Chapter  119:  Perinatal Depression: Prevalence, Screening Accuracy, and Screening Outcomes

Inclusion and Exclusion Criteria

To identify relevant studies, we generated a list of inclusion and exclusion criteria for each key question. We made the criteria fairly restrictive to ensure that our conclusions would be based on the highest quality data available with the lowest risk of bias. Some criteria were common across the three key questions; others were specific to the question. Table 2 summarizes the criteria.

For all key questions, studies had to report on original data, be in English, and be published from January 1980 through March 2004. This time frame ensured that the applied reference standards were consistent with the Diagnostic and Statistical Manual for Mental Disorders, Third Edition (DSM-III), or later criteria for the diagnosis of depression. The study could be conducted in any clinical setting or home but had to be from a developed country to increase the likelihood of being generalizable to the US population. In our original criteria submitted in the research proposal, we proposed including only studies done in the United States, the United Kingdom and other Commonwealth/English-speaking countries, Europe, and Scandinavia. However, we determined after abstract review that such limitations would leave out a large number of relevant studies. Therefore, we modified our inclusion criteria to accept any study conducted in developed countries where the population could be generalized to pregnant and postpartum women in the United States, regardless of the language spoken. We excluded studies published before 1980 or in a language other than English and those on women in less developed countries. We also excluded studies of women with major or minor depression in which the outcomes of interest were not distinguishable from those for women with bipolar disorder, primary psychotic disorders, or maternity blues.

In addition, studies for all key questions had to assess women for major depression either alone or together with minor depression during pregnancy or the first year postpartum by means of a clinical assessment or structured clinical interview. For Key Question (KQ) 1, we excluded studies of the prevalence and incidence of perinatal depression that relied solely on self-report screens to identify depression. For KQs 2 and 3, we excluded studies that included women with known depressive disorders at the outset. In KQ 2, study investigators used the clinical assessment or structured clinical interview as the criterion or gold standard with which to assess the properties of the screening instrument. In many KQ 3 studies, investigators used the clinical assessment to measure the depression outcomes from screening with subsequent intervention among women found to be at elevated risk of depression. Studies that measured women's mood using self-report measures only were also included in KQ 3.

For KQ 1, we included both prospective and retrospective studies of the prevalence and incidence of perinatal depression and studies that were conducted for purposes other than determining the prevalence and incidence of perinatal depression but nevertheless included a population-based estimate meeting the other inclusion criteria (e.g., studies of the properties of screening instruments). Furthermore, to answer the second part of KQ 1, we included both clinical trials and case-control studies comparing the incidence or prevalence of depression among pregnant women and newly delivered mothers to prevalence among women of similar age during other nonchildbearing periods of their lives. We included only prospective studies in those reviewed for KQs 2 and 3.

Literature Search and Retrieval Process

We used three strategies to identify studies providing evidence related to the key questions: systematic searches of electronic databases using both search terms and author names, hand searches of reference lists of included articles, and consultation with the TEAG. First, we generated a list of Medical Subject Heading (MeSH) search terms for each key question in the feasibility study. We used these terms to search standard electronic databases: MEDLINE, Cumulative Index to Nursing & Allied Health Literature (CINAHL), PsycINFO, Sociofile, and the Cochrane Library.

We conducted the electronic database searches twice. We initially did them in April 2003 for the feasibility study.²⁴ That study included three additional key questions, including questions on natural history, risk factors, and treatment effectiveness for perinatal depression. We found relevant articles for the three key questions of the current study under the natural history and treatment effectiveness searches. We therefore conducted these and the incidence or prevalence and mass screening searches again in March 2004 to capture any studies published and posted in the interim.

The subject headings used and the total yield from each source are shown in Table 3 by key question. We found a total of 837 unduplicated citations in the electronic searches and picked up an additional 9 citations through the hand searches and discussion with the TEAG, for a total of 846 citations. We also searched the Cochrane Collaboration database for prior systematic reviews using the keywords “perinatal” and “depression.” This search yielded 38 reviews.

Three senior reviewers with clinical expertise in perinatal depression reviewed the abstracts of articles identified during the literature search. Two clinicians evaluated each abstract against the inclusion criteria and resolved any differences in inclusion by consensus. In several instances, the abstracts did not provide enough information to make an inclusion decision; we pulled full articles to review for those studies. Of the 846 articles identified, 729 did not meet the inclusion criteria for any of the key questions and were therefore excluded, 8 studies were pulled for background only, and the remaining 109 articles were pulled for a full review.

Among the 109 studies pulled for full review, 50 did not meet our inclusion/exclusion criteria for any of the three key questions. The most common reason for exclusion was the absence of a gold standard (i.e., either a clinical assessment or structured clinical interview) for assessing depression, which eliminated 26 studies. Ten of the studies pulled for the evaluation of the properties of screening instruments were excluded because they did not report sensitivity and specificity or data from which these statistics could be computed. Other reasons for exclusion were depression assessed after the first year postpartum, no depression outcome measure, a retrospective study design, and restriction of the study sample to specific population subgroups (e.g., teens, patients of psychiatric hospitals). We based the last exclusion on two lines of reasoning. First, although groups such as adolescents are a key subgroup, our charge was to ensure that our results were generalizable to the broader US population. Second, these specific subpopulations are different enough from the remainder of the population that they warrant separate consideration. We excluded only one study because it was limited to an adolescent population.

We included the remaining 59 studies in our review, and some met the inclusion criteria for more than one key question. We abstracted 30 studies for KQ 1, 23 for KQ 2, and 15 for KQ3. We provide a graphical presentation of the disposition of the citations in Figure 2.

Data Collection and Assessment

The data collection process involved abstracting relevant information from the eligible articles and generating evidence tables that present the key details of the study design and the major findings from the articles. A trained member of the study team read and abstracted each article; a second member checked the table entries for accuracy against the original article.

Appendix C contains the final evidence tables in their entirety. They provide the study design details and major findings. The dimensions of each study design abstracted vary by key question, but they contain some common elements, such as author, year of publication, study location (e.g., country, state), population description, and sample size. We also collected information on the clinical interview instrument and diagnostic criteria used to diagnose depression and the age and racial and ethnic distribution of study subjects in each study.

The study results are recorded in the form reported in the article. However, for assessing consistency of results across the studies and for combining study results in a meta-analysis (see below), we also transformed the study results when necessary into consistent outcome measures using the appropriate statistical formulas. These computed data elements are shown in bold in the evidence tables (Appendix C).

We conducted data abstraction electronically in a word processing program and in such a way that study identifiers and results were easily transferred from the forms to electronic files for input into programs for meta-analysis.

Meta-analysis

We conducted a meta-analysis of the different prevalence and incidence estimates from studies abstracted for KQ 1 to arrive at single prevalence and incidence estimates for particular periods and points in time. We elaborate on these methods in Chapter 3. We also conducted meta-analyses of the different estimates of the receiver operating characteristics (ROC) curves for screening instruments evaluated for KQ 2, as described in Chapter 4. Because of the diversity of screening instruments and prevention interventions in the studies found for KQ 3, we did not conduct a meta-analysis for this key question.

Evaluation of the Quality and Strength of the Evidence

Appendix B presents the quality rating form used for articles considered for KQ 1. The total possible score for these studies was 20 for studies without a comparison group and 25 for studies with a comparison group. For both types of studies, we considered those articles with a score of 16 or greater to be good, those with scores between 10 and 15 to be fair, and those with scores of 9 and below to be poor. The domains and maximum points possible for each domain are as follows:

• Reporting (domain score of 9): Eight items covering study aims, measures, patient populations, findings, and statistical presentation; each scored yes or no (1 or 0), except for an item concerning principal confounders that was scored yes, partially, or no (2, 1, or 0, respectively).

• External validity (domain score of 3): Three items relating to the representativeness of populations from which people were recruited and of settings and clinicians that treat such patients; each scored yes, no, or unable to determine (1, 0, or 0, respectively).

• Internal validity—bias (domain score of 3): Three items relating to issues such as validation of the depression diagnosis through clinical interview, follow-up periods, and appropriate statistical tests; each scored yes, no, or unable to determine (1, 0, or 0, respectively).

• Internal validity—confounding (domain score of 2 for studies without a comparison group and 4 for studies with a comparison group): Two items relating to sources of comparison groups, one for the adequacy of adjustments for confounding, and one for the handling of loss to follow-up; each scored yes, no, or unable to determine (1, 0, or 0, respectively).

• Precision (domain score of 3 for studies without a comparison group and 6 with a comparison group): One item relating to the number of pregnant or postpartum women assessed for depression, with scores of 3 for more than 1,000 women, 2 for 250 to 1,000 women, 1 for 30 to 250 women, and 0 for fewer than 30 women. For studies with a comparison group, a second item gave points based on the size of the smallest comparison group: a score of 3 for more than 2,000 women, 2 for 1,000 to 2,000 women, 1 for 500 to 1,000 women, and 0 for fewer than 500 women.

Best Estimates of Prevalence and Incidence

We abstracted all estimates of the prevalence and incidence of major and minor depression together and major depression alone. We distinguished prevalence estimates by whether they were point or period estimates and both prevalence and incidence estimates by the time period covered. Time periods for point prevalence estimates were defined as trimesters during pregnancy and months during the first postpartum year. Estimates taken at different weeks of gestation but within the same trimester of pregnancy were considered as being conducted in the same time period (e.g., estimates taken week 14 through week 27 of gestation were considered the second trimester). Similarly, estimates taken at different weeks postpartum but within the same month postpartum were considered within the same time period (e.g., estimates taken during week 1 through week 4 postpartum would be considered month 1; week 5 through week 9 postpartum, month 2). Where we found two or more estimates within the same trimester of pregnancy or month postpartum, we used meta-analysis to obtain a combined estimate for that trimester or month. We then graphed the resulting estimates to determine how they changed throughout pregnancy and the first postpartum year.

We conducted similar procedures for period prevalence and incidence estimates. The relevant time periods were either single trimesters and months or multiple trimesters and months. Because we found fewer estimates of these types, however, we graphed period prevalence and incidence estimates for only the first 3 months postpartum.

We combined all estimates with the same diagnosis, estimate type, and time period using the meta command in Stata. This procedure uses the inverse-variance weighting method to calculate random effects summary estimates. It also produces (1) Q tests of the homogeneity of the estimates and (2) forest plots of the individual study and combined estimates and their CIs. To satisfy the normalcy assumptions of these methods, we first transformed the prevalence estimates into log odds estimates.

We reviewed the forest plots of the studies in each summary estimate to determine whether we could identify the source of any heterogeneity between studies. We then reran the meta-analyses excluding studies that were obvious outliers and for which we could identify the source of the bias. The new summary estimates are considered our best estimates of the prevalence and incidence of perinatal depression for the general female population in the United States.

Analysis of Confounders

To analyze associations between the prevalence of depression and study characteristics, we conducted cumulative meta-analysis and a series of meta-regressions on the point prevalence estimates for major and minor depression together and major depression alone. In the cumulative meta-analysis, we added studies one by one, based on publication year, to produce a new combined estimate with the cumulative evidence for each year. This procedure allowed us to see trends in the estimate over time. We conducted cumulative meta-analysis on the 2-month point prevalence estimates using the metacum command in Stata.

We then used the Stata metareg command to estimate several different meta-regression models. For all models, we used the log odds as the dependent variable and included the time point at which depression was assessed and indicators for whether the study enrolled only low-risk women and only women of low socioeconomic status (SES) as explanatory variables. The time point was represented by a categorical variable with included values for the first, second, and third trimesters and the first, second, and third months postpartum. The reference category for this variable was 4 to 12 months postpartum.

We estimated seven different models. Each had a different set of additional explanatory variables:

1. No additional explanatory variables;

2. Publication year;

3. Study country, categorized as the United States (the reference category), other western countries, and Asian countries;

4. Interview type, categorized as the Schedule for Affective Disorders and Schizophrenia (SADS) (the reference category), the Structured Clinical Interview for DSM Diagnoses (SCID), and other interview types;

5. Diagnostic criteria, categorized as Research Diagnostic Criteria (RDC) (the reference category), DSM III-R, DSM IV, and other criteria;

6. Whether depression was assessed only for women who were designated as at risk based on a screening instrument; and

7. The quality rating score.

Comparison with Other Women

To answer the second part of KQ 1, whether the prevalence and incidence of depression is higher during pregnancy and the first year postpartum compared to nonchildbearing periods, we computed odds ratios for studies with a comparison group of women of similar age during nonchildbearing times. Because the types and timing of prevalence and incidence estimates did not overlap in these studies, except for one time point, we did not conduct meta-analyses of the log odds ratios.

Study Characteristics

The major characteristics of the 30 studies are summarized in Table 4 by study type and alphabetically within type. The 25 prospective studies without a comparison group are shown first,^{19, 23, 29–51} followed by the three prospective studies with a comparison group,^{20, 52, 53} and, finally, the two retrospective studies.^{54, 55} Important study characteristics include the precision or size of the studies, the representativeness of the study populations, the methods and timing used to assess the mother's mood, and the quality rating of study design. Each of these characteristics is addressed in turn below.

Precision. The study sample sizes ranged from 54 to 4,964 women; the median sample size was 202 women. Although all the studies had an adequate sample size to provide a prevalence estimate of 10 percent with 80 percent power at a 95% confidence level, most were not large enough to allow subgroup analyses.

The three studies with comparison groups included 313, 232, and 179 women in the comparison groups.^{20, 52, 53} These sample sizes are inadequate to detect a difference as large as 5 percentage points in incidence or prevalence at 80 percent power and a 95% confidence level; a minimum sample size of more than 500 per group is required.

Representativeness. Included studies represented a wide array of developed nations, but the study subjects were not a good representation of the racial and ethnic mix of the US population (Table 4). Seven of the prospective studies were located in England; six in the United States; three in Japan; two each in Canada, Australia, and Hong Kong; and one each in the Netherlands, Norway, Portugal, and Scotland. The two retrospective studies investigated depression diagnoses documented in the Olmsted County, Minnesota, population-based databases during two different 12-month periods—the 12 months following deliveries occurring in 1993 and the 12 months following deliveries among women visiting the Olmsted County and Mayo clinics in 1997 and 1998.^{54, 55}

None of the studies was designed to compare rates of depression among women of different racial and ethnic groups. Sixteen of the 30 studies did not even specify the racial and ethnic composition of the study subjects. Among the other 14 studies, 5 included only white non-Hispanic women;^{19, 32, 38, 44, 47} two studies included only Chinese women;^{42, 43} and two others included only Japanese women.^{50, 51} The remaining five studies noted a racially mixed population, but all had a predominant race or ethnicity. In four of these studies, 73 percent to 90 percent of the women were white non-Hispanic,^{29, 36, 37, 48} and, in the fourth, 75 percent were Hispanic.²³

Depression Assessment. Our inclusion criteria required that the study use a clinical interview or assessment to validate depression diagnoses. The prospective studies differed in who received a clinical interview, the interview instrument, the diagnostic criteria used to identify a depressive episode from the interview responses, and when the interview was conducted. These differences can affect the resulting estimates of prevalence and incidence.

Eighteen of the 28 prospective studies conducted a clinical interview on all study women. The remaining 10 studies first had study subjects complete a self-report depression screening instrument, such as the EPDS, the BDI, or the General Health Questionnaire (GHQ), a broader measure designed to assess the presence of psychiatric distress related to general medical illness. These studies then administered a clinical interview to women scoring over a predetermined cutoff on the screening instrument. Seven of the 10 studies also interviewed a small sample (e.g., 10 percent) of the women scoring below the cutoff, but few of the studies used the results from these interviews to adjust the final prevalence estimates for false negatives. Most studies used low enough cutoff scores that the resulting downward bias in the estimates was minimal. The one exception was the Lucas et al. study, which used a high cutoff of 21 on the BDI and did not interview any women scoring below the cutoff or adjust the resulting prevalence rates in any way, thereby introducing a significant, uncorrected downward bias.⁴⁴

Different interview instruments have been developed for identifying depression diagnoses. These different instruments use different criteria for diagnosing depression. Little is known about how these different instruments and diagnostic criteria affect the prevalence and incidence estimates.

The most frequently used instrument among our studies was the SADS. This semistructured interview is widely used in clinical research and has well-established reliability and validity.⁵⁶ O'Hara et al. adapted the SADS for use with pregnant and postpartum women.¹⁹ Twelve of the 28 prospective studies used this interview instrument.

Five of the studies used the section of the SCID that covers depressive disorders.^{57, 58} The SCID allows the interviewer to use additional questions to inquire about idioms of distress that are specific to the local context. Lee et al. used this feature of the SCID to incorporate questions about traditional Chinese customs used during the puerperium that may affect the clinical presentation of postpartum depression.^{42, 43} They also modified the instrument to identify cases of minor depression.

Five other studies used the Standardized Psychiatric Interview (SPI) of Goldberg et al.⁵⁹ The SPI includes 10 five-point scales that rate the severity of neurotic symptoms in the 7 days preceding the interview and a rating of 12 abnormalities observed during the interview.

Other interview instruments used include the Composite International Diagnostic Interview (CIDI-A),⁶⁰ the Diagnostic Interview Schedule,⁶¹ the Mini International Neuropsychiatric Interview (MINI-V4.4),⁶² the Present State Examination (PSE),⁶³ and the Montgomery and Asberg Depression Rating Scale (MADRS).⁶⁴

All studies that used the SADS and three of the studies that used the SPI based depression diagnosis on the RDC.⁶⁵ To be diagnosed with depression, women had to have reported that they felt sad, tearful, or blue for at least 2 weeks. The 2-week criterion serves to rule out women who were experiencing postpartum blues only. In addition, for a diagnosis of major depression, the women had to have reported at least three or four additional symptoms, such as sleeping disturbances, loss of appetite, fatigue, loss of interest in usual activities or the ability to concentrate, psychomotor retardation, and suicidal thoughts. Women with only two to four of these symptoms were classified as having minor depression. The RDC attempts to differentiate between normal physical effects of pregnancy and the puerperium and actual symptoms of depression.

Five of the prospective studies based diagnoses of depression on DSM-III-R criteria and four based diagnoses on DSM-IV criteria. A diagnosis of major depression based on the DSM-III-R criteria is comparable with the RDC for definite major depression.⁶⁶ However, the RDC includes criteria for minor depression, which, as mentioned above, received its first DSM mention in the fourth edition (DSM-IV)²⁸ as a proposed category for further study. Other criteria used for diagnoses of depression included Pitt's criteria;⁶⁷ the International Classification of Diseases, Ninth Edition (ICD-9); and PSE Index of Definition (PSE ID) and Catego Class.⁶³

Finally, because the prevalence and incidence of depression may not be constant throughout pregnancy and the first postpartum year, the timing of the clinical interview is also very important. Most of the studies we reviewed administered the clinical interview at multiple points in time throughout pregnancy and the first postpartum year, allowing for multiple estimates of prevalence and incidence. The 28 prospective studies provided 80 estimates of the prevalence and incidence of major and minor depression and 70 estimates of the prevalence and incidence of major depression alone. Clinical assessments of depression were taken at different points in time throughout pregnancy and the first postpartum year. Graphical presentations of the timing of each of the estimates by diagnosis and estimate type are shown in Figures 3 through 8 as follows:

• Point prevalence estimates

  42 for major and minor depression (Figure 3)
  

  

  Figure 4. Estimates of point prevalence of major depression (more...)

  

     Figure 4. Estimates of point prevalence of major depression by time of assessment

  46 for major depression alone (Figure 4)

• Period prevalence estimates

  

  

  Figure 5. Estimates of period prevalence of major and (more...)

  

     Figure 5. Estimates of period prevalence of major and minor depression by time period of assessment

  17 for major and minor depression (Figure 5)
  

  

  Figure 6. Estimates of period prevalence of major depression (more...)

  

     Figure 6. Estimates of period prevalence of major depression by time period of assessment

  12 for major depression alone (Figure 6)

• Incidence estimates

  

  

  Figure 7. Estimates of incidence of major and minor (more...)

  

     Figure 7. Estimates of incidence of major and minor depression by time period of assessment

  21 for major and minor depression (Figure 7)
  12 for major depression alone (Figure 8).

The numbers in parentheses in these figures are the number of estimates found in the 28 studies for that point or period of time.

For the two retrospective studies, the investigators had abstracted information on symptoms and diagnoses of depression from medical records beginning at delivery and extending to 1 year postpartum. Both studies provided only estimates of 1-year period prevalence. Bryan et al.⁵⁴ provided estimates of the prevalence for both major and minor depression and major depression alone, whereas Georgiopoulos et al.⁵⁵ provided only the prevalence of major depression alone. Bryan et al. identified a woman as having postpartum depression if any of the following criteria were found in her medical records:⁵⁴ (1) two notations at least 2 weeks apart of symptoms of depression; (2) a documented diagnosis of depression by a physician, psychologist, nurse practitioner, or midwife; (3) a new prescription for an antidepressant with no evidence that it was for chronic pain or for any indication other than depression; and (4) documentation of symptoms sufficient to meet the DSM-IV criteria of major depression. Georgiopoulos et al.⁵⁵ based their prevalence estimate solely on a documented diagnosis of postpartum depression.

Quality Rating. We show the results of the quality rating of the included articles in Table 5 by study type. Studies were rated on reporting completeness, external validity, internal validity, and precision. The average overall quality rating score, out of a possible 20 points, was 11.1 for prospective studies without comparison groups and 12.0 for the retrospective studies. For prospective studies with comparison groups, which had 25 possible points, the average overall quality score was 11.7. Thus, we would rate the overall body of evidence for the prevalence and incidence of perinatal depression as fair at best.

In general, studies ranked good on reporting. The 28 prospective studies, both those with and those without comparison groups, scored an average of 6.0 out of 9 possible points for reporting. The retrospective studies scored 5.0 on average. Most studies clearly described the purpose of the study, the method of assessing depression, the characteristics of the patients in the study, and the study findings. Most studies also provided adequate information to estimate the random variability in the estimates and reported actual probability values for the statistical significance of the main outcomes. Fewer studies provided the distribution of the major principal confounders and described the characteristics of patients lost to follow-up. In particular, studies often did not discuss whether the women had prior depressive episodes or obstetrical complications and frequently did not report the women's socioeconomic status or race and ethnicity. Most studies also did not specifically exclude cases of bipolar disorder or psychosis.

Virtually all prospective studies rated poor on external validity. Prospective studies without a comparison group averaged 0.8 points out of 3 possible points; those with a comparison group averaged 0.3 points. These studies seldom supplied adequate information to determine whether study subjects were representative of the patient population of the facilities from which they were recruited and whether the recruitment facilities were representative of the facilities frequented by the general population in the geographic area. In contrast, the two retrospective studies, which were conducted using the Olmsted County Health Department and Mayo Clinic databases, included the majority of all newly delivered women in the county and therefore scored an average of 2.5 points on external validity.

We separated scores for internal validity into two sets of study design characteristics: those that may bias the prevalence estimates and those that reflect possible confounding factors, which relate to the comparability of the comparison groups and whether losses of patients to follow-up were taken into consideration. The prospective studies scored high on the first measure of internal validity; the studies without a comparison group averaged 2.4 of 3 points and the studies with a comparison group averaged 2.3 points. Virtually all prospective studies assessed the mood of study women within 2 weeks of designated times during pregnancy and postpartum and applied appropriate statistical tests for measuring incidence or prevalence. However, as noted above, 10 studies introduced potential bias by not administering the clinical interview to all study women.

The retrospective studies averaged a lower 1.5 points. Diagnoses were not validated through clinical interview for all women, and Georgiopoulos et al. did not provide adequate information to determine whether they used appropriate statistical techniques to compute the prevalence estimate.⁵⁵

Studies with comparison groups could get 4 possible points for the internal validity confounding score. We awarded 2 additional points if the cases and controls were recruited from the same population and over the same period of time. Only two of the three prospective studies with comparison groups met these criteria. The comparison group in the Cooper et al. study comprised women interviewed by another researcher over a different time period in a different city. Study women were recruited from the appointments diary of the prenatal clinic and the delivery booking diary of the general practitioner unit of the John Radcliffe Hospital in Oxford; the comparison group was derived from a community sample of Edinburgh women of similar age but who were not pregnant and had not delivered in the previous 12 months.⁵²

By contrast, in the Cox et al. study, both cases and controls resided in the North Staffordshire Health District.²⁰ Cases were recruited from the prenatal clinic lists of the North Staffordshire Maternity Hospital; controls matching cases on marital status, number of children, and age (within 5 years) were recruited from four general practice registers. The O'Hara et al. study recruited cases from a public obstetrics and gynecology clinic and two private practices at the University of Iowa Hospitals and Clinics.⁵³ Each subject was asked to provide the names of five acquaintances similar in age, marital status, work status, and number of children. The acquaintance most similar to the subject was selected as a control.

We also gave points for the internal validity confounding measure if the investigators made adjustments or discussed the possible direction and magnitude of any biases from confounding factors and if they took the loss of patients to follow-up into account in their prevalence or incidence estimate. A minority of studies met either of these criteria, resulting in an average score on this measure of 0.4 out of 4 possible points for prospective studies without comparison groups, 1.7 for prospective studies with comparison groups, and 1.0 for the retrospective studies.

Finally, we gave 17 studies with 30 to 250 pregnant or recently delivered women a precision score of 1, 10 studies with 250 to 1,000 women a precision score of 2, and 3 studies with more than 1,000 women a precision score of 3. None of the studies had a comparison group of at least 500 women; therefore, we awarded no additional points for precision. The average precision score was 1.5 for prospective studies without comparison groups, 1.3 for prospective studies with comparison groups, and 2.0 for the retrospective studies.

In summary, the included studies generally were rated as good on reporting and internal validity for bias, poor on external validity and internal validity for confounding, and only fair on precision.

Results from Prospective Studies

Our original estimates of point prevalence, period prevalence, and incidence rates computed from all of the estimates in the included studies are shown by time period in Table 6 for major and minor depression together and in Table 7 for major depression alone. For time periods for which we had more than one estimate, we show the combined estimate from the meta-analysis and the P-value for the Q test of homogeneity. This statistic tests the null hypothesis that the estimates come from the same distribution—that is, whether or not the studies appear statistically to measure the same phenomenon. A P-value < 0.05 suggests that they do not.

The results of these tests indicate that considerable heterogeneity exists across the studies included in many of the pooled estimates, particularly among the point prevalence estimates. Therefore, we first discuss the results of our analysis of outliers and then discuss the results of the revised meta-analyses. We finish this section by presenting the findings from the studies with comparison groups of nonchildbearing women.

Outliers. In a review of the forest plots of the meta-analyses of the prevalence and incidence estimates, we found estimates from several studies consistently to be outliers for all time periods at which they assessed the women's mood. Two studies included only women at low risk of depression.^{29, 32} Affonso et al.²⁹ included only primigravida women with a viable fetus who were married or living with the infant's father and who had no recent depression episodes. Campbell and Cohn³² included only primiparous women who delivered full-term, single infants without major complications and who were Caucasian, married, over 17 years of age, and had at least a high school education. The estimates from these studies were consistently lower than the estimates from the other studies.

Two additional studies included only women of lower socioeconomic status.^{23, 37} These studies generally provided higher estimates of depression prevalence and incidence than the other studies.

The Lucas et al. study included only women who screened positive for depression on the BDI.⁴⁴ The cutoff used (> 21) was so high that the bias from false negatives produced consistently lower prevalence estimates compared to the other studies.

Finally, because of its size, the Cooper et al. study dominated the combined 2-month point prevalence estimate for major depression alone.³³ However, the 15.3 percent estimated point prevalence from this study is outside the 95% CI of the combined estimate for major and minor depression. The purpose of the study was not to produce a prevalence estimate but rather to develop a predictive index for postpartum depression. Furthermore, many of the clinical interviews were conducted by telephone and the article did not state whether a clinician or lay person conducted the interview. Thus, the procedures for assessing depression in this study may have introduced significant bias in the prevalence estimate.

We reran the meta-analyses excluding these six studies to produce “best estimates” of the prevalence of perinatal depression. The final best estimates are shown in Table 8 for major and minor depression together and in Table 9 for major depression alone.

Point Prevalence. We show the best estimates for the point prevalence of major and minor depression graphically in Figure 9. This figure graphs the mean estimate and corresponding 95% CI for each trimester of pregnancy and month postpartum in the first year following delivery. The number of studies that we used to compute the estimate and the P-value for the Q test of homogeneity among the studies are shown above each estimate. For points in time for which no numbers are shown, we found only a single estimate.

As shown in Figure 9, prevalence in the first trimester is 11.0 percent but drops to 8.5 percent in the second and third trimesters. Following delivery, prevalence of major and minor depression begins to rise and is highest in the third month at 12.9 percent. In the fourth through seventh month postpartum, prevalence declines slightly, staying in the range of 9.9 percent to 10.6 percent, after which it declines to 6.5 percent. However, all of these estimates have broad 95% CIs, suggesting that a considerable amount of uncertainty remains in the precise values of the estimates and that the differences in the estimates over time may be attributed to chance or to uncontrolled factors. We cannot say with certainty from these data that perinatal depression is higher at any particular trimester during pregnancy or month in the first postpartum year.

The best estimates for the point prevalence of major depression alone (Figure 10) are more variable and no more precise than those for major and minor depression together. Episodes of major depression comprise less than half of all cases of depression in the perinatal period, except during three seemingly peak times. As shown in Figure 10, the prevalence of major depression is highest in the second trimester (4.9 percent), 2 months postpartum (5.7 percent), and 6 months postpartum (5.6 percent). However, the 95% CIs for these estimates are very wide and overlap those at other times. Thus, we cannot say with certainty that major depression peaks at these points in time. Furthermore, the tests for homogeneity show that considerable heterogeneity persists among studies in the combined estimates.

Period Prevalence. The many fewer estimates of period prevalence allow us to say little about the period prevalence for major and minor depression. As shown in Tables 8 and 9, the best estimates suggest that as many as 18.4 percent of pregnant women are depressed during their pregnancy (i.e., from conception to birth), with as many as 12.7 percent having an episode of major depression. Furthermore, as many as 19.2 percent of new mothers may have major or minor depression in the first 3 months following delivery (Table 8), with as many as 7.1 percent having major depression (Table 9).

However, all estimates have wide 95% CIs. Moreover, as shown in Figure 11, the best estimates of different durations are not consistent over longer periods of time. We would expect the period prevalence for major and minor depression from birth to 2 months postpartum to be higher than the period prevalence from birth to 1 month postpartum and the period prevalence for major depression from birth to 3 months postpartum to be higher than the period prevalence from birth to 2 months postpartum, but we do not see these patterns.

Incidence. We also found few estimates of the incidence of depression—the percentage of women with depressive episodes that begin during pregnancy or the first year postpartum. The studies we found suggest that as many as 14.5 percent of pregnant women have a new episode of major or minor depression during pregnancy, and 14.5 percent have a new episode during the first 3 months postpartum (Table 8). Considering major depression alone, 7.5 percent of women may have a new episode during pregnancy and 6.5 percent during the first 3 months after delivery (Table 9). Figure 12 shows that, although the incidence estimates for major and minor depression in the first 3 months postpartum follow the expected upward trend, the incidence estimates of major depression alone do not.

Analysis of Confounders

The results of the cumulative meta-analysis are graphed in Figure 13. They clearly show the impact of the more precise diagnostic criteria in more recent studies. For both major and minor depression together (left panel) and major depression alone (right panel), the cumulative combined 2-month point prevalence estimate drifts downward as more recent studies are added. Thus, the more precise criteria in the more recent studies identify fewer women as depressed. However, we did not find a statistically significant effect of the year of publication in our meta-regression.

We provide the estimated coefficients, their standard errors, and P-values from the different meta-regression models in Table 10 for major and minor depression together and in Table 11 for major depression alone. We have bolded coefficients significantly different from zero at the α = 0.05 level. The results for major and minor depression show large, positive coefficients for the 2-month postpartum and 3-month postpartum time periods compared to the 4- to 12-month postpartum period (Table 10). These findings suggest a higher prevalence of depression during these 2 months. However, the coefficients are both significant only in the equation that includes diagnostic criteria (Model 5, Table 10). The 2-month postpartum time period is also large and positive for major depression alone, but significant only in the equations including diagnostic criteria (Model 5, Table 11) and whether only women who screened positive for depression were interviewed (Model 6, Table 11). None of the coefficients for the trimesters of pregnancy is statistically significant, suggesting that the prevalence of depression during pregnancy is similar to that during the last three quarters of the first postpartum year.

The low-risk indicator has a statistically significant, negative coefficient for both sets of diagnoses, as expected (Tables 10 and 11). Low SES has a statistically significant, positive coefficient only for major and minor depression together (Table 10). The latter result suggests that the prevalence of major depression is similar among SES groups but that minor depression may be more prevalent among lower SES groups.

The meta-regression results also suggest that prevalence can vary by the clinical instrument and diagnostic criteria used to assess depression. The SCID instrument defined fewer women with major and minor depression than did the SADS interview (Table 10), but the coefficient for this variable is not significant in the equation for major depression alone (Table 11), suggesting that the difference is in the identification of women with minor depression. DSM-IV and other diagnostic criteria (e.g., Pitt, ICD-9) defined fewer women as depressed than did the RDC in the equation for major and minor depression (Table 10), and DSM-III-R and other criteria defined significantly more women as suffering from major depression than did the RDC (Table 11).

Finally, studies with higher quality rating scores have lower log odds, suggesting lower prevalence of depression, but the coefficient of this variable is only marginally significant (P = 0.072) in the equation for major depression alone (Model 7, Table 11) and is not significant in the equation for major and minor depression together (Model 7, Table 10). No statistically significant results were found for study country or whether the study interviewed only women who screened positive for depression, although the signs of the coefficients for these variables are as predicted.

Comparison with Other Women

The three prospective studies with comparison groups of women of similar age in nonchildbearing periods had adequate data to compute 13 estimates of the relative prevalence and incidence of depression. The estimated odds ratios and corresponding 95% CIs are shown in Table 12.

None of the odds ratios for prevalence, which covered different time periods in the first postpartum year, indicated a statistically significant difference. In addition, the National Comorbidity Survey fielded in 1990-1992 found a 5.9 percent current 30-day prevalence of major depressive episodes among women ages 15 to 54 years using the CIDI instrument and DSM-III-R criteria.⁶⁸ This finding is approximately equivalent to our best 1-month postpartum period prevalence of 5.7 percent and to the point prevalence at 2 months and 6 months postpartum (5.7 percent and 5.6 percent, respectively) shown in Table 9. Thus, the evidence indicates no difference in the prevalence of postpartum depression among pregnant or newly delivered women and women at other times in their childbearing years.

The single estimate of the incidence of major and minor depression (Table 12) shows a significant 3-fold difference in the odds of having a new episode of major or minor depression among women in their first 5 weeks postpartum compared to women who were not pregnant and had not recently given birth.²⁰ However, by 6 months postpartum, the difference in the incidence had narrowed and was no longer significant (Table 12).

Results from Retrospective Studies

The prevalence estimates from the retrospective studies measure something different than the prospective studies. In the prospective studies, all study women recruited from prenatal clinics or maternity wards were screened and interviewed for depression. Thus, all (or nearly all) women with depression in the populations so defined are identified. In the retrospective studies, only those women with depression detected through the course of medical contacts during the year were identified.

The two retrospective studies that met our inclusion criteria provided estimates of the period prevalence of major depression in the first postpartum year. The first study conducted in 1993 found that 1.2 percent of postpartum women in Olmsted County, Minnesota, had a major depressive episode during their first postpartum year and that 2.5 percent had a major or minor episode.⁵⁴ These rates are significantly below the 3-month period prevalence of 7.1 percent for major depression alone and the 19.2 percent for major and minor depression reported in Tables 8 and 9.

In 1997-1998, universal screening for depression with the EPDS at the 6-week postpartum visit was implemented in Olmsted County. As a result, the prevalence of a diagnosis of major depression among postpartum women rose to 10.7 percent, suggesting that the screening score posted in medical charts led clinicians to become more aware of their patients' mental state.⁵⁵

Inclusion and Exclusion Criteria

Studies to be retained had to report directly or to provide data allowing us to calculate our primary outcomes of interests—sensitivity and specificity. We required that the screening instrument be compared to a reference standard for a diagnosis of depression. Reference standards could be one of two types. The first includes a clinical assessment by a mental health professional based on criteria from the Diagnostic and Statistical Manual of Mental Disorders (DSM), the Research Diagnostic Criteria (RDC), the Bedford College Checklist,⁶⁹ or the International Classification of Diseases (ICD). The second involves a research-based diagnosis obtained by structured or semistructured clinical interview, such as the Structured Clinical Interview for Depression (SCID), the Diagnostic Interview Schedule (DIS), the Schedule for Affective Disorders and Schizophrenia (SADS), or Goldberg's Standardized Psychiatric Interview (SPI); each of these confirms a diagnosis based on one of the above systems of criteria.

Depressive illness can be either a major depressive disorder or a minor depression. The latter is understood to be an impairing, episodic depression with clear symptoms exceeding a normal state but without severity reaching the diagnostic criteria for major depressive disorder. For this chapter, we are concerned with the ability of screening tools to detect either major depression or minor depression in a given individual (because an individual can have only one or the other of these diagnoses), so the terminology intentionally differs from that used in Chapter 3.

We excluded studies that included patients with a known current depressive illness (for whom a screen would not provide new information). Furthermore, we excluded studies on women with bipolar disorder or a primary psychotic disorder and studies in which women with diagnosed depression could not be distinguished from women with maternity blues, a transient, subthreshold cluster of depressive symptoms commonly described in up to 50 percent of postpartum women.

Data Analysis

Our main outcomes of interest were sensitivity and specificity of the screening approaches or instruments as described in the selected articles. When calculating outcomes ourselves or doing other analyses, we used Stata, version 8. For each reported instrument and associated cutoff, we calculated sensitivity and specificity from the published data. We constructed 95% confidence intervals (CIs) using exact methods. For instruments with three or more outcome values reported, we created plots of the sensitivity or specificity with associated 95% CIs to provide a graphic description of the degree of consistency of results. In addition, where possible we estimated pooled sensitivity and specificity values using meta-analytic methods for fixed effects. We evaluated heterogeneity using the Q statistic test for homogeneity. In several circumstances, pooled estimates were impossible to calculate because of perfect estimates of sensitivity (i.e., 100 percent) with associated variance estimates equal to 0.

Evaluation of Quality and Strength of Evidence

We developed a quality rating form for these articles on screening accuracy from criteria identified by the Cochrane Methods Working Group on Systematic Review of Screening and Diagnostic Tests.²⁵ The quality rating forms, provided in Appendix B, rated reporting, external validity, and internal validity. The senior abstractor completed the quality rating form for each article; another project team member reviewed a sample of the completed forms for accuracy and completeness.

We rated retained studies on three separate categories of quality then summed the individual category scores for a total score. The domains and maximum points possible for each domain are as follows:

• Reporting (domain score of 10): Nine items covering study aims, description of depression assessment, potential confounders described, and instrument procedures described, each scored yes or no (1 or 0), except for an item concerning principal confounders that was scored yes, partially, or no (2, 1, or 0). We considered 0 to 3 as poor, 4 to 7 as fair, and 8 to 10 as good.

• External validity (domain score of 3): Three items relating to representativeness of populations from which people were recruited and of settings and clinicians that treat such patients, each scored yes or no (1 or 0). We considered 0 or 1 as poor, 2 as fair, and 3 as good.

• Internal validity (domain score of 8): Six items relating to both bias and confounding in the use of the screen and reference standard, each scored yes or no (1 or 0), except for an item assessing whether all screens were done independently on each person, all tests done on each person but not independently, or different tests done on different persons and not randomly allocated (2, 1, or 0, respectively). We considered 0 to 2 as poor, 3 to 5 as fair, and 6 to 8 as good.

The maximum total quality score was 21. We considered 0 to 7 as poor, 8 to 14 as fair, and 15 to 21 as good.

Study Characteristics

Our literature review of screening tools for detecting depression during pregnancy and the postpartum period identified no relevant systematic reviews. We did find 23 studies meeting our inclusion criteria. Of these, 10 were studies involving screening instruments in English;^{32, 46, 70–77} 13 involved non-English screening instruments.^{31, 35, 42, 43, 50, 51, 78–84}

The major characteristics of these studies are summarized in Table 13 and detailed in Evidence Table 3 (Appendix C). The studies represent a wide variety of countries. Of the 10 studies using an English-language screening instrument, two were conducted in US populations,^{32, 71} six were performed in the United Kingdom,^{46, 70, 73–76} and one each was conducted in Canada⁷⁷ and Australia.⁷² Of the 13 studies using a non-English screening instrument, four were conducted in Chinese,^{42, 43, 81, 82} three were in Japanese,^{50, 51, 80} and one each was in German (Austria),⁸³ Swedish,⁸⁴ French,⁷⁸ Spanish (Spain),³⁵ Norwegian,³¹ and English/Africaans.⁷⁹ We will focus on the screening instruments used in the English language, given their greater relevance to our population of interest.

Unfortunately, the racial and ethnic mix of the study populations for the studies using English language screening instruments was poorly representative of the US population (our target of interest). Of the 10 studies, only the two studies conducted in the United States reported race and ethnicity.^{32, 71} These populations were overwhelmingly Caucasian; in by far the largest study,³² 100 percent of the 1,007 women enrolled were white, and, in the other, 87 percent of the women were white.⁷¹

When reported, the mean age of women in these studies ranged from approximately 24 to 31 years. Of these 10 studies, only one was conducted during pregnancy.⁴⁶ The remaining nine studies were conducted postpartum between 2 weeks and 6 months after delivery, with most occurring between weeks 8 and 12. Individual study sizes ranged from 103 to 1,007, with an aggregate sample size of 2,800.

Studies might use one or more screening tools; the selected articles evaluated four different screening tools.

Screening Instruments Used

The key features of the four different types of screening instruments used are summarized in Table 14. Eight studies assessed the Edinburgh Postnatal Depression Scale (EPDS); seven of these used the 10-item version,^{46, 71–76} and one used a 13-item version.⁷⁰ The EPDS had been developed specifically for assessing postpartum depression and relies much less than standard depression screens on somatic, or physical, questions. In its most common form, it is a 10-item self-report screening scale for postpartum depression that is specifically aimed at exploring mood symptoms in the postpartum period.⁸⁵ Questions on the EPDS scale are framed within the “past seven days” and the response format is frequency-based. Each item is scored on a 4-point scale (0 to 3); the minimum and maximum scores are 0 and 30, respectively. It takes less than 5 minutes to administer. The responses to the 10 items are summed to obtain a score.

Three studies assessed the Beck Depression Inventory (BDI).^{71, 74, 77} The BDI is a list of 21 symptoms and attitudes that are each rated on intensity.⁸⁶ Versions include the BDI, which uses “last week, including today” as the time frame for symptoms;⁸⁶ the BDI-II, which uses 2 weeks as the time frame for symptoms;⁸⁷ and the BDI-PC, which also has a 2-week time frame.⁸⁸ The versions used most often (BDI or BDI-II) are scored by summing the ratings that respondents give to the 21 items. Although originally designed to be administered by trained interviewers, it is most often self-administered and takes 5 to 10 minutes to complete. This instrument has been used to measure severity of depression in depressed samples and also to assess depression in general population samples. Because of its reliance on somatic symptoms, some experts worry that it may produce higher scores and more false-positive results in pregnant women than in other respondents.

One study used the Postpartum Depression Screening Scale (PDSS).⁷¹ The PDSS is a 35-item Likert-type self-report instrument created specifically for new mothers that can be administered in 5 to 10 minutes. Written at a third-grade reading level, PDSS items are brief and easy to understand. Mothers respond using a 5-point scale ranging from “strongly disagree” to “strongly agree.” The test yields an overall severity score falling into one of three ranges: normal adjustment, significant symptoms of postpartum depression, and positive screen for major postpartum depression. The PDSS also provides scores for seven symptom areas: Sleeping/Eating Disturbances, Anxiety/Insecurity, Emotional Lability, Mental Confusion, Loss of Self, Guilt/Shame, and Suicidal Thoughts.

Another study used the Center for Epidemiological Studies Depression Scale (CES-D).³² The CES-D was designed to measure current level of depressive symptomatology and especially depressive affect.⁸⁹ The 20 items were chosen from five previously used depression scales to represent all major components of depressive symptomatology, and it was designed to apply to a general population. Each item is rated on 4-point scales indicating the degree of its occurrence during the past week. The scales range from “rarely or none of the time” to “most all of the time.” The scale can distinguish between clinical groups and general community groups. It takes approximately 5 to 10 minutes to complete; scoring takes about 1 to 2 minutes. Although it is usually scored continuously, various cutoff scores for clinical depression have reasonable associations with a clinical diagnosis. A cutoff score of 16 or higher has been suggested as a positive screen for depression.⁸⁹

Reference Standards Used

Investigators used a variety of strategies to confirm the diagnosis of depression. Six studies used the RDC⁶⁵ for depressive illness as the reference standard but employed different instruments to identify patients meeting this standard. Three studies used the Standardized Psychiatric Interview,^{46, 73, 76} two studies used a version of the Schedule for Affective Disorders and Schizophrenia,^{32, 77} and one study used the Psychiatric Assessment Schedule (an adaptation of the Present State Examination).⁷⁰

Other reference standards were also employed. Beck and Gable used the Structured Clinical Interview for DSM IV to confirm the diagnosis of depressive illness per DSM IV criteria;⁷¹ Boyce et al. used the Diagnostic Interview Schedule, based on DSM III-R criteria, as the reference standard to confirm depressive illness;⁷² Harris et al. used a clinical assessment of whether a patient's presentation met DSM III criteria for depressive illness;⁷⁴ and Leverton and Elliott used the Present State Examination to identify whether patients met depressive illness criteria by either the Bedford College Criteria or the Catego criteria (based on ICD-8 criteria).⁷⁵

Classifications of Depressive Illness

Investigators classified depressive illness into one of two categories that reflected how perinatal depression is described in the scientific literature: major depression alone or major or minor depression. Patients identified as major depression alone met criteria for an episode of severe depressive illness according to the standardized criteria. In this report, we refer to major depressive episodes as major depression. For major depressive disorders, clearly effective interventions have been identified in clinical trials. Seven studies provided this classification.^{46, 70–74, 76}

The point prevalence for major depression alone was 6 percent in the single prenatal study,⁴⁶ somewhat higher than the 3.1 percent “best estimate” that we discussed in Chapter 3. For the postpartum studies, the point prevalence for the six studies reporting on major depression alone ranged from 6 percent to 15.5 percent;^{70–74, 76} this frequency is somewhat higher than the postpartum results from KQ 1 showing a best estimate prevalence between 1 and 3 months postpartum of 3.8 percent and 4.7 percent, respectively.

The major or minor depression category of depressive illness requires that patients meet diagnostic criteria for either a major depressive episode or a minor depressive episode. Minor depression is an impairing yet less severe constellation of depressive symptoms¹³ for which controlled trials have not consistently indicated that particular interventions are more effective than placebo.^{14, 15} In this report, we refer to this grouping as major or minor depression, or by the more general terms of “depression” or “depressive illness.” Seven studies classified depression in this way.^{32, 46, 71, 73, 75–77}

In the single prenatal screening study, the point prevalence of major or minor depression in the third trimester (14 percent) was greater than our best estimate from KQ 1 for this time period (8.5 percent).⁷³ For the postpartum studies, prevalence rates ranged from 5 percent to 19 percent; these figures are somewhat higher than our best estimate range for point prevalence of 9.7 percent to 12.9 percent in the first 3 months postpartum. Given that this distinction substantially affects screening accuracy at a particular cutoff, we sort the results below by these two case definitions.

Quality Rating

Table 15 documents the details of our grading of individual studies. For reporting completeness, we rated studies as fair; they averaged 6.1 of a possible 10 points. Three studies scored in the good range (8 or above).^{32, 70, 72} For external validity, studies ranged from poor to good, averaging a poor-to-fair rating of 1.6 overall (of a possible 3 points), suggesting that at best they were a fair representation of each individual study's target population. Given that only two studies (Campbell and Cohn with an external validity score of 3³² and Beck and Gable with an external validity score of 1⁷¹) were conducted on US populations, the generalizability of these results to our target population appears limited. For internal validity, studies scored better, ranging from 4 to 8, with an overall average in the good range (7.0). Total scores for the three categories ranged from fair to good, with an overall average of 14.7 of a possible 21 points; of these 10 studies, six scored in the good overall quality range (15 or higher).

Prenatal Screening Results

One English study of 100 subjects used the10-item EPDS to screen women in their third trimester of pregnancy (Table 13).⁴⁶ For major depressive disorder alone (n = 6 depressed patients), sensitivity and specificity point estimates at all cutoff points (12, 13, 14, 15) were quite good, although the sensitivity estimates were imprecise (as demonstrated by the wide CIs). At all cutoff points used, sensitivity was 100 percent, and each cutoff had a wide CI from 0.54 to 1.0 (Figure 14a). Specificity varied among the different cutoff points, with means varying from 0.79 (at a cutoff of 12) up to 0.96 (at a cutoff of 15), and all CIs were more precise, reflecting the larger number of subjects (n = 94) without major depression alone (Figure 14b). At the traditional postpartum cutoff of ≥ 13, sensitivity was 100 percent and specificity was 87 percent.

As a screening instrument for major or minor depression (n = 14 depressed patients), overall test performance was worse. Sensitivity was much lower, ranging from 0.71 (at a cutoff of 11 or greater) to 0.57 (at a cutoff of 14 or greater), and CIs remained wide (Figure 15a). Specificity remained relatively good, varying from 0.72 (cutoff of 11 or greater) to 0.95 (cutoff of 14 or greater), with reasonable precision (Figure 15b). At the same ≥ 13 cutoff, sensitivity was 64 percent and specificity was 90 percent.

In summary, one prenatal screening study is of good quality. However, the inclusion of only six women with major depression substantially limits conclusions about the accuracy of prenatal depression screens. Indeed, the sensitivity results at 100 percent for each cutoff dramatically underscore the small number of depressed patients involved.

Results for major or minor depression from this one study are similarly limited. Only 14 depression cases are involved. Sensitivity and specificity estimates appeared to be lower than those for major depression alone. In particular, sensitivity estimates appeared worse than those for major depression alone, but again CIs are wide.

Postpartum Screening Results

Nine studies provided sensitivity and specificity estimates in the postpartum period (Table 16). These studies used one of four screening instruments—EPDS, BDI, PDSS, and CES-D—at a variety of cutoff points. We review the results separately for each scale for major depression alone and for major or minor depression.

Edinburgh Postnatal Depression Scale. The EPDS was the most common tool reported, involving 1,573 patients from eight studies;^{46, 70–76} 80 patients had major depression alone, and 83 patients had major or minor depression. Murray and Carothers reported test characteristics for both major depression alone and major and minor depression together (not in Table 16), but they did not give information allowing us to calculate CIs for the results;⁷⁶ we address their work separately.

Major Depression Alone. Figures 16a and 16b present the sensitivity and specificity estimates for the five studies reporting on major depression alone (a total of 927 patients).^70–74 We show data according to the version and cutoff point used. The sensitivity graphs show the number of depressed patients; the specificity graphs use the number of nondepressed patients.

For the Ballard et al. study employing the 13-item version (n = 23 depressed women),⁷⁰ we used only the cutoff of ≥ 13. Mean sensitivity was 0.96 and mean specificity was 0.70, with relatively wide CIs for both point estimates.

Results for the remaining four major depression alone studies are listed below the solid line in Figures 16a and 16b.^71–74 All used a cutoff point of 13. Sensitivities in these studies range from 0.75 to 1.0, with very wide CIs.

Specificities ranged from 0.84 to 0.99 and appeared to be more precise than sensitivities, as indicated by the much narrower CIs. Of note, results at this threshold from these individual studies of the 10-item screen indicated that sensitivities were similar to the value reported in the one 13-item screen study, but specificities were higher with the 10-item version.

We attempted to conduct a meta-analysis of the sensitivity results from the four studies using the cutoff point of 13 or greater. The Boyce et al. study⁷² reported a sensitivity point estimate of 1.0, thus we were unable to generate a meaningful standard error; consequently, we could not include this result in the sensitivity meta-analysis. Leaving this study out, our meta-analysis produced a sensitivity point estimate of 0.91 (95% CI, 0.84 to 0.99); the test for heterogeneity was not significant (P= 0.141). We were able to include all four studies in our meta-analysis of specificity, but heterogeneity was significant (P < 0.001), precluding a pooled specificity estimate.

One study assessed a cutoff point of ≥ 12.⁷³ It reported a sensitivity of 0.88 (with a wide CI) and a specificity of 0.76 (with a narrow CI).

Three studies reported a cutoff of ≥ 10, all producing estimates with imprecise sensitivities yet relatively precise specificities.^72–74 Point estimates for sensitivity ranged from 0.88⁷³ to 1.0.^{72, 74} Because two studies reported a perfect sensitivity of 1.0, we could not determine a pooled sensitivity estimate. Specificity ranged from 0.71 to 0.89, but heterogeneity was significant (P = 0.002), precluding a pooled estimate.

Major or Minor Depression. For major or minor depression (1,343 patients), four studies^{71, 73, 75, 76} reported test characteristics for the 10-item EPDS (Table 16). All but Murray and Carothers⁷⁶ allowed a calculation of confidence intervals and are presented in Figures 17a and 17b.

Two studies report a cutoff score of ≥ 13.^{73, 75} Sensitivities were low (0.62⁷³ and 0.44⁷⁵) and imprecise (wide CIs). Specificities were high (0.89 and 0.92, respectively) and quite precise. A meta-analysis at this cutoff produced a pooled sensitivity estimate of 0.54 (95% CI, 0.39 to 0.70) without significant heterogeneity (P = 0.266) and a pooled specificity estimate of 0.91 (95% CI, 0.88 to 0.94) without significant heterogeneity (P = 0.410).

One study reported a cutoff score of 12 or greater.⁷³ Relative to a threshold of 13 or more, this score appeared to improve sensitivity and decrease specificity, with the precision remaining unchanged.

Three studies reported results with a cutoff score of ≥ 10.^{71, 73, 75} Reported sensitivities ranged from 0.59 to 0.81, and specificities ranged from 0.77 to 0.88. Again, sensitivity estimates were quite imprecise, whereas specificity estimates were quite precise. A meta-analysis of these results produced a pooled sensitivity estimate of 0.68 (95% CI, 0.58 to 0.78) without significant heterogeneity (P = 0.140). Specificities could not be pooled because of significant heterogeneity (P = 0.068).

Murray and Carothers reported sensitivities and specificities at various cutoff points as estimated by logistic regression analyses on results from 646 subjects;⁷⁶ they addressed both major depression alone and major or minor depression. Because we could not calculate CIs from their reported results, we do not show them in Table 16; their reported test characteristics are listed in Evidence Table 4 (Appendix C). For major depression alone, their sensitivity and specificity results mirrored the other studies' point estimates in Figures 16a and 16b. For major or minor depression, although specificities were similar to those of other studies, sensitivities were slightly higher than those reported in Figures 17a and 17b. For example, at a cutoff of ≥ 10, sensitivity was reported as 0.89; at a cutoff of ≥ 13, sensitivity was 0.68.

Beck Depression Index. Three studies involving the BDI (417 patients in all) reported lower sensitivity and slightly higher specificity than did the EPDS studies.^{71, 74, 77} Two studies^{71, 74} reported results for major depression only using the BDI-II and BDI, respectively (Figures 18a and 18b). Using a cutoff of ≥ 21, Beck and Gable reported a sensitivity of 0.56 and a specificity of 1.0,⁷¹ Harris et al. reported a sensitivity of 0.32 and a specificity of 0.99 at this cutoff.⁷⁴ Harris et al. also reported test characteristics for cutoff points of ≥ 13 and ≥ 11;⁷⁴ these cutoff scores each produced slightly higher sensitivity and slightly lower specificity than did a cutoff of 21. Relative to the EPDS for major depression alone, Beck and Gable results showed sensitivities that remained substantially lower and specificities that appeared to be slightly higher, although wide CIs preclude a confident comparison.

For major or minor depression, two articles reported BDI test characteristics using different thresholds.^{71, 77} Beck and Gable,⁷¹ using a cutoff of ≥ 15 on the BDI-II, reported a sensitivity of 0.57 and a specificity of 0.97. The BDI study by Whiffen employed a cutoff of ≥ 10 and reported a sensitivity of 0.48 and a specificity of 0.86.⁷⁷

Postpartum Depression Screening Scale. One study of the PDSS (150 patients) reported high sensitivity (0.94) and high specificity (0.98) for major depressive disorder alone at a cutoff of ≥ 80.⁷¹ The investigators also reported lower sensitivity (0.91) and lower specificity (0.72) for major or minor depression using a cutoff of ≥ 60.

Center for Epidemiological Studies - Depression Scale. One study of the CES-D (1,007 patients) used two cutoff points (≥ 21 and ≥ 16).³² It reported low sensitivity (0.60 and 0.43, respectively) and high specificity (0.92 and 0.97, respectively) for major or minor depression.

Summary of Results of Screening Instrument Review

The available evidence for both major depression alone and major or minor depression together is characterized by studies including markedly low numbers of depressed patients, a narrow racial and ethnic mix, varying cutoff points, and varying reference standards. These factors combine to preclude definitive conclusions or recommendations about screening instruments or thresholds.

Screening Instruments. For major depression alone, all screening instruments investigated (EPDS, BDI, PDSS) provided similarly high degrees of specificity at various cutoffs. Because of wide CIs, however, conclusions about sensitivity are more restricted. Heterogeneity among the studies limited our ability to synthesize these results quantitatively. In most instances, we could not obtain a more precise estimate. For an EPDS cutoff of ≥ 13 for patients with major depression alone, sensitivity estimates were combined in a meta-analysis to produce a point estimate of 0.91; however, heterogeneity precluded a meta-analysis for a specificity point estimate.

The EPDS and PDSS (with point estimates ranging from 0.75 to 1.0 at various cutoffs) appeared to be more sensitive than the BDI instruments (0.32 to 0.68 at various cutoffs), but the wide CIs overlapped nearly completely. A recent meta-analysis of prevalence estimates found that, compared with structured clinical interviews, the EPDS produced statistically equivalent prevalence estimates whereas the BDI produced significantly higher estimates.²⁷ Together, these findings suggest that a positive screen with EPDS may be more clinically useful than screens with the other instruments.

For major or minor depression, sensitivity point estimates for each tool at each cutoff were consistently lower than those for major depression alone, although specificities were quite similar to those for major depression alone. We were able to synthesize EPDS results quantitatively at a cutoff of ≥ 13, producing a sensitivity point estimate of 0.54 with a wide CI (95% CI, 0.39 to 0.70) and a specificity estimate of 0.91 with a narrow CI (95% CI, 0.88 to 0.94). At an EPDS cutoff of ≥ 10, we were able to produce a pooled sensitivity estimate of 0.68 (95% CI, 0.58 to 0.78), but heterogeneity precluded a pooled analysis for specificity.

In short, estimates of specificity are relatively precise, but estimates of sensitivity are imprecise. This pattern of results prevents any substantive conclusions about the accuracy of these tools for identifying true positives. This imprecision can be attributed to the consistently low number of patients with a depression diagnosis, a fact reflected by a number of studies reporting 100 percent sensitivity, and it is a major limitation of the currently available data. Because of this imprecision, we cannot meaningfully compare sensitivities of screening instruments.

Cutoff Points. For an individual screening instrument, we cannot make any substantive conclusions about the use of a particular cutoff point. As noted above, the wide CIs for sensitivity prevent one from confidently distinguishing one sensitivity result from another. However, two further guides that bear directly on the choice of a threshold need to be considered before a particular threshold could be suggested.

First, the relative cost, or value, of errors in screening tests (false-negative compared to false-positive results) needs to be clarified. False-negative results (miss true depression) can lead to bad outcomes such as continued morbidity, costs of unnecessary tests, and similar effects. By contrast, false-positive results (identifying depression when it is not there), can lead to unnecessary time, effort, and financial cost for diagnostic workup as well as potential side effects of a treatment that is not indicated. If false-negative and false-positive results are equally bad, then a screening test should try to minimize both equally to identify the most effective cutoff.

If missing depression in a patient is worse than falsely identifying depression in a patient (i.e., a false-negative classification is worse than a false-positive one), then one would want a test that maximizes sensitivity and has the highest negative predictive value. Said another way, the preferred test would be one in which the greatest proportion of those screening negative do not have the disease. By contrast, if falsely identifying a patient as having depression (a false positive) is worse, then one would want a test that maximizes specificity and has the highest positive predictive value. Clinical intuition suggests that missing a diagnosis is worse than making an incorrect diagnosis. We could find no literature addressing the trade-off of false-positive versus false-negative diagnoses in this clinical situation.

A second important guide in choosing a cutoff is the prevalence of a disease in a particular population. Regardless of test characteristics, in populations in which the prevalence of depression is relatively high, the number of false-negative results is higher; in populations in which the prevalence is relatively low, the number of false-positive results is higher. Therefore, the choice of a test and cutoff may differ depending on whether the population has a higher prevalence of depression (e.g., a high-risk postpartum clinic) or a lower prevalence (e.g., a healthy baby clinic). As a result, these three variables—sensitivity and specificity, the predictive value of screening errors (false positives versus false negatives), and the prevalence of the disease—must be clarified before clinicians or researchers can choose a specific test and related cutoff.

The above limitations notwithstanding, the tools we have reviewed above appear to be able to identify depressive illness in pregnant and postpartum women with a degree of accuracy similar to that for depression screen results in other nonpsychiatric settings. Screening results in primary care for a combined major or minor depression group are not available, but the results in primary care settings for major depression alone are similar to those reported for perinatal depression. For example, in a synthesis of depression case-identifying instruments in primary care settings using selection criteria similar to ours, Williams et al. reported a median sensitivity for major depression of 85 percent (range, 50 percent to 97 percent), and a median specificity of 74 percent (range, 51 percent to 98 percent).⁹⁰ This review included both women and men, which might explain the lower measures of accuracy; female gender appears to improve the accuracy of depression screens in primary care settings.⁹¹

Interpretation of Results

The small numbers of relevant articles limits our interpretation of the results. Given that most of the articles address the EPDS, we will use this instrument as an example. Because of the reports of 100 percent sensitivity in the prenatal tests of the EPDS (underscoring the very small number of prenatal depressed patients involved), we consider application of our results only to the postpartum population, and we draw on the prevalence data reported in Chapter 3 for KQ 1. We caution that, given the low numbers of depressed patients in the postpartum studies, the sensitivity estimates are likely to be inaccurate. Also, the majority of postpartum screens were performed 6 to 8 weeks after delivery, so the examples below apply only to that time period.

For major depression alone, the estimated point prevalence for the 6- to 8-week postpartum period is 6.8 percent, although the confidence interval around this estimate is wide. EPDS screens using the most commonly cited cutoff of 13 have a sensitivity of 91 percent and a specificity of approximately 95 percent. To illustrate this scenario, consider using this tool and cutoff for 1,000 patients. This EPDS screen would produce 62 true-positive cases and 6 false-negative cases, and 47 false-positive cases and 885 true-negative cases. The positive predictive value is 57 percent, meaning that the probability that a woman with a positive screen truly has major depressive disorder is slightly more than half. The negative predictive value (i.e., the probability that a woman with a negative screen would not have depressive illness) is 99 percent.

For major or minor depression, the estimated point prevalence from KQ 1 is 11.3 percent. EPDS screens tested for this population most commonly reported a cutoff of 10. This threshold at 6 to 8 weeks postpartum has a sensitivity of 68 percent and a specificity of approximately 80 percent. For 1,000 patients, the screen would produce 77 true-positive cases and 36 false-negative results, and 177 false-positive cases and 710 true-negative cases. The positive predictive value is 30 percent, and the negative predictive value is 95 percent.

Conclusions

Very little is known about the accuracy of depression screening tests in pregnant and postpartum women. The available evidence is limited in several ways. It has a very narrow racial and ethnic mix. Study samples have prevalence rates of depression that are, by design, somewhat higher than our best estimate prevalence rates from KQ 1 (which would produce a higher positive predictive value). Most important, the available data involve small numbers of depressed patients. We could not address the limits of the small numbers of depressed patients using meta-analytic procedures. Case definitions, reference standards, screening tools, and screening thresholds all varied across the studies, and the heterogeneity of study methods constrained our ability to synthesize the data and obtain pooled estimates.

Despite these limitations, the available evidence does indicate that depression screens are feasible to administer in perinatal settings. It also suggests that the estimates of sensitivity and specificity, although limited, appear equivalent to those that have been reported in primary care settings. In particular, specificity is relatively good, suggesting a relatively good positive predictive value.

Future Research

Further studies in this area need to standardize the above parameters we have examined in this chapter (instruments and, in particular, cutoff points), involve a more representative mix of racial and ethnic groups, test the screening tools in populations with a frequency of depression more reflective of the actual prevalence, and include a larger number of depressed patients to clarify the accuracy of depression screening tools and make them more relevant to the population of interest. Given the currently available evidence, we offer six future research recommendations.

First, subsequent studies on the test characteristics of screeners must be designed with sample size estimates that take into account prevalence and that project a reasonable width of sensitivity confidence intervals for the particular illness. For example, studies would need to screen 1,000 women to identify 34 with major depression or 110 with major or minor depression. This sample size might be enough for precise estimates for women with major or minor depression as a group, but it may not be enough for precise estimates for major depression alone.

Second, the sample should represent the target population. Specifically, subsequent studies need to provide a more representative racial and ethnic mix. In addition, studies should incorporate a range of other demographic variables that could influence screening performance, such as socioeconomic status measures, and assess the screening tools in these subpopulations.

Third, as in the Beck and Gable study,⁷¹ subsequent studies should assess and directly compare multiple screening instruments. This design provides a head-to-head comparison that allows researchers and clinicians to understand which screening instruments are more accurate than others in different settings.

Fourth, studies evaluating both the risks and benefits of screening, specifically assessing the relative cost of false-negatives and false-positive results, will provide insights on how to consider target sensitivity and specificity when attempting to maximize cost-effectiveness.

Fifth, subsequent depression screening studies should carefully consider whether to target major depression alone, for which beneficial treatments clearly exist, or the traditional combined category of major or minor depression, a heterogeneous group for which treatment benefit is unclear. Our results suggest that the sensitivity of screening instruments is generally greater for the major depression alone group.

Sixth, the bulk of the screening studies we reviewed were conducted in the first 3 months postpartum. Subsequent studies should examine screening not just in the first 3 months postpartum but also at 6 weeks, 6 months, and 12 months postpartum. If peak prevalence and incidence occur within the first 6 weeks, the obstetrics clinic is a prime place to target resources for such a program. If, however, peaks occur after this time, most postpartum women will have completed follow-up care with an obstetrician, so programs in an obstetrics clinic may be less helpful. In this case, it is possible that programs targeting new mothers in family medicine, internal medicine, or pediatric clinics might be more effective.

Use in Clinical Settings

In the interim, what is a clinician to do? The best available evidence supports the conclusion that screening instruments with reasonable test characteristics appear feasible to use in a perinatal population with a depression prevalence between 5 percent and 10 percent. Given that use of the tools likely carries low risk, and that they all have reasonable specificity (and, thus, a reasonable positive predictive value), the selection of a tool would be guided by an interest in maximizing sensitivity. For the category of major or minor depression, sensitivity estimates were quite similar for all instruments. However, for major depression alone, sensitivity estimates for the EPDS and PDSS appear to be higher than those for the BDI. The standard cutoffs of ≥ 13 for the EPDS and ≥ 81 for the PDSS appear to be reasonable thresholds.

Having an instrument that can accurately identify women at risk of having perinatal depression is an important and necessary link in improving the clinical outcomes of women with perinatal depression: women who may benefit from a depression intervention first need to be recognized. Nonetheless, it remains merely an initial step. A more important question is whether screening pregnant or postpartum women to identify those at risk of having depression, and subsequently providing an intervention, ultimately leads to improved outcome. We address this key question in our next chapter.

Prenatal Studies

Of the four studies examining screening, interventions, and outcomes in the prenatal period,^94–97 three were RCTs^{94, 95, 97} and one was a nonrandomized controlled trial.⁹⁶ All four studies (published between 1995 and 2001) were set in prenatal clinics. Sample sizes for screening ranged from 37 to 209, for a total population of 473 women. The types of screening instruments used to identify patients with depressive symptoms differed among these studies; similarly, the outcome measures differed, although three studies used the Edinburgh Postnatal Depression Scale (EPDS) as one measure. All four studies implemented some type of psychological intervention, generally characterized as group classes or sessions relating to prenatal preparation, skills, and perinatal support. One study was considered fair; the other three were poor.

Brugha et al. screened 209 women with a modified General Health Questionnaire Depression Score (GHQ-D) to study the effect of six weekly prenatal group therapy classes called “Preparing for Parenthood” compared to routine prenatal care.⁹⁵ In this study, which we graded as fair, the program aimed to increase social support and problem-solving skills. Outcome measures that assessed maternal mood and depressive symptoms at 3 months postpartum included the GHQ-D (cutoff ≥ 2), the EPDS (cutoff score of > 11), and a Schedule for Clinical Assessment in Neuropsychiatry (SCAN, related to the International Classification of Diseases [ICD], version 10). Assignment to the intervention group did not significantly improve postpartum depression. On the GHQ-D, 26 percent of the intervention group and 22 percent of the control group scored at or above 2, with an adjusted odds ratio of 1.19 (95% confidence interval [CI], 0.59 to 2.37). On the EPDS, 16 percent of the intervention group and 19 percent of the control group scored above 11 ( an adjusted odds ratio of 0.83 (95% CI, 0.39 to 1.79).

In the earliest study we included in this group (1995), Stamp et al. used a study-specific, modified prenatal questionnaire as a screening instrument and assigned 129 patients to either two prenatal group classes plus one postpartum class (at 6 weeks) or routine care.⁹⁴ Outcome measures were the EPDS at 6 weeks and 6 months postpartum, using a cutoff point of > 12 for major depression and > 9 for major or minor depression. The intervention did not significantly reduce rates of postpartum depression on either measure. For example, at 6 weeks, 13 percent of the intervention group and 17 percent of the control group had EPDS scores greater than 12; at 6 months, the figures were 15 percent and 10 percent, respectively.

Elliott et al. screened 98 women with the Leverton Questionnaire and the depression, anxiety, and somatic subscales of the Crown Crisp Experiential Index.⁹⁶ The authors studied a preventive group of psychosocial intervention versus routine care; they also looked at differences between first-and second-time mothers. The structured group intervention was conducted once per month for 5 months during the prenatal period (starting at 24 weeks) and for 6 months postpartum. Outcome measures included the EPDS and the Present State Examination (PSE), as well as a self-rating questionnaire, at 3 and 12 months postpartum. For first-time mothers, the median EPDS score was significantly lower in the intervention group (Mann-Whitney one-tailed test, P = 0.005); for second-time mothers, the median EPDS did not differ significantly between the two groups. The PSE served as a formal diagnosis of depression, and the investigators reported no significant differences in diagnosis of major depression. When the authors included cases of borderline depression or “minor depression” in the analysis, first-time mothers in the intervention group were significantly less likely to have a diagnosis of depression than controls (19 percent and 39 percent, respectively, Chi-square = 2.64, one-tailed test, P < 0.05). PSE scores did not differ significantly between groups in second-time mothers.

In the only study in this category done in the United States, Zlotnick et al. used the Beck Depression Inventory (BDI) and a Diagnostic and Statistical Manual (DSM-IV) Structured Clinical Interview for Depression (SCID) as a positive screen among women of low socioeconomic status (SES).⁹⁷ They excluded patients who met criteria for a current episode of major depression based on the SCID. A total of 37 patients with a positive screen were assigned to either a four-session Interpersonal-Therapy-Oriented Group (given weekly) or to a usual-care group. Outcome measures included the BDI before and after the intervention and the SCID at 3 months postpartum. Women in the intervention group had a significantly greater change in their BDI scores from baseline than did those in the control group (“pre” versus “post” intervention Beck scores were 13.0 and 8.4, respectively, for the treatment group). In contrast, the control group “pre” versus “post” intervention scores were 9.2 and 11.3, respectively, suggesting that they got worse over time. The change between the intervention and the control group was significant (t-test = 3.50; df = 33; P = 0.001). In addition, women in the intervention group had a significantly decreased rate of major depression during the postpartum period as measured by the SCID at 3 months postpartum; no women in the intervention group and 33 percent of women in the usual-care group developed postpartum depression (P > 0.02).

These four small studies of programs for women identified by screening prenatally did not, collectively, produce many positive results from the various psychosocial interventions as compared with usual care. All of these studies scored poor on external validity (0 of 3 points), and two of the four had 0 of 2 points for power. The four studies did, at best, only a fair job of reporting data (from 5 to 7 of 11 points). For bias, the study scores ranged from 1 to 4 of 7 possible points; for confounding, they ranged from 3 to 5 of 6 points. Given the heterogeneity in populations, the screening instruments and cutoff points for defining “at-risk” individuals, the interventions themselves, and the outcome measures used, we cannot draw any overall conclusions about the utility of such programs.

Postpartum Studies

Of the 11 studies examining screening and intervention outcomes only in the postpartum period, ^98–108 eight were RCTs published between 1992 and 2003,^{98, 100, 101, 103–106, 108} and three were controlled trials published between 1992 and 2002.^{99, 102, 107} Sample sizes ranged from 23 to 1, 215, for a total population of 4,289 women.

As with the prenatal screening studies, the screening instruments used to identify patients with depressive symptoms differed among the postnatal studies, although the EPDS was used in the majority of studies and two studies by the same investigator team used “prior history of postpartum depression.” The treatment interventions also differed considerablyNine of these studies involved various behavioral and psycho-educational programs or other innovative activities (e.g., infant massage or infant sleep interventions); two involved tests of antidepressants. Unlike the prenatal studies, the settings varied from postpartum hospital wards to child-health and immunization clinics. Finally, the outcome measures also varied across these studies, but the EPDS was most commonly used (in seven studies). We graded one study good, seven fair, and three poor.

Behavioral and Psychosocial Interventions. Of the nine studies in this subgroup, one was conducted in the United States; the remainder were in Australia (two studies), Canada (two studies), the United Kingdom (not otherwise specified, two studies), and France and Taiwan (one study each). Using “number randomized or enrolled” as the metric, the sample sizes ranged from 45 to 859. We describe the studies below according to quality grade and sample size.

In a recent study rated good that randomized participants for the intervention (not screening), Dennis et al. screened 501 women recruited from child immunization clinics between 8 and 12 weeks postpartum.¹⁰¹ Inclusion criteria included the mother's being at least 18 years of age, having a singleton birth, and delivering a full-term infant. Women were screened using the EPDS (cutoff score > 9). The 44 women with a positive screen were randomized to a “mother-to-mother” peer support telephone intervention or to routine care. The outcome measures were the EPDS at 4 and 8 weeks after randomization. The women in the intervention group had significantly lower EPDS scores than those in the control group: 15 percent of the intervention group and 52.4 percent of the control group had an EPDS > 12 at 8 weeks (P = 0.02).

In the largest of the seven studies rated fair, Chabrol et al. screened 859 women and identified 258 who were at risk based on an EPDS > 9 on day 2 or 3 postpartum.⁹⁹ They assigned these 258 women randomly to receive a cognitive behavioral therapy (CBT) (n = 130) intervention or to routine care (n = 128) during the postpartum hospitalization. CBT is a form of psychotherapy that actively examines how cognitions influence emotions or affect and involves active exploration, clarification, and testing of the patient's perceptions and beliefs.¹⁰⁹ Outcome measures for the prevention intervention included the EPDS (cutoff ≥ 11) taken at 4 to 6 weeks postpartum.

Women in the CBT group who continued to have positive screens on the EPDS (defined as EPDS score ≥ 11) at 4 to 6 weeks were assessed for major depression in a clinical interview using the Mini-Neuropsychiatric Interview (MINI) and DSM-IV criteria. Those with major depression were offered an at-home CBT program for five to eight additional sessions. These women were then compared with women with probable major depression in the control group at 10 to 12 weeks using the EPDS, Hamilton Depression Rating Scale (HAM-D), and the BDI. Women in the control group received one initial home visit assessment but then received only weekly telephone checks.

The study results demonstrated that women in the prevention intervention group had significant reductions in the frequency of depressive symptoms. At 4 to 6 weeks postpartum, 30.2 percent of those in the CBT group versus 48.2 percent in the control group (P = 0.0067) were still depressed (based on an EPDS score of ≥ 11). Additionally, the intensity of depressive symptoms measured by the mean score on the EPDS was significantly lower in the prevention group than in the control group: mean EPDS scores, respectively, of 8.5 (standard deviation [SD] 4) and 10.3 (SD 4.4) (t-test = 3.06, df = 209, P = 0.0024); the analyses indicate a medium effect size (ES, 0.42). At 10 to 12 weeks postpartum after completion of the home-based CBT intervention, women in the intervention group had significantly lower scores on all measures of depressive symptoms (HAM-D, BDI, EPDS) than did those in the control group. Specifically, the intervention and the control group mean scores were as follows: HAM-D, 5.7 versus 16.2 (t-test = 8.4, P < 0.0001); BDI, 4.7 versus 15.7 (t-test = 9, P < 0.0001); and EPDS, 5.9 versus 13.7 (t-test = 7.7; P < 0.0001).

Armstrong et al. screened 181 women with good literacy skills in the immediate postpartum period by asking about a history of trauma or abuse or a positive screen for adverse family characteristics on the Brisbane Evaluation of Needs Questionnaire.⁹⁸ Women were randomized to receive 6 months of home visits by a child-health nurse or routine primary care. Primary outcome measures involved measures of child health;, parental and family functioning (measured by the EPDS [cutoff > 12] and the Parenting Stress Index [PSI]), quality of the home environment (HOME assessment), and satisfaction with community services. All assessments were administered immediately postpartum.

If we focus primarily on scores of maternal depression and functioning at 6 weeks postpartum, women in the intervention group had significantly lower depression scores than the control group: 5.8 percent in the intervention group and 20.7 percent in the control group (P = 0.003) with EPDS > 12. Additionally, women in the intervention group had significantly lower (better) PSI scores at 6 weeks than controls (15.3 versus 38.4, P < 0.001). The investigators also reported that the total HOME score differed significantly between groups: 28.34 for the intervention group versus 25.51 for the control group (P < 0.001), providing evidence for the positive effect the intervention had on influencing parent-infant interaction and the home environment for the child.

Hiscock and Wake recruited 155 women from a child-health center at 7 to 8 months postpartum and screened with the EPDS (cutoff > 12).¹⁰³ Other inclusion criteria included reported child sleep problems. Of these women, 99 were considered depressed (baseline EPDS ≥ 10) and randomly assigned to either an infant sleep intervention group or a usual-care group. The infant sleep intervention comprised three private sessions (one session every 2 weeks) held at the local child-health center where sleep management plans were discussed, including an emphasis on controlled crying (where parents responded to their infants' crying at increasing time intervals, allowing the infant to fall asleep unaided). At the 10- to 12-month follow-up assessment, outcome measures included the EPDS, measures of sleep quality, and measures of maternal stress and coping.

The results of this study were mixed. Women who began with higher (worse) scores of depression at baseline had a significantly greater improvement in their EPDS scores than did those in the control group. At the 10-month follow-up, women in the intervention group had a 6.0 point decrease (95% CI, 7.5 to 4.0) in EPDS score compared to a 3.7 point decrease (95% CI, 4.9 to 2.6) in the control group (P = 0.01). At the 12-month follow-up visit, the intervention group had a 6.5 point decrease (95% CI, 7.9 to 5.1) in EPDS score compared to a 4.2 point decrease (95% CI, 5.9 to 2.5) in the control group (P = 0.04). Also, at the 10-month follow-up, women in the intervention group reported improvements in their own sleep quality, including being more likely than control mothers to rate their own sleep quality as “very good” and less likely to rate it as “very bad” (Chi square = 9.93; P = 0.02). They also reported having “enough sleep” and were less likely to have “not enough” sleep (Chi square = 8.11, P = 0.04).

Horowitz et al. screened 1,215 women at 2 to 4 weeks postpartum with the EPDS (cutoff > 10). Women with positive screens (n = 122) were randomly assigned to either an interactive coaching intervention or a control group. The coached behavioral intervention was designed to promote maternal-infant responsiveness. All women in the study (both intervention and control groups) received three home visits when their infants were 4 to 8 weeks, 10 to 14 weeks, and 14 to 18 weeks of age; the women in the intervention group practiced the coaching intervention during these visits. Outcome measures included the BDI-II for maternal depression and, secondarily, the Dyadic Mutuality Code (DMC), a measure of the level of responsiveness in the maternal-infant relationship. Responsiveness was defined as “the mother's ability to accommodate to her infant's behavior and to give it meaning through regulation of her own behavioral responses” (p. 326). The intervention and control groups did not differ significantly in terms of maternal depression scores (BDI-II) at any time period. The DMC showed a significantly better outcome for mother-infant responsiveness for the treatment group (P = 0.06).

Onozawa et al. screened 581 primiparous women with the EPDS (cutoff ≥ 13) at 4 weeks postpartum.¹⁰⁶ Of 91 women who had a positive screen, 59 agreed to participate in the study. Participants were randomized to either a 5-week infant massage class with a support group or the support group only. The 1-hour infant massage class (approved by the International Association of Infant Massage) taught parents the techniques of infant massage by encouraging parents to observe and respond to their infants' body language and cues and to adjust their touch accordingly. Outcome measures included maternal depression on the EPDS (cutoff ≥ 13) at 4 weeks and 2 months postpartum and a videotaped mother-infant interaction that assessed the mother's attitude toward the infant, the infant's response to the mother, and the overall quality of the interaction. At 14 weeks postpartum, EPDS scores had fallen for both groups (reported as a change in median EPDS score from baseline to final visit), but the intervention group demonstrated a significantly greater change in scores than did the control group (intervention group baseline of 15.0 and final visit score of 5.0, versus the control group score of 16.0 at baseline to 10.0 at the final visit; P = 0.03). Additionally, significant improvements in all aspects of mother-infant interaction as measured by the videotape were seen only in the massage group (P = 0.0004).

Honey et al. used the EPDS (cutoff > 12) to screen postpartum women recruited through mother-baby clinics but assessed at home.¹⁰⁴ The 45 women with a positive screen on the EPDS were randomly assigned to either an 8-week psycho-educational group (PEG) or to a routine care group. Outcome measures included the EPDS (cutoff > 12) after completion of the PEG and at a 6-month follow-up. At the end of the 8-week assessment interval, the women in the PEG did not differ significantly from those in the routine-care group. By contrast, at the 6-month follow-up assessment, the percentage of women scoring below the EPDS cutoff for a probable major depressive episode was significantly higher in the PEG group than in the routine-care group (65 percent versus 36 percent, Chi square = 3.75; P ≤ 0.05). An additional analysis demonstrated that the use of antidepressant medication during the study had no impact on the improvement in mood observed at the 6-month follow-up assessment.

Fleming et al. screened 781 primiparous women with full-term deliveries and no psychiatric history during their first 2 weeks postpartum using the EPDS (cutoff ≥ 13), the Current Experience Scale (CES, cutoff ≥ 35), and the Multiple Affect Adjective Checklist (MAACL, cutoff ≥ 21).¹⁰² Women with a positive screen (n = 142) were assigned (not randomly) to either a postpartum social support group that included both depressed and nondepressed women or a usual-care group.

Outcome measures included the EPDS and the CES at the same cutoff scores used for screening. At the 6-week and 5-month follow-up assessments, the groups did not differ significantly with respect to rates of maternal depression, and the support groups had no apparent effect on the mothers' general affective mood. However, women in the social support group had a statistically significant increase in the number of maternal-infant interactions and noted decreased infant crying compared to women in the routine-care group.

Chen et al. screened 414 women at 3 weeks postpartum using the Taiwanese BDI (cutoff ≥ 10).¹⁰⁰ Of these, 115 women with positive screens were randomized to weekly support groups or to a routine-care group; 60 patients were available for analysis. Outcome measures included the BDI (cutoff ≥ 10), Perceived Stress Scale (PSS), and measures of interpersonal support. At the 15-week follow-up assessment, women in the intervention group had significantly lower rates of depression: 33.3 percent of the intervention group and 60.0 percent of the control group had BDI values equal to or greater than 10 (P < 0.05). The rate of perceived stress was also significantly lower in the intervention group than the control group (t-test = 3.75, P < 0.01). Finally, women receiving the intervention reported significantly more interpersonal support as measured by the Interpersonal Support Evaluation List than those in the control group (t-test = 2.81, P < 0.01).

Pharmacologic Studies. Two of the studies were psychotropic medication trials to prevent the occurrence of postpartum depression. The women were not directly screened with any instrument, but rather were included if they had a previous history of postpartum depression. The same research team conducted both of these pharmacologic trials. In the first trial, Wisner and Wheeler studied the efficacy of antidepressant treatment in women with a previous history of postpartum depression (i.e., at high risk of maternal depression but no history of psychosis or bipolar disorder).¹⁰⁷ At-risk postpartum women (n = 23), who had had at least one episode of postpartum depression were treated in an open clinical trial with the tricyclic antidepressant nortriptyline and postpartum monitoring or with postpartum monitoring only. Outcome measures included a clinical assessment of major depression and the Inventory to Diagnose Depression scale. After 3 months, study results demonstrated a significantly greater proportion of new-episode major depression in those patients who received monitoring alone than in those in the medication group (62.5 percent of those in the monitoring group; 6.7 percent in the medication group; P = 0.0086).

In a later Wisner et al. RCT, 56 women with a prior history of postpartum depression within the past 5 years but no depressive episode upon enrollment, as diagnosed by standardized research diagnostic criteria, were randomized to either a nortriptyline group or a placebo group immediately postpartum.¹⁰⁸ Outcome measures of recurrence of perinatal depression included the HAM-D and Research Diagnostic Criteria (RDC). In contrast to the earlier open-label trial, the investigators reported no difference in the rate of recurrence of depression (one-fourth) between women treated with nortriptyline and those receiving placebo (23 percent versus 24 percent, respectively).

None of the studies used treatment interventions that are recognized as the gold standard treatment for major depressive illness according to current American Psychiatric Association guidelines. These guidelines specify that the gold standard include antidepressant medication plus psychotherapy.¹².

The overall quality of these 11 postpartum studies was fair; one study was rated good; two, poor. Of a possible quality score of 29, one study scored 22, seven scored between 15 and 20, and two at or below 14 (Table 18). External validity was generally poor; the majority of studies scored 0 of 3 points. The bias measure of internal validity ranged between scores of 1 and 6 (of a possible 7); 6 of 11 studies scored 4 or better. For confounding, scores ranged from 1 to 6 (of a possible 6); 6 of 11 studies scored 4 or better. Power was generally poor, reflecting the small sample sizes; 9 of 11 studies scored 0 (of a possible 2).

Only three studies had quality scores of 18 or higher: Wisner et al.,¹⁰⁸ Armstrong et al.,⁹⁸ and Dennis et al.¹⁰¹ All three enrolled women in the postpartum period and had a fairly intensive treatment approach consisting of weekly interventions (Wisner et al. for 20 weeks, Dennis et al. for 8 weeks, and Armstrong et al. for 6 weeks). The Wisner et al. study had a pharmacologic intervention with weekly assessments of efficacy but no psychotherapeutic intervention; by contrast, the Dennis et al. and Armstrong et al. studies had weekly psychotherapeutic interventions but no pharmacologic intervention. Interestingly, although Wisner et al. treated patients for 20 weeks postpartum with antidepressant medication, their study did not have a significant result. This finding may suggest that psychosocial support and psychotherapeutic intervention are both critical as part of a treatment plan for women with postpartum depression.

Conclusions

The 15 studies examined a variety of screening and treatment interventions for women identified as being at risk (sometimes at high risk) for postpartum depression. The majority of these studies focused on intervention strategies in the postpartum period; all but two dealt with a wide array of psychosocial, education, skill-building, and other mother-child behavioral activities. Generally, the more successful efforts occurred in the studies in which screening and interventions were carried out in the postpartum, not the prenatal, period. Once again, none of the studies had a treatment intervention with both psychotherapeutic and pharmacologic components that would be considered “gold standard therapy” for the treatment of major depression.

Overall, many of the studies suggest that providing the mother with some form of psychosocial program to increase maternal support or improve maternal-child interaction may decrease the rate of postpartum depression. Across the nine nonpharmacologic studies, about 20 outcomes were assessed; of these, 12 showed significant effects for the intervention group. Taking only the outcomes dealing specifically with depression, nine significant effects were reported. The two small pharmacologic trials from the same group yielded conflicting results about the impact of nortriptyline in reducing recurrence of maternal depression.

Only one study⁹⁷ specifically studied low SES women—a matter of some interest to the Safe Motherhood Group. Low SES women with at least one risk factor for postpartum depression who participated in weekly prenatal survival skills classes were less likely to develop postpartum depression compared to controls. This small study suggested that increasing support and parenting skills may help to decrease postpartum depression in this particular population.

Study Limitations

This set of studies, however, has several limitations, and it can be regarded as offering, at best, only fair evidence about the utility of screening plus prevention or treatment programs or even interventions alone. Although a variety of interventions may be helpful in treating women with or at risk of perinatal depression, the available evidence does not directly address whether screening with subsequent intervention improves outcomes. Screening, in the classic sense, implies “examination of a group of usually asymptomatic individuals to detect those with a high probability of having a given disease” (italics added; http://dictionary.reference.com/search?q=screening); this meaning can be extended to using appropriate screening or diagnostic tools within populations with known risk factors. These studies provide little guidance in answering the practical question of whether clinicians should screen all women in the perinatal period (i.e., essentially an asymptomatic population with respect to depression) for risk factors or latent depression, or whether they should screen only women who have known prior histories or risk factors for depression.

The studies are generally small, with poor generalizability (especially to the heterogeneous childbearing population of the United States). We contemplated and rejected the idea of any quantitative analyses: populations, settings, and screening and outcome measures—let alone interventions—were simply too disparate for anything but qualitative synthesis.

Future Research

To overcome some of these problems in understanding the impact of programs designed to prevent the problems of perinatal depression or to mitigate the considerable deleterious effects of this disorder on mothers, infants, and families, considerably more and better research needs to be conducted. Possibly the most important issue is for future studies to enroll adequate samples of women and, if screening is the question, to screen quite large numbers of women to produce sample sizes with adequate power to detect relevant differences between treatment and control groups in later phases of these studies. Virtually all studies appeared to be underpowered to start with, and some lost participants along the way. This deficiency hampers investigators and policymakers in making sense of, or decisions based on, much of this work.

Moreover, a greater effort should be made, at least in US-based studies, to focus on ethnic and disadvantaged populations, such as low-income women. Even if the incidence and prevalence of perinatal depression were “evenly” spread over population groups in this country, the underfunding of health care for many (e.g., lack of insurance, poor coverage of mental health benefits in insurance plans, unavailability of publicly funded services) and the more precarious economic resources and family support for some populations means that additional attention needs to be paid to them. For example, programs may need to be designed to take lack of transportation, child care, or telephone access into account.

In addition, researchers might direct attention to several other variables that appear to be important. They include first-time versus second-time mothers, maternal comorbidities and lifestyle behaviors, family structure (make-up) and available support, and status of infant at birth (e.g., full term or not, healthy or not). Another gap may be programs intended to assist the mother-father dyad or, indeed, to assist fathers in providing the emotional or physical support needed to forestall depression in new mothers.

Ideally, researchers would employ similar screening measures with similar cutoff points so that some elements of separate studies could be compared more readily. Not all of the screening instruments used appear to be sufficiently well-targeted to perinatal depression (i.e., even if they are reliable, their validity for this purpose may be called into question). Moreover, some instruments may be relatively infeasible for use in certain populations (e.g., immigrants) or in cases in which patient self-report is important and literacy may be low. For these situations, some work to calibrate well-known instruments that have been specifically designed for this disorder and that have acceptable test properties against each other might be useful. Calibrating less well-known or well-proven instruments against some agreed-upon reference (“gold”) standard instrument in this area might also be valuable. Testing these in different settings, trying to use shorter instruments, attempting to take literacy levels into account, and in other ways improving the screening armamentarium are also important steps. In that way, investigators and clinicians can have a better selection of proven screening tools for future research or clinical practice applications.

Another element warranting more clarity is the purpose of the screening-cum-intervention effort. All appear to relate to populations of women at risk of perinatal depression (particularly postpartum depression that goes beyond “maternal blues”), but the severity level of being at risk differed in these studies. Moreover, women could have had no prior history of depression (or perinatal depression) and be at risk; alternatively, they could have had some history, especially of postpartum depression, and be, essentially, at “high” risk. These distinctions did not seem to be well or consistently described across these studies. They also have implications for the goals of the interventions themselves: for example, preventing any “first episode” of depression, mitigating the effects of a first episode that is not wholly prevented, or preventing a recurrence.

Interventions tested in the future would, ideally, be those shown to have some promise so far (e.g., as reflected in some of the studies reported here). The components of the programs should be of appropriate length and intensity, and published articles should describe them thoroughly. In addition, interventions should be consistent with current evidence-based practice standards for the treatment of major depression. Multiple studies of the same interventions, perhaps at different time periods or different settings and populations, might be helpful in completing the picture of the impact of screening and interventions on occurrence or reoccurrence of perinatal depression. Finally, outcome measures should be appropriate to the research questions and preferably selected from among the more reliable, valid, and widely used instruments. These steps might help fill the gaps in this knowledge base and permit those performing systematic reviews to compare and synthesize studies more readily.

Key Question 1: Prevalence and Incidence of Perinatal Depression

For KQ 1, we identified 30 studies of generally moderate size that provide estimates of the prevalence of perinatal depression; 13 of these inquiries provide estimates of incidence. Studies were generally of good quality for reporting completeness and internal validity for bias; by contrast, they were of fair quality for precision and only poor quality for external validity and internal validity for confounding. In particular, the study populations were not representative of the racial and ethnic mix of the countries in which the studies were performed and especially not of the United States.

Our final best estimates of prevalence and incidence were somewhat lower than those reported in prior systematic reviews because we excluded studies that assessed depression based on self-report screens alone, which have been found to overestimate prevalence. Also, we separated out estimates of major and minor depression from estimates of major depression alone. Finally, we included more recent studies that use more precise criteria to identify major depression.

For major depression alone, our final combined point prevalence estimates ranged from 3.1 percent to 4.9 percent at different times during pregnancy and from 1.0 percent to 5.9 percent at different times during the first postpartum year. For major and minor depression, our final combined estimates of point prevalence ranged from 8.5 percent to 11.0 percent at different times during pregnancy and between 6.5 percent and 12.9 percent at different times during the first year postpartum. This nearly 2-fold higher rate suggests that approximately half of the women experience a major depressive episode and half a minor depressive episode at any given time. Confidence intervals surrounding all these estimates remain wide, suggesting that a fair amount of uncertainty remains in the combined estimates.

Fewer estimates were available for the incidence of depression. These limited data suggest that as many as 14.5 percent of pregnant women have a new episode of either major or minor depression during pregnancy, and 14.5 percent have a new episode during the first 3 months postpartum. Considering only major depression, 7.5 percent may have a new episode during pregnancy, with 6.5 percent having a new episode in the first 3 months postpartum.

Are the prevalence and incidence of depression during the perinatal period higher than the rates during nonchildbearing periods? We found three studies that measured the prevalence of major or minor depression and major depression alone for women at different times during these two periods . None of these estimates shows a statistically significant difference. Only one study²⁰ directly compared the incidence (new onset) of perinatal depression to that of nonchildbearing women of similar age; women at 5 weeks postpartum were more than three times as likely as the comparison group to have a new episode of major or minor depression. By 6 months postpartum, this difference had disappeared. An incidence for major depression alone was not reported.

That these estimates did not appear significantly different from those of nonchildbearing women of the same age does not reduce the dramatic burden experienced by women postpartum. Indeed, these estimates, based on the best available evidence, suggest that perinatal depression, whether major or minor depression, is a very common complication of pregnancy. Furthermore, and arguably more important, after labor and delivery this dramatically common complication, rather than primarily affecting one individual, now directly affects two: mother and child.

Key Question 2: Accuracy of Perinatal Depression Screening Tools

For our analysis of the accuracy of screening tools (KQ 2), we identified 10 studies reporting test characteristics for English-language screeners. In general, studies were of fair to good quality, although external validity was only poor to fair. In particular, the study populations were nearly entirely white, so the accuracy of these screeners in nonwhite perinatal populations is not clear. A major limitation in the available evidence is the very small number of depressed patients involved, a fact that results in substantial imprecision in the point estimate of sensitivity and prevents one from reasonably determining an ideal cutoff point.

For depression during pregnancy, we found only one study reporting on screening accuracy in a population with 6 patients with major depression and 14 patients with either major or minor depression. For major depression, sensitivities for the Edinburgh Postnatal Depression Scale (EPDS) at all evaluated thresholds (12, 13, 14, 15) were 1.0, underscoring the markedly small number of depressed patients involved; specificities ranged from 0.79 (at EPDS ≥ 12) to 0.96 (at EPDS ≥ 15). For major or minor depression, sensitivity was much poorer (0.57 to 0.71); specificity remained fairly high (0.72 to 0.95).

For postpartum depression also, the small number of depressed patients involved in the studies precluded identifying an optimal screener or an optimal threshold for screening. Our ability to conduct a meta-analysis of the results of different studies was limited by the use of multiple cutoffs and other differences across studies that precluded a meaningful interpretation of the results. Where we were able to combine the results , the pooled estimates did not add to what one could conclude from individual studies.

For women with major depression alone, specificity for all screeners (the Beck Depression Inventory [BDI], the Postpartum Depression Screening Scale [PDSS], and the EPDS) was relatively high. This finding suggests that a positive screen was accurate in ruling major depression in; that is, the risk that a screen with one of these instruments would be falsely positive was low. By contrast, sensitivities varied much moreThe EPDS and the PDSS appeared to be more sensitive (with estimates ranging from 0.75 to 1.0 at different thresholds) than the BDI instruments (with estimates from 0.32 to 0.68), but the wide confidence intervals (CIs) overlapped nearly completely. This means that we could not say with confidence that the specificity estimates using the different tools were different.

The point estimates are consistent with what is reported for depression screeners in primary care settings.⁹⁰ Still, the imprecision is important to clarify. If falsely missing depression (a false negative) is worse than falsely identifying it, as may be the case with this disorder, clinicians must be able to feel confident that the screen is usually positive if the disease is there and that a negative result can help rule out the illness.

For patients with major or minor depression, results were reported for EPDS, BDI, PDSS, and the Center for Epidemiologic Studies-Depression (CES-D). Specificity estimates remained relatively high, but sensitivity results were much lower (ranging from 0.43 to 0.71) than for major depression alone. This means that the ability of the screening instrument to score women as positive for this condition when the disease is present was poorer than for major depression alone. Again, neither any particular cutoff nor any particular screening instrument performed differently from the others. No available comparators were found for primary care populations.

Our results suggest that various screening instruments can identify perinatal depression, most accurately major depression, but clinicians need to know more about the precision of individual instruments. If one assumes that the risk of a false-negative depression screen is worse than the risk of a false-positive screen, perinatal depression is a condition in which sensitivity is likely to be more important than specificity. Whether as a screen for major depression alone or for major or minor depression, specificities appear high and relatively precise. By contrast, sensitivity for identifying either category is imprecise and differs by diagnostic category. For major depression alone, point estimates are equivalent to those in primary care medical settings. For major or minor depression, however, sensitivity is quite low. At this time, these screens do not appear to be useful for identifying patients in this latter category of illness.

Key Question 3: Screening and Treatment Outcomes

KQ 3 concerned issues of whether screening ultimately leads to improved patient outcomes. Although it is the most vital question from the public health perspective, it is the one with the most limited evidence. Indeed, the studies that we identified were not designed to test whether screening for depression (versus not screening) improved patient outcomes. Such a design would randomize patients to be screened or not to be screened and then compare subsequent outcomes. We found no studies designed in this way.

Instead, we made use of studies in which women were screened by formal depression screening or the presence of a risk factor associated with perinatal depression to identify those at risk of having a depressive illness; then, for those screening positive, the investigators compared the outcomes of women receiving a treatment intervention to those in a control group. This design tests whether, among women identified as at risk of depression by a screen, an intervention improves outcomes compared to the outcomes in a control group. This is an important intermediary step, but it does not directly test whether screening itself improves outcome compared to not screening. All the trials included are treatment studies that had a screening component (either a formal depression screening instrument, or other type of screen that identified women at risk of a depressive illness) but did not have diagnostic confirmation of depression.

We attempted to synthesize the results of the included studies quantitatively, but the study methods (screening instruments, type of intervention, intensity of intervention, outcomes measured) were so heterogeneous that a meta-analytic synthesis would not be meaningful. We also attempted to compare effect sizes to attempt an exploratory analysis of the various studies, but the data necessary to compute these were not available.

For patients whose screening results identified them as at risk of perinatal depression and for whom a subsequent intervention was provided, we identified 15 studies. Four small prenatal studies involved various psychosocial interventions. Quality was poor for three of these studies and fair for one. Overall, the effects of the interventions in these studies were not consistently superior to those in the control groups.

The 11 postpartum studies were of overall fair quality and had larger sample sizes than the prenatal trials. Study populations reflected only a limited racial and ethnic mix, and both external validity and the power to demonstrate statistically significant differences were generally poor. Again, screening tools and interventions varied considerably; the latter involved both psychosocial and pharmaceutical interventions.

Results were mixed. Of the nine trials that employed a psychosocial intervention, six studies^{98–101, 103, 104} reported significant benefit for depression outcomes in the experimental group compared to those in the control group. The one RCT involving pharmacologic intervention did not show benefit relative to the control group.¹⁰⁸ Overall, the evidence available is not sufficient to draw conclusions about this key question. These results, although limited, do suggest that providing some form of psychosocial support to pregnant women at risk of having a depressive illness may decrease depressive symptoms.