, and emotional stress. An increase in the prevalence of asthma has been reported in some countries during the second half of the 20^th century.⁴⁹ In 2002-2003, about six percent of the children in the United States had an asthma attack (www.cdc.gov/nchs/data/hus/hus05.pdf). In addition, the hygiene hypothesis (a lack of early childhood exposure to infectious agents increases susceptibility to allergic diseases^{50, 51}) has been proffered as a possible explanation for some of the increase in incidence and prevalence of asthma, but that hypothesis remains a matter of debate.⁴⁹ Potential confounders considered in the studies of breastfeeding and asthma include age, socioeconomic status, family history of atopy, and parental smoking.

Findings from primary studies on the relationship between breastfeeding and the development of asthma have been conflicting. One meta-analysis published in 2001 aimed to examine this relationship.⁵² We identified three relevant prospective longitudinal cohort studies and one followup study (to one of the primary studies in the meta-analysis) published subsequent to the meta-analysis.⁵³–⁵⁶ Three of the four studies met the inclusion criteria of the original meta-analysis and an updated meta-analysis stratified by family history of asthma was conducted.

Published Systematic Review/Meta-Analysis (Table 6)

Gdalevich 2001 conducted a meta-analysis that included 12 prospective observational studies with 8,183 term infants followed for a mean of 4.1 years. Study inclusion criteria included subjects from developed countries, exclusive of breastfeeding for at least 3 months, blinding of diagnosing physician to feeding status, and maternal recall of child's feeding history of not more than 12 months. Exclusive breastfeeding was defined as having no substitutes or additions. Critical appraisal of the studies was conducted based on the standards suggested by Kramer.¹⁶ Potential confounders including age, socioeconomic status, family history of atopy, and parental smoking were controlled for by means of multivariate analysis. The outcome of asthma was diagnosis by a physician.

Meta-analysis of the 12 studies with followup of 2 or more years reported a summary odds ratio of 0.70 (95% CI 0.60 – 0.81), suggesting an association of breastfeeding and a reduction in the risk of the development of asthma. Subgroup analysis reported that children with a family history of asthma or atopy benefited more from breastfeeding (OR 0.52, 95% CI 0.35 – 0.79) in the risk reduction of the development of asthma, compared with children who did not have a family history of asthma or atopy. There was no statistical heterogeneity in the studies. Sensitivity analysis indicated the exclusion of any one study did not change the overall results. Methodological quality for the meta-analysis was rated grade A.

Studies Identified after the Published Systematic Review/Meta-Analysis (Table 7)

We identified three prospective cohort studies, including one that had both prospective and retrospective analyses, and one followup publication examining the relationship between breastfeeding and risk of the development of asthma.⁵³–⁵⁵ Methodological quality of these four studies ranged from grade C to grade A.

Subjects selected in these studies ranged from healthy full-term infants to any live births from mainly population-based samples. Sample sizes at followup ranged from 1,037 to 4,964 with zero to 31 percent dropouts or withdrawals. Verification of asthma varied from parental confirmation through questionnaires to clinic assessment and physician diagnosis. Although all four studies adjusted for confounding, the only variable common to all studies was maternal smoking, either during pregnancy or after birth.

Two of the studies were population-based, one was hospital-based, and one study was done in a large health maintenance organization (HMO). Wright 2001 and Kull 2004 looked at exclusive breastfeeding whereas Burgess 2006 did not define breastfeeding but acknowledged that exclusive breastfeeding was in practice difficult to verify. Another issue is that there were no gold standards for the diagnosis of asthma in population studies. To confirm the diagnosis of asthma, Kull 2004 relied on clinical examinations as well as blood sampling and pulmonary function tests. Burgess 2006 used data from a questionnaire that asked about medications, symptoms, hospitalization, and family history to formulate the diagnosis. Sears 2002 based the diagnosis on pulmonary function tests with the addition of maternal description of symptoms. Wright 2001 and Kull 2004 appeared to have the most rigorous criteria for diagnosis of asthma by considering recent symptoms, results from testings, and confirmation by a physician. Two studies gathered feeding data post-delivery by recall, first study at 6 months,⁵³ and the second study at 2 months with another followup at 12 months.⁵⁴ Two studies relied on objective data for feeding history; Sears 2002 relied on data from nursing program records,⁵⁵ and Wright 2001 relied on breastfeeding data from clinic visits.⁵⁶ The methodological quality of the studies by Kull, Wright, and Sears were rated grade B, and Burgess was rated grade C.

Study findings. Kull 2004 followed 3,384 newborns for the first 4 years of life and reported that there was a statistically significant association of exclusive breastfeeding for 4 months or more and a reduction in the risk of the development of asthma (OR 0.72, 95%CI 0.53 – 0.97).⁵⁴ Furthermore, subgroup analysis showed that the association was stronger for children whose parents did not have a history of allergic diseases.

Sears 2002 enrolled 1,037 subjects at age 3 years and a retrospective record review found that approximately half were breastfed and half were not (533 breastfed, 504 not breastfed).⁵⁵ The subjects were followed prospectively until 21 or 26 years of age. The study found an increased risk of asthma in those subjects who were breastfed compared with those who were not for all time points assessed. Length of breastfeeding duration had no protective effect against asthma. Family history of asthma did not significantly affect these results. Even though “exclusive breastfeeding” rate was reported in the study, the authors acknowledged that it was common practice for the hospital staff to feed the newborns with formula for the first few nights post-delivery to allow the mothers to sleep.

Wright 2001 is a followup publication to one of the primary studies from the Gdalevich 2001 meta-analysis. This study reported that there was no association between duration of exclusive breastfeeding and asthma for 926 children by age 13 years followed since birth, except in those children who were atopic and whose mothers had a history of asthma.⁵⁶ This was the only study to report that a family history of asthma is an effect modifier of breastfeeding in the increased risk of the development of asthma in children.

Burgess 2006 investigated 4,964 children at 14 years of age concerning their history of asthma (“yes” or “no”) as reported by their mothers. This study found that there was no significant relationship between the duration of breastfeeding and the prevalence of asthma. The rates of asthma were the same for any given rates of breastfeeding regardless of the maternal history concerning asthma.⁵³ Data on feeding were collected at 6 months after birth.

Updating the Previous Meta-Analysis

We performed an update meta-analysis stratified by family history of asthma using the random effects model including three of the four recent publications.^{53, 54, 56} The fourth study did not qualify for inclusion because it did not have a comparison group of at least 3 months duration of breastfeeding. In the subgroup analysis of those children with a positive family history of asthma, two studies (Wright⁵⁶ and Kull⁵⁴) met the inclusion criteria set by Gdalevich. One of the studies reported a very large adjusted odds ratio (OR 8.7, 95%CI 3.4 – 22.2) in a followup of 13 year olds, compared with the other studies.⁵⁶ Sensitivity analyses were conducted to examine the source of heterogeneity by including or excluding this study. Excluding Wright, a history of exclusive breastfeeding for more than 3 months was associated with a reduction in the risk of asthma (OR_adj 0.60; 95%CI 0.43 – 0.82), compared with no breastfeeding. Including Wright, there was no longer a statistically significant association between a history of breastfeeding and the risk of asthma (OR_adj 0.81, 95%CI 0.41 – 1.60). This result suggests that the heterogeneity can be explained by a single study. Compared with the other studies in the analyses, the age of followup was 13 years in Wright 2001, while it ranged from 2 to 9 years in the other studies.

In the analysis of those children without a family history of asthma, the addition of the two subsequent studies (Kull⁵⁴ and Burgess⁵³) did not alter the statistically significant association of breastfeeding and the reduction in the risk of asthma (OR 0.73, 95%CI 0.59 – 0.92) compared with the original results reported by Gdalevich. It should be noted that Burgess reported only the unadjusted odds ratio, stating that the adjusted was “minimally altered”.

Conclusion

A well-performed meta-analysis from 2001 concluded that breastfeeding was associated with a reduction in the risk of developing asthma. This association was stronger in those subjects with a positive family history. However, three new primary studies and one followup study reported conflicting results. With our new meta-analyses including three of these studies, it is clear that there remained an association between breastfeeding and a reduction in the risk of asthma in those subjects without a family history of asthma. This association was also found in subjects under 10 years of age with a positive family history of asthma. It is unclear whether this association changes for older children. It should also be noted that the fourth study, which did not qualify for inclusion in our new meta-analyses, reported an increase in asthma risk with increased duration of breastfeeding in those subjects with a maternal history of asthma. Further studies concerning the effect of a family history of asthma on long-term outcome of asthma is warranted.

Background

Many studies have examined the relationship between breastfeeding and cognitive development. Results have been conflicting. Many of them did not have a clear definition of breastfeeding or breast milk exposure. Different cognitive assessment tools were used. Outcomes were measured anywhere from less than 2 years of age to adulthoods. Confounders commonly considered in these studies were socioeconomic status, maternal education, birthweight, gestational age, birth order and gender.⁵⁷ Jain et al. considered SES and quality and quantity of stimulation of the child (including social interactions) to be crucial confounders. ⁵⁸ These authors did not consider maternal or paternal intelligence, marital status, number of children, and maternal age to be crucial because to “some degree, they are markers of socioeconomic status, and are not clearly related to both feeding method and intelligence independent of socioeconomic status”. Three systematic reviews from 1999 to 2002 have tried to either establish methodological standards to assess the observational studies or adjust for covariates in pooled analysis. Since the last systematic review by Jain et al. in 2002,⁵⁸ there have been eight prospective cohort studies on healthy term infants that examined the relationship of breastfeeding to some aspects of cognitive development.

Additional Methodological Comments

We searched for systematic reviews on breastfeeding and cognitive development using terms like “cognitive”, “neurodevelopment”, “intelligence”...etc. Using similar terms, we searched MEDLINE and CINAHL in January and April of 2006, respectively, for additional primary studies published after 2000. We also identified additional articles based on reviews of the bibliographies cited in the relevant retrieved studies from the search and from suggestions by the reviewers for this report. For primary studies, qualifying study designs included prospective cohort and case-control studies; only studies from developed countries were included. For the healthy term infants, subgroups like infants of diabetic mothers and small-for-gestational age infants were not included. For preterm infants, no subgroups were excluded. Studies concerned exclusively with individual breast milk factor supplements (e.g., long chain polyunsaturated fatty acids, nucleotides) were not included. There was no restriction on timing of and the tools used for cognitive assessment. Only data on cognitive outcomes were extracted, data from motor, psychomotor, or visual development were not extracted.

Published Meta-Analyses/Systematic Reviews (Table 8)

Jain 2002 identified 40 relevant publications (30 birth cohorts, five RCTs, five school registry cohorts, and three case-control studies) from 1929 to 2001. Most of them studied term infants. A few studies included only preterm infants in their analyses. Each study was assessed according to a set of eight clinical epidemiological standards: study design, target population, sample size, quality of feeding data (suitable definition and duration of breastfeeding, and appropriate timing and source of feeding data), whether studies controlled for socioeconomic status and stimulation of the child, whether observers of the outcome were blind to feeding status, whether a standardized individual test of general intelligence at older than 2 years of age was used, and whether studies reported an effect size. Two studies on term infants met all the proposed methodological standards. One concluded that the effect of breastfeeding on intellect was significant (4.6 points in IQ at age 3 years in children who breastfed compared with those who bottle-fed),⁵⁹ and the other concluded that the effect on cognitive performance was statistically non-significant.⁶⁰

Drane 2000 examined 24 studies from 1966 to 1998.⁵⁷ Twenty-one included term infants, two included low birthweight infants, and one involved small-for-gestational age infants. Each study was assessed according to a set of three methodological standards: definition of outcome, correct classification of type of infant feeding, and control of potential confounding variables (socioeconomic status, maternal education, birthweight, gestational age, birth order, gender). Five of the 24 studies met all three methodological standards. These studies indicated an advantage in IQ to breastfed infants in the range of two to five points for term infants and eight points for low birth weight infants.

Interestingly, of the two studies that were identified by Jain 2002 to have met all the proposed methodological reporting standards, the feeding data from one did not meet the feeding data reporting standard put forth in the review by Drane 2000. The study at issue was Wigg 1998, in which the following statement could be found: “Feeding methods (i.e., breast, bottle-fed or mixed) and duration of breastfeeding in infancy were recorded at age 6 months by the trained research nurse”.⁶⁰ Jain 2002 interpreted that study to have met their requirement of adequate reporting of feeding data (“whether infants received breast milk exclusively or with supplemental formula or other foods”), while Drane 2000 interpreted that study not to have adequately distinguished between partial and exclusive breastfeeding. The other study that met all the methodological standards according to Jain 2002 was published in 1996,⁵⁹ but it was not included in Drane 2000's review; reason of which was not readily apparent.

Anderson 1999 examined 11 studies from 1978 to 1995 that controlled for at least five covariates (breastfeeding duration, sex, maternal smoking history, maternal age, maternal intelligence, maternal education, maternal training, paternal education, race or ethnicity, socioeconomic status, family size, birth order, birth weight, gestational age, and childhood experiences) and presented unadjusted and adjusted results.⁶¹ Eight studies included term infants; three studies included only preterm infants. The results were combined in a meta-analysis. The adjusted (fixed effects) pooled mean difference was 3.16 points (standardized effect estimate of cognitive developmental mean score, 95% CI 2.35 – 3.98) in favor of the breastfeeding group. Low birthweight infants showed larger differences (5.18 points in cognitive developmental score; 95%CI 3.59 – 6.77) than did normal birth weight infants (2.66 points; 95%CI 2.15 – 3.17). Other methodological quality of the studies was not assessed.

Studies Identified after the Published Meta-Analysis/Systematic Review in Term Infants (Table 9)

Prospective cohort. One secondary analysis of a prospectively collected data set⁶² and seven prospective cohort studies ranging from 44 to 3880 term subjects reported on the relationship between breast milk feeding and some aspects of cognitive development since 2002. ⁶³–⁷¹

None of the cohort studies made a clear distinction between partial and exclusive breastfeeding. Two studies stated to have a prospective design, but the breastfeeding information was collected retrospectively. Subjects were breastfed from less than 3 weeks to 12 months. Three studies reported the proportion of subjects who breastfed for more than 6 to 7 months, they ranged from 9 percent to 62 percent.

Bayley Mental Development Index Scale (MDI) was the cognitive assessment tool for subjects under 2 years of age. Peabody Picture Vocabulary Test (PPVT-R) and Wechsler Preschool and Primary Scale of Intelligence (WPPSI-R) were used in subjects under 7 years of age. Wechsler Adult Intelligence Scale (WAIS), Raven's standard progressive matrices, standard reading, verbal, and mathematical reasoning tests were used in adults up to 27 years of age. Time of assessment ranged from 1 to 27 years.

Der 2006 analyzed the database from the US National Longitudinal Survey of Youth 1979 (NLSY79).⁶² This database has information on the participants and their offsprings. The study reported that the adjusted effect of breastfeeding on Peabody individual achievement test (standardized mean of 100 and SD of 15) at 14 years was reduced to +0.52 from +4.7 after adjustment for maternal IQ, education, age, family poverty, home stimulation, and birth order. Further analysis of 332 pairs of siblings discordant for breastfeeding status found non-significant difference between groups in both status and duration of breastfeeding. Meta-regression of nine unique studies (including the data from NLSY79) reported an advantage of breastfeeding of 0.16 cognitive points after controlling for maternal IQ and other confounders. This study was rated good methodological quality (grade A).

Der 2006 also combined its estimate with the estimate from the only other sibling analysis in the literature to date (Evenhouse 2005,⁷² this study did not qualify for inclusion in this review because it was an analysis of data obtained from a cross-sectional design study). Evenhouse 2005 analyzed the database from the National Longitudinal Study of Adolescent Health (Add Health) 1994 (This database oversamples low-income, African-American, and Hispanic children and provided information on an abbreviated Peabody Picture Vocabulary Test results. There were 523 pairs of siblings with different breastfeeding history in this data set). The combined estimate from NLSY79 and Add Health was 0.025 (P=0.54) for breastfeeding status and 0.04 (P=0.271) for duration of breastfeeding.

Of the six prospective cohort studies of moderate (grade B) methodological quality, five reported an advantage in cognitive development in subjects who breastfed. Specifically, Lawlor 2006 reported that the adjusted score (for sex, parental characteristics, birthweight, and perinatal characteristics) in Raven's standard progressive matrices at age 14 years showed a mean difference of 6.79 (95%CI, 5.33–8.26) in less than 4 months of breastfeeding versus never breastfeeding (compared to an unadjusted mean of 8.20).⁷¹ Quinn 2001 reported that the mean PPVT-R at 5 years for those breastfed for at least 6 months was 8.2 points (95%CI, 6.5–9.9) higher for females and 5.8 points (95%CI, 4.1–7.5) higher for males when compared to those never breastfed. PPVT-R was adjusted for birthweight, poverty, maternal education, maternal age, time in daycare or preschool, the number of children in the household at 5 years, English speaking background in parents, and infant stimulation.⁷⁰ Similarly, Oddy 2003 and 2004, reported that the PPVT-R score at 6 years was 3.56 point higher for children breastfed more than 6 months compared with children never breastfed (F=8.59, P=0.003).^{68, 69} The result was adjusted for gender, gestational age, maternal age and education, parental smoking, and the presence of older siblings. Mortensen 2002 reported that the duration of breastfeeding was associated with significantly higher scores on the verbal, performance, and full scale WAIS at 27 years.⁶⁷ With regression adjustment for parental social status and education, single mother status, mother's height, age, and weight gain during pregnancy, cigarette smoking during 3^rd trimester, number of pregnancies, estimated gestational age, birth weight, birth length, and indices of pregnancy and delivery complications, the mean full scale WAIS were 99.4, 101.7, 102.3, 106.0, and 104.0 for breastfeeding durations of < 1 month, 2 to 3 months, 4 to 6 months, 7 to 9 months, and > 9 months, respectively (P=0.003). GomezSanchiz 2004 reported that the Bayley MDI at 24 months was 4.3 points higher in those breastfed more than 4 months compared with those breastfed less than 4 months after multiple linear regression adjusting for parental IQ.⁶⁶ Angelsen 2001 reported that adjustment for differences in maternal intelligence reduced the odds ratio of having a low IQ score among children who were breastfed for <3 months compared to ≥ 6 months from 2.8 (95%CI 1.4–5.3) to 1.5 (95%CI 1.0–2.1).⁶⁴

Conclusion

One well-performed sibling analysis and prospective studies that controlled specifically for maternal intelligence demonstrated that there is either little or no evidence for an association between breastfeeding and cognitive performance in children. It is clear that maternal intelligence is a major confounder in the studies on relationship between breastfeeding and cognitive development. For those studies that still reported a significant effect after adjustment for maternal intelligence, residual confounding from other factors like different home environments cannot be ruled out. Many studies controlled for socioeconomic status and maternal education but not specifically for maternal intelligence. It is clear that maternal intelligence should be controlled for separately from socioeconomic status and maternal education in any studies of breastfeeding and cognitive development. As cautioned by Der et al., “The generalizability of the results presented here must be considered carefully. This study and the others included in the meta-analysis are all based on samples from developed countries. Generalization of the findings beyond these and similar societies would be unwise. We have also excluded premature and low birthweight infants for whom the effect may be different.”⁶²

Background

The prevalence of overweight or obesity among children and adolescents has rapidly increased in the past two decades. Results from the 1999-2002 National Health and Nutrition Examination Survey (NHANES) showed that an estimated 16 percent of children and adolescents ages 6–19 years were overweight. This represents a 45 percent increase from the overweight estimate of 11 percent obtained from NHANES III (1988-94). (www.cdc.gov/nchs/products/pubs/pubd/hestats/overwght99.htm) It is increasingly recognized that nutrition in early life may have long-term physiologic effects. Relationships between the types of postnatal feeding and the subsequent development of fat and fat-free mass are quite complex and are dependent on multiple factors including differences in food composition (human milk versus formula), food delivery (breast versus bottle), food “lifestyle” (breastfeeding versus formula feeding) and food behavior (self-regulation and feeding on demand versus set schedules of feeding of predetermined amounts).⁷³ It is known that infants fed breast milk differ in their growth kinetics from formula-fed infants. Formula-fed infants demonstrate higher weight and length gains compared with breastfed infants. A systematic review of 19 studies in developed countries concluded that by the age of 12 months, the cumulative difference in body weight amounted to approximately 400 g less in infants breastfed for 9 months compared with formula-infants, and as much as 600–650 g less in infants breastfed for 12 months compared with formula-fed infants.⁷⁴ Differences in feeding behavior and mother-child interaction between breast- and formula-fed infants may account for some of the differences reported. For instance, breastfed infants showed a different suckling pattern, and appeared to have greater degree of control on meal sizes and feeding intervals than infants who were formula-fed.⁷⁵ Diet-related differences in the circulating levels of biochemical markers (such as leptin, ghrelin, insulin-like growth factors, and other compounds) implicated in energy metabolism during infancy might explain some of the anthropometric and behavioral differences between breastfed and formula-fed infants. These observed differences may have potential long-term consequences.⁷³

Commonly considered confounders in the studies of relationship between obesity or overweight and breastfeeding were birth weight, parental overweight, parental smoking, dietary factors, physical activity, socioeconomic status (SES), age, sex, birth order, and number of siblings.

Additional Methodological Comments

We identified three systematic reviews and meta-analyses that examined the relationship between breastfeeding and childhood obesity or obesity across all ages.⁷⁶–⁷⁸ Although the outcomes of interest were similar among these systematic reviews and meta-analyses, they answered slightly different research questions because of the differences in their study eligibility criteria and analyses. Thus, we have summarized and discussed these systematic reviews and meta-analyses separately.

Published Systematic Reviews/Meta-Analyses (Table 10)

Arenz 2004 was a meta-analysis of studies from 1966 to December 2003 that examined the relationship between breastfeeding and childhood obesity in children at least one year of age. Inclusion criteria for the meta-analysis were: obesity defined by a body mass index (BMI) greater than 90^th, 95^th or 97^th percentile; adjustment for at least three potential confounding or interacting factors; reported either odds ratio or relative risk; and last followup between 5 and 18 years of age. Nine of 28 studies reviewed met the eligibility criteria for meta-analysis. There were two prospective cohort and seven cross-sectional studies totaling more than 69,000 children from developed countries. The meta-analysis used both fixed- and random-effects models and pooled crude and adjusted odds ratios from the individual studies. Definitions of breastfeeding and comparative feedings were heterogeneous across studies. Sensitivity analyses were performed to assess for heterogeneity. The factors analyzed were cohort study or cross-sectional study, different definitions of breastfeeding, different definitions of obesity, different age groups and number of potential confounders considered for adjustment. The methodological quality of this meta-analysis was grade A.

The pooled crude odds ratio for breastfeeding and obesity defined as a BMI > 90th, 95th or 97th percentile could be calculated for six of the nine studies included. The odds ratio was 0.67 (95% CI 0.62 – 0.73). The adjusted odds ratio for the nine studies was 0.78 (95% CI 0.71 – 0.85) for both the fixed and random-effects models, suggesting that there was no heterogeneity between the studies. Sensitivity analyses showed that the protective effect of breastfeeding was more pronounced in studies with adjustment for less than seven potential confounding factors compared with adjustment for seven or more potential confounding factors (adjusted OR 0.69 vs. 0.78, respectively). Other criteria (e.g., cohort study or cross-sectional study, different definitions of breastfeeding, different definitions of obesity, different age-groups) did not affect the summary estimates significantly. For example, the pooled adjusted odds ratio of obesity was 0.76 (95%CI 0.67–0.86) in studies comparing ever breastfeeding to never breastfeeding, versus 0.74 (95%CI 0.64–0.85) in studies that used other definitions of breastfeeding and comparative feedings.

Eight studies analyzed the relationship between breastfeeding duration and the risk of overweight or obesity in later childhood. Exclusivity of breastfeeding was not reported. Four studies reported an inverse association of breastfeeding duration and the prevalence of obesity both in the crude and the adjusted estimates. One of the studies lost statistical significance after adjustment. Three studies found no significant effect of duration of breastfeeding on obesity.

Harder 2005 was a meta-analysis of 17 qualifying studies published from 1966 to December 2003. A total of 120,831 subjects (66 to 32,200 subjects per study) from developed countries were included. Eligibility criteria included any original report comparing breastfed subjects with exclusively formula-fed subjects at any age, the reports must have either reported odds ratio or contained data for the calculation of odds ratio for the risk of overweight or obesity in relationship to the feeding history, and the duration of breastfeeding must have been reported. All definitions of overweight or obesity were included. Three different meta-analytic techniques that specifically required the use of crude odds ratios and 95% confidence intervals were employed. Because of suboptimal consideration for potential confounding, we rated the methodological quality of this meta-analyses grade B.

Fourteen studies provided data for more than one category of duration of breastfeeding, leading to 52 estimates included in the meta-regression analysis. In the analysis, duration of breastfeeding was significantly negatively related to the risk of overweight (regression coefficient: 0.94, 95%CI 0.89 – 0.98). Categorical analysis showed that from 1 month of breastfeeding onward (the reference group), the risk of subsequent overweight continued to decrease, reaching a plateau of more than 30 percent risk reduction at 9 months of breastfeeding. Using the “pool-first method” (that is to calculate a study-specific regression coefficient and corresponding 95 percent confidence interval for each study using a log-linear model and then pooled all studies with a random effects model) to quantify the dose-response relationship, the results showed that each month of breastfeeding was associated with a four percent decrease in risk of overweight per month of breastfeeding exposure (OR 0.96/month of breastfeeding, 95%CI 0.94 – 0.98). A fixed effect model reported a similar pooled odds ratio (OR 0.96/month of breastfeeding, 95%CI 0.95 – 0.98). The age at examination had little influence on the magnitude of the effect of duration of breastfeeding on the risk of overweight. The pooled odds ratio from five studies investigating subjects up to 5 years of age was 0.97 (95%CI 0.94 – 0.99); while for six studies on subjects 6 or more years of age, it was 0.96 (95%CI 0.93 – 0.99).

The effect of the duration of exclusive breastfeeding was analyzed in two studies. The pooled odds ratio for the risk of overweight per month of exclusive breastfeeding was 0.94 (95%CI 0.89 – 0.99, random effects model).

Subgroup analyses showed that the different definitions of overweight influenced the estimate of odds ratio only slightly. In eight studies that used BMI to define overweight, the pooled odds ratio was 0.96 (95%CI 0.94 – 0.98); while in three studies that used another measures (e.g., percentile of weight for length, or weight for age) to define overweight or obesity, the odds ratio was 0.93 (95%CI 0.87 – 0.99).

Lastly, Owen 2005 was a systematic review of 61 observational studies from 1966 to September 2003 that examined the effects of infant feeding on a measure of adiposity (quantitatively or narratively) in later life. Twenty-eight studies (totaling 298,900 subjects) that provided 29 unadjusted odds ratios relating the initial infant feeding method and obesity were included in a meta-analysis. A fixed effect model was used. Meta-regression and sensitivity analyses were used to examine the influence of various factors defined a priori, including the effects of adjustment for factors such as parental body size (mostly BMI), SES, and maternal smoking. Because of suboptimal consideration for potential confounding, we rated the methodological quality of this systematic review and meta-analyses grade B.

Twenty-eight of 29 estimates related breastfeeding to a lower risk of obesity in later life. Four estimates were for infants, 23 for children, and two for adults. There was evidence of marked heterogeneity among studies (P < 0.001). In a fixed-effect meta-analysis, breastfed subjects were less likely to be defined as obese than were formula-fed subjects (OR 0.87, 95% CI 0.85 – 0.89). In six studies, it was possible to examine the effect of adjustment for the following potential confounders: SES (based on parental education in two studies), parental BMI, and current maternal smoking or maternal smoking in early life. The pooled odds ratio in these studies changed from 0.86 (95% CI 0.81 – 0.91) before adjustment to 0.93 (95% CI 0.88 – 0.99) after combined adjustment. The effect of adjustment for birth weight (based on either actual birth weight or prevalence of low birth weight) was examined in 10 studies; this had no appreciable effect on the odds ratios.

There was no clear evidence that the protective effect of breastfeeding altered with increasing age of outcome assessment. Odds ratios of 0.50 (95% CI 0.26 – 0.94) for infants, 0.90 (95% CI 0.87 – 0.92) for young children, 0.66 (95% CI 0.60 – 0.72) for older children, and 0.80 (95% CI 0.71 – 0.91) for adults were observed (test for trend, P = .85, adjusted for study size; P = .99 with the exclusion of infants). The protective effect of breastfeeding on obesity was stronger and more homogeneous among four studies in which initial feeding groups were exclusive (OR 0.76; 95% CI 0.70 – 0.83; test for heterogeneity between estimates, P= .143), compared with all other studies. In 14 studies that provided data on breastfeeding duration, the protective effect of breastfeeding over formula feeding was greater among subjects breastfed for at least 2 months (OR 0.81, 95% CI 0.77 – 0.84), compared with those breastfed for any duration (OR 0.89, 95% CI 0.86 – 0.91). In six studies, it was possible to examine the effect of adjustment for the following potentially important confounders: socioeconomic status, parental BMI, and current maternal smoking or maternal smoking in early life. The pooled odds ratio in these studies was reduced from 0.86 (95% CI: 0.81–0.91) before adjustment to 0.93 (95% CI: 0.88–0.99) after adjustment.

Thirty-three studies totaling 12,505 subjects explored the relationship between breastfeeding and obesity even though they did not provide odds ratio. However, they provided 35 reports of directions of association; of these, breastfeeding was unrelated to the risk of obesity in 33, related to a reduced risk in one, and related to an increased risk in another. Studies that did not provide odds ratios were much less likely to report that breastfeeding was associated with a reduced risk of obesity, compared with studies that did provide odds ratios (1 of 35 studies and 18 of 29 studies, respectively; P < .001).

Conclusion

Findings from three systematic reviews and meta-analyses of good and moderate methodological quality suggest that a history of breastfeeding is associated with a reduction in the risk of obesity in later life. The pooled adjusted odds ratio of overweight/obesity comparing ever breastfeeders to never breastfeeders was 0.76 (95%CI 0.67–0.86) and 0.93 (95%CI: 0.88–0.99) in Arenz 2004 and Owen 2006 meta-analysis, respectively. In Harder 2005 meta-analysis, duration of breastfeeding was significantly negatively related to the unadjusted risk of overweight (regression coefficient: 0.94, 95%CI 0.89 – 0.98), and each month of breastfeeding was found to be associated with a four percent decrease in risk (unadjusted OR 0.96/month of breastfeeding, 95%CI 0.94 – 0.98). However, the results from Harder 2005 meta-analysis employed techniques that required the use of crude odds ratios from the primary studies for its summary estimates. Therefore, those estimates may not be accurate because potential confounders could not be accounted for in the analysis. As demonstrated in the sensitivity analyses in both Arenz 2004 and Owen 2005, the magnitude of effects was reduced when more confounders were adjusted for in the analyses. The observed association between breastfeeding and a reduced risk of obesity could also reflect selective reporting and/or publication bias. The exclusivity of breastfeeding was not described in the majority of the studies.

Background

Abnormal levels of total cholesterol, low-density lipoprotein (LDL) cholesterol, and blood pressure in adults are major risk factors for cardiovascular disease (CVD). Observational studies of large cohorts of men and women have consistently reported that serum cholesterol level >200 mg/dL was associated with an increased risk of all-cause and CVD mortality.^{79, 80} Each increment of 20 mm Hg of systolic blood pressure and 10 mm Hg of diastolic blood pressure doubles the risk for CVD.⁸¹ Diet modification, weight reduction, and pharmacotherapy can reduce the risk of CVD. We identified one meta-analysis that evaluated the relationship of breastfeeding during infancy with cholesterol levels in adolescents and adults,⁸² two meta-analyses that evaluated the relationship of breastfeeding with adult blood pressure,^{83, 84} and one systematic review and meta-analysis that examined the relationship between breastfeeding and cardiovascular disease mortality in later life.⁸⁵

Published Systematic Reviews/Meta-Analyses (Table 11)

One meta-analysis of 37 cohort and cross-sectional studies evaluated the effect of breastfeeding on total and LDL cholesterol levels among infants, adolescents, and adults.⁸² All primary studies published in English language from both developed and developing countries that reported estimates of a mean difference and standard error in cholesterol levels between breastfed and formula-fed infants were included in the meta-analysis. The meta-analysis utilized a random effects model. A total of 5,829 breastfed and 4,852 formula-fed subjects were evaluated. Data from the included primary studies were categorized into three age strata: infancy (< 1 year of age), children and adolescents (1 to 16 years), and adults (17 years to 65 years). Outcomes included total and LDL cholesterol levels. No information was provided on the timing of the sample collection in relation to fasting or not fasting. Of the 37 studies, there were 26 outcomes (total or LDL cholesterol levels) in infants, 17 in children and adolescents, and nine in adults. The analysis combined data from a broad age category for the adult participants. It was unclear if there were adjustments for potential confounders such as body mass index, height, and socioeconomic status for the data on cholesterol. The methodological quality of the meta-analysis was rated grade C.

In 25 of 26 observations, infants who were breastfed reported higher mean total cholesterol levels compared with infants who were formula-fed. The overall mean difference was +24.75 mg/dL (95% CI 18.95 to 30.55). There was a statistically significant heterogeneity across the studies. The meta-analysis did not find an association between total cholesterol level and age or gender. There were only seven reported observations on LDL cholesterol levels in infants, six of which reported higher mean levels of LDL cholesterol in breastfed infants compared with formula-fed infants. The mean difference was +22 mg/dL (95% CI 15.47 to 29.0 mg/dL). There was no statistical heterogeneity.

In 16 of 17 observations, the mean total cholesterol levels in children or adolescents who were breastfed in their infancy were similar to those who were formula-fed. The overall mean difference was 0.0 mg/dL (95% CI -2.7 to 2.7 mg/dL). A statistically significant heterogeneity was observed across studies. There was no association between total cholesterol level and age or gender. There were only four observations of LDL cholesterol levels in children or adolescents. The mean levels of LDL cholesterol in children or adolescents who were breastfed in their infancy were similar to those who were formula-fed. The mean difference was +0.39 mg/dL (95% CI -2.7 to 3.09 mg/dL). There was no statistical heterogeneity between the studies.

The mean age of adults evaluated in the primary studies ranged from 17 to 64 years. Lower mean total cholesterol levels in adults who were breastfed in their infancy compared with those who were formula-fed in their infancy were reported in seven of nine observations. The overall mean difference was -6.96 mg/dL (95% CI -2.32 to -11.6 mg/dL). There was no statistically significant heterogeneity between the studies. There were only four observations of LDL cholesterol reported in adults. Adults who were breastfed in their infancy had lower mean LDL cholesterol levels compared with those who were formula-fed. The mean difference was -7.7 mg/dL (95% CI -3.09 to -12.37 mg/dL). There was no statistical heterogeneity between the studies.

Conclusion

Results from the meta-analysis of cohort and case-control studies reported that there was a reduction in total and LDL cholesterol levels in adults who were breastfed during infancy compared with those who were formula-fed. While higher serum lipid levels were observed in infancy, the meta-analysis found that breastfeeding was associated with a reduction in serum lipid level in adult life. The significance of higher serum lipid levels observed in infancy is unclear and studies have neither shown benefit nor harm from such high levels.⁸⁶ These findings were based on data from adults with a wide age range. The analysis did not segregate the data according to gender. Potential confounders were not explicitly analyzed. Detailed information (e.g., fasting or non-fasting) on the collection of specimen for cholesterol testing was not included. The methodological quality of the meta-analysis was rated grade C. Because of the poor methodological quality of the meta-analysis, we find that the conclusions drawn by the authors were suspect. We conclude that the relationship between breastfeeding and adult cholesterol levels cannot be correctly characterized at this time.

Published Systematic Reviews/Meta-Analyses (Table 12)

Two meta-analyses evaluated a total of 26 studies of various designs for the effect of breastfeeding and formula feeding in infancy on systolic and diastolic blood pressure levels in adult life. The primary studies included in these meta-analyses were conducted in developed and developing countries. Outcomes assessed were differences in systolic and diastolic blood pressures in adulthood. Both meta-analyses examined potential confounders in the studies and also explored the possibility of publication bias. Potential confounders considered in the studies were age, gender, race, height, and body mass index. Among the 26 primary studies evaluated, 13 studies were common to both meta-analyses. Both meta-analyses included at least one study that included preterm infants. Duration and exclusivity of breastfeeding or formula feeding were heterogeneous in the studies. The methodological quality of both meta-analyses were rated grade B.

The meta-analysis by Martin 2005 included 15 studies with 17 observations published until 2004 that evaluated a total of 17,503 eligible subjects.⁸³ In these studies, blood pressure was measured in subjects whose age ranged from 1 to 60 years. The meta-analysis utilized a random effects model. Meta-regression was performed to evaluate the effect of heterogeneity in study size, age at measurement of blood pressure (<10 years, 11–45 years, >45 years), maternal recall, exclusivity of the feeds, methods of blood pressure measurement, and other variables on the summary estimates. Mean systolic blood pressures were reduced by 1.4 mm Hg (95% CI 0.6 to 2.2) in adulthood among subjects who were breastfed in infancy compared with those who were formula-fed. There was a statistically significant heterogeneity across studies for this outcome. Mean diastolic blood pressures were reduced by 0.5 mm Hg (95% CI 0.04 to 0.9) in adulthood among subjects who were breastfed in infancy compared with those who were formula-fed, with no heterogeneity for this outcome.

Owen 2003 included 24 studies published until 2003 with 26 observations for systolic blood pressure and 23 observations for diastolic blood pressure.⁸⁴ The meta-analysis evaluated 8,471 subjects who were breastfed in infancy and 11,292 who were formula-fed. In these studies, blood pressure was measured in subjects whose age ranged from 1 to 71 years. The meta-analyses utilized a random effects model. Subgroup analyses evaluated the effect of heterogeneity in study size, age of assessment of outcomes, and year of birth. Mean systolic blood pressures were reduced by 1.10 mm Hg (95% CI 0.42 to 1.78) in adulthood among subjects who were breastfed in infancy compared with those who were formula-fed. When stratified by age groups (≤1 year, >1 to 16 years, ≥17 years), a similar association was observed among age groups that ranged from more than 1 year to 16 years. There was a statistically significant heterogeneity for this outcome. Mean diastolic blood pressures were reduced by 0.36 mm Hg (95% CI - 0.08 to 0.79) among subjects who were breastfed compared with subjects who were formula-fed. There was no statistical heterogeneity for this outcome.

Both meta-analyses observed smaller association of breastfeeding on systolic blood pressure in large studies (>300 and/or >1000 participants) compared with smaller studies. Both meta-analyses attributed this observation to publication bias. Similar effect size reduction of breastfeeding on diastolic blood pressure was not observed. In studies that adjusted for potential confounders, results remained similar before and after adjustment for potential confounders.

Conclusion

Results from both meta-analyses concluded that there was a small reduction in systolic blood pressures among adults who were breastfed in their infancy compared with those who were formula-fed. The association weakened after stratification by study size, suggesting the possibility of bias. Although both analyses had moderate methodological quality and reported similar findings, the authors had different appraisals of the public health importance of the small reduction in systolic blood pressure. In conclusion, there is an association between a history of breastfeeding during infancy and a small reduction in adult blood pressure, but the clinical or public health implication of this finding is unclear.

Published Systematic Review/Meta-Analysis (Table 13)

We identified one systematic review and meta-analysis that examined the relationship between breastfeeding and cardiovascular disease (CVD) mortality in later life.⁸⁵ Articles were included in the systematic review if breastfed infants were compared with bottle-fed infants, if the outcome was cardiovascular disease or ischemic heart disease mortality, and if estimates of the association between having been breastfed and cardiovascular disease or ischemic heart disease mortality could be obtained from the paper or after correspondence with the authors. A total of four historical cohort studies from developed countries were identified, involving 25,166 subjects at baseline and 10,785 subjects at followup. The studies were not graded for their methodological quality. The meta-analysis used a random-effects model. Heterogeneity across studies was assessed. The methodological quality of this systematic review and meta-analysis was rated grade B due to incomplete consideration of the heterogeneity across studies in the meta-analyses.

The four historical cohorts included in the systematic review and meta-analysis were Wingard cohort, 1,373 birth children in California (85 percent in followup); Hertfordshire cohort, 5,908 women and 10,374 men born in Hertfordshire (43 percent in followup); Boyd Orr cohort, 4,999 men and women from a survey of diet and health in pre-war Britain (71 percent in followup); and Caerphilly cohort, 2,512 middle-aged men living in Caerphilly, South Wales (63 percent in followup). Subjects from these four cohorts were born between 1904 and 1939. Potential confounders considered in the association between the risk of CVD mortality and breastfeeding in the four studies were age, birth weight, infant health, socioeconomic status, and/or birth order.

Random-effects model showed little difference in all cause mortality between breast- and bottle-fed subjects (pooled rate ratio = 1.01; 95% CI: 0.91 – 1.13, P=0.8), and there was little evidence of heterogeneity. Five observations from three studies suggested little or no association between breastfeeding and cardiovascular disease mortality in both males and females, and one suggested a possible adverse effect (Caerphilly cohort). In random effects meta-analysis, CVD mortality was similar in breastfed versus bottle-fed subjects (pooled rate ratio = 1.06; 95% CI: 0.94 – 1.20), and there was no statistical evidence of between-study heterogeneity.

Ischemic heart disease mortality was 6 percent lower among males who had been breastfed in the Hertfordshire cohort, but 56 percent higher among breastfed females. This result was in agreement with point estimates from the Boyd Orr cohort, suggesting that ischemic heart disease mortality was 10 percent lower among males who had been breastfed, but 40 percent higher among breastfed females (although there was little statistical evidence of interaction: P = 0.2). In Caerphilly, however, ischemic heart disease mortality was 73 percent higher among breastfed males. In a random effects meta-analysis (pooled rate ratio = 1.19; 95% CI 0.89 – 1.58, P = 0.3), and there was evidence of heterogeneity.

Similar analyses were also performed to examine the association between prolonged breastfeeding (> 1 year duration) and the risk of all-cause, CVD, and ischemic heart disease mortality in later life. There was little evidence that prolonged breastfeeding was associated with all-cause mortality (pooled rate ratio: 0.94; 95% CI 0.71 – 1.24), although there was moderate statistical evidence of heterogeneity. There was some evidence that prolonged breastfeeding was associated with a 16 percent increase (95% CI 0.99 – 1.36; P = 0.06) in CVD mortality, and no evidence of inconsistency in estimates. There was little evidence that prolonged breastfeeding was associated with ischemic heart disease mortality (rate ratio: 1.08; 95% CI 0.88 – 1.31; P = 0.5) and there was no heterogeneity.

Conclusion

The authors concluded that the data reviewed did not provide evidence that breastfeeding was related to all-cause or CVD mortality. The confidence limits around the point estimates and the observed between-study heterogeneity for associations between breastfeeding and ischemic heart disease, however, do not rule out important beneficial or adverse cardiovascular effects of breastfeeding.

There were some possible sources of bias and limitations in the studies reviewed. Two of the four studies had followup rate of less than 70 percent of the original population; therefore, selection bias cannot be ruled out. Recall bias was possible in the three studies where breastfeeding data were collect retrospectively. As confounding and biases may have distorted results from individual studies, the statistical combination of estimates into a single rate ratio needs to be interpreted with caution. All four studies in the meta-analyses were historical cohorts (born between 1904 and 1939). Given the statistical heterogeneity across studies, combining study results might not be appropriate. For the outcome of ischemic heart disease mortality, it may not be appropriate to combine results for men and women into a single analysis because of apparent effect modification by gender.

Because of the above limitations, no definitive conclusions can be drawn regarding the relationship between breastfeeding and CVD mortality. Further investigation is warranted.

Background

Type 1 diabetes results from destruction of the insulin-producing β cells of the pancreatic islets. Various exogenous triggers, such as certain dietary factors and viruses, are thought to induce the immune-mediated process leading to extensive β cell destruction.⁸⁷ Putative mechanisms of protection against type 1 diabetes afforded by breast milk include passive immunity provided by secretory immunoglobulin A antibodies, increased β cell proliferation observed in breastfed compared with formula-fed infants, or delayed exposure to foreign food antigens in exclusively breastfed infants.⁸⁷ In addition, hypotheses have been proposed to explain the putative diabetogenicity of cow milk.⁸⁷ For example, β-lactoglobulin, a cow milk-specific protein, has been implicated as a possible trigger of the immune defect, leading to type 1 diabetes. Therefore, concerns regarding the safety and advisability of feeding cow milk-based products to infants have been raised.⁸⁸ Gersterin 1994 conducted a systematic review of the epidemiological and clinical literature that explored a possible link between cow milk and type 1 diabetes.⁸⁹ Subsequently, Norris and Scott 1996 performed meta-analyses on all published case-control studies from 1966 to 1994 that examined infant diet exposures and type 1 diabetes.⁹⁰ Since this meta-analysis, we identified six case-control studies that reported outcome of type 1 diabetes in relation to breastfeeding during infancy.

Commonly considered confounders in the relationship between type 1 diabetes and breastfeeding were maternal/parental education, maternal age at birth, birth order, household income, race/ethnicity, social class, family history of type 1 diabetes, neonatal illness, and type of delivery.

Published Systematic Reviews/Meta-Analyses (Table 14)

Gersterin 1994 conducted a systematic review on literature that explored a possible link between cow milk and type 1 diabetes. Articles were excluded if they exclusively used surrogate markers for either type 1 diabetes or cow milk exposure. A total of three ecological and time-series studies, 13 case-control studies, one cohort study, and one case series were included. Meta-analysis of all included case-control studies was performed, using a fixed effect model. Both adjusted and unadjusted odds ratios were used, but there was no further analysis or discussion on potential impacts of confounding by combining unadjusted odds ratios. The methodological quality of the systematic review and meta-analysis was rated grade B, due to insufficient consideration for the potential confounding in the primary studies.

In the included time-series and ecological studies, the results showed an inverse association (geographical and temporal) between the rate/prevalence of breastfeeding and the rate/prevalence of type 1 diabetes.

The cohort study was an analysis of two groups of children born in the UK in 1958 and 1970, who were followed for 16 and 10 years, respectively. This study did not find any association between breastfeeding for less than 1 month and the risk of development of type 1 diabetes.

In the 13 case-control studies in which the neonatal feeding history of patients with type 1 diabetes and individually matched non-diabetic control subjects were compared, the results were mixed. A total of 3,708 type 1 diabetes cases and 20,340 non-diabetic controls were included. None of these studies satisfied all six methodological criteria defined by the author. Four of the 13 case-control studies fulfilled five of six methodological criteria (e.g., inclusion of ≥75 percent of eligible diabetic patients, unbiased selection of unrelated nondiabetic control subjects, control subjects derived from the same population as diabetic subjects, an identical means of determination of infant feeding practices in both diabetic and nondiabetic groups, blind determination of early feeding history, and identification of diabetic patients from incident cases). When these four studies were combined, the overall odds ratio (adjusted odds ratios were used) for type 1 diabetes in patients exposed to less than 3 months of breastfeeding was 1.43 (95%CI 1.15 – 1.77; P=0.3 for homogeneity). When all 13 included case-control studies were combined, the overall odds ratio (mixed crude and adjusted odds ratios were used) for type 1 diabetes in patients exposed to less than 3 months of breastfeeding was 1.37 (95%CI 1.22–1.53; P=0.11 for homogeneity).

In 1996, Norris and Scott performed a meta-analysis of infant diet and type 1 diabetes to examine further the inconsistent results reported in the literature. A total of 17 case-control studies with appropriate data for meta-analysis were included. The analysis included 4,656 type 1 diabetes cases and 16,383 non-diabetic controls. The authors abstracted case and control data for four separate exposures: breastfeeding status (ever/never), total breastfeeding duration, exposure to breast-milk substitutes, and exposure to cow milk-based substitutes. Due to the limitation of the meta-analytic technique, only unadjusted odds ratios were calculated and combined, although some primary studies had adjustments for maternal education, maternal age at birth, birth order, household income, race/ethnicity, and/or social class. Authors performed sensitivity analyses on various characteristics of study methodological quality to explore the impacts of potential biases on the summary odds ratios. The methodological quality of this meta-analysis was rated grade B, due to insufficient consideration for the potential confounding in the primary studies.

The overall odds ratio of all included case-control studies that examined the association between never breastfed and type 1 diabetes was 1.13 (95%CI 1.04 – 1.23). Among these studies, fourteen also examined type 1 diabetes risk by months of breastfeeding duration. The duration categories in the analysis were cumulative, rather than mutually exclusive (i.e., some studies provided data on both 3-month and 6 month breastfeeding data in the same subjects). The summary odds ratios showed consistently elevated risks of type 1 diabetes associated with age at first exposure to any breast milk substitutes before 6 months of age. Since the majority of the studies reported odds ratios using a cutoff of 3 months when examining continuous exposures, this cutoff was used for the meta-analysis. The summary odds ratio for type 1 diabetes in subjects who were breastfed for less than 3 months compared with those who were breastfed for at least 3 months was 1.23 (95%CI 1.12 – 1.35).

Stratified analyses of studies by methodological and study population characteristics were performed to see whether differences in these characteristics might explain the heterogeneity. The characteristics were prevalent versus incident case-control study design, adequate versus inadequate response rates of the cases and controls, the breastfeeding prevalence in the background population, the type 1 diabetes risk in the background population, and retrospective versus concurrent infant diet assessment. All of these factors had differential impacts on the summary odds ratios for the risk of type 1 diabetes associated with infant diet exposures.

Norris and Scott concluded that their meta-analysis showed that the increased risk of type 1 diabetes associated with any of the infant diet exposures was small. According to these authors, interpretation of weak associations (i.e., an odds ratio of less than 2.0) can be problematic, since weak associations can more readily be explained by biases.

Studies Identified after the Published Systematic Reviews/Meta-Analyses/Systematic Review (Table 15)

We only included all studies that examined the outcome of type 1 diabetes in relation to breastfeeding in developed countries. To be consistent with previous meta-analyses, articles were excluded if they only used surrogate markers for type 1 diabetes (e.g., the presence of islet cell antibodies). A total of six case-control studies were identified,⁹¹–⁹⁶ The studies included 1,293 patients with type 1 diabetes and 3,262 control subjects. Five studies were conducted in Europe, and one in Taiwan. Four studies were rated grade B in methodological quality; two studies were rated grade C. Commonly considered confounders in these studies were family history of type 1 diabetes, neonatal illness, maternal age at birth, birth order, maternal/parental education, and type of delivery.

In four studies, the definition of type 1 diabetes cases was children with juvenile-onset diabetes (or developing diabetes before 17 years of age). In the other two studies, the cases were registered type 1 diabetes patients who were younger than 30 years of age and children with diabetes who were identified through hospital records. Matched controls were those without type 1 diabetes selected from various sources in the same population as the cases.

Three studies reported odds ratio of type 1 diabetes comparing subjects who were ever breastfed with those who were never breastfed. Two studies found a reduced risk of type 1 diabetes in subjects who were ever breastfed (ORs 0.56⁹⁵ and 0.75⁹¹), while the third study reported an increased risk (OR 2.44⁹⁴).

Three studies compared subjects who were breastfed for more than 3 months or 6 months with those who were never breastfed. Two studies reported a reduced risk of type 1 diabetes in subjects who were breastfed for more than 3 or 6 months (adjusted OR 0.57⁹⁶ and 0.25⁹⁵, respectively). The third study⁹⁴ reported an opposite finding (this was the same study that reported an increased risk with ever breastfed). A reduced risk of type 1 diabetes was found when comparing subjects who were never breastfed with those who were breastfed for more than 6 months (adjusted OR 0.36, 95%CI 0.14 – 0.94). The data also showed a slight increase in the risk of type 1 diabetes with longer duration of breastfeeding (1 month increment). The control subjects in this study were selected from children admitted to the same hospital as the cases, whether this explained the finding was unclear.

One study reported a reduced risk of type 1 diabetes comparing subjects who were initially exclusively breastfed with those who were not (adjusted OR 0.6, 95%CI 0.41 – 0.89).⁹³ Another study found a small, but non-significant increased risk of type 1 diabetes with not breastfeeding at discharge (RR 1.33, 95%CI 0.76 – 2.31).⁹² This study was rated to have poor methodological quality because only univariate analysis was performed and potential confounders were not considered.

Conclusion

Our findings from the six additional case-control studies are similar to the findings from the two meta-analyses. However, the exclusivity of breastfeeding was not addressed in all studies, and the assessment of infant diet was based on long-term recall in five of six studies. We elected not to perform a meta-analysis, because it is unlikely to change the pooled estimates from the previous meta-analyses by adding additional three studies from the updates that compared subjects who were breastfed for more than 3 or 6 months with those who were never breastfed.

Two meta-analyses of moderate methodological quality reported statistically significant odds ratios of 1.23 and 1.43, respectively, for the risk of type 1 diabetes in subjects exposed to less than 3 months compared with more than 3 months of breastfeeding. Since case-control studies are prone to recall biases, Norris and Scott compared the odds ratios in studies relied on long-term recall to assess infant diet with studies that did not. The results showed that studies using existing infant records to determine breastfeeding initiation and duration failed to show the associations reported in the studies relying on long-term recall for their exposure data. This suggests that subjects with type 1 diabetes were more likely to report shorter duration of breastfeeding than control subjects.

In conclusion, even though there is some evidence to suggest that breastfeeding for more than 3 months is associated with a reduced risk of type 1 diabetes, this evidence must be interpreted with caution because of the possibility of recall biases and suboptimal adjustments for potential confounders in the primary studies.

Background

In 2002, it was estimated that a total of 18.2 million people, or 6.3 percent of the US population carried a diagnosis of diabetes (cdc.gov/diabetes/pubs/estimates.htm#prev). No data are currently available on the prevalence of type 2 diabetes in children and adolescents. The Centers for Disease Control and Prevention (CDC) estimated that among new cases of childhood diabetes, the proportion of those with type 2 diabetes ranges between eight percent and 43 percent (www.cdc.gov/diabetes/pubs/factsheets/search.htm). Type 2 diabetes begins when the body develops a resistance to insulin and no longer uses insulin properly. Adults and children who develop type 2 diabetes are typically overweight or obese and have a family history of the disease. Furthermore, offsprings of mothers who had diabetes during pregnancy had higher rates of type 2 diabetes and obesity.⁹⁷ It is increasingly recognized that nutrition in early postnatal life may have long-term physiologic effects. Studies have suggested that breastfeeding may be protective against later obesity.^{77, 78} Therefore, it seems biologically plausible that there may be a relationship between breastfeeding and long-term glucose and insulin metabolism.

We identified one systematic review by Taylor et al.⁹⁸ in our initial literature search and we also identified one additional systematic review by Owen et al.⁹⁹ during final phase for the preparation of this report. We did not identify any additional systematic reviews.

Commonly considered confounders in the studies of relationship between type 2 diabetes and breastfeeding were age, sex, BMI, birth weight, socioeconomic status, history of parental diabetes, maternal diabetes during pregnancy, maternal diet and smoking, and prepregnancy BMI.

Additional Methodological Comments

We have elected to describe in details only the Owen 2006 systematic review because it superseded the Taylor 2005 systematic review. Owen 2006 included two of the three primary studies that were covered in Taylor 2005. In addition, Taylor 2005 was of poor methodological quality because there was no synthesis of results and it was unclear how conclusions were drawn.

Published Systematic Reviews/Meta-Analyses (Table 16)

Owen 2006 conducted a systematic review and meta-analyses to examine the influence of breastfeeding on type 2 diabetes and blood glucose and insulin concentrations.⁹⁹ Studies that did not provide the odds ratios of type 2 diabetes comparing breastfed to formula-fed subjects were excluded. In addition to two of the three studies included in the Taylor 2005 review (the third study was excluded from Owen 2006 because it did not provide odds ratio of the disease), five additional studies were identified in the Owen 2006 review. Seven studies (six in adults and one in adolescents) totaling 76,744 subjects provided odds ratios that related initial infant feeding methods and type 2 diabetes were included in the meta-analyses. There were three historical cohort, two cross-sectional, one prospective cohort, and one case-control studies. All seven studies were conducted in developed countries. The effects of study size, year of birth, the method of ascertainment of infant feeding status (whether contemporary or recalled up to 71 years after birth), type of formula feeding, study response rate, study design, and whether infants were born pre- or full-term were examined by using meta-regression and sensitivity analyses. Sensitivity analyses were also used to examine the effect of adjustment for important confounders and of fasting status. The methodological quality of this systematic review and meta-analyses was rated grade A.

Six of seven studies related breastfeeding to a lower risk of type 2 diabetes, and there was no evidence of heterogeneity across studies. Overall, the subjects who were breastfed showed a lower risk of type 2 diabetes than those who were formula-fed (pooled adjusted OR 0.61; 95%CI 0.44–0.85, P=0.003). Three studies considered the effects of potential confounding by birth weight, parental diabetes, socioeconomic status, and individual or maternal body size, while the other four studies only considered the effects of confounding by age, sex and/or birth weight. However, the odds ratio relating breastfeeding and diabetes risk was similar before (OR 0.55; 95% CI: 0.35–0.86; P=0.009) and after adjustment for all the important confounders (OR 0.55, 95% CI 0.34–0.90; P=0.017) in the three studies. The method of ascertaining feeding exposure was unrelated to the odds ratios, although there was insufficient power to detect appreciable differences in examining the effect of potential biases (such as study size, year of birth, the method of ascertaining infant feeding status, type of formula feeding, study response rate, study design, and whether infants were born pre- or full-term) by using meta-regression and sensitivity analyses.

Studies Identified after the Published Meta-Analysis/Systematic Review

None was found.

Conclusion

Results from a high-quality systematic review and meta-analyses of seven studies suggest that breastfeeding is associated with a lower risk of type 2 diabetes in later life, compared with formula feeding. Comparing subjects who were ever breastfed to those who were formula fed, the pooled adjusted odds ratio of type 2 diabetes in later life was 0.61 (95%CI 0.44–0.85). However, only three studies appropriately adjusted for all the important confounders, including birth weight, parental diabetes, socioeconomic status, and individual or maternal body size. Even though these three studies found that adjustment did not alter the crude estimate, we cannot be completely confident that potential confounding by birth weight and maternal factors has been ruled out for the overall pooled estimate. This could lead to an overestimate of the association. Publication bias is also a possible explanation for the consistent associations observed in these studies.

Background

Leukemia, the most common cancer in children, encompasses multiple diseases including three types: acute lymphocytic leukemia (ALL), acute myelogenous leukemia (AML), and chronic myelogenous leukemia. In the United States, approximately 3,250 children are diagnosed with leukemia each year and 2,400 (74 percent) of them have ALL. With the exception of prenatal exposure to x-rays and specific genetic syndromes, little is known about the causes of childhood ALL (National Cancer Institute: seer.cancer.gov/publications/childhood/leukemia.pdf). Although the majority of human leukemias or lymphomas have no readily identifiable infectious etiologies, viral causes have been identified for Burkitt's lymphoma and a rare form of adult leukemia/lymphoma.^{100, 101} Because breast milk is noted for providing passive immunity and protection of newborns from some early infections, investigators have hypothesized that breastfeeding may reduce the risk of childhood leukemia.¹⁰²

Some of the confounders that had been considered in the studies of the relationship between childhood leukemia and breastfeeding were age, sex, race, socioeconomic status (such as mother's education, region of residence at the time of diagnosis, parental occupation, deprivation index, and/or annual household income), parental alcohol consumption, parental smoking, birth order, birth weight, parity, and age of mother at birth of index child.

Additional Methodological Comments

We identified four systematic reviews or meta-analyses that examined the relationship between breastfeeding and childhood leukemia¹⁰³–¹⁰⁶ We have elected to describe in details only the Guise 2005¹⁰⁴ systematic review and Kwan 2004¹⁰⁶ meta-analysis because they superseded the Beral 2001¹⁰³ meta-analysis and Davis 1998¹⁰⁵ systematic review. All studies included in Beral 2001 and Davis 10998 were also included in Kwan 2004 or Guise 2005, except that Guise 2005 review excluded studies published before 1990, unpublished data, and studies conducted in developing countries. Kwan 2004 also included three additional studies published after 2001. In addition, the reporting and analysis in Beral 2001 and Davis 1998 were poor due to deficiency in reporting of the meta-analysis methods, search strategy and/or study eligibility criteria, and there was a lack of consideration of potential confounding in the included studies.

Guise 2005's systematic review was the only study that reported methodological grading for individual studies. Guise 2005 did not perform a meta-analysis. Kwan 2004 combined all case-control studies regardless of their study quality into a meta-analyses. To understand further how study quality can affect the effect estimate, we decided to perform meta-analyses on the effect estimates from the four studies graded as good or fair quality in Guise 2005's systematic review.

Published Systematic Reviews/Meta-Analyses (Table 18)

Guise 2005 conducted a systematic review of case-control studies related to breastfeeding and the risk of childhood leukemia. Ten case-control studies totaling 9,653 subjects with leukemia were included. All the studies were conducted in developed countries. Study quality was rated by the authors in a three-levels scale: good, fair, or poor. The following aspects of the study quality were assessed: reliability of the diagnoses of leukemia, comparability of the case and control groups, differences in nonrespondents in cases versus controls, and the conduct of controlling for confounding. There were two good, two fair, and six poor quality studies. All studies but one focused solely on childhood leukemia. There was no meta-analysis or statistical analysis performed. The methodological quality of this systematic review was rated grade A.

Many of the studies in Guise 2005 systematic review did not provide data on exclusivity of breastfeeding and did not consider potential confounders such as infectious exposures from household or school contacts. Six of the ten studies explicitly sought to characterize the relationship between breastfeeding and leukemia.

Table 17

Combined SES-adjusted ORs of ALL for the three case-control (more...)

Table 17

Combined SES-adjusted ORs of ALL for the three case-control studies rated as good and fair methodological quality in the systematic review by Guise et al. (2005)

Study	OR (≤ 6 mo vs. never BF)	Lower CI	Upper CI
UKCCS 2001
<1 mo	0.90	0.77	1.04
1–6 mo	0.98	0.82	1.17
CCG study 1999	0.86	0.73	1.01
Dockerty 1999	1.24	0.47	3.23
Pooled	0.91	0.83	1.00
Kwan 2004 meta-analysis results	0.88	0.80	0.96
Study	OR (> 6 mo vs. never BF)	Lower CI	Upper CI
UKCCS 2001	0.89	0.75	1.05
CCG study 1999	0.72	0.60	0.87
Dockerty 1999
>6 mo to 1 yr	0.82	0.29	2.27
>1 yr	0.47	0.15	1.43
Pooled	0.80	0.71	0.91
Kwan 2004meta-analysis results	0.76	0.68	0.84

Kwan 2004 conducted a meta-analysis of 14 case-control studies on breastfeeding and the risk of childhood leukemia. A total of 8,051 subjects with leukemia were included in the analysis. The meta-analysis examined the relationship between short-term breastfeeding (defined as breastfeeding for 6 months or less) or long-term breastfeeding (defined as breastfeeding for more than 6 months) and the risk of childhood leukemias. Twelve studies from developed countries and two studies from developing countries, including any type of leukemia in children 15 years or younger that reported odds ratio and duration of breastfeeding, were included in the meta-analysis. Kwan 2004 did not formally assess the individual study quality. However, the potential for confounding in each study was considered in the meta-analysis. The methodological quality of this meta-analysis was rated grade A.

For each of the 14 primary studies in the meta-analysis, Kwan 2004 selected odds ratio adjusted for SES when available. The meta-analysis showed a statistically significant reduced risk of ALL with short- and long-term breastfeeding (OR 0.88, 95%CI 0.80 – 0.96; OR 0.76, 95%CI 0.68 – 0.84, respectively). The analysis reported a statistically significant reduction in AML for long-term breastfeeding (OR 0.85, 95%CI 0.73 – 0.98) but not for short-term breastfeeding (OR 0.90, 95%CI 0.80 – 1.02). The unadjusted and adjusted odds ratios for reduction of the risk of ALL in short-term breastfeeding were both statistically significant. For reduction of the risk of ALL in long-term breastfeeding, only the adjusted odds ratio was statistically significant. The same sensitivity analyses were performed for the reduction of the risk of AML in short- and long-term breastfeeding, and similar results were found.

Studies included in the additional analysis. Guise 2005 identified two good and two fair methodological quality studies. We have analyzed these studies further as detailed below.

The two good studies were the UKCCS and the CCG studies. The UKCCS included 1,401 (87 percent) ALL and 214 (13 percent) AML case patients recruited from health programs that enrolled 98 percent of the total childhood cancers throughout England, Scotland and Wales over the period 1991-1998; control subjects were selected from population-based health rosters. In contrast, the CCG study included 1,744 (79 percent) ALL and 456 (21 percent) AML case patients enrolled from specific CCG centers, and control subjects were selected via random-digit-dialing. Both studies excluded leukemia diagnoses less than 1 year of age because most leukemias occurring during infancy are thought to have different etiologies from childhood leukemias. Both studies found that long-term breastfeeding (> 6 months duration) was protective for ALL, but the confidence interval for the risk estimate from the UKCCS did not exclude unity (1.00), whereas the confidence interval for the risk estimate reported by the CCG study clearly excluded unity (1.00), indicating statistical significance. The UKCCS, however, found no association between short- or long-term breastfeeding and the risk of AML, while the CCG study found a significant protective effect of long-term breastfeeding for AML.

Table 22

Meta-Analysis of four RCTs on the effects of breast (more...)

Table 22

Meta-Analysis of four RCTs on the effects of breast milk feeding and NEC in preterm infants: Random Effects Model (D&L)

Study, year	Breast milk feeding		Control		Risk Ratio	95% CI
	Event	Total	Event	Total		Low	High
Gross, 1983	1	41	3	26	0.21	0.02	1.93
Tyson, 1983	0	37	1	44	0.39	0.02	9.41
Lucas. 1984	1	86	4	76	0.22	0.03	1.93
Schanler, 2005	5	78	10	88	0.56	0.20	1.58
Total patients = 476	7	242	18	234	0.42	0.18	0.96

z = -2.0629 2P = 0.039

Overall Heterogeneity: Q = 1.02 Tau^2 = 0.0000

Conclusion

Further evaluation of the biological mechanisms underlying this relationship while taking into consideration potential biases can be achieved with more large-scale case-control studies utilizing population-based and socioeconomic status-matched controls.

Background

Infant mortality (both neonatal and post-neonatal mortality) is on the decline in both developing and developed countries during the past four decades. In the United States, the national average of infant mortality was 7.0 deaths per 1000 live births in 2002.¹¹⁰ Leading causes of infant death in a developed country include congenital abnormalities, pre-term births, low birth weight, Sudden Infant Death Syndrome (SIDS), problems related to complications of pregnancy, and respiratory distress syndrome.¹¹⁰ The common risk factors that are associated with infant mortality include infant and maternal age, gender, race, socioeconomic status, birth order, birth weight, congenital malformation at birth, and maternal smoking during pregnancy. Breastfeeding protects against infectious diarrhea and respiratory diseases, which are the leading causes of infant mortality in developing countries.¹¹¹ However, the role of breastfeeding in preventing infant deaths in developed countries is less clear. We elect to review the relationship between breastfeeding and post-neonatal mortality.

Additional Methodological Comments

For this section, infant mortality is defined as any death that occurred more than 1 month but less than 12 months after birth. Neonatal mortality (death < 1 month of age) is not considered in this review. No published meta-analysis was identified that evaluated the relationship between breastfeeding and infant mortality. We included only studies conducted in developed countries that evaluated the relationship between breastfeeding and infant mortality (excluding SIDS).

Results (Table 19)

We identified two studies conducted in the United States that evaluated the role of breastfeeding in infant deaths other than SIDS.^{112, 113}; one study did not qualify for inclusion in this review because it was an analysis based on data collected from cross-sectional surveys.¹¹³ The study that qualified for inclusion was a case-control study that selected subjects from nationally representative samples.¹¹² The study excluded infants who died at less than one month of age. The methodological quality of the study was rated grade B.

Chen 2004 analyzed the 1988 US National Maternal and Infant Health Survey (NMIHS) data using a case-control study design. Their sample included 1,204 infants who died between 28 days and 1 year of age. The authors excluded infants who died from congenital anomalies or malignant tumors. In an attempt to address the issue of reverse causality, only infants who survived beyond the neonatal period were included and feeding status was categorized based on the assessments undertaken some time before death occurred. The controls included 7,740 live births who were alive and older than 1 year of age at the time of survey. Mothers were surveyed using a mailed questionnaire. Breastfeeding was assessed as “ever breastfed” or “never breastfed.” The study did not report the mean or the range of duration of maternal recall for breastfeeding. Overall and cause-specific odds ratios were calculated to evaluate “ever” versus “never” breastfeeding. Since there was an oversampling of black and low birth weight infants in the original sample, in order for the study population to be more representative of the US population, additional analyses were performed with SUDAAN software adjusted samples. This adjusted sample included a total of 9,145 cases and 3,186,497 controls. The analyses were adjusted for sampling strategy and potential confounders such as maternal age, education, smoking, infant gender, race, birth weight, congenital malformation, live birth order, plurality, and status of enrollment in the Special Supplemental Nutrition Program for Women, Infant, and Children program (WIC). The study reported that the odds of death in the postneonatal period were 21 percent lower for “ever breastfed” compared with “never breastfed” infants. However, in subgroup analyses of cause-specific death, the only statistically significant association was reported between SIDS (in the original sample) or injury-related death (in the SUDAAN-adjusted sample) and “never breastfed” status.

Conclusion

One study of moderate methodological quality using a large sample of infants reported a protective effect of breastfeeding in reducing infant mortality after controlling for some of the potential confounders. However, in subgroup analyses of the study, the only statistically significant association reported was between “never breastfeeding” and SIDS or the risk of injury-related deaths. Because of the limited data in this area, the relationship between breastfeeding and post-neonatal infant mortality remains unclear. Further research is warranted.

Background

Sudden infant death syndrome (SIDS) is the leading cause of mortality among infants aged 1 to 12 months in the United States.¹¹⁴ SIDS accounted for a death rate of 0.55 per 1000 live births for the year 2004 according to the National Center for Health Statistics (NCHS) at CDC (www.cdc.gov/nchs/deaths.htm.). Several modifiable risk factors for SIDS are sleeping positions, maternal smoking, and bed sharing. Other potential risk factors include birth weight, gender, and socioeconomic status. The relationship of breastfeeding and SIDS has been evaluated among a broad range of potential risk factors. However, the role of breastfeeding as a protective factor in SIDS is unclear. One meta-analysis published in 2000 assessed the relationship of breastfeeding and SIDS.

Additional Methodological Comments

We identified four eligible studies conducted in developed countries that evaluated the role of breastfeeding in SIDS, and published since 1997, the cutoff date for the literature search of the published meta-analysis.

Published Systematic Review/Meta-Analysis (Table 20)

One meta-analysis¹¹⁵ of 23 studies (18 case-control; four nested case-control; and one observational cohort) evaluated the relationship of breastfeeding and SIDS. All studies were conducted in developed countries from 1965 to 1997. Studies that reported a minimal definition of SIDS - sudden unexplained death of an infant or young child - met the eligibility criteria. The meta-analysis analyzed a total of 4,251 cases of SIDS and 58,055 controls. Seventeen studies included subjects from birth to 2 years old when SIDS occurred; eight studies did not provide age data. The studies differed in their definition of breastfeeding exposure. Also, the studies varied in their description of SIDS. Of these, only 14 studies reported autopsy-confirmed diagnoses of SIDS. Three studies were specifically designed to examine the relationship of breastfeeding and SIDS. The rest of the studies examined multiple risk factors and their association with SIDS, of which feeding history would be one.

The meta-analysis utilized a random effects model. It reported an overall risk of SIDS twice as great for formula-fed infants compared with breastfed infants (crude odds ratio (OR) of 2.11; 95% CI 1.66 to 2.68). The methodological quality of individual studies was appraised. The authors conducted a separate meta-analysis for those studies published since 1988 when more advanced epidemiological and autopsy procedures were available. A separate analysis was also performed for those studies with “high” quality scores. The results from these meta-analyses concurred with the overall result. Heterogeneity was not explored, i.e., no subgroup analyses were performed to account for the different definitions of interventions or outcomes. The authors reported “no publication bias”. Differences in case matching precluded them from combining adjusted odds ratios in the meta-analysis. Four of the 16 studies showed a dose response trend with the risk of SIDS increasing with increasing formula feeding. The overall methodological quality of the meta-analysis was rated grade C.

Studies Identified after the Published Meta-Analysis (Table 21)

We identified four eligible studies from five publications conducted in developed countries and published since 1997 that evaluated the relationship of breastfeeding and SIDS.¹¹⁶–¹²⁰ Three were case-control studies and the fourth, a case-cohort study (a cohort study analyzed as a case-control study). The methodological quality ranged from grades B to C. There were a total of 769 cases of SIDS and 2,681 controls.

All studies were designed to evaluate a broad range of potential risk factors for SIDS. The studies differed in their description of the duration of breastfeeding. Similarly, the studies varied in their definition of the time interval when SIDS occurs in infants, but all studies reported autopsy-confirmed diagnoses. The mean age of the infants with SIDS ranged from 2 to 19 weeks. All studies provided adjusted odds ratios for the association of breastfeeding and SIDS. Three of the four studies identified statistically significant increased risk of SIDS in bottle-fed infants.^{117, 119, 120} Two studies reported that the risk of SIDS was twice or more for non-breastfed infants compared with some or ever breastfed infants.^{117, 119} One study reported an approximately two times increased risk of SIDS among those breastfed less than 2 weeks compared with those breastfed more than 2 weeks.¹²⁰ One case-cohort study (a cohort study analyzed as case-control study) did not find a statistically significant increased risk of SIDS in bottle fed infants.¹¹⁸

Meta-analysis results. Because of the limitations of the previous meta-analysis, we elected to conduct our own meta-analysis using only studies that provided an objective definition of SIDS (autopsy confirmed SIDS among infants 1 week to 1 year of age), clear reporting of breastfeeding data, and outcomes adjusted for important confounders or risk factors (e.g., sleeping positions, maternal smoking, and socioeconomic status). Four studies included in the previously published meta-analysis¹²¹–¹²⁴ and two studies published since 1997 met the eligibility criteria.^{118, 120} The majority of the studies provided data on ever versus never breastfeeding and this was combined using a random effects model. The results from our meta-analysis found that ever breastfeeding was associated with a reduction in both crude and adjusted risk of SIDS (crude OR 0.41; 95%CI (0.28, 0.58), and adjusted OR 0.64; 95%CI (0.51, 0.81), respectively); both estimates were statistically significant with a reduction in SIDS for the ever breastfed infants.

Conclusion

Results from the previously published meta-analysis of case-control studies concluded that an overall crude risk of SIDS was twice as great for formula-fed infants compared with breastfed infants. The conclusion may be biased because the reported association was not adjusted for potential confounders. Misclassification biases may occur because of differences among studies with regard to definitions of breastfeeding exposure, definitions of SIDS, and the wide age range of population included in the studies.

Findings from the four studies published subsequent to the meta-analysis in developed countries concurred with the findings from the meta-analysis. All studies reported autopsy-confirmed diagnoses of SIDS and adjusted for potential confounders. However, the definitions of breastfeeding exposure and the time intervals accepted for defining SIDS varied across studies. Three of four studies reported statistically significant increased risk of SIDS associated with non-breastfeeding or reduced duration of breastfeeding and the fourth study reported a statistically non-significant increased risk.

Our meta-analysis included only studies that reported clear definitions of exposure, outcomes, and results adjusted for well-known confounders or risk factors for SIDS. The summary estimate found a statistically significant adjusted odds ratio for an association between breastfeeding and a reduced risk of SIDS (adjusted OR 0.64, 95%CI 0.51 – 0.81). We conclude that there is a relationship between breastfeeding and a reduced risk of SIDS.

Background

Necrotizing enterocolitis (NEC) is a serious gastrointestinal disease in the preterm infants. No definitive causes have been identified. A population-based epidemiological study published in 2002 reported that the highest incidence occurred in infants with birth weights 750 to 1000 g and decreased with increasing birth weights.¹²⁵ Observational studies have suggested that breast milk might be protective. There have been very few randomized controlled trials (RCTs) that examined this issue. McGuire 2001 performed a meta-analysis of RCTs of breast milk comparing with formula milk in preterm infants to reduce the risk of NEC.¹²⁶ Since that meta-analysis, we identified one RCT¹²⁷ and two prospective cohort studies^{128, 129} that reported outcome of NEC in preterm infants in relation to a history of breast milk feeding. Some of the potential confounders that may affect the results of neonatal morbidity and mortality include birth weight, ethnicity, and sex.¹²⁸

Published Systematic Review/Meta-Analysis (Table 23)

McGuire 2001's meta-analysis compared formula feeding with term breast milk feeding in low birth weight or preterm infants. Three RCTs published in 1983 and 1984 totaling 308 preterm infants were included: Gross 1983 compared formula with unfortified term donor breast milk;¹² Tyson 1983 compared preterm formula with pooled banked term breast milk;¹³ and Lucas 1984130 (results for NEC reported in 1990¹³¹) compared preterm formula with banked term breast milk as the sole diet. In the meta-analysis comparing formula with breast milk, the risk ratio for developing NEC was 2.5 (95% CI 0.9 – 7.3); risk difference was 0.05 (95% CI 0.00 – 0.1). The authors concluded that there was no statistically significant difference in the risk of NEC with either form of milk feeding.

Studies Identified after the Published Systematic Review/Meta-Analysis (Table 24)

Randomized controlled trial. Schanler 2005 enrolled 243 infants ≤ 29-week gestation whose mothers were expected to breastfeed.¹²⁷ If these infants' own mothers' milk were unavailable, the infants were then randomly assigned to receive either pasteurized donor milk or preterm formula. However, both groups continued to receive mother's milk partially if they were available during the study. The infants who were fed mother's milk exclusively were not randomized and served as a reference group. The incidence of NEC in Donor milk versus Preterm formula was 5/78 versus 10/88 (P = 0.27). The non-randomized group “mother's milk” had fewer repeated episodes of late-onset sepsis and/or NEC (OR 0.18, 95% CI 0.04 – 0.79) compared with combined groups “donor milk” and “preterm formula”. The methodological quality of this study was rated B.

Prospective cohort. Furman 2003 was a prospective cohort study on 119 infants with gestational age < 33 weeks and birth weight 600–1499 g.¹²⁸ Enteral feeding was begun by day 2 or 3 of life, parenteral nutrition was continued until a daily enteral intake of 120 mL/kg of body weight was reached. Infants received their mother's milk in the sequence it was expressed, except that fresh rather than frozen milk was given if available. Maternal milk was fortified, and preterm infant formula was offered when the infant reached a daily oral intake of at least 110 mL/kg. Limited availability of maternal milk was the sole reason infants were fed preterm formula in addition to maternal milk. Four subgroups were analyzed: no maternal milk, daily maternal milk of 1–24 mL/kg, 25–49 mL/kg, and ≥ 50 mL/kg. Rates of NEC did not differ according to the amounts of maternal milk received. The results of the regression analysis were adjusted for birth weight, ethnicity, and sex. The methodological quality of this study was rated B.

Ronnestad 2005 was a prospective cohort study of late-onset sepsis on 462 infants with gestational age <28 weeks or birth weight < 1000 g in Norway.¹²⁹ NEC was not the primary outcome of interest; it was studied as a potential confounder in the analysis of late-onset sepsis. Four hundred five survived until day 7. Participating centers had a common policy of achieving full enteral feeding with the mother's milk or banked donor milk as early as possible, although there was no uniformity in a detailed protocol for feeding strategies. Enteral feeding with breast milk was commenced within 1, 2, or 3 days for 61 percent, 92 percent, and 96 percent of the infants, respectively. Nine of 405 (2.2 percent) patients had confirmed NEC. There was no concurrent comparison reported in this study. The methodological quality of this study was rate C with respect to the outcome of NEC.

Figure 4. Primary studies available to assess the relationship (more...)

   Figure 4. Primary studies available to assess the relationship between breastfeeding and infant health outcomes

Figure 5. Primary studies available to assess the relationship (more...)

   Figure 5. Primary studies available to assess the relationship between breastfeeding and maternal health outcomes

Figure 6. Meta-analysis of the association between (more...)

   Figure 6. Meta-analysis of the association between breastfeeding and the risk of AOM compared to exclusive bottle-feeding in cohort studies

* Exclusive breastfeeding ≥ 6 months vs. exclusive bottle-feeding and breastfeeding < 4 mo

Figure 7. Meta-Analysis of prospective cohort studies (more...)

   Figure 7. Meta-Analysis of prospective cohort studies of the association between asthma risk and breastfeeding ≥ 3 months for children with positive family history of asthma or atopy (excluding Wright 2001)

Figure 8. Meta-Analysis of prospective cohort studies (more...)

   Figure 8. Meta-Analysis of prospective cohort studies of the association between asthma risk and breastfeeding ≥ 3 months for children with positive family history of asthma or atopy (including Wright 2001)

Figure 9. Meta-Analysis of prospective cohort studies (more...)

   Figure 9. Meta-Analysis of prospective cohort studies of the association between asthma risk and breastfeeding ≥ 3 months for children without family history of asthma or atopy

Figure 10. Random effects model of summary estimate (more...)

   Figure 10. Random effects model of summary estimate evaluating the association of breastfeeding and SIDS

Figure 11. Meta-Analysis of four RCTs on the effects (more...)

   Figure 11. Meta-Analysis of four RCTs on the effects of breast milk feeding and NEC in preterm infants

Figure 11

Conclusion

Even though the observational study by Furman et al. did not find a difference in the rates of NEC according to the amount of maternal milk received by the infants, our meta-analysis of four RCTs demonstrated that there was a marginally statistically significant association between breast milk feeding and the reduction in the risk of NEC. The confidence interval for the estimate in the relative risk reduction ranged from four percent to 82 percent. The absolute risk difference was five percent. The wide confidence of the estimate reflects the relatively small number of total subjects in the studies and the small number of events. One must be cognizant of the clinical heterogeneity underlying these RCTs in interpreting the findings of the meta-analysis. Some of them were: different time periods when the studies were conducted; different preterm formulas as comparators; wide range of gestational ages and birth weights in the subjects; different degree of illnesses in the subjects; and others. How the heterogeneity in the studies affected the findings is unclear. Lastly, one may question the importance of an absolute risk difference of five percent between groups. Taking into account the high case-fatality rate of NEC, we consider this estimate is of meaningful clinical difference. In conclusion, there is evidence to support an association between breast milk feeding and a reduction in the risk of NEC in preterm infants.

Background

Many studies have examined the relationship between breastfeeding and cognitive development. Results have been conflicting. Most of the studies were observational in design. Many of them did not have a clear definition of breastfeeding or breast milk exposure. Different cognitive assessment tools were used. Outcomes were measured anywhere from less than 2 years of age to adulthood.

Three systematic reviews from 1999 to 2002 have tried to either establish methodological standards to assess the observational studies or adjust for covariates in pooled analysis (see discussion of confounders in section on cognitive outcomes in term infants). Since the last systematic review by Jain et al. in 2002,⁵⁸ there have been eight cohort studies on preterm infants that examined the relationship of breast milk feeding to some aspects of cognitive development.

A note of caution is in order here. The Mental Developmental Index of the Bayley Scales of Infant Development is widely used to assess the cognitive ability of young children in the studies reviewed in this report. One must keep in mind that the primary purpose of the Bayley Scales is to identify children who may be at risk from developmental delay; it was not the primary purpose to use the results of the Bayley Scales to predict IQ at a later age. Even though some recent studies on preterm infants have shown some predictive ability of the Bayley Scales, the ability to do so is imperfect.^{132, 133} Comorbidities (e.g., neurological impairment, extremely low birth weight, other neonatal illnesses), early intervention, environmental, and socioeconomic factors are some of the additional important variables that could affect the prediction of future cognitive function.

Published Systematic Review/Meta-Analysis (Please Refer To Table 8 In Part I)

Studies Identified after the Published Systematic Review/Meta-Analysis (Table 25)

Since 2002, five prospective cohort studies,¹³⁴–¹³⁹ two nested case control studies,^{140, 141} and one secondary data analysis of a previous randomized controlled trial on supplemental arachidonic and docosahexaenoic acid¹⁴² reported on the relationship between breast milk feeding and some aspects of cognitive development in preterm infants. Sample size of the studies ranged from 39 to 1,035. Five of the studies were of moderate methodological quality,^{135, 137, 140}–¹⁴² and three studies were of poor methodological quality.^{134, 136, 138, 139} Each study was graded within its own study design stratum and only with respect to the data on the relationship of breast milk feeding and cognitive development.

Gestational age of the infants ranged from 26 to 33 weeks. Except for one study that specifically enrolled children who had cerebral ultrasound abnormalities, including echodensity, echolucency, and/or ventriculomegaly,¹⁴⁰ the rest of the studies excluded infants with severe congenital abnormalities. Some also excluded infants with perinatal asphyxia and sensorineural abnormalities.

One study reported that half of the subjects had received breast milk exclusively.¹³⁷ Most studies provided information on the amount of breast milk intake and whether the milk had added cow-milk based fortifiers or not while the subjects were in the neonatal wards, but the information on breast milk intake was less informative after discharge to home. Three studies reported the proportion of subjects who breastfed for more than 6 to 7 months; they were 20 percent,¹⁴⁰ 27 percent,¹³⁷ and 29 percent,¹⁴¹ respectively.

Bayley Mental Development Index Scale (MDI) was the cognitive assessment tool for subjects up to 2 years of age. Wechsler Preschool and Primary Scale of Intelligence (WPPSI-R) was used in subjects under 7 years of age. Weschler Intelligence Scale for Children (WISC-R) was used in 7 years and 11 years old. Kaufman Assessment Battery for children (test for overall intellectual function) and Peabody Picture Vocabulary Test (PPVT-R) were administered to 6 to 8 year old children in one study.¹⁴⁰

Prospective cohort. Elgen 2003 prospectively studied 130 low birth weight children at 5 years and 11 years of age using the Wechsler Preschool and Primary Scale of Intelligence (WPPSI-R) and the Weschler Intelligence Scale for Children (WISC-R), respectively.¹³⁵ Twenty-seven percent of them received less than 30 percent breast milk in the neonatal ward. After adjustment for parental confounding (paternal and maternal education), breast milk was no longer a statistically significant predictor of IQ.

Pinelli 2003 prospectively studied 128 infants with birth weight <1,500 g at 6 months and 12 months corrected age using Bayley Mental Development Index Scale (MDI).¹³⁷ Fifty percent of them received breast milk exclusively in the neonatal ward. Twenty-seven percent of infants were breastfeeding at 6 months. After adjustment for sex, SES, birth weight and maternal age, no statistically significant difference in MDI at 12 months was found between the predominantly breastfed group and the predominantly formula-fed group.

The other three lesser quality studies suffered from selection biases (e.g., convenience sample), lack of adjustment for potential confounders (e.g., lack of adjustment for maternal intelligence), small sample sizes, or had issues with incomplete reporting.

One prospective study (in two publications) with post hoc comparisons reported significant differences (P < 0.05) in Bayley MDI at 6 months between substantial breast milk feeding group (> 75 percent of nutrition as breast milk) and intermediate (25–75 percent) or minimal feeding groups (< 25 percent) (94.2±8.8 vs. 91.7±7.2 vs. 90.5±8.5, respectively).^{134, 136} Maternal education or SES were not adjusted for in these results.

In a cohort study of subjects from a convenience sample (29 in breast milk group; 10 in formula group), a regression analysis showed an association between the amount of milk infants received in the special care nursery and the Bayley MDI at 7 months (r=0.4, R²=0.1, P < 0.05) and 12 months (r= 0.4, R²=0.2, P < 0.025).¹³⁹

In a prospective study of 775 preterm infants who were fed breast milk and 260 preterm infants who were not fed breast milk, the adjusted Bayley MDI at 18 to 22 months of age was 79.9 ± 18 (SD) in the breast milk group versus 75.8 ± 16 (SD) in the no breast milk group (P = 0.07).¹³⁸ The result was adjusted for maternal (education, age, marital status) and perinatal (sex, gestational age, oxygen need, birth weight, and illnesses) factors. Mothers in the breast milk group were more likely to have private health insurance, be white and married, and have a college degree. Mothers who had low household income, higher parity, or were black were less likely to provide breast milk feeds. It was unclear if the results were actually adjusted for household income or not. Further analysis of breast milk intake by quintile relative to the no breast milk group showed that there was a 13.1 point difference in MDI between ≤ 20^th quintile and >80^th quintile (P<0.0044).

Re-analysis of a previous RCT on supplemental arachidonic and docosahexaenoic acid. O'Connor 2003 re-analyzed data on 463 subjects from a randomized controlled trial on supplemental formulas in infants less than 33 weeks gestation.¹⁴² Bayley MDI was evaluated at 12 months corrected age. There were no differences in the Bayley MDI among feeding groups. After controlling for home environment and maternal intelligence, there was a significant positive association between duration of breastfeeding and the Bayley MDI at 12 months corrected age in a “full” statistical model (P = 0.03, adjusted for large number of preplanned covariates for developmental outcomes), but not in a “reduced” statistical model (P = 0.07, adjusted only for those preplanned covariates with a P value of < 0.15).¹⁴²

Nested case-control. Smith 2003 studied 119 preterm subjects with cerebral ultrasound abnormalities and 320 subjects (presumably without ultrasound abnormalities) matched for gestational age in a nested case-control design.¹⁴⁰ The Kaufman Assessment Battery for children were administered to these 6 to 8 years old subjects. In the regression model that included adjustment for maternal verbal ability, home environment, and composite socioeconomic status, the advantage in overall intellectual function associated with direct breastfeeding was 3.6 points (95% CI -0.3 to 7.5; outcome measure was standardized with a mean of 100 points and a standard deviation of 15 points).

Horwood 2001 studied 280 subjects with very low birth weight at 7 years using WISC-R.¹⁴¹ After adjustment for perinatal (sex, gestation, birth weight, multiple birth, Apgar score), socio-demographic (family income, single/two parent family, child ethnicity), and maternal factors (age, education, smoking), there remained a significant association between duration of receipt of breast milk and verbal IQ, with a 6 point advantage for infants who received breast milk for ≥ 8 months compared with no breastfeeding (P<0.001).

Conclusion

No definitive conclusion can be made regarding the relationship between breast milk feeding and cognitive development in preterm infants. One meta-analysis reported a five points advantage in standardized mean score and one systematic review identified one primary study that reported an eight points advantage in IQ in preterm or low birth weight infants who received breast milk feeding. In three of four primary studies of moderate quality that controlled for either maternal education or maternal intelligence, the advantage from breastfeeding was reduced to a statistically non-significant level after adjustment; the fourth study reported a positive association between duration of breastfeeding and the Bayley MDI at 12 months after controlling for maternal intelligence and home environment.

The roles of maternal intelligence and home environment should be accounted for in future studies on breastfeeding and cognitive development. Keeping in mind that cognitive function measured at an early age is not necessarily predictive of later cognitive ability, one should also consider carefully the timing and the selection of appropriate testing instrument in future studies. In addition, clear subject selection criteria, controlling for differences in early complications of prematurity and its relation to receiving breast milk, accounting for subjects lost to follow up, clear distinction between direct breastfeeding and bottle/gavage feeding of breast milk, collect data on breast milk fortifiers or supplemental preterm formulas, and better data collection on breast milk feeding after discharge from the neonatal units will improve the quality of these studies.

Background

Return to pre-pregnancy weight is desirable since postpartum weight retention is a possible risk factor for obesity and its ensuing medical complications.¹⁴³ The change in weight results from changes in energy metabolism during pregnancy and lactation. This is mediated through the complex neuroendocrine and biochemical stimuli that follow from conception.

Despite the fact that the average weight retention associated with child bearing is modest, estimated at approximately 1.51 kg (s.d.=5.95 kg)¹⁴⁴–¹⁴⁶ for some, there is some risk of major weight gain with pregnancy. Ohlin and Rossner 1990 reported changes in body weight that ranged from -12.3 to +26.5 kg from preconception to 1 year postpartum.¹⁴⁴ In various studies, the proportion of women retaining 5 kg or more after 6 months postpartum ranged from 14 to 20%.¹⁴⁶–¹⁴⁸ Studies of the impact of physiological and behavioral influences, such as dietary intake, physical activity, and lactation on postpartum weight change reported mixed results. Studies of postpartum weight changes in lactating and non-lactating women also were equivocal within and across populations, with some showing that the length and intensity of breastfeeding were associated with less weight retention after pregnancy, while other studies reported that women who fed their infants formula lost more weight than women who nursed their infants.

Commonly considered confounders in the relationship between return to pre-pregnancy weight or post-partum weight change and breastfeeding were pre-pregnancy weight or BMI, age, educational level, physical activity, parity, smoking status, dieting practice, and ethnicity.

Additional Methodological Comments

To assess the relationship between breastfeeding and return to pre-pregnancy weight and postpartum weight/BMI changes, we relied on Fraser 2003,¹⁴⁹ the only published systematic review of the effect of lactation on maternal body weight. Since Fraser 2003 did not provide any descriptive or summary tables for further analysis, we contacted the authors to obtain copies of their evidence tables. However, the tables made available lacked the appropriate data for additional clarification of the authors' conclusions. Therefore, we decided to perform a primary analysis of all the studies cited in Fraser 2003. We also screened the references in the primary studies to identify additional studies for inclusion in our systematic review. In addition, we also evaluated primary studies that were published after Fraser 2003 and from suggestions by the reviewers of this report. These primary studies had been identified by our general literature search on the intervention of breastfeeding (see Methods chapter). We did not perform a search on maternal weight change by itself without any reference to breastfeeding.

Because of the known methodological problems in the studies of the relationship between breastfeeding and maternal weight,^{15, 150, 151} we adopted the following inclusion criteria for our review: prospective cohort studies conducted in developed countries which directly compared weight changes of nonlactating women with that of lactating women and for whom the exclusivity or the amount of breastfeeding was clear. The sample size criterion was at least 50 women per feeding group in the final analyses (e.g., lactating versus nonlactating).

For studies of the relationship between postpartum weight change and breastfeeding, studies need to control for subjects' gestational weight gain or pre-pregnancy weight and have at least 3-months postpartum followup to be included.

Results (Tables 26–27)

A total of 54 potentially relevant articles were retrieved for full-text screening. Forty-five articles were excluded because they were cross-sectional design, non-comparative studies, review articles, sample size less than 50 subjects per group, or had unclear breastfeeding data (exclusivity or amount of breastfeeding not clearly reported).

We included a total of three prospective cohort studies that examined the relationship between exclusive or full breastfeeding and return to pre-pregnancy weight^{145, 152, 153} and five prospective cohort studies (in six publications) that examined postpartum weight changes in relation to exclusive breastfeeding.^{144, 154}–¹⁵⁷ One study, Linne 2003, was a 15-year followup of the two earlier Ohlin and Rossner studies of 1990 and 1996.

The reporting of breastfeeding data varied across studies. Some studies reported explicit and quantifiable amount of breastfeeding (e.g., providing at least two-thirds of the needed energy intake per kilogram of the infant's weight in breast milk, or infants received 120 mL/day or less of other milk until at least 1 year of age), while others did not quantify the amount of breastfeeding the infants received.

Conclusion

Based on the results from three prospective cohort studies, we concluded that the overall effect of breastfeeding on return-to-pre-pregnancy weight (weight change from pre-pregnancy or first trimester to 1 to 2 year postpartum) was negligible (less than 1 kg). Results from four prospective cohort studies showed that the effects of breastfeeding on postpartum weight loss were unclear. All seven studies consistently suggested that many other factors have larger effects on weight retention or postpartum weight loss than breastfeeding. Examples of which included annual household income, baseline BMI, ethnicity, gestational weight gain, and energy intake. Undoubtedly, all these factors need to be carefully considered in any future investigation of the relationship between breastfeeding and postpartum weight changes.

Background

Studies have shown that lactation has a beneficial effect on glucose and lipid metabolism; and improved pancreatic beta-cell function in women with gestational diabetes.^{159, 160} Thus, it is plausible that lactation could reduce the risk of the development of type 2 diabetes.

Commonly considered confounders in the studies of relationship between maternal type 2 diabetes and breastfeeding were parity, body mass index (BMI), diet, physical activity, family history of diabetes, and smoking status.

Published Systematic Review/Meta-Analysis (Table 28)

One systematic review examined the effect of breastfeeding on maternal risk of subsequent diabetes.⁹⁸ The authors identified three studies that evaluated the effects of breastfeeding on glucose tolerance and insulin levels; one study that evaluated the risk factors for recurrent gestational diabetes (GDM); and one study that examined the relationship between lactation and the risk of developing type 2 diabetes in women who had GDM. For this report, we focused only on the one study that examined the relationship between breastfeeding and the risk of developing type 2 diabetes in women with GDM.

Kjos 1998 conducted a retrospective cohort study of 904 Hispanic American women with GDM.¹⁶¹ Kjos 1998 found that the use of progestin-only oral contraceptives was associated with an almost three-fold risk of developing type 2 diabetes compared with the use of combination estrogen-progestin oral contraceptives for breastfeeding Latina women with recent GDM. Since progestin-only oral contraceptive use was invariably associated with breastfeeding in this cohort, the authors then looked for an independent effect of breastfeeding on the risk of developing diabetes among women who initially elected nonhormonal forms of contraception and who were breastfeeding at the time of their initial postpartum examination. The results showed that the risk of developing diabetes in these women was not significantly different from the risk in women who elected nonhormonal contraception but did not breastfeed (unadjusted RR 0.90, 95%CI 0.56 – 1.46; adjusted RR 1.16, 95%CI 0.70 – 1.92). The authors of the systematic review did not assess the quality of this study.

The authors of the systematic review concluded that although no study has ever reported an increased risk of developing diabetes from breastfeeding, a single study did show a potential harm from the use of progestin-only oral contraceptive among breastfeeding Latino women with recent GDM.

Studies Identified after the Published Systematic Review/Meta-Analysis (Table 29)

One prospective, longitudinal cohort study on the association between the duration of lactation and the incidence of type 2 diabetes was identified.¹⁶² The study consisted of two large cohorts in the United States, including participants from the Nurses' Health Study (NHS) and Nurses' Health Study II (NHS II). The Nurses' Health Study (NHS) was initiated in 1976 and enrolled 121,700 women from 11 states. Participants were between 30 and 55 years of age at baseline, and each woman completed a detailed baseline questionnaire regarding diseases and health related topics. The Nurses' Health Study II (NHS II), begun in 1989, enrolled 116,671 women from 14 states. Participants were between 25 and 42 years of age and completed a similar baseline questionnaire as well as biennial followup questionnaires. This study was graded as high methodological quality (grade A).

The assessments of lactation history and type 2 diabetes were performed longitudinally. Women in the NHS II were also asked to report the diagnosis of GDM on each biennial questionnaire. Cox proportional hazards model was used to calculate the hazard ratio (HR) for type 2 diabetes by lactation history. All models were age-adjusted. Potential confounders including parity, BMI at age 18 years, diet, physical activity, family history of diabetes, and smoking status were included in the multivariate model a priori. Lifetime lactation history among parous women was stratified into six groups: none (reference group), 0 to 3 months, more than 3 to 6 months, more than 6 to 11 months, more than 11 to 23 months, and more than 23 months. Lifetime duration was updated every 2 years. Linear trend was assessed using midpoints of lactation categories. In the analysis of HR per year of lactation, the midpoints of reporting categories to calculate total lifetime lactation were used because this was the closest approximation of the original reported duration.

In the NHS, women who had ever breastfed had a covariate-adjusted HR for type 2 diabetes of 0.97 (95%CI 0.91 – 1.02) compared with women who never breastfed. There was a modest but statistically significant inverse association between duration of lactation and the risk of type 2 diabetes. In the multivariate-adjusted model including current BMI, each additional year of lactation was associated with an HR of 0.96 (95%CI 0.92 – 0.99) for type 2 diabetes.

Among women who had ever breastfed in the NHS II, the covariate-adjusted HR for type 2 diabetes was 0.90 (95%CI 0.77 – 1.04). Each year of lactation was associated with a covariate-adjusted HR of 0.84 (95%CI 0.78 – 0.89). When BMI was added to this model, the HR was 0.88 (95%CI 0.82 – 0.94) for each additional year of lactation.

Women with a history of GDM had a markedly increased risk of type 2 diabetes in the NHS II cohort, with 624 cases per 100,000 person years compared with 118 cases per 100,000 person-years among those without such a history. Lactation had no effect on diabetes risk in the GDM group, with a covariate-adjusted HR of 0.96 (95% CI 0.84 – 1.09) per additional year of lactation.

The effects of exclusive versus total breastfeeding could be compared in the NHS II cohort data. In models controlling for age and parity, each year of lifetime exclusive breastfeeding was associated with an HR for type 2 diabetes of 0.63 (95%CI 0.54 – 0.73), while each year of total breastfeeding was associated with an HR of 0.76 (95%CI 0.71 – 0.81).

Conclusion

Based on the longitudinal study of two large cohorts in the United States with over 150,000 parous women, we conclude that a longer duration of lifetime breastfeeding is associated with a reduced risk of developing type 2 diabetes among parous women who did not have a history of GDM. There was a difference in the risk of developing type 2 diabetes between women with and without GDM in relation to lactation. Compared with women who did not have a history of GDM, women with a history of GDM had a markedly increased risk of type 2 diabetes; and lactation showed no significant relationship with diabetes risk among this group of women. One must be cautious in interpreting these findings, as they are only generalizable to population with characteristics similar to that of the Nurses' Health cohort.

Background

Osteoporosis is a condition of decreased bone mass. This leads to fragile bones that are at an increased risk for fractures. The World Health Organization (WHO) has established criteria for making the diagnosis of osteoporosis, and for determining levels that predict higher chances of fractures. These criteria are based on comparing bone mineral density (BMD) in a particular patient with those of a healthy 25-year-old female (T-scores). BMD values (T-scores) which fall well below the average for the 25-year-old female (stated statistically as 2.5 standard deviations below the average) are diagnosed as osteoporosis. Although BMD T-scores were based originally on assessment of BMD at the hip by dual-energy X-ray absorptiometry (DXA), they have been applied to define diagnostic thresholds at other skeletal sites and for other techniques. Experts have expressed concern that this approach may not produce comparable data between sites and techniques. Of the various sampling sites, measurements of BMD made at the hip predict hip fracture better than measurements made at other sites, while BMD measurement at the spine predicts spine fracture better than measurements at other sites (consensus.nih.gov/2000/2000Osteoporosis111html.htm).

Calcium and bone metabolism is substantially altered during pregnancy and lactation. The typical daily loss of calcium in breast milk has been estimated to range from 280 to 400 mg, although daily losses as great as 1000 mg calcium have been reported.¹⁶³ Physiologically, during nursing, the body could theoretically meet this demand by increasing the intestinal absorption of calcium, decreasing renal calcium losses, and increasing the resorption of calcium from the maternal skeleton. Bone densities can decrease and increase 3 to 10 percent in the span of a few months in healthy mothers.¹⁶⁴ Prospective cohort studies have reported that lactation is associated with bone mineral loss in the first 6 months to 1 year postpartum, but such loss rebounds overtime.¹⁶⁵–¹⁶⁹

Confounders commonly considered in the studies of relationship between fracture risk and breastfeeding were age, hormone replacement therapy, parity and BMI.

Inclusion and Exclusion Criteria

To evaluate the relationship between breastfeeding and the development of osteoporosis, we included all studies that examined the link between breastfeeding and fracture. We also included long-term prospective cohort studies (greater than 1 year of followup duration) that examined the relationship between the duration of breastfeeding and changes in bone mineral densities or bone mineral contents. Articles were excluded if they only used surrogate measures of fracture (e.g., fracture risk score or index) or bone turnover markers.

Results (Tables 30–31)

A total of 44 potential relevant articles were retrieved for full-text screening. Thirty-four articles were excluded due to various reasons (e.g., cross-sectional design, relatively short duration of follow up (< 2 years), studies done in developing countries). We included a total of six case-control studies that examined the risk of fractures in relation to a history of breastfeeding,¹⁷⁰–¹⁷⁵ and four long-term prospective cohort studies that examined the changes of bone mineral densities or bone mineral contents in relation to the duration of breastfeeding.¹⁷⁶–¹⁷⁹

Incident cases of hip and forearm fractures were identified from hospital or clinical records (with or without radiography confirmations) in five studies. Matched or unmatched general population controls living in the same area were used in three studies, while hospital controls were used in the other two studies. In the remaining one study, all women who were living in three housing blocks were enrolled. Cases of definite vertebral fractures were classified according to the radiological diagnoses and the remaining enrolled subjects without fractures were classified as control subjects. Assessments of breastfeeding history were based on subjects' long-term recalls in all studies. Only one study reported blinded interviewers.

Overall, there was no significant association between a history of breastfeeding and the risk of hip, forearm, or vertebral fractures after adjustment for potential confounders. In four of six studies, parity was considered a potential effect modifier or confounder. Other confounders examined included age, body mass index or weight, hormone replacement therapies, and bilateral oophorectomy. None of the studies provided data on the exclusivity of breastfeeding.

Matsushita 2002 examined the effects of multiple pregnancies on BMD of lumbar spine (L2–L4) in 110 parous Japanese women. The outcome was the percent change in BMD, calculated by subtracting the value at the time of the initial pregnancy from the value at the time of the second pregnancy. The results showed that the BMD after the subsequent delivery was significantly higher than the BMD after the initial delivery (P = 0.001), with a percent change in BMD of 1.4 percent. Independent determinants of the percent change in BMD were explored by multiple regression analysis. The length of lactation between the deliveries showed no correlation with the percent change in BMD (correlation coefficient = -0.06, P = 0.702). Age was the most significant predictor for the percent change in BMD in the model.

Sowers 1992 examined various risk factors for 5-year radial BMD changes in 169 pre- and post-menopausal parous women with European heritage living in the United States. The authors reported that “a recalled history of breastfeeding in parous women did not predict significant differences in BMD level or amount of BMD change”.

Uusi-Rasi 2002 examined the relationship between physical activity, calcium intake, and the maintenance of bone mass in 92 non-smoking premenopausal women living in Finland. The effects of total breastfeeding duration on changes in bone mineral contents (BMC) were also analyzed. According to the multiple stepwise regression analyses, the statistically significant independent predictors for site-specific bone loss were low calcium intake at the baseline and change in body weight both at the proximal femur and at the distal radius sites. In addition, breastfeeding was associated with radial bone loss; the longer the duration of breastfeeding the greater the bone loss (correlation coefficient = -0.34, P = 0.015).

Hansen 1991 examined the risk factors for the development of postmenopausal osteoporosis over a 12-year period in 121 postmenopausal women living in Denmark. One hundred eleven of them (92 percent) had one or more pregnancies (mean 2.6) and breastfed their infants for a mean total breastfeeding duration of 13 months. Comparing postmenopausal women who never breastfed their infants with those who did, there was no significant difference in lumbar spine BMD between groups. In addition, there was no significant difference in the annual rate of postmenopausal bone loss between these two groups.

Conclusion

There is no evidence of an association between lifetime breastfeeding duration and osteoporosis. In six case-control studies, there was no significant relationship between a history of lactation and the risk of fractures in postmenopausal women. In two of three moderate or good quality prospective cohort studies using bone mineral density as a surrogate for osteoporosis, lactation does not appear to have an effect on long-term changes in bone mineral densities. The third study found a small decrease in the bone mineral contents in the distal radius with increased duration of breastfeeding, but no significant changes in bone mineral contents in the femoral neck or the trochanter. However, these findings should be interpreted with caution because the feeding history was obtained by maternal recall and data on exclusivity of breastfeeding were not provided. Further investigation with accurate breastfeeding data is warranted.

Background

Postpartum depression is a serious health problem. The prevalence has been estimated at around 13%.¹⁸⁰ It not only affects mother's health, it also affects her ability to care for her infant. Breastfeeding plays a role in affecting an infant's health and in maternal-infant bonding. It is important to understand the nature of the relationship between postpartum depression and the decision to initiate and continue breastfeeding.

Many studies have examined the relationship between breastfeeding and the development of postpartum depression. The results have been quite variable. This may be explained by the lack of a uniform standard in arriving at a diagnosis of postpartum depression. Some studies used questionnaires and some used clinical interviews and different criteria of depression. Many of the studies had relatively small number of subjects. Furthermore, the items commonly used to assess clinical depression like fatigue and sleep problems are to be expected in caring for a newborn. Some of the potential confounders thought to be important in studies of depressive symptoms and feeding practices were marital status, employment status, and whether or not the pregnancy was planned.¹⁸¹

Additional Methodological Comments

We screened the abstracts identified from the MEDLINE general search on breastfeeding in November 2005. Abstracts qualifying for full text retrieval included studies on the relationship between breastfeeding and postpartum depression, psychological disorders, psychiatric illnesses, or mental health issues. We also identified additional articles based on reviews of the bibliographies cited in the relevant retrieved studies from the search. We only included studies that had at least 100 nursing mothers. Qualifying study designs included prospective cohort studies and case-control studies. All methods of assessment of depression were included. Only data pertaining to the relationship of breastfeeding and postpartum depression were extracted from the studies.

Results (Table 32)

Prospective cohort. A total of six prospective cohort studies qualified for inclusion.¹⁸¹–¹⁸⁶ There were no case-control studies. The number of women in each study ranged from 113 to 2,375. Three of six studies were rated methodological quality grade B within their respective study design hierarchy and with respect to only the data on the relationship of breastfeeding and postpartum depression. Studies of methodological quality grade C suffered from a combination of incomplete reporting of relevant data, inadequate blinding, lack of or suboptimal adjustment for confounding factors.

Four of six studies did not have specific inclusion criteria based on baseline mental health status. Four studies screened for depression using the Edinburgh Postnatal Depression Scale (EPDS), but the cut off point ranged from 9 to 13 (lowest severity score was zero, highest severity score was 30). Four studies established the diagnosis of depression after clinical interviews. None of the studies provided a clear definition of breastfeeding.

Four prospective cohort studies of moderate methodological quality totaling 4,941 subjects reported postpartum depression rates of 6% to 18%.¹⁸¹–^{183, 185} In addition to a history of breastfeeding, all of the studies also considered socio-demographic and obstetric variables as independent predictors of postpartum depression. Assessment of depression by self-reported questionnaires or interviews took place from 1 to 12 months after birth. Except for one study,¹⁸² all of them reported that not breastfeeding or early cessation of breastfeeding was associated with postpartum depression. One study reported that onset of postpartum depression occurred before cessation of breastfeeding in most cases.¹⁸⁵ One study reported that depressed mothers were less likely to continue breastfeeding beyond 2 to 4 months compared with mothers who were not depressed.¹⁸³ In the one study that reported no significant difference in the development of depression in mothers who breastfed versus those who did not breastfeed, women who were breastfeeding at 1 month and were worried about breastfeeding were significantly more likely to become depressed than those who did not worry (RR 3.0, 95%CI 1.041 – 9.216).¹⁸²

Despite the poor methodological quality of the remaining studies, their findings were also consistent with those from studies of moderate quality. One study reported that high EPDS scorers experienced breastfeeding more negatively than the low EPDS scorers (51% versus 16%, P < 0.0001).¹⁸⁶ One study reported that mothers who were depressed were less likely to initiate breastfeeding.¹⁸⁴

Conclusion

Studies of moderate quality reported an association between not breastfeeding or short duration of breastfeeding and postpartum depression. More investigation will be needed to determine the nature of this association. It is plausible that postpartum depression led to early cessation of breastfeeding, as opposed to breastfeeding altering the risk of depression. Both effects might occur concurrently. Additional factors that may have a bearing on both postpartum depression and the decision to initiate or terminate breastfeeding should also be sought. Moreover, documentation of baseline mental health status before the initiation of breastfeeding and detailed recording of breastfeeding data will improve the quality of the studies and help understand the nature of the association.

Background

Breast cancer is the second most frequently diagnosed and second most deadly cancer among women.¹⁸⁷ Risk factors associated with increased risk of breast cancer include: family history, nulliparity, early menarche, hormone replacement therapy, obesity, and advanced age. Ovarian cancer ranks seventh in the most frequently diagnosed and fourth in the most deadly cancer among women.¹⁸⁷ Risk factors for ovarian cancers are similar to those of breast cancer. While breast and ovarian cancers are closely associated with parity, women with increased parity also have increased lifetime duration of breastfeeding. Therefore, it would be instructive to examine the relationship of breastfeeding and the risk of developing breast or ovarian cancer.

Methods

We identified two meta-analyses and one systematic review that evaluated the relationship between breastfeeding and maternal breast cancer.¹⁸⁸–¹⁹⁰ We also identified 23 primary studies published since 2001, the cut-off date for literature search used in the latest meta-analysis. Primary studies were screened based on the same inclusion criteria described in the latest meta-analysis. In addition, we elected to include only primary studies conducted in developed countries. Twenty of the 23 primary studies were excluded for the following reasons: studies published in developing countries (5); studies that did not meet the eligibility criteria (e.g., studies that did not provide data on the following: incident invasive breast cancers, reproductive factors, and use of hormonal preparation) (13); duplicate study (1); and review that was not systematically conducted (1). A total of three studies from developed countries published subsequent to the latest meta-analysis were included in the update.¹⁹¹–¹⁹³

Published Systematic Reviews and/or Meta-Analyses (Table 33)

The most recent meta-analysis published by the Collaborative Group on Hormonal Factors in breast cancer combined 45 studies published through 2001 and two unpublished studies.¹⁸⁹ The meta-analysis evaluated a total of 50,302 parous women with incident invasive breast cancer and 96,973 controls. Included in the meta-analysis were primary studies that analyzed at least 100 cases of incident invasive breast cancers per study regardless of the menopausal status with additional information on reproductive factors and use of hormonal preparations. Both cohort and case-control studies from developed and developing countries were included. Individual subject data from the primary studies were analyzed for homogeneity across study definitions. The majority of the primary studies did not differentiate between exclusive and partial breastfeeding, and some studies varied in the definition of “ever breastfeeding”. The average age at diagnosis of breast cancer in the studies was 50 years. There was a higher proportion of women with either nulliparity or low parity in the breast cancer group compared with the control group. In addition, lower number of the parous women in the breast cancer group had ever breastfed their infants compared with the control group. There was a statistically significant reduction in risk of breast cancer by 4.3% (95% CI 2.9–5.8) for each year of breastfeeding. The reduction in the risk of breast cancer with breastfeeding remained unaltered even after stratification for potential confounders such as parity, number of children breastfed, menopausal status, and lifetime duration of breastfeeding. The results were also adjusted for ethnic origin, education, family history of breast cancer, age at menarche, height, weight, body mass index, and use of hormonal contraceptives, alcohol, and tobacco. Decrease in the relative risk of breast cancer associated with each year of breastfeeding remained homogeneous across the studies with regards to developed versus developing countries, age at diagnosis, menopausal status, family history of breast cancer, and study designs. In addition, the decrease in relative risk of breast cancer in parous women, according to breastfeeding history (ever versus never) and number of births, was more pronounced after four or more births. The methodological quality of the meta-analysis was rated grade A.

In the meta-analysis by Bernier 2000 evaluated the relationship of breastfeeding with histologically diagnosed breast cancer in parous women.¹⁸⁸ The meta-analysis combined 25,871 cases and 44,910 controls from 23 primary case-control studies. Criteria for inclusion are studies from developed and developing countries published in English or French languages between 1980 and 1998 that provided usable data for the calculation of odds ratio of breastfeeding and breast cancer risk. The meta-analysis used both fixed and random effects model. There was a small, but statistically significant decreased risk of breast cancer in women who had breastfed their infants compared with women who had not. The decreased risk was further explored by examining the menopausal status at the time of diagnosis of breast cancer and the duration of breastfeeding. Women who were pre-menopausal at the time of diagnosis of breast cancer had a small but statistically significant decreased risk of breast cancer compared with menopausal women. The duration of breastfeeding was divided into categories of 1 to 6 months, 7 to 12 months, and more than 12 months. When these categories of breastfeeding were compared with non-breastfeeding, only those whose lifetime duration of breastfeeding was longer than 12 months (compared to never) had a small but statistically significant reduction in the risk of breast cancer in subgroup analysis. The authors reported that there was no “publication bias”. The methodological quality of the meta-analysis was rated grade B.

Another systematic review included all English language studies published through 1998 and enrolled at least 200 women with breast cancer.¹⁹⁰ In addition, only studies that explicitly adjusted for the number of full-term pregnancies and age at first birth were included. The data included 19,482 cases and 37,627 controls from 24 case-control studies, and 3,857 cases identified from three longitudinal followup studies that comprised of 229,574 subjects. The case-control studies were conducted in hospital- or population-based settings. Studies conducted in developed and developing countries were included. Based on qualitative appraisal, the authors concluded that either there was no relationship between breastfeeding and the risk of development of breast cancer or there was a weak protective effect of ever breastfeeding against the development of breast cancer. The authors did report some reduction in the risk of development of breast cancer in premenopausal women who had breastfed their infants for long duration. The methodological quality of the systematic review was rated grade B.

Studies Published after the Systematic Reviews/Meta-Analyses (Table 34)

We identified three eligible studies from developed countries that were published subsequent to the latest meta-analysis.¹⁹¹–¹⁹³ One study was multi-center;¹⁹² the remainders were single center.^{191, 193} One study was a prospective cohort of Korean women and evaluated only premenopausal women.¹⁹³ The other two were case-control studies; one evaluated the relationship between breast cancer and breastfeeding among carriers of deleterious BRCA1 and BRCA2 mutations,¹⁹² the other evaluated incident invasive breast cancer patients.¹⁹¹ Two studies included only women whose mean age at diagnosis of breast cancer was in the premenopausal range.^{192, 193} All studies reported a statistically significant reduced risk or odds of breast cancer with increased duration of breastfeeding, which varied across the studies. The duration of breastfeeding ranged from 12 to 24 or more months. The methodological quality of the studies ranged from grade B to C.

Conclusion

Results from both meta-analyses concluded that there was a reduction in the risk of breast cancer in women who breastfed their infants. No studies evaluated exclusive breastfeeding. Studies also reported decreased risk or odds of breast cancer in women with a lifetime breastfeeding of more than 12 months. Neither one of the meta-analyses detected any publication bias. In addition, one of the meta-analyses and the systematic review reported decreased odds of breast cancer primarily in premenopausal women. Findings from primary studies published subsequent to the meta-analyses concurred with the findings from the meta-analyses. In conclusion, there is evidence to support the observation that breastfeeding is associated with a reduction in the risk of breast cancer.

Methods

No meta-analysis was identified on this topic. Eligible designs included prospective cohorts, case-cohort studies or nested case-control studies. Cross-sectional studies were excluded, as described in the Methods section. Eligible studies were conducted in developed countries. We quantified the association between breastfeeding and any type of histopathologically defined maternal ovarian cancer. No specific exclusion criteria were used for the participants in the primary studies.

We recorded or estimated odds ratios for the association between breastfeeding and ovarian cancer for the following comparisons: women who ever breastfed versus never breastfed; women who breastfed less than 12 months (cumulative duration) versus those who never breastfed; and women who breastfed at least 12 months (cumulative duration) versus never. The 12-months cutoff was arbitrary, and was chosen for symmetry with analyses on maternal breast cancer outcomes. At a minimum, studies that were included in the meta-analyses must have been adjusted for parity (or used matching for parity). Adjustment for the use of oral contraceptives (or appropriate matching) was desirable but not mandatory. Studies that reported unadjusted effects only were not included in the meta-analyses, but are reported in the tables and text.

For each study, we estimated the odds ratios for the comparisons of interest when they were not directly reported. This was done for studies that analyzed breastfeeding as an ordinal categorical predictor using cutoffs other than 12 months, provided that mothers who never breastfed were the reference category. In such studies, the odds ratios for the comparisons of interest were estimated by combining odds ratios for different durations of breastfeeding using a random effects model. For example, Riman 2002¹⁹⁴ did not report the odds ratio of ever versus never breastfeeding, but reported odds ratios for specific durations of breastfeeding versus never breastfeeding. Compared with women who never breastfed, the adjusted odds ratio in the Riman study for 1 to 5 months of cumulative breastfeeding was 0.99 (95%CI 0.64 –1.52); for 6 to 11 months, it was 0.77 (95%CI 0.50 – 1.19) and for at least 12 months of cumulative breastfeeding, it was 0.87 (95%CI 0.56 – 1.35). We can estimate the odds ratio of ever versus never breastfeeding in the Riman 2002 study with a random effects meta-analysis of the aforementioned duration-specific odds ratios (estimated adjusted odds ratio was 0.87, 95%CI 0.68 – 1.12).

Results (Tables 35–38)

We did not identify any prospective studies. We found 15 eligible case-control studies that examined the relationship between breastfeeding and maternal ovarian cancer.¹⁹⁴–²¹⁰ In addition, we identified three studies that conducted secondary analyses on population subgroups including white, black, and Jewish women by pooling data from case-control studies.²¹¹–²¹³ Population-based controls were used in nine of the 15 studies and hospital-based controls were used in the remaining six. A total of 6,006 subjects with ovarian cancer were studied. The smallest sample included 76 subjects with ovarian cancer matched with 76 hospital-based controls,²⁰⁸ and the largest included 1,028 subjects matched with 2,390 hospital-based controls.¹⁹⁵

Meta-analyses

Figure 12. Meta-Analysis of case-control studies on (more...)

   Figure 12. Meta-Analysis of case-control studies on the relationship between breastfeeding and maternal ovarian cancer risk: ever breastfed versus never breastfed

Ever breastfed versus never breastfed. Nine studies with a total of 4,387 cases and 10,574 controls were included in the comparison of ever versus never breastfeeding (Figure 12). The random effects meta-analysis found an association between breastfeeding and reduced ovarian cancer risk (OR_adj 0.79, 95%CI 0.68 – 0.91).^{194, 195, 197, 199}–^{201, 204}–^{206, 210} There was a statistically significant between-study heterogeneity (P=0.02), which was mainly due to the outlying study of Chiaffarino 2005. Excluding the study yielded very similar estimates without a statistically significant heterogeneity.

For five of the nine studies we, estimated the odds ratio of ever versus never breastfeeding.^{194, 197, 199}–^{201, 210} If we exclude these five studies from the meta-analysis, a total of 2,582 cases and 4,138 controls remained; and the association between breastfeeding and reduced ovarian cancer risk was no longer statistically significant (OR_adj 0.80, 95%CI 0.59 – 1.09)^{195, 204}–²⁰⁶ Again, there was a statistically significant heterogeneity among the four remaining studies (p<0.01), but the heterogeneity was not readily explained by the characteristics of the primary studies. Excluding the study by Chiaffarino 2005,¹⁹⁵, which was an obvious outlier, the summary odds ratio suggested a slightly stronger (and formally statistically significant) association (0.70 [95%CI, 0.59–0.83]).

Figure 13. Meta-Analysis of case-control studies on (more...)

   Figure 13. Meta-Analysis of case-control studies on the relationship between breastfeeding and maternal ovarian cancer risk: breastfed less than 12 months (cumulative duration) versus never breastfed

Two studies (Chiaffarino 2005¹⁹⁵ and Hartge 1989¹⁹⁹) were included in this meta-analysis despite the fact that they used as cutoffs shorter cumulative different than 12 months (10 and 9 months, respectively). A sensitivity analysis that excluded them yielded similar inferences (summary of adjusted odds ratios 0.87 [95%CI 0.74, 1.02]).

Less than 12 months of cumulative breastfeeding versus never breastfed. Cumulative breastfeeding for less than 12 months was not statistically significantly associated with a decreased risk of ovarian cancer in a meta-analysis of six studies, including 1,911 cases and 5,007 controls in total (OR_adj 0.95; 95%CI: 0.80 – 1.12). (Figure 13) The odds ratios for this comparison were estimated for three of the six studies.^{194, 195, 200, 201} There was no statistically significant heterogeneity between the six studies. Two studies (Chiaffarino 2005¹⁹⁵ and Hartge 1989¹⁹⁹) were included in this meta-analysis despite the fact that they used as cutoffs shorter cumulative duration of breastfeeding (10 and 9 months, respectively). Excluding them, the summary of the adjusted odds ratio remained statistically non-significant (0.87, 95%CI 0.74 – 1.02).

Figure 14. Meta-Analysis of case-control studies on (more...)

   Figure 14. Meta-Analysis of case-control studies on the relationship between breastfeeding and maternal ovarian cancer risk: breastfed at least 12 months (cumulative duration) versus never breastfed

Two studies (Chiaffarino 2005¹⁹⁵ and Hartge 1989¹⁹⁹) were included in this meta-analysis despite the fact that they used as cutoffs shorter cumulative different than 12 months (10 and 9 months, respectively). A sensitivity analysis that excluded them yielded similar inferences (summary of adjusted odds ratios 0.63 [95%CI 0.50, 0.79]).

At least 12 months of cumulative breastfeeding versus never breastfed. Six studies including 1,650 cases and 4,575 controls provided adjusted odds ratios for this comparison,^{194, 197, 210} or allowed an approximation.^{195, 199}–²⁰¹ (Figure 14) Breastfeeding of at least 12 months cumulative duration was associated with 28% lower odds for ovarian cancer (OR_adj 0.72, 95%CI 0.54 – 0.97). There was a statistically significant heterogeneity across the six studies. Two studies (Chiaffarino 2005¹⁹⁵ and Hartge 1989¹⁹⁹) were included in this meta-analysis despite the fact that they used as cutoffs shorter cumulative duration of breastfeeding (10 and 9 months, respectively). Excluding them, the summary adjusted odds ratio was even more suggestive of a protective association (0.63 [95%CI 0.50, 0.79]).

Subgroup analyses and additional results

Table 36

Summary of case-control studies on the relationship (more...)

Table 36

Summary of case-control studies on the relationship between breastfeeding and maternal ovarian cancer by menopausal status

Author Year Country	Cases (N)	Controls (N)	Definition of disease	Mean or median age at dx of disease (year)	Breastfeeding group	Comparator group	Confounders adjusted	ORadj (95% CI)	p for trend	Quality
Tung 2003; 2005	Premenopausal	Premenopausal	Epithelial ovarian cancer based on histology	Cases 55	BF ≤6 mo	Never BF	Age, ethnicity, study site, education, tubal ligation, HRT, ovulation	0.96 (0.49–1.85)		B
USA	217	256		Controls 55	BF 6–12 mo			0.81 (0.43–1.54)	0.003
					BF >12 mo			0.46 (0.22–0.97)
	Postmenopausal	Postmenopausal			BF ≤6 mo	Never BF		0.64 (0.41–1.02)
	341	351			BF 6–12 mo			0.60 (0.36–1.00)	0.08
					BF >12 mo			0.69 (0.43–1.12)
Siskind 1997	Premenopausal	Premenopausal	Epithelial ovarian cancer based on histology	Cases 58	BF 1–6 mo	Never BF	Age, parity, age at 1st birth, education, oral contraceptive use, smoking history, menopausal status	0.75 (0.46–1.21)		B
Australia	215	264		Controls 57	BF 7–12 mo			0.53 (0.31–0.94)	ND
					BF 13–24 mo			1.03 (0.54–1.95)
					BF >24 mo			0.29 (0.08–1.04)
	Postmenopausal	Postmenopausal			BF 1–6 mo	Never BF		0.98 (0.65–1.47)
	403	460			BF 7–12 mo			0.83 (0.54–1.26)	ND
					BF 13–24 mo			0.88 (0.56–1.38)
					BF 25–36 mo			0.93 (0.46–1.88)
					BF >36 mo			1.27 (0.50–3.2)
Wynder 1969	Premenopausal	Premenopausal	Epithelial ovarian cancer based on histology	Cases 52	BF >12 mo	---	ND	---	---	C
USA	55	95		Control ND	9% Cases
					7% Controls
	Postmenopausal	Postmenopausal			BF >12 mo
	95	205			24% Cases
					21% Controls

Dx, diagnosis; BF, breastfeeding; Mo, month(s); Yr, year(s)

Pre- versus post-menopausal women (Table 36). Three studies reported on the risk of ovarian cancer stratified by menopausal status. There were sporadic statistically significant associations between specific categories of breastfeeding duration and reduced risk for ovarian cancer in Sikind 1997²⁰⁴ and Tung 2005.²⁰⁷ Tung 2005 reported a significant dose-response trend between increasing breastfeeding duration and reduction in maternal ovarian cancer risk for post-menopausal women (P value for trend = 0.003). No statistically significant dose-response trends were found in the postmenopausal stratum of Tung 2005 and in both strata in the other two studies. Wynder 1969 reported that in premenopausal women, 10 percent of the cases versus seven percent of the controls breastfed for 12 months or more;²⁰⁹ for postmenopausal women, 24 percent of the cases versus 21 percent of the controls breastfed for 12 months or more. No statistical comparison was reported.

Table 37

Summary of case-control studies on the relationship (more...)

Table 37

Summary of case-control studies on the relationship between breastfeeding and maternal ovarian cancer by histologic type

Author Year Country	Cases (N)	Controls (N)	Mean age at dx of disease (year)	Histologic type (n)	Breastfeeding group	Comparator group	Confounders adjusted	Adjusted odds ratio (95% CI)	p for trend	Quality
Chiaffarino 2005	1028	2390	Cases 56	Serous (492)	Ever BF	Never BF	Age, center, education, parity, oral contraceptive use, family hx of ovarian/ breast cancer in 1° relatives	1.1 (0.85–1.48)	0.71	B
Italy			Controls 57	Mucinous (81)				1.59 (0.82–3.07)	ND
Titus-Ernstoff 2001	528	523	Cases 20–74	Serous borderline (86)	Ever BF	Never BF	Age, state, parity	0.8 (0.4–1.6)	0.61	B
USA			Controls matched within 4 years	Serous invasive (229)				1.0 (0.6–1.4)	0.34
				Mucinous (83)				0.6 (0.3–1.2)	0.88
				Endometrioid/clear cell (130)				0.4 (0.2–0.7)	0.04
Tung 2003	558	607	Cases	Mucinous (109)	Ever BF	Never BF	Age, ethnicity, study site, education, oral contraceptive use, tubal ligation	0.8 (0.5–1.4)	0.99	B
USA			Mucinous 50		BF ≤5 mo			0.6 (0.4–1.4)
			Nonmucinous 55		BF 6–16 mo			0.7 (0.3–1.3)
			Controls 56		BF >16 mo			0.9 (0.8–1.8)
				Nonmucinous (449)	Ever BF	Never BF		0.8 (0.5–1.4)	0.0005
					BF ≤5 mo			0.7 (0.4–0.9)
					BF 6–16 mo			0.5 (0.3–0.7)
					BF >16 mo			0.4 (0.3–0.7)
Ness 2000	531	1,190	Cases	Borderline	Months BF		Age, number of live births, years of oral contraceptive use, years of non-contraceptive estrogen use and months breastfed, tubal ligation, hysterectomy, family history of ovarian and breast cancer, ethnicity	0.95* (0.92–1.00)	ND	B
Modugno 2001			Invasive 53	Serous (79)				0.99 (0.96–1.02)
USA			Borderline 45	Mucinous (60)
			Controls 49	Invasive	Months BF			1.00 (0.99–1.01)	ND
				Serous (278)				1.01 (0.98–1.03)
				Mucinous (52)				0.98 (0.95–1.00)
				Endometrioid (136)				0.97* (0.94–1.00)
				Other (150)
Riman 2002	655	3,899	Cases 62	Invasive	BF 1–5 mo	BF <1 mo	Age, BMI, parity, age at menopause, duration of oral contraceptive use, HRT use	0.87 (0.50–1.53)	ND	B
Sweden			Controls 63	Serous (240)	BF 6–11 mo			0.61 (0.35–1.09)
					BF ≥12 mo			0.87 (0.49–1.54)
				Mucinous (44)	BF 1–5 mo	BF <1 mo		2.19 (0.49–9.87)	ND
					BF 6–11 mo			1.75 (0.39–7.87)
					BF ≥12 mo			0.83 (0.17–4.14)
				Endometrioid (126)	BF 1–5 mo	BF <1 mo		1.05 (0.47–2.34)	ND
					BF 6–11 mo			1.10 (0.50–2.46)
					BF ≥12 mo			1.02 (0.44–2.37)
				Clear cell (25)	BF 1–5 mo	BF <1 mo		0.54 (0.16–1.87)	ND
					BF 6–11 mo			0.23 (0.06–0.88)
					BF ≥12 mo			0.24 (0.06–0.97)

*

p<0.05

Risk by histologic type (Table 37). Five studies reported the relationship between breastfeeding and the risk of ovarian cancer subgrouped by tumor histology.^{194, 195, 201, 205}–²⁰⁷ Comparisons between tumor types included mucinous versus nonmucinous, and invasive versus borderline tumors. Data were limited to allow robust conclusions per histologic type. Moreover, these analyses were subgroup analyses, and therefore, should be viewed as hypothesis forming observations and be interpreted conservatively.

Sporadic associations between specific durations or breastfeeding and reduced risk for histological subtypes were reported in all studies except for Chiaffarino 2005. Breastfeeding was not protective for mucinous or serous cancers consistently across the five studies. Increased duration of breastfeeding was associated with reduced risk for non-mucinous cancers in Tung 2003 (P for trend < 0.001). Titus-Ernstoff 2001 reported an association of breastfeeding with reduced risk for combined endometrioid and clear cell carcinomas, and Riman 2002 reported an association with reduced risk for clear cell ovarian cancer for a specific duration of cumulative breastfeeding (5 months).

Pooled analysis of subgroups (Table 38). Secondary subgroup analyses were conducted on different racial groups in three studies. John 1993 studied 53 cases of ovarian cancer in black women using data from seven case-control studies,²¹¹ whereas Whittemore 1992 examined data of white women from the same studies and added the cases from five additional studies for a combined population of 1,071 cases. Ever having breastfed was not associated with the risk of development of ovarian cancer in black women but was reported to reduce the risk for white women. Whittemore 1992 divided the sample into hospital and population subjects, the odds ratio for those who breastfed their infants compared with those who did not in the development of ovarian cancer were 0.73 (95%CI 0.51–1.0) and 0.81 (95%CI 0.68–0.95), respectively.²¹² Modugno 2003 analyzed 242 Jewish women from five studies conducted in Israel and the United States.²¹³ This was a cohort study comparing carriers and noncarriers of BRCA1 or BRCA2. The study did not find a difference between carriers and noncarriers of BRCA1 or BRCA2 mutation in the breastfeeding status or breastfeeding duration and the risk of ovarian cancer.

Conclusion

We reviewed 15 case-control studies that examined the relationship between breastfeeding and the risk of ovarian cancer, and performed quantitative syntheses using data from nine studies that adjusted for potential confounders. The overall result from the nine studies suggests an association between breastfeeding and a reduction in the risk of ovarian cancer. Because the reporting in these studies was inconsistent, we needed to estimate the odds ratios in five of nine studies for the meta-analysis. Excluding these five studies results in loss of statistical significance for this association.

There was indirect evidence for a dose-response relationship between breastfeeding and a reduction in the risk of ovarian cancer. Breastfeeding of less than 12 months (cumulative duration) was not statistically significantly associated with a reduction in the risk of ovarian cancer in a meta-analysis of six studies. However, breastfeeding of more than 12 months (cumulative duration) was associated with a reduction in the risk of ovarian cancer, compared with never breastfeeding. We caution that the cutoff of 12 months was arbitrary, and the odds ratios were estimated in half of these studies. Therefore, the interpretation of the postulated dose-response relationship should be done with caution.

Finally, several studies assessed subgroups of pre- and post-menopausal women and ovarian cancer histology. Overall, few sporadic statistically significant associations were identified in some subgroups and for some specific durations of breastfeeding. These findings do not constitute robust evidence.

We conclude that there is some evidence to suggest an association between breastfeeding and a reduction in the risk of maternal ovarian cancer. However, one must be cautious in interpreting this association because it was largely based on estimations of the odds ratios from retrospective studies.