Chapter  162:  Therapeutic Management, Delivery, and Postpartum Risk Assessment and Screening in Gestational Diabetes

Approach to Evaluating the Literature

We identified the primary literature on labor and postpartum management of gestational diabetes and the association with maternal and neonatal outcomes through a comprehensive search plan that included electronic and hand searching. We ran searches of the following databases for the specified periods of time: MEDLINE^® (1950 through January 2007), EMBASE^® (1974 through January 2007), The Cochrane Central Register of Controlled Trials (CENTRAL; Issue 1, 2007), and the Cumulative Index to Nursing & Allied Health Literature (CINAHL^®; 1982 through January 2007). Hand searching for relevant citations took several forms. From our electronic search, we identified the 13 journals (see Appendix B ^a) that were most likely to publish articles on this topic. We scanned the table of contents of each issue of these journals for relevant articles from August 2006 through January 2007. For the second form of hand searching, reviewers received eligible articles and flagged references of interest for the team to compare to the existing database.

Two independent reviewers conducted title scans in a parallel fashion. If either reviewer thought that a title was potentially eligible, its abstract was reviewed. If the abstract was deemed to meet the inclusion criteria by two reviewers, the abstract was included in our article review. Any differences of opinion were resolved by the two primary reviewers or by a third independent reviewer.

Each eligible article underwent double review by study investigators. A primary reviewer completed all data abstraction forms, and a second reviewer confirmed the first reviewer's data abstraction forms for completeness and accuracy. The reviewers assessed study quality independently. Reviewer pairs were formed to include personnel with both clinical and methodological expertise. A third reviewer re-reviewed a random sample of articles by the first two reviewers to ensure consistency in the abstraction of the articles.

Quality Assessment

We used several study quality assessment tools, based on the study design of the articles included in the review. Our dual, independent review of article quality judged articles on several aspects of each study type's internal validity. Quality assessment of trials for Key Questions 1 and 2 was based on the Jadad criteria and included: (1) whether the study was randomized, (2) the appropriateness of the randomization scheme, (3) whether the study was blinded, (4) the appropriateness of the blinding, and (5) the description of withdrawals and drop-outs. For each trial, we created a score between 5 (high quality) and 0 (low quality). Quality assessment of observational studies for Key Questions 1, 2, and 3 was designed by selecting key elements from the Standards for Reporting of Observational Studies (STROBE) checklist for reporting observational studies. The STROBE checklist is based on the consensus of 27 participants of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group and includes recommendations for standards for individual studies with regard to the presentation of the study hypotheses, eligibility criteria, study population, power and sample size calculations, definitions of outcomes, and description of loss to followup and missing data. Quality assessment of the diagnostic test studies for Key Question 4 was designed by selecting key elements from the Standards for Reporting of Diagnostic Accuracy (STARD) Initiative and included items about reporting the sampling design, describing the lost-to-followup, reporting diagnostic accuracy, verifying positive and negative tests equally, interpreting the tests independently, reporting reproducibility, and reporting subgroup analyses.

Based on the quantity, quality, and consistency of the studies, we graded the overall body of evidence for each of the key questions using the evidence-grading scheme recommended by the GRADE Working Group.

Analysis

We conducted meta-analyses when there were sufficient data (three or more studies) and the studies were homogeneous with respect to key variables (population characteristics, study duration, intervention/exposure/comparison tests, and length of followup). When the data were not sufficient to allow us combine the studies in a meta-analysis, we prepared a qualitative summary of the results.

Search Results

We retrieved 11,400 unique citations from our original search. After reviewing the titles and abstracts, 552 were deemed eligible for further review, and the full articles were retrieved. A total of 45 articles were ultimately included in this review.

Key Question 1

What is the evidence for the risks and benefits of oral diabetes agents (e.g., second-generation sulfonylureas and metformin), as compared to all types of insulin, for both the mother and neonate in the treatment of women with gestational diabetes?

• a

  How does maternal outcome vary based on the level of glucose at the initiation of a medication?

• b

  How does neonatal outcome vary based on the level of glucose at the initiation of a medication?

• We identified eight randomized controlled trials (RCTs) with a total of 845 participants that met our inclusion criteria for review: Three trials compared insulin to glyburide; two trials compared insulin to insulin lispro; one trial compared long-acting to short-acting insulin; one trial compared four-times-daily insulin to two-times daily insulin; and one trial compared diet to insulin.

• Two small trials and one large trial (404 women) reported no significant difference in maternal glucose control or rates of cesarean delivery between the insulin and glyburide groups.

• A meta-analysis of the three RCTs comparing insulin and glyburide showed that treatment with insulin was associated with a lower mean infant birth weight when compared to glyburide (weighted mean difference: -93 grams [gm]) (95 percent confidence interval [CI]: -191 to 5 gm), but the difference was small and not statistically significant.

• The largest trial reported no difference in the proportion of infants with hypoglycemia (9 percent with glyburide as compared to 6 percent with insulin therapy [p = 0.25]). A smaller trial reported a significantly higher percentage of infants with hypoglycemia in the glyburide group than in the insulin or acarbose groups (33 percent compared to 4 percent and 5 percent, respectively; p = 0.006).

• Four observational studies (N = 911 women) compared the effects of insulin and glyburide on maternal and neonatal outcomes.

• Due to potential selection bias, loss to followup, and the lack of any power analysis to estimate detectable effect sizes, none of the observational studies was deemed strong enough to justify a modification of the conclusions drawn from the RCTs.

• We graded the overall evidence comparing insulin and glyburide as very low.

• We identified two RCTs that compared insulin lispro to insulin. It appeared that insulin lispro might be associated with tighter maternal glucose control than regular insulin, but there were only limited data to support this conclusion.

• Both RCTs reported similar rates of cesarean delivery among women in the insulin lispro group, as compared to the insulin group.

• No evidence existed to suggest that neonatal outcomes differ between women treated with insulin lispro and those treated with regular insulin.

• We graded the strength of the evidence comparing insulin to insulin lispro as very low.

• One RCT (N = 23 women) reported that long-acting insulin was associated with a higher proportion of infants with macrosomia when compared to short-acting insulin. No difference in birth trauma or metabolic abnormalities was found, but this study was not adequately powered to detect differences in these outcomes.

• There was insufficient evidence to allow us to draw any conclusions regarding maternal outcomes.

• One RCT (N = 274 women) reported that twice-daily insulin was associated with a higher proportion of hypoglycemia (6 percent versus 1 percent; p = 0.002) and hyperbilirubinemia (21 percent versus 11 percent; p = 0.002) when compared to four-times-daily insulin. No evidence existed to suggest a difference in maternal glucose levels or cesarean delivery between twice-daily and four-times-daily use of insulin.

• We identified only one RCT (N = 95 women) that reported lower rates of macrosomia (5.9 percent versus 26.5 percent, respectively; p = 0.005) and lower infant birth weights (p = 0.002) for those using insulin plus dietary management versus those treated with diet alone.

• We graded the overall evidence regarding comparisons of diet plus insulin to diet alone as very low, given that only one small RCT met our inclusion criteria.

• We found no evidence to indicate whether the relative effect of different treatment approaches on maternal and infant outcomes varied with the level of glucose at the initiation of medical therapy.

• We expect that the ongoing Hyperglycemia and Adverse Pregnancy Outcomes (HAPO) Study, an observational study of 23,325 pregnant women, should provide data on maternal and neonatal outcomes at various maternal glucose levels and potentially indicate thresholds at which medical therapy should be initiated.

• We found no RCTs or observational studies comparing metformin to insulin in gestational diabetics that met our criteria for review. The Metformin in Gestational Diabetes (MiG) Trial is an ongoing randomized trial of over 500 women that should provide future insight into the benefits and risks of metformin use throughout pregnancy.

• There were insufficient data regarding the teratogenic effects of intrauterine exposure to metformin or its potential effect on infant growth and motor development.

Key Question 2

What is the evidence that elective cesarean delivery or the choice of timing of induction in women with gestational diabetes results in beneficial or harmful maternal and neonatal outcomes?

• One RCT and seven observational studies evaluated two of our five maternal outcomes of interest and 11 of the 12 neonatal outcomes of interest.

• One RCT of 200 women reported that elective induction at 38 weeks of gestation, as compared to expectant management (induction at 42 weeks' gestation or if the EFW was 4,200 gm or greater), reduced infant birth weight and the rate of macrosomia but did not alter other maternal or neonatal outcomes, including the rate of cesarean delivery.

• Two observational studies of low quality also reported a reduction in infant birth weight or rates of macrosomia in women induced at 38 weeks of gestation, as compared to historical controls.

• Five additional observational studies examined the effects of various delivery management protocols, but each had serious limitations, including reliance on historical controls, no adjustment for potential confounders, or no adjustment or stratified analysis based on severity of gestational diabetes (class A1 [diet-controlled] versus class A2 [insulin-controlled]). In addition, the studies covered a wide time period (4–19 years), with no adjustment for changes in clinical practice.

• We were unable to draw firm conclusions from the limited data available.

• We graded the overall strength of the evidence as very low, given the limited number of RCTs and the serious design limitations in the conduct of the observational studies.

Key Question 3

What risk factors, including but not limited to family history, physical activity, pre-pregnancy weight, and gestational weight gain, are associated with short-term and long-term development of type 2 diabetes following a pregnancy with gestational diabetes?

• We developed an a priori list of risk factors for type 2 diabetes, based on guidance from the AHRQ, the ACOG, and members of our technical panel, and then grouped these risk factors into nine categories:

  • 1

    Family history of type 2 diabetes

  • 2

    Sociodemographics

  • 3

    Lifestyle factors

  • 4

    Parity

  • 5

    Pregnancy-related factors

  • 6

    Postpartum factors

  • 7

    Measures of anthropometry

  • 8

    Oral contraceptives

  • 9

    Physiological measures

• Sixteen studies evaluated at least one risk factor and reported adjusted measures of association.

• We identified 11 cohort studies that evaluated the relationship between 11 different anthropometric measures and the development of type 2 diabetes; 8 studies reported adjusted measures of association from multivariate models.

• Seven of the eight studies that evaluated anthropometric measures (pre-pregnancy body mass index [BMI], pregnancy BMI, weight, waist-to-hip ratio) using multivariate analysis reported that these measures were positively associated with the risk of developing type 2 diabetes.

• We graded the evidence on anthropometric measures and the risk of developing type 2 diabetes as moderate because of the inconsistency in the anthropometric measures used across the studies.

• We identified five studies that included family history of type 2 diabetes in the multivariate analysis, but only one study reported the actual magnitude of the association of family history with the risk of type 2 diabetes, and this association was not statistically significant (relative risk [RR] = 1.7; 95 percent CI: 0.6 to 4.6).

• We graded the evidence on family history as very low because only one study reported the actual measure of association.

• Five studies assessed age as a risk factor for developing type 2 diabetes among gestational diabetics, but only one study reported the actual measure of association; women who were 30 years of age and older at diagnosis of gestational diabetes had a higher likelihood of developing type 2 diabetes, but the relative risk was not statistically significant.

• We did not identify any studies of lifestyle behaviors that met our criteria for inclusion in this review.

• Gestational age at diagnosis of gestational diabetes was inversely associated with a higher likelihood of the development of type 2 diabetes, but the modeling of gestational age varied across studies and therefore limited our ability to synthesize the data.

• Two studies evaluated the association between the use of progesterone-only contraception or combination oral contraception (estrogen and progesterone) and the risk of developing type 2 diabetes in women with gestational diabetes. One study reported a two-fold increase in the risk of developing diabetes with the use of progestin-only oral contraceptives as compared to combination oral contraception; one study reported no increased risk in women using depo-medroxyprogesterone acetate as compared to combined oral contraceptives.

• FBG, 2-hour glucose value, and the area under the curve from the diagnostic antepartum oral glucose tolerance test (OGTT) were associated with a significantly higher risk of developing type 2 diabetes in women with gestational diabetes.

Key Question 4

What are the performance characteristics (sensitivity, specificity, and reproducibility) of tests for diagnosing type 2 diabetes after pregnancy in patients with a history of gestational diabetes? Are there differences in the performance characteristics of the test results based on subgroup analysis?

• We identified eight studies that reported 10 evaluations of the performance of a reference test versus a comparison (screening) test for the diagnosis of type 2 diabetes in the postpartum period.

• Our review yielded three general comparisons: (1) two different diagnostic threshold values applied to the 75-gm OGTT (the World Health Organization [WHO] 1985 criterion compared with the WHO 1999 criterion), (2) FBG level greater than 7.0 mmol/L (126 mg/dL) (the American Diabetes Association [ADA] 1997) compared to the 75-gm OGTT (WHO 1999), and (3) FBG greater than 7.0 mmol/L (126 mg/dL) (ADA 1997) compared to the 75-gm OGTT (WHO 1985).

• The sensitivity for the FBG greater than 7.0 mmol/L (126 mg/dL) alone as compared with a complete OGTT using the same FBG threshold (FBG greater than 7.0 mmol/L (126 mg/dL) or a 2-hr plasma glucose level after 75-g OGTT greater than 11.1 mmol/L (200 mg/dL) varied across the three studies, ranging from 46 to 89 percent.

• With a threshold greater than 7.0 mmol/L, the FBG had high specificity when compared to the 75-gm OGTT but had highly variable sensitivity.

• No studies included in this review reported measures of reproducibility.

• We graded the strength of the evidence regarding postpartum screening for type 2 diabetes as very low because of the limited number of studies within each category of comparisons and the heterogeneity in the study populations.

Conclusions

Although the overall quality of the evidence was very low, we were able to draw some conclusions regarding treatment options for maternal glucose control, the timing and method of delivery, the risk factors for the development of type 2 diabetes, and the performance characteristics of screening tests conducted in the postpartum period to identify those who have developed type 2 diabetes.

When patients ask about the effect of the use of insulin analogues or glyburide as compared to insulin, clinicians should be aware that little clinical difference has been demonstrated in the infant birth weights associated with the use of these three regimens. Clinicians should also be aware that while the use of an alternative to regular insulin is unlikely to result in any adverse maternal or infant outcomes, there were insufficient data to allow us to determine whether insulin analogues or glyburide are more efficacious than regular insulin in achieving maternal glucose targets. Also, there was no evidence supporting a difference in terms of the prevention of episodes of maternal or neonatal hypoglycemia. To date, only insulin has been approved by the Food and Drug Administration for use in gestational diabetes. Because of the limited data available, it was unclear what glucose thresholds should be used to initiate treatment with insulin, insulin analogues, or glyburide in patients who are being treated with diet alone. Furthermore, there were insufficient data regarding the potential benefits or risk of metformin use.

There was also insufficient evidence to permit us to develop guidelines for elective labor induction or cesarean delivery in women with gestational diabetes. Well-designed clinical trials are needed to provide a stronger base of evidence for the management of gestational diabetes.

Based on multivariate models, measures of obesity appeared to be the strongest risk factor for type 2 diabetes in women with gestational diabetes. We have concluded that there are insufficient data to justify recommending alternative tests to the 75-gm OGTT for the detection of type 2 diabetes in women with gestational diabetes. Further studies conducted in diverse populations and high-risk subgroups and incorporating measures of reproducibility will help to move this area of investigation toward the development of clinically acceptable testing guidelines.

Background and Conceptual Framework

Understanding the risks and benefits of the use of insulins or oral diabetes agents during pregnancy for both maternal and neonatal outcomes is essential to the care of women with gestational diabetes and their offspring.^{28 29} As shown in the conceptual framework (see Figure 3), our objective for Key Question 1 was to review RCTs and observational studies to compare the risks and benefits of medical treatment for the management of glucose levels in women with gestational diabetes. As previously highlighted, pregnancies with gestational diabetes are often characterized by many maternal and neonatal complications, including poor maternal glucose control, cesarean delivery, and neonatal hypoglycemia. Our primary goal was to summarize the maternal and neonatal outcomes across treatment modalities, to derive pooled estimates where possible, and to summarize the relevant conclusions based on the available literature.

Results

Overview and population characteristics of eight RCTs comparing insulins, insulin analogues, and oral diabetes medications. We identified eight RCTs with a total of 845 participants that met our inclusion criteria for review.³⁰–³⁷ Evidence Table 1 describes the primary characteristics of each of the trials. Four of the studies reported study durations of 8 months to 4 years,^{30 34 35 37} while the other four studies did not report a study duration.³¹–^{33 36} The studies were published between 1990 and 2006. For the four studies that reported the starting year of the study,^{30 34 35 37} the earliest starting year was 1985. The trials were conducted in diverse countries and populations: Three trials were conducted in the United States,^{30 32 36} one in Italy,³⁴ one in Finland,³¹ one in India,³³ one in Brazil,³⁷ and one in Israel.³⁵ The trials also compared different treatment interventions: Two clinical trials^{32 33} compared insulin to glyburide; one trial³⁷ compared insulin, glyburide, and acarbose; two studies^{34 36} compared regular human insulin and insulin lispro; one study³¹ compared long-acting and short-acting insulins; one study³⁸ compared insulin administered two-times-daily and four-times-daily; and one study³⁰ compared diet and insulin.

The average maternal age ranged from 25 to 34 years and did not substantially differ across groups. Only three studies reported the racial distribution of the study participants:^{33 34 36} Anjalakshi et al.³³ reported that 100 percent of the study participants were Indian. Most participants (95 percent) in the study by Jovanovic et al.³⁶ were reported as Hispanic. All of the participants in the study by Mecacci et al.³⁴ were reported as Caucasian.

In the studies that reported maternal weight, the weight measures were similar between groups. Five studies^{30 34}–³⁷ reported gravidity, and three studies^{30 32 36} reported the parity of study participants, which ranged from nulliparity to 2.5 prior births.

Consistent with our study selection criteria, each of the eight RCTs reported the test used to diagnose gestational diabetes. Three studies^{31 33 37} used the 75-gm OGTT World Health Organization (WHO) criterion. Two studies^{30 35} used the National Diabetes Data Group (NDDG) criterion, and three studies^{32 34 36} used the 3-hr, 100-gm OGTT with threshold values based on the Carpenter and Coustan criterion. Langer et al.³² used the FBG threshold of 95 mg/deciliter (dL) based on the 100-gm OGTT Carpenter and Coustan criteria to determine eligibility and as the threshold value for treatment with insulin or glyburide. Bertini et al. used a FBG greater than 90 mg/dL or a 2-hr PPG greater than 100 mg/dL as threshold values for initiation of treatment with glyburide or insulin. Anjalakshi et al.³³ initiated medical therapy if the 2-hr PPG was 120 mg/dL or greater after two weeks of nutritional therapy. The average gestational age at screening and diagnosis of gestational diabetes varied across studies from 22 to 28 gestational weeks. Mecacci et al.³⁴ reported a median gestational age at diagnosis of 28 weeks (range: 25 to 32). Polyhonen-Alho³¹ reported a gestational age range of 24 to 28 weeks.

Maternal and neonatal outcomes in eight RCTs of insulin, insulin analogues, and oral diabetes medications. Data were available for abstraction for five of the eight maternal outcomes of interest and 11 of the 13 neonatal outcomes in Key Question 1. As shown in Table 2, data were abstracted on several maternal outcomes, including: (1) average glycemic control (mg/dL), (2) episodes of maternal hypoglycemia, (3) mean difference in maternal weight; (4) cesarean delivery, and (5) episodes of pre-eclampsia. Neonatal outcomes included: 1) infant birth weight, (2) macrosomia, (3) LGA, (4) SGA, (5) hypoglycemia, (6) hyperbilirubinemia, (7) perinatal mortality, (8) respiratory distress syndrome, (9) congenital malformations, (10) birth trauma, and (11) neonatal intensive care admissions.

Conclusions

We found limited evidence on the risks and benefits of oral diabetes agents, insulin analogues, and insulin. The available evidence, to date, suggested little difference in maternal or neonatal outcomes for treatment with oral agents versus any type of insulin, but inconsistencies in clinical outcomes measures across studies and lack of data make it difficult to draw firm conclusions. No studies compared metformin to insulin or other oral agents. Our meta-analysis showed a small, non-significant lower infant birth weight in pregnant women treated with insulin as compared with those treated with glyburide. This small difference of 93 gm is unlikely to have significant clinical relevance. Further studies are needed to determine whether there is a consistent and clinically definable difference in infant birth weight. There appeared to be little difference in various reported measures of maternal glucose control in women treated with glyburide versus insulin (FBG and 2-hr PPG) or in women treated with insulin lispro versus regular insulin (glycosylated hemoglobin and 1-hr PPG). It is unclear whether differences in maternal hypoglycemia are associated with different treatment regimens: Only one study of glyburide and insulin³⁷ defined threshold values for maternal hypoglycemia as part of the study protocol. In one study comparing insulin lispro to regular insulin, maternal hypoglycemia was based on the need for hospitalization rather than threshold glucose values. No available evidence met our inclusion criteria for variations in maternal or neonatal outcomes being based on glucose levels at the initiation of oral agents or insulin. However, as we have indicated above, ongoing investigations, such as the HAPO Study, may provide evidence to suggest threshold values at which clinicians should add oral diabetic agents, insulin analogues, or insulin to diet therapy. The results of the MiG trial should provide evidence regarding the relative benefits and harms of treatment with metformin versus insulin. Finally, additional data regarding congenital anomalies, the long-term consequences of glyburide use, and the effects of metformin transport across the placenta should inform clinical practice and clinical guidelines for the use of oral diabetic agents in pregnancy.

Background and Conceptual Framework

Clinicians use a variety of clinical parameters in their clinical decisionmaking for intrapartum management. Estimates of fetal weight, gestational age, and maternal glucose control are measures of particular importance in pregnancies complicated by gestational diabetes. Clinical management can also be influenced by patient preference and provider perception. Management options include expectant management, labor induction, or “elective” cesarean delivery. In the context of diabetic pregnancies, we refer to “elective” cesarean delivery as a procedure performed following discussion between the patient and clinician, with the goal of avoiding adverse neonatal outcomes such as shoulder dystocia, nerve palsy, or fracture.

Medical institutions have traditionally developed protocols for labor management of women with gestational diabetes, incorporating a combination of anecdotal experience, published literature, and recommendations by national clinical organizations. Both the ACOG and the ADA^{7 13} have provided guidance with regard to labor management of pregnancies complicated by gestational diabetes. The current guidelines, however, are based primarily on retrospective studies that summarize individual hospitals' experiences with maternal and neonatal outcomes. Limitations in the available literature on the management of women with gestational diabetes may have contributed to delays in the development of broadly accepted guidelines for clinical management and to the current variation in practice patterns and clinical outcomes.

Our objective was to conduct a systematic review of the available literature on the effect of EFW and gestational age on maternal and neonatal outcomes in pregnancies involving gestational diabetes. We also focused on the effect of delivery options (i.e., expectant management, induction, and elective cesarean delivery). We developed a conceptual framework to guide the review of Key Question 2, incorporating the key steps in clinical decisionmaking for labor management (see Figure 5). We focused on gestational age and EFW and the potential influence of these measures on options for delivery. Although they are outside the scope of this review, we include contributing maternal and metabolic factors in the conceptual framework, since these are key elements in the broader context of labor management of women with gestational diabetes.

Results

Overview and population characteristics of studies of the effect of labor management on outcomes. Evidence Table 10 describes the characteristics of each of the eight studies that met our criteria for review (see Appendix F). Five studies were conducted in the United States⁵²–⁵⁶ and three in Israel.⁵⁷–⁵⁹ The studies were conducted between 1983 and 2004, and the study periods ranged from 4 to 19 years. We identified one RCT that compared the effect of two labor induction protocols on maternal and perinatal outcomes.⁵⁵ We also identified four observational studies that examined the effect of EFW and/or gestational age on delivery management and outcomes.^{53 57}–⁵⁹ One observational study⁵⁶ compared the effect of labor induction in a class A2 gestational diabetes sample at 38 weeks of gestation to expectant management of a class A1 sample. Gestational diabetes class A1 is managed with diet alone, while gestational diabetes class A2 requires insulin or glyburide in addition to diet to manage glucose levels. One retrospective cohort study compared a trial of labor to repeat cesarean delivery in a sample of women with gestational diabetes and a prior cesarean delivery.⁵² Another study⁵⁴ examined the risk of shoulder dystocia in gestational diabetes patients undergoing a trial of labor.

Outcomes from eight studies. The eight studies identified for this review were heterogeneous with regard to methodology, comparison groups, the time period in which the study was conducted, the length of the study period, the populations included, and the outcome measures of maternal and infant well-being (see Table 5 and Appendix F, Evidence Table 10). Because of the extent of this heterogeneity, we were unable to provide any quantitative synthesis of the literature. We have summarized each study individually, incorporating a summary of the objectives, study design, results, and conclusions presented by the authors. Also, we identify methodological issues that might influence these conclusions. We have categorized our summary of the studies first in terms of study design (RCTs followed by observational studies) and then in terms of the primary exposure (i.e., fetal weight, gestational age, delivery method) under study. The categories we considered were: (1) gestational age and timing of induction, (2) EFW and elective cesarean or timing of labor induction, (3) gestational age or EFW and timing of labor induction, (4) gestational age at delivery, and (5) gestational age and/or EFW and timing of labor induction and/or elective cesarean delivery.

Impact of gestational age on the timing of labor induction. We identified one RCT that addressed the impact of labor induction at term, as compared to expectant management, on maternal and neonatal outcomes in gestational diabetes.⁵⁵ Kjos 1993 recruited 200 women from one tertiary care center. The study sample included 187 women with class A2 gestational diabetes and 13 women with pre-existing (class B non-insulin-requiring) diabetes. Inclusion criteria were clearly stated: good glucose control in at least 90 percent of measured levels, 38 completed weeks of gestation, good compliance with clinic appointments and home glucose monitoring, no antepartum testing abnormalities, singleton gestation with cephalic presentation, EFW less than 3800 gm at 38 weeks with no evidence of fetal growth restriction, no other medical or obstetrical complications, and no more than two previous cesarean deliveries. Women who met the inclusion criteria, agreed to randomization, and had an established diagnosis of diabetes were eligible to participate in the study. Women were randomized to either expectant management or induction of labor at 38 weeks. Of those with pre-existing diabetes, nine were in the active induction group and four were in the expectant management group. The two treatment groups did not differ significantly in terms of maternal age, gravidity, parity, maternal weight, or gestational age at entry into the study. The racial distribution of the study participants was not reported. Gestational age was calculated from the first day of the last menstrual period and adjusted if ultrasound estimation (before 22 weeks) differed from the menstrual age by 10 days or more. Amniocentesis and measurement of the lecithin-to-sphingomyelin (L/S ratio) was used if gestational age could not be accurately determined. Labor was induced with intravenous oxytocin at 38 weeks or in the presence of fetal lung maturity. Vaginal prostaglandin was used for cervical ripening if indicated (Bishop's score less than four) and if the patient had no contraindications to therapy.

Maternal outcomes. Thirty of 100 women in the active induction group had spontaneous labor or cesarean delivery prior to scheduled induction, and 56 of 100 women in the expectant management group required induction or cesarean delivery prior to the onset of labor for medical indications (see Appendix F, Evidence Table 11).⁵⁵

In the final intention-to-treat analysis, there was no difference in cesarean delivery rates between the two groups (25 percent in the active induction group versus 31 percent in the expectant management group; p = 0.43). The average gestational age at delivery in the induction group was 1 week less than the gestational age in the expectant management group (39 weeks versus 40 weeks; p < 0.05).

Neonatal outcomes. Even after adjustment for gestational age at delivery, maternal weight, and maternal age, the average infant birth weight in the expectant management group (3,672 gm; 95 percent CI: 3,595 to 3,749 gm) was significantly greater than that in the active induction group (3,446 gm; 95 percent CI: 3,368 to 3,522 gm; p < 0.01) (see Appendix F, Evidence Table 12).⁵⁵ The proportion of infants with macrosomia, defined as a birth weight of 4,000 gm or more, was higher in the expectant management than in the active induction group (27 percent versus 15 percent; p = 0.05). When defined as a birth weight greater than the 90^th percentile, the proportion of infants with macrosomia was also higher in the expectant management than in the induction group (23 percent versus 10 percent; p = 0.02). There was no significant difference in the number of cases of shoulder dystocia or in the average 5-minute Apgar score between the two groups. Also, there were no episodes of neonatal hypoglycemia requiring treatment and no perinatal deaths in either treatment group.

The findings of this RCT suggested that infants born to women undergoing induction at 38 weeks have significantly lower average birth weights and perhaps a lower risk of macrosomia than do those born to women treated with expectant management. The absence of any difference in cesarean delivery rates suggested that maternal morbidity among women undergoing 38-week induction is similar to that of women undergoing expectant management. The similarity in demographics of the two groups suggested appropriate randomization. Adjustment for key covariates, including gestational at delivery, maternal weight, and age strengthened our confidence in the observed associations.

Impact of EFW on elective cesarean delivery and timing of labor induction. We identified one observational study on the effect of EFW on maternal and neonatal outcomes related to elective cesarean delivery and the timing of induction of labor.⁵³ Conway et al. prospectively followed diabetic women (91.4 percent with gestational diabetes) who were delivered at a tertiary care institution between 1993 and 1995 according to an institutional protocol. Based on this protocol, women with diabetes underwent ultrasonographic estimates of fetal weight between 37 and 38 weeks of gestation. Women whose EFW was greater than or equal to 4,250 gm underwent cesarean delivery; those in whom the EFW was estimated at less than 4,250 gm but considered LGA (defined as 90^th percentile or greater for the gestational age in their population) underwent labor induction. We will refer to this group who delivered between 1993 and 1995 as the study group. Outcomes for this study group were compared to those of a historical control group of diabetic women who delivered between 1990 and 1992, prior to the implementation of the new protocol. The study and control groups did not differ significantly in terms of their mean maternal age, racial composition, gestational age at delivery, or proportion of women with gestational diabetes or pre-gestational diabetes. Twenty-seven percent of the patients in the study group did not undergo ultrasound evaluation.

Maternal outcomes. As shown in Evidence Table 10 (see Appendix F), the authors reported that the average gestational age at delivery was similar for the study group and the historical control group (39.2 weeks versus 39.3 weeks; p > 0.05).⁵³ The cesarean delivery rate, however, was significantly higher in the study group (25.1 percent versus 21.7 percent; p < 0.04) (see Appendix F, Evidence Table 11). The authors suggested that the higher proportion of cesarean deliveries in the study group could be attributed to the implementation of the new protocol. When the elective cesarean deliveries for EFW of 4,250 gm or more (53/343) and cesarean deliveries for failed induction for LGA (7/343) were excluded from the study group, there was no difference in cesarean delivery rate between groups.

Neonatal outcomes. There were significantly fewer macrosomic infants (defined as weighing 4,000 gm or more) in the study group than in the control group (8.9 percent versus 11.6 percent; p = 0.04) (see Appendix F, Evidence Table 12). There was a greater likelihood of shoulder dystocia in the control group (OR = 1.9, 95 percent CI: 1.0 to 3.5) than in the study group. In a subgroup analysis of the macrosomic infants delivered vaginally, there was also a statistically significant greater likelihood (OR = 2.9, 95 percent CI: 1.0 to 8.4) of shoulder dystocia in the control than in the study group.

Based on this prospective, observational study, it appears that in women with gestational diabetes, a protocol involving elective cesarean delivery for macrosomia and induction at 38 weeks for LGA may reduce the number of macrosomic infants and the risk of shoulder dystocia, but it may also be associated with an increase in the number of cesarean deliveries. However, the lack of adjustment for the severity of the diabetes or other potentially confounding variables in this study may have resulted in an overestimate of the effect of the protocol on outcomes. Furthermore, temporal changes in the management of women with gestational diabetes may have also influenced the outcomes reported.

Relationship of gestational age and fetal weight to the timing of labor induction. We identified one cohort study⁵⁸ that examined the relationship of gestational age and EFW to the timing of induction. Lurie et al⁵⁸ prospectively followed a sample of women and compared outcomes with a historical control group in order to determine whether labor induction at 38 to 39 weeks of gestation might reduce the incidence of shoulder dystocia in women with gestational diabetes class A2. The study group (n = 96) was induced at 38 weeks or, if the EFW was greater than 4,500 gm, underwent elective cesarean delivery. The study group was compared to a historical cohort of women (n = 164) who delivered between 1983 and 1989 and in whom labor was induced only if the EFW was greater than 4,000 gm or, if the EFW was greater than 4,500 gm, underwent elective cesarean delivery. This historical cohort was the same study population described by Lurie et al. in an earlier paper,⁵⁷ which will be discussed subsequently. Gestational age was based on the first day of the last menstrual period and serial crown rump measurements in the first trimester. Amniocentesis was performed to assess fetal lung maturity, using the L/S ratio prior to induction. Baseline participant characteristics, including maternal age and parity, were similar between the two groups. There were no reported data on maternal race, weight, or glucose control.

Maternal outcomes. Women in the study group delivered significantly earlier than did women in the control group (38.4 weeks versus 39.2 weeks; p < 0.001) (see Appendix F; Evidence Table 11). A slightly higher proportion of women in the study group than in the control group underwent cesarean delivery, but the difference was not statistically significant (23 percent versus 19 percent; p > 0.05). The rates of vacuum-assisted delivery were similar for the two groups (5.2 percent versus 5.5 percent; p > 0.05).⁵⁸

Neonatal outcomes. Neither infant birth weight nor the proportion of macrosomic infants was significantly different between the two groups (see Appendix F, Evidence Table 12). The proportion of infants with shoulder dystocia was lower in the elective induction group than in the historic control group, but the difference did not reach statistical significance (1.4 percent versus 5.3 percent; p > 0.05). Clavicular fractures, nerve palsies, mortality, and respiratory distress were rare events overall, and their incidence was not significantly different between groups.

Additional analysis. The authors conducted a second analysis in which the outcomes in the study group were compared to those in a subset of the historical cohort of women who delivered after 40 weeks of gestation (n = 62). The proportion of deliveries complicated by shoulder dystocia was significantly reduced (from 10.2 percent to 1.4 percent; p < 0.05) in the study group when compared to this subset of the historical control group. Also, only nine percent of the infants in the study group had a weight greater than 4,000 gm, as compared to 24 percent in the historical control group (p < 0.05).

In summary, the authors of this paper found that the decrease in shoulder dystocia in the study group was only statistically significant if the study group was compared to the subgroup of control patients that delivered after 40 weeks. In addition to a lack of adjustment for severity of diabetes and a consideration of the temporal changes that had occurred in the management of patients with gestational diabetes, this study was further limited by its small population size.

Impact of gestational age at delivery. In their 1992 paper, Lurie et al⁵⁷ conducted a retrospective chart review of all gestational diabetic women who delivered over a 5-year period, examining maternal and neonatal outcomes for women with gestational diabetes class A1 and A2 who delivered after 40 weeks of gestation or prior to 40 weeks. The groups were matched with regard to age, parity, and fetal presentation. Gestational age was based on the date of the last menstrual period and ultrasound measurements of crown rump lengths in the first trimester. Outcomes were reported separately for gestational diabetes classes A1 and A2.

Maternal outcomes. Among women with gestational diabetes class A1 (diet-controlled gestational diabetes), the mean gestational age at delivery was 40.9 weeks for those who delivered after 40 weeks and 38.2 weeks for those who delivered before 40 weeks (p not reported). There were no differences in the numbers of vacuum-assisted vaginal deliveries (0/65 versus 4/65) or cesarean deliveries (7/65 versus 9/65; p = 0.0997) between women delivering after 40 weeks of gestation and those delivering prior to 40 weeks of gestation (see Appendix F, Evidence Table 11).

Similar findings were obtained for the women with gestational diabetes class A2 (insulin-requiring gestational diabetes). The mean gestational age at delivery was 40.5 weeks in the group delivering after 40 weeks and 37.5 weeks in the group delivering before 40 weeks (p not reported). There were no differences in the number of vacuum-assisted deliveries (1/59 versus 4/59) or cesarean deliveries (15/59 versus 13/59; p = 0.6216).

Neonatal outcomes. For women with either class A1 or A2 gestational diabetes, the rate of macrosomia (defined as birth weight greater than 4,000 gm) was higher in the group of women delivering after 40 weeks than in those delivering prior to 40 weeks, but the difference was not statistically significant: for gestational diabetes A1, 24.6 percent in the group delivering after 40 weeks versus 15.4 percent for those delivering before 40 weeks (p = 0.1853); for gestational diabetes A2, 20.3 percent in the group delivering after 40 weeks versus 6.8 percent for those delivering before 40 weeks (p = 0.057) (see Appendix F, Evidence Table 12). The mean birth weights in the two groups were not significantly different in the case of gestational diabetes A1 patients (3,439.00 gm versus 3,617.85 gm; p = 0.0619). However, infants of gestational diabetes A2 patients who delivered after 40 weeks had a significantly higher mean birth weight (3,639 gm) than did infants born to those who delivered before 40 weeks (3,275 gm) (p = 0.0003). There was no significant difference in the rate of shoulder dystocia, birth trauma, neonatal metabolic complications, respiratory distress syndrome, or mortality between the two groups in either population.

In this retrospective cohort study, the only significant difference between patients delivering after 40 weeks and those delivering before 40 weeks was a higher mean birth weight in the subset of class A2 gestational diabetes patients, which was to be expected, given that gestational age is a strong predictor of birth weight. The authors concluded that the timing of delivery does not have a significant impact on clinically important maternal or neonatal outcomes. However, although the authors did perform a stratified analysis for class of gestational diabetes, the study did not adjust for other potential confounders or for delivery management in the groups.

Impact of gestational age and/or EFW on the timing of labor induction and/or elective cesarean delivery. Peled⁵⁹ conducted a protocol-based chart review to evaluate the effect of gestational age and EFW on labor management. In this study, the charts of 2,060 patients with gestational diabetes treated over a 19-year period were abstracted for maternal and neonatal outcomes. The investigators compared four time periods, each with a distinct management protocol for the timing of labor induction or elective cesarean delivery, based on EFW and gestational age and target thresholds for maternal glycemia (Period A: 1980-1989; Period B: 1990-1992; Period C: 1993-1995; Period D: 1996-1999). Gestational age was calculated from the first day of the last menstrual period and confirmed by first trimester ultrasound when possible. EFW was estimated either clinically or by ultrasound. Outcomes among women in Period D (the study group) were compared with outcomes among women in the three prior periods (historical control groups). Women in the study group were induced at 38 weeks of gestation if the EFW was consistent with LGA (defined as greater than 90^th percentile) or underwent elective cesarean if the EFW was greater than 4,000 gm. In Period A, patients underwent elective cesarean if the EFW was greater than 4,500 gm; otherwise, they were induced at 42 weeks. In both Periods B and C, patients underwent elective cesarean delivery if the EFW was greater than 4,000 gm, and they were induced at 40 weeks if LGA was diagnosed. It is noteworthy that the groups differed in terms of the level of glycemic control in the institution's protocol. For patients in Periods C and D, insulin was started at lower fasting glucose levels (> 5.3 mmol/L) and 2-hr postprandial levels (> 6.6 mmol/L) than in Periods A and B (> 5.8 mmol/L and > 7.8 mmol/L, respectively). Furthermore, patients had lower glycemic goals in Periods C and D (< 5.3 mmol/L) than in Period B (< 5.8 mmol/L) or Period A (no goal set). Thus, although glycemic control did not alter decisions regarding delivery, it is important to keep in mind that patients in the four periods differed in terms of their level of glucose control. Prostaglandin E2 gel or tablets was used for labor inductions over the 19-year period of the study. The authors also included both class A1 and A2 gestational diabetes patients but did not report outcomes separately for the two groups. The proportions of women treated with insulin during the four study periods were variable: 13 percent in Period A, 16.4 percent in Period B, 28 percent in Period C, and 32 percent in Period D. There was no other comparison of baseline characteristics (e.g., age, race, parity) in the four groups.

Maternal outcomes. The mean gestational age at delivery was similar for all four groups (between 38 and 39 weeks). The cesarean delivery rate decreased over time, from 21 percent in Period A to 18 percent in Period B and 16 percent in Period C, but it increased to 34 percent in Period D (see Appendix F, Evidence Table 11). A similar increase in the cesarean delivery rate was noted by the author in a concurrent non-gestational diabetes population included in the same study.⁵⁹

Neonatal outcomes. There was a reduction in the proportion of infants with birth weights greater than 4,000 gm (3.86 percent in the study group versus 20.6 percent in Period A, 16.3 percent in Period B, and 11.7 percent in Period C). The proportion of deliveries complicated by shoulder dystocia (none in Period D, versus 1.5 percent in Period A, 1.2 percent in Period B, and 0.6 percent in Period C) also decreased over the study period. Perinatal mortality rates also decreased from 8 percent in Period A to 3 percent in Period B, to 0 percent in Period C, and 0.77 percent in Period D. While p values were reported for comparisons between the gestational diabetes population and the non-gestational diabetes population, they were not reported for comparisons between time periods (the relevant comparison groups for this analysis).

Although this study provides data on a large population of patients with gestational diabetes, the lack of information on baseline characteristics (e.g., age, race, parity, severity of disease) in the four groups and the lack of adjustment for any differences between groups severely limited our ability to draw any substantial conclusions from this study. Also, the authors did not adjust for or discuss the influence of other potential obstetrical management patterns over the 19-year period. Clinical management of diabetic patients had changed substantially over the 19-year period of the study. Modifications in practice patterns have likely influenced the outcomes reported in these investigations. While examining trends in outcomes is useful, it is not possible to fully adjust for changes in practice patterns, leading to some level of bias in the reported associations.

Additional studies. We identified three additional studies that met our initial inclusion criteria but which focused on aspects of labor management that are outside our primary area of evidence review for Key Question 2. Nevertheless, given the paucity of data addressing labor management among women with gestational diabetes, we believe the findings of these studies and their relevance to delivery management deserve limited discussion.

Impact of gestational age on the timing of induction of labor in patients with different levels of disease severity. In a retrospective cohort study, Rayburn examined maternal and neonatal outcomes under an institutional protocol in which class A2 gestational diabetes patients were routinely induced at 38 weeks and class A1 gestational diabetes patients were managed expectantly.⁵⁶ It is important to note that the control group, the gestational diabetes A1 patients who were managed expectantly (n = 137), underwent induction if there were any obstetrical indications for delivery, including pre-eclampsia, gestational hypertension, or poor glucose control; if the cervix was “favorable” at 40 weeks; or if the patient reached 42 weeks of gestation. The authors reported that only 53 percent of patients in the control group required induction, a rate that was significantly different from that in the study group (90 percent, p < 0.001). The gestational age at delivery was significantly different between groups (38.1 weeks in the study group as compared to 39 weeks in the control group, p < 0.001). The study found no differences in the rates of cesarean delivery or shoulder dystocia, macrosomia, respiratory difficulties in the neonate, or neonatal intensive care admissions.

The significant limitation of this investigation is that the study and control groups by definition had different severity levels of disease (class A1 versus class A2). There were also significant differences in the racial composition (the study group was 70 percent Hispanic, versus 60 percent in the control group; p < 0.01) and parity in each group (18 percent were nulliparous in the study group, versus 31 percent in the control group, p = 0.01).⁵⁶

Impact of elective cesarean delivery versus a trial of labor in patients with previous cesarean delivery. Marchiano conducted a retrospective cohort study to examine outcomes related to elective repeat cesarean delivery versus a trial of labor in a population of women with gestational diabetes;⁵² 423 women with class A1 gestational diabetes and singleton pregnancy who had undergone one previous cesarean delivery were included in the study.

The repeat cesarean delivery rate was 30 percent for those who attempted a trial of labor. The rate of macrosomia (defined as infant birth weight > 4,000 gm) for those who attempted a trial of labor was 18 percent, as compared to 33 percent for those who underwent elective cesarean (p < 0.0001) delivery. A sub-group analysis of women who attempted a trial of labor indicated a cesarean delivery rate of 43 percent for those whose infants weighed 4,000 gm or more, as compared to 28 percent for those with infants weighing less than 4,000 gm.

Although these results are relevant to the management of women with gestational diabetes, the results are only generalizable to those with prior cesarean delivery. Furthermore, the authors used actual infant birth weight rather than EFW in the analysis. Because EFW can vary from actual weight at delivery, it is difficult to draw useful conclusions from these results in terms of clinical decisionmaking for elective cesarean delivery versus an attempt at vaginal delivery.

Shoulder dystocia in patients with gestational diabetes. Keller 1991⁵⁴ performed a retrospective chart review of 210 patients with gestational diabetes from a tertiary care center in Chicago. Of the 210 patients, 173 underwent a trial of labor, 34 had elective repeat cesarean delivery, and 3 had an elective cesarean delivery for EFW greater than 4,000 gm (individual patient/provider decision). In those who underwent a trial of labor, the rate of cesarean delivery was 30.6 percent and the rate of forceps use was 4.6 percent. When birth weight categories were examined, the cesarean delivery rate was 33 percent in the greater than 4,500 gm group, 34 percent in the 4,000 to 4,499 gm group, and 29 percent in the 3,500 to 3,999 gm group.

The risk of shoulder dystocia in those patients who delivered vaginally was 12.5 percent overall and ranged from 9 percent in the lowest birth weight group to 14 percent in those weighing 4,000 to 4,499 gm and 38 percent in those infants weighing over 4,500 gm. Fractures and nerve injuries were rare (seven total) and were not related to birth weight category. The study also reported that the risk of shoulder dystocia in patients with class A1 gestational diabetes was not significantly different (OR = 0.78, 95 percent CI: 0.25 to 2.27) from that in patients with class A2 gestational diabetes.⁵⁴

These findings by Keller offer a descriptive analysis of labor outcomes in women with gestational diabetes. Given the lack of a comparison group and any adjustment for confounders, as well as the limited sociodemographic and clinical information on the study sample, it is difficult to draw any reasonable conclusions from this study regarding labor management in women with gestational diabetes.

Quality assessment. We assessed the quality of the single RCT⁵⁵ identified for this review using the Jadad criteria.²² The study reported pre-specified hypotheses, the inclusion criteria were clearly stated, power calculations were presented with effect sizes, the outcomes were clearly described, and adjustment was performed for several potential confounders. However, the methods for randomization, including sequence generation and assurance of allocation concealment, were not clearly described (see Appendix F, Evidence Table 13).

No observational studies met all of our quality criteria (see Appendix F, Evidence Table 13). Of the seven observational studies, only two had pre-specified hypotheses.^{53 58} Of the eight studies, all but one⁵⁹ reported inclusion and/or exclusion criteria, and sampling was consecutive in all eight studies. Outcomes were not clearly defined in two of the studies.^{56 57} Power calculations were only performed in two studies.^{56 58} The analysis was adjusted for potential confounders in two studies.^{52 56}

Limitations. Several limitations to these studies deserve further comment. First, there was heterogeneity in the severity of the gestational diabetes reported in the one RCT and four primary observational studies, making it difficult to assess the magnitude and direction of any association of the effect of gestational age or EFW with labor management. All four of the primary observational studies included women with gestational diabetes A1 and gestational diabetes A2, but only one reported outcomes stratified by insulin requirement.⁵⁷ Furthermore, the RCT⁵⁵ and one of the observational studies⁵³ included pre-gestational diabetics, even though this condition represented only a small proportion of the sample (less than 10 percent). The results of these studies might have varied substantially if the study population had been limited to women with gestational diabetes class A1 or A2 or if the outcomes were stratified by severity.

Second, the four primary observational studies were conducted over a wide timeframe. It is difficult to account for the rise in the prevalence of gestational diabetes during this timeframe or the modifications in physician practice patterns and obstetrical technology that have certainly influenced maternal and neonatal outcomes. For example, while the intention of the study by Peled⁵⁹ was to assess the impact of different management approaches over the 19-year period, it was impossible to discern the potential contribution of changes in glycemic target levels to delivery management over the four time periods.

Third, none of the four primary observational studies adjusted for potential confounders. Therefore, the magnitude of the associations between gestational age or EFW and outcomes may have been overestimated.

Fourth, the high rates of induction of labor in the expectant management group (49 percent) and of cesarean delivery in both groups in the RCT by Kjos et al⁵⁵ illustrate the low threshold for intervention in current practice for patients with diabetes. They also highlight the potential role of medical liability in the design of studies of labor management. Physicians' concerns regarding medical liability, provider perception of risk, and maternal demand for cesarean delivery may limit the ability to conduct well-designed clinical trials of labor management.

Conclusions

One experimental study in this field suggested that active induction of labor at 38 weeks of gestation reduces birth weight, macrosomia, and LGA without increasing the rate of cesarean section. It was difficult to fully assess these outcomes, however, on the basis of a single clinical trial of only 200 patients. The current body of observational studies also suggested a potential reduction in macrosomia and shoulder dystocia with elective labor induction and elective cesarean delivery for macrosomia or LGA infants. We systematically searched the literature for evidence that the choice of timing of induction or elective cesarean delivery resulted in beneficial or harmful maternal or neonatal outcomes, as described in detail in the Key Question. Given the substantial heterogeneity in the studies reviewed and the serious limitations in study design and analysis of the existing literature, we were unable to draw any firm conclusions about the role of elective induction or cesarean delivery in the management of gestational diabetes.

Taking into consideration the quantity, quality, and consistency of the studies comparing the effects of labor management on maternal and neonatal outcomes, we graded the strength of evidence as very low (see Appendix F, Evidence Table 14).

Background and Conceptual Framework

We conducted our systematic review of this question according to the framework outlined in Figure 6. Our objective was to include a range of risk factors that incorporated sociodemographics and pre-pregnancy measures as well as antenatal and delivery factors in both the immediate and long-term postpartum periods. The risk factors included were based on (1) traditional, established epidemiologic and physiologic risk factors for type 2 diabetes and (2) risk factors identified in the literature during our initial review of titles and abstracts.

We identified a number of studies that examined the risk factors for type 2 diabetes among women with previous gestational diabetes. These studies varied widely in terms of their design, population, measurement of risk factors, and method of analysis. No single study included all the risk factors we enumerated. Although longitudinal studies and well-done case-control studies that use multiple regression methods provide the best evidence about the independent contribution of risk factors, we also included studies that used univariate analytic methods if they reported a relative measure of association.

Based on our conceptual model in Figure 6, we grouped the risk factors into the following nine categories:

• 1

  Anthropometry

• 2

  Pregnancy-related factors

• 3

  Postpartum factors

• 4

  Parity

• 5

  Family history of type 2 diabetes

• 6

  Maternal lifestyle factors

• 7

  Sociodemographics

• 8

  Oral contraceptive use

• 9

  Physiologic factors

Results

Overview and population characteristics of 16 observational studies of risk factors for the development of type 2 diabetes. We identified 16 prospective or retrospective/non-concurrent cohort studies that evaluated at least one risk factor in our categories (see Appendix F, Evidence Table 15). However, none of the studies addressed the lifestyle factors depicted in our conceptual model. The studies were conducted in diverse populations and included 10 studies in North America; three studies were conducted in Asia, two in Europe, and one in Australia. Patients were recruited from a hospital or hospital-based clinic in all cases. The followup time for the studies ranged from 6 weeks to 12 years (see Appendix F, Evidence Table 16).

The studies varied with respect to quality (see Appendix F, Evidence Table 17). All of the studies reported inclusion and exclusion criteria, and most stated how the outcome of type 2 diabetes was defined (93.3 percent). Most reported loss to followup (75 percent), with 75 percent of these having a loss to followup of greater than 20 percent. Comparisons between those participants who were successfully followed up and those who were lost to followup were reported in 33 percent of the studies. Only 50 percent of the studies stated pre-specified hypotheses. None of the studies reported power or sample size calculations or the strategy used to handle missing data.

Studies varied in terms of their reporting of the baseline characteristics of the participants. Fifty-six percent of the studies reported the racial makeup of the population, 75 percent reported parity status, and all of them reported the ages of the participants (see Appendix F, Evidence Table 16).

Family history of type 2 diabetes. We identified five studies that evaluated family history of type 2 diabetes as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes (see Appendix F, Evidence Table 18).⁶⁰–⁶⁴ The duration of followup for 102 to 909 participants ranged from 6 weeks to 8 years. All five studies conducted multivariate analyses, but only one study reported a relative measure of association.⁶¹ Cho et al. reported that after adjusting for age, gestational age at diagnosis, pre-pregnancy BMI, FBG at diagnosis, and homocysteine level, women with a family history of type 2 diabetes were more likely to develop type 2 diabetes than were women without such a history (RR = 1.7; 95 percent CI: 0.6 to 4.6), but the relative risk was not statistically significant. Because of the limited data, we were unable to draw firm conclusions regarding the magnitude of the association between a family history of diabetes and the risk of type 2 diabetes in women with gestational diabetes.

Sociodemographics. We identified six studies that evaluated a total of four sociodemographic factors as risk factors for the development of type 2 diabetes in women with previous gestational diabetes (see Appendix F, Evidence Table 19).⁶⁰–^{63 65 66} The four sociodemographic factors examined were age, race, working status, and hospital. The duration of followup for the six samples of 100 to 909 participants ranged from 6 weeks to 11 years.

Age. Six studies⁶⁰–^{63 65 66} assessed age as a risk factor; five of the six studies used multivariate analysis.⁶⁰–^{63 66} Only one study reported the relative measure of association resulting from the multivariate analysis: Cho et al. reported that after adjustment for gestational age at the time of diagnosis, pre-pregnancy BMI, family history of type 2 diabetes, FBG at diagnosis, and homocysteine level, women greater than 30 years of age had a two-fold increased likelihood of developing type 2 diabetes (RR = 2.0; 95 percent CI: 0.68 to 6.0), but this association was not statistically significant, as evidenced by the 95 percent CI that included one.⁶¹ In one univariate analysis, Dacus et al. observed that older age did not appear to be associated with the risk of type 2 diabetes (RR = 0.68; 95 percent CI: 0.24 to 1.9).⁶⁵

Hospital location. Cheung et al. were able to evaluate the hospital attended for antenatal clinic visits as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes, since they had recruited women from two hospitals.⁶⁰ Although they included age, parity, FBG at gestational diabetes diagnosis, BMI during pregnancy, 2-hr OGTT, number of prior gestational diabetes pregnancies, method of glucose control, and family history of type 2 diabetes, these investigators did not report a relative measure of association for the hospital attended and type of diabetes.

Work status. Cho et al. evaluated working status as a risk factor for the development of type 2 diabetes in eight multivariate models, including age, parity, family history of type 2 diabetes, working status, blood pressure, lipid profile, and one of eight measures of adiposity (postpartum BMI, waist circumference, weight, subscapular skin fold thickness, suprailiac skin fold thickness, tricep skin fold thickness, body fat weight, or waist-to-hip ratio).⁶² However, the relative measure of association was not reported for the association between working status and development of type 2 diabetes for any of the eight models.

Race. In a univariate analysis, Dacus et al. evaluated race as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes. They reported that as compared to other race groups, blacks had a 50 percent increased risk of developing type 2 diabetes, but this association was not statistically significant (RR = 1.5; 95 percent CI: 0.45 to 5.0).⁶⁵

We concluded that there are only limited data on which to base any meaningful conclusions regarding sociodemographic factors and the short- or long-term risk of type 2 diabetes among women with gestational diabetes.

Maternal lifestyle factors. We did not identify any studies that examined the relationship between lifestyle factors, such as physical activity and diet, and the development of type 2 diabetes in women with prior gestational diabetes. We therefore concluded that no evidence exists to determine whether maternal lifestyle affects the risk of developing type 2 diabetes after having gestational diabetes.

Parity. We identified four studies that evaluated parity as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes (see Appendix F, Evidence Table 20).^{60 62 66 67} The duration of followup for the samples of 102 to 909 participants ranged from 6 weeks to 11 years. All four studies conducted multivariate analyses, but only two studies reported a relative measure of association for parity with type 2 diabetes.^{66 67} After adjustment for GAD and insulinoma antigen-2 (IA-2) antibody status, method of glucose control, BMI, age, and serum C-reactive protein (CRP) at 9 months, Lobner et al. found that compared to gestational diabetics who were nulliparous, gestational diabetics with more than two previous births had an almost three-fold increased risk of developing type 2 diabetes (relative hazard [RH] = 2.5; 95 percent CI: 1.1 to 5.3).⁶⁶ There was a 20 percent increased risk of developing type 2 diabetes associated with having had one to two previous births, as compared to nulliparity, but this association was not statistically significant (RH = 1.2; 95 percent CI: 0.8 to 1.7).⁶⁶ Metzger et al. evaluated parity as a continuous variable and reported that for each unit increase in parity, there was no statistically significant change in the log odds of developing type 2 diabetes (β = 0.19; p = 0.09).⁶⁷ We concluded that higher parity may be associated with an increased risk for type 2 diabetes among women with gestational diabetes, but further evidence is needed to draw firm conclusions regarding this potential association.

Pregnancy-related factors. We identified nine studies that evaluated seven pregnancy-related factors as risk factors for the development of type 2 diabetes in women with previous gestational diabetes (see Appendix F, Evidence Table 21).^{60 61 63 65 66 68}–⁷¹ These factors were: gestational age at diagnosis, method of glucose control, dosage of bedtime intermediate-acting insulin required, class A2 gestational diabetes (defined as any FBG ≥ 105 mg/dL), previous gestational diabetes, number of prior gestational diabetes pregnancies, 50-gm glucose challenge test (GCT), and spontaneous abortions. The duration of followup for the 88 to 1,636 participants ranged from 6 weeks to 12 years.

Gestational age at diagnosis of gestational diabetes. Five studies^{61 63 65 68 70} assessed gestational age at diagnosis of gestational diabetes as a risk factor, and four of the five studies used multivariate analysis.^{61 63 68 70} The studies varied in terms of their categorization of gestational age at gestational diabetes diagnosis: Two studies divided gestational age at gestational diabetes diagnosis into quartiles and used the first quartile as the reference:^{68 70} Both Kjos et al.⁶⁸ and Schaefer-Graf et al.⁷⁰ reported a protective effect of gestational age at gestational diabetes diagnosis in the fourth quartile as compared to gestational age at gestational diabetes diagnosis in the first quartile, with the effect ranging from a 52 percent to a 65 percent reduction in the likelihood of developing type 2 diabetes (RH = 0.48; 95 percent CI: 0.29 to 0.82; and OR = 0.35; 95 percent CI: 0.23 to 0.54) respectively. Both studies varied with respect to the covariates included in the multivariate model, and they did not share any common covariates. Schaefer-Graf et al.⁷⁰ included FBG at gestational diabetes diagnosis, class A2 gestational diabetes, area under the glucose curve of pregnancy OGTT, previous gestational diabetes and 50-gm GCT, while Kjos et al.⁶⁸ included postpartum OGTT glucose area under the curve, antepartum OGTT glucose area under the curve, and highest antepartum FBG. When third-quartile gestational age at gestational diabetes diagnosis was compared to the first quartile, a smaller protective effect was observed in both studies. Schaefer-Graf et al. reported a 55 percent reduction in the likelihood of developing type 2 diabetes (OR = 0.45; 95 percent CI: 0.27 to 0.76).⁷⁰ Kjos et al. reported a 27 percent reduction in the likelihood of developing diabetes, but this association was not statistically significant (RH = 0.73; 95 percent CI: 0.45 to 1.2).⁶⁸ For both studies, when second-quartile gestational age at gestational diabetes diagnosis was compared to the first quartile, no significant difference in the development of type 2 diabetes was found (Schaefer-Graf et al., OR = 1.1; 95 percent CI: 0.72 to 1.7; and Kjos et al., RH = 0.66; 95 percent CI: 0.39 – 1.1).

Cho et al. categorized gestational age at gestational diabetes diagnosis into two groups, women who were diagnosed with gestational diabetes at greater than or equal to 26 weeks and women who were diagnosed at less than 26 weeks. There was no significant difference in the development of type 2 diabetes between the two groups after adjusting for age, pre-pregnancy BMI, family history of type 2 diabetes, FBG at diagnosis, and homocysteine level (RR = 2.4; 95 percent CI: 0.88 to 6.6).⁶¹

Jang et al. assessed gestational age at gestational diabetes diagnosis as a continuous variable and found that for each week of increase in gestational age at gestational diabetes diagnosis, there was a 0.01 decrease in the log odds of developing type 2 diabetes (β = -0.01; SE = 0.05; p = 0.008).⁶³

In a univariate analysis, Dacus et al. categorized gestational age at gestational diabetes diagnosis into two groups, comparing women who were diagnosed with gestational diabetes at less than 24 weeks and those diagnosed with gestational diabetes greater than or equal to 24 weeks. No significant difference was observed between the two groups in terms of the development of type 2 diabetes (RR = 2.5; 95 percent CI: 0.9 to 6.9).⁶⁵

Method of glucose control. Five studies evaluated the method of glucose control during pregnancy as a risk factor for the development of type 2 diabetes.^{60 65 66 69 71} Three of these studies^{60 66 69} included a multivariate analysis, but only two of them^{60 66} reported a relative measure of association for this risk factor. These two studies varied considerably. Cheung et al. found that as compared to women who did not use insulin, those that did use insulin during pregnancy had a three-fold higher risk of developing type 2 diabetes after adjusting for age, parity, FBG at diagnosis, BMI at index pregnancy, 2-hr OGTT, number of prior pregnancies complicated by gestational diabetes, family history of type 2 diabetes, and hospital location (RR = 3.2; 95 percent CI: 1.6 to 7.0).⁶⁰ Lobner et al. reported that as compared to women who were diet-controlled, women who received insulin during pregnancy had an almost five-fold increased risk of developing type 2 diabetes after adjustment for age, parity, GAD and IA-2 antibody status, BMI during pregnancy, and serum CRP (RH = 4.7; 95 percent CI: 3.2 to 7.1; p < 0.0001).⁶⁶

Two studies included a univariate analysis, but only Steinhart et al. reported a relative measure of association for the method of glucose control. This study reported an almost three-fold increased likelihood of developing type 2 diabetes in women requiring insulin as compared to those not on insulin, but this association was not statistically significant (OR = 2.8; 95 percent CI: 0.8 to 11.2).⁷¹

One study by Cheung et al. examined the required dosage of bedtime intermediate-acting insulin as a risk factor. For each unit (unspecified) increase in dosage, there was a 9 percent increased likelihood of developing type 2 diabetes after adjustment for FBG (RR = 1.1; 95 percent CI: 1.0 to 1.2).⁶⁰ The clinical relevance of this finding, however, is unclear, given that it is based on data from one study and is of borderline statistical significance.

50-gm GCT. The 50-gm GCT is routinely performed during pregnancy as the baseline screening test for gestational diabetes. Only one study evaluated the results of the 50-gm GCT performed during pregnancy as a risk factor for the development of type 2 diabetes.⁷⁰ Schaefer-Graf et al. categorized the GCT results into quartiles, using the first quartile as the reference. They reported that as compared to women with 50-gm GCT results in the first quartile, women with results in the second, third, and fourth quartiles had an increasingly higher risk of developing type 2 diabetes (OR = 2.9; 95 percent CI: 1.2 to 6.6; OR = 3.8; 95 percent CI: 1.7 to 8.5; and OR = 3.5; 95 percent CI: 1.6 to 7.6 for the second, third, and fourth quartiles, respectively), after adjusting for FBG at diagnosis, class A2 gestational diabetes, area under the glucose challenge curve of pregnancy OGTT, gestational age at diagnosis of gestational diabetes, and previous pregnancy complicated by gestational diabetes.

Class A-2 (insulin-requiring gestational diabetes). One study evaluated class A2 gestational diabetes, defined as requiring insulin therapy because of FBG levels greater than or equal to 105 mg/dL, as a risk factor for the development of type 2 diabetes.⁷⁰ Schaefer-Graf et al. reported that as compared to women with gestational diabetes class A1, women with gestational diabetes class A2 were 2.4 times more likely to develop type 2 diabetes, after adjusting for FBG at diagnosis, 50-gm GCT, area under the curve for a pregnancy OGTT, gestational age at diagnosis of gestational diabetes, and previous pregnancy complicated by gestational diabetes (OR = 2.4; 95 percent CI: 1.2 to 4.7).

Previous pregnancies complicated by gestational diabetes. Two studies evaluated previous pregnancies complicated by gestational diabetes as a risk factor for the development of type 2 diabetes.^{60 70} These studies included multivariate analysis, but only one study reported a relative measure of association for this risk factor.⁷⁰ Schaefer-Graf et al. reported that as compared to women without a previous pregnancy complicated by gestational diabetes, those with a such a pregnancy were 60 percent more likely to develop type 2 diabetes, after adjusting for FBG at diagnosis, 50-gm GCT, area under the glucose challenge curve of pregnancy OGTT, gestational age at gestational diabetes diagnosis, and previous pregnancy complicated by gestational diabetes (OR = 1.6; 95 percent CI: 1.1 to 2.5).

Spontaneous abortion. One study that included a univariate analysis examined spontaneous abortion as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes.⁷¹ Steinhart et al. reported that as compared to women without spontaneous abortions, those with spontaneous abortions were 36 percent more likely to develop type 2 diabetes, but this association was not statistically significant (OR = 1.4; 95 percent CI: 0.5 to 3.5).

We concluded that the overall grade of evidence for pregnancy-related factors was moderate.

Postpartum factors. We identified five studies that evaluated a total of four postpartum factors as risk factors for the development of type 2 diabetes in women with previous gestational diabetes (see Appendix F, Evidence Table 22).^{62 64 69 71 72} The four postpartum factors examined were additional pregnancy, breastfeeding, duration of followup, and recurrent gestational diabetes. Duration of followup for the 88 to 909 participants ranged from 6 weeks to 12 years.

Additional pregnancy. Two studies assessed additional pregnancy as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes, and both used multivariate analysis.^{69 72} However, only one study reported a relative measure of association for this risk factor.⁷² After adjusting for postpartum weight change (per 10 pounds), OGTT glucose area, postpartum BMI, and breastfeeding, Peters et al. found that as compared to women with no additional pregnancy, those with an additional pregnancy had a three-fold increased risk of developing type 2 diabetes (RH = 3.3; 95 percent CI: 1.8 to 6.2).⁷²

Breastfeeding. Two studies assessed breastfeeding as a risk factor for the development of type 2 diabetes, and both constructed multivariate models.^{64 72} However, neither of these studies reported relative measures of association for this risk factor.

Duration of followup. One study evaluated the duration of followup as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes and constructed eight multivariate models, involving age, parity, family history of type 2 diabetes, working status, blood pressure, lipid profile, and one of eight measures of adiposity (postpartum BMI, waist circumference, weight, subscapular skin fold thickness, suprailiac skin fold thickness, tricep skin fold thickness, body fat weight, or waist-to-hip ratio).⁶² However, the relative measure of association was not reported for the association between duration of followup and development of type 2 diabetes for any of the eight models.

Recurrent gestational diabetes. One study evaluated recurrent gestational diabetes as a risk factor for the development of type 2 diabetes in women with previous gestational diabetes and conducted a univariate analysis.⁷¹ Steinhart et al. reported that as compared to women without recurrent gestational diabetes, those with recurrent gestational diabetes had a 24-fold increased risk of developing type 2 diabetes (OR = 24.8; 95 percent CI: 3.0 to 1132.2). The width of this confidence interval, however, suggests substantial variability in the point estimate and makes it impossible for us to draw any firm conclusions from these data.

We concluded that the overall grade of the evidence for postpartum factors was very low.

Measures of anthropometry. We identified 11 cohort studies that evaluated a total of 11 different anthropometric measures: weight, height, BMI, body fat weight, subscapular skin fold thickness, suprailiac skin fold thickness, tricep skin fold thickness, waist circumference, waist-to-hip ratio, percent ideal body weight, and weight change (see Appendix F, Evidence Table 23).⁶⁰–^{67 69 72 73} The number of participants ranged from 170⁶¹ to 909.⁶² Followup of participants ranged from 6 weeks to 12 years. Of the 11 studies, 9 reported a relative measure of association.⁶⁰–^{63 65}–^{67 72 73} Eight studies ⁶⁰–^{63 66 67 72 73} reported an adjusted relative measure of association. We have included these adjusted relative measures in Figure 7. One study reported an unadjusted relative measure from a univariate model.⁶⁵ The studies varied in terms of the time period in which the assessment of anthropometry was conducted.

Of the 11 studies, three used pre-pregnancy measures of obesity.^{61 63 73} Two of these studies reported a significant positive association between pre-pregnancy anthropometric measures and the development of type 2 diabetes.^{63 73} One study reported a protective effect of a higher anthropometric measure,⁶¹ and one study did not report the measure of association.⁶³ Pallardo et al. found that as compared to women with a pre-pregnancy BMI less than or equal to 27 kg/m², women with a BMI greater than 27 kg/m² had an eight-fold increased risk of developing type 2 diabetes, after adjusting for the number of abnormal glucose results from the OGTT and C-peptide glucose score (OR = 8.7; 95 percent CI: 2.3 to 32.9).⁷³ Jang et al. reported that for every 1-kg increase in pre-pregnancy weight, there was a 0.36 increase in the log odds of developing type 2 diabetes, although this relationship was not statistically significant (β = 0.36, SE = 0.10).⁶³ One study⁶¹ reported a reduction in the likelihood of type 2 diabetes with higher BMI: Cho et al.⁶¹ reported that as compared to women with a pre-pregnancy BMI less than or equal to 23 kg/m², women with a pre-pregnancy BMI greater than 23 kg/m² were less likely (RR = 0.78; 95 percent CI: 0.27 to 2.2) to develop type 2 diabetes, after adjusting for age, gestational age at diagnosis of gestational diabetes, family history of type 2 diabetes, FBG at diagnosis, and homocysteine level. This reported association, however, was not statistically significant. We concluded that pre-pregnancy measures of obesity are associated with an increased likelihood of type 2 diabetes.

Three of the 11 studies used anthropometric measures during pregnancy. These studies reported a positive association between anthropometric measures and the development of type 2 diabetes.^{60 66 67} For example, Lobner et al. reported that women with a BMI greater than 30 kg/m² were 50 percent more likely to develop type 2 diabetes than were women with a BMI less than 30 kg/m², after adjusting for GAD and IA-2 antibody status, method of glucose control, parity, age, and serum CRP at 9 months (RH = 1.5; 95 percent CI: 1.0 to 2.2).⁶⁶ In addition, Metzger et al. reported that as compared to women who were non-obese (<120 percent ideal body weight), women who were obese (≥ 120 percent ideal body weight) had an almost three-fold increased likelihood of developing type 2 diabetes, after adjusting for 3-hr integrated insulin level and parity.⁶⁷ For each kg/m² increase in BMI, Cheung et al. reported a 10 percent increase in the risk of developing type 2 diabetes (relative risk [RR] = 1.1; 95 percent CI: 1.0 to 1.2), after adjusting for age, parity, FBG at diagnosis, 2-hr OGTT, the number of prior pregnancies complicated by gestational diabetes, method of glucose control, family history of type 2 diabetes, and hospital location.⁶⁰

Five studies⁶²–^{65 72}evaluated anthropometric measures assessed during the postpartum period, but only four studies^{62 64 65 72} reported a relative measure of association. As shown in Figure 7, Cho et al. assessed eight anthropometric measures, comparing women in the highest quartile to those in the lowest quartile.⁶² Each of the eight measures was positively associated with the development of type 2 diabetes, after adjusting for age, duration of followup, parity, family history of type 2 diabetes, working status, blood pressure, and lipid profile, including triglycerides, high-density lipoprotein (HDL) cholesterol, and low-density lipoprotein (LDL) cholesterol. As compared to women in the lowest quartile of postpartum BMI and weight, women in the highest quartile had a three-fold increased likelihood (OR = 3.3; 95 percent CI: 1.7 to 6.5 and OR = 3.1; 95 percent CI: 1.6 to 6.0, respectively) of developing type 2 diabetes. In the same sample, Cho et al. reported that women in the highest quartile were 3.8 times more likely to develop type 2 diabetes (OR = 3.8; 95 percent CI: 1.8 to 7.6) than were women in the lowest quartile of body fat weight. The direction and magnitude of the association with type 2 diabetes were similar across several additional anthropometric measures. As compared to women in the lowest quartile, women in the highest quartile of (1) subscapular skin fold thickness, (2) suprailiac skin fold thickness, and (3) tricep skin fold thickness had a 2.0- to 2.8-fold higher likelihood of developing type 2 diabetes. Women in the highest quartile were over two times more likely to develop type 2 diabetes (OR = 2.8; 95 percent CI: 1.4 to 5.6; OR = 2.1; 95 percent CI: 1.2 to 3.7; and OR = 2.0; 95 percent CI: 1.1 to 3.6, respectively). Also, Cho et al. reported that as compared to women in the lowest quartile of waist circumference and waist-to-hip ratio, women in the highest quartile were over three times as likely to develop type 2 diabetes (OR = 3.9; 95 percent CI: 1.8 to 8.2 and OR = 3.1; 95 percent CI: 1.7 to 5.6, respectively). Two additional studies (Peters et al. and Xiang et al.) assessed postpartum BMI^{64 72} and postpartum weight⁶⁴ in multivariate models but did not report the measure of association. In an unadjusted analysis, Dacus et al. reported a four-fold increased risk (RR = 4.1; 95 percent CI: 0.6 – 29.8) in the development of type 2 diabetes in women with a BMI of 27 kg/m² or greater as compared to women with a BMI of less than 27 kg/m², but this difference was not statistically significant.⁶⁵

Three studies evaluated anthropometric measures as time-dependent covariates, assessing the association of the change in these measures between delivery and followup with type 2 diabetes.^{64 69 72} Peters et al. showed that for every 10-pound change in weight, there was a 95 percent increase in the risk of developing type 2 diabetes, after adjusting for additional pregnancy, OGTT glucose area, postpartum BMI, and breastfeeding (RH = 2.0; 95 percent CI: 1.6 to 2.3).⁷² Although Kjos et al. and Xiang et al. included weight change in their multivariate analyses, the relative association of weight change with type 2 diabetes was not reported.^{64 69} Height was examined in one study, but the measure of association from the multivariate model was not reported.⁶³ Because of multiple cohort studies and measures of association, we graded the overall evidence for anthropometric measures as moderate.

Oral contraceptive use. Two studies evaluated oral contraceptive use and the risk of type 2 diabetes in women with a prior history of gestational diabetes (see Appendix F, Evidence Table 24).^{64 69} Kjos et al. found that as compared to women using a combination oral contraceptive pill, those using a progestin-only pill had a greater than two-fold increased risk of developing type 2 diabetes, following adjustment for the area under the postpartum glucose tolerance test curve, prior oral contraceptive use, method of glucose control, completion of a second pregnancy, postpartum weight loss, and duration of oral contraceptive use (RH = 2.9; 95 percent CI: 1.6 to 5.3).⁶⁹ In that same study, duration of oral contraceptive use was also a significant predictor of type 2 diabetes risk. Women using oral contraceptives for 4 to 8 months and more than 8 months had, respectively, a three-fold (RH = 3.0; 95 percent CI: 1.4 to 6.5) and nearly five-fold (RH = 4.9; 95 percent CI: 1.8 to 13.7) increased risk of developing type 2 diabetes when compared to those with lower-duration use, following multivariable adjustment for the same variables.⁶⁹

Xiang et al. did not find that progesterone-based contraceptives were consistently associated with an increased risk of type 2 diabetes. As compared to women who used combination oral contraceptives, those using depo-medroxyprogesterone acetate did not have an increased risk of type 2 diabetes in the entire cohort, after adjusting for postpartum BMI, breastfeeding, family history of type 2 diabetes, HDL cholesterol, triglycerides, and weight gain during followup (RH = 1.1; 95 percent CI: 0.6 to 1.9).⁶⁴ This association did not differ by breastfeeding status.⁶⁴ However, after adjusting for postpartum BMI, family history of type 2 diabetes, breastfeeding, HDL cholesterol, and weight gain during followup, use of depo-medroxyprogesterone acetate in women with triglycerides above the median of the population was associated with a two-fold greater risk of developing type 2 diabetes when compared to the use of a combination oral contraceptive and triglyceride levels below the population median (RH = 2.3; 95 percent CI: 1.1 to 4.8).⁶⁴ We concluded that the limited number of studies available and the overall very low grade of evidence made it difficult to draw any firm conclusions regarding the relationship between progestin-only contraception and the development of type 2 diabetes among women with gestational diabetes.

Background and Conceptual Framework

The prevalence of type 2 diabetes is increasing in the United States and globally.¹ Early detection and treatment of diabetes has been associated with improved outcomes related to microvascular complications and may prevent macrovascular complications as well.⁸⁸ Women with gestational diabetes are at an increased risk of developing type 2 diabetes. An estimated 16 to 63 percent of women with gestational diabetes will develop type 2 diabetes in the 5 to 10 years immediately following pregnancy.²⁹ While postpartum screening for type 2 diabetes among women with gestational diabetes has been supported by the ADA¹⁷ and ACOG,⁷ there is debate about which screening test to use and at what interval to screen. These are important questions for both clinical providers and public health officials. The fact that only limited evidence is available with regard to screening test performance in women with a history of gestational diabetes has prolonged the debate and perhaps delayed a consensus on appropriate screening. To further define our efforts in addressing this topic, we developed a conceptual framework (see Figure 9). Our model incorporates test performance, as measured by sensitivity, specificity, and reproducibility, as well as the time interval for screening.

Despite the known risk of type 2 diabetes among women with gestational diabetes, only 75 percent of ACOG fellows reported that they routinely perform postpartum screening with the 75-gm OGTT. Followup varies widely, and many women do not receive the recommended screening for type 2 diabetes.^{19 89} The barriers to use of the OGTT include the cost and inconvenience for a new mother. However, there is insufficient evidence supporting the use of an alternative screening test, such as the FBG. A recent cost-effectiveness analysis examined models for screening and found the OGTT to be cost-effective if used at 3-year intervals. Screening with the FBG was cost-effective if used at 1-year intervals.⁹⁰ More precise knowledge of the performance characteristics of these tests may help improve our estimates of the effectiveness and total costs associated with screening.

In this report, we summarize and critically appraise the literature on the performance of currently available screening tests for postpartum glucose screening in order to support the development of clinical guidelines for postpartum glucose surveillance.

Table 6 summarizes the current tests available for postpartum glucose screening and their threshold values.

Results

Overview and population characteristics for screening tests for type 2 diabetes. Our literature search identified 8 studies and 10 evaluations of a reference test and comparison test. Each of the eight studies had a cohort design. Four studies collected data retrospectively.⁹⁴–⁹⁷ Each of these studies retrospectively applied the threshold glucose values of the comparison test to previously collected postpartum OGTT results. These studies used a clinic convenience sample, including all women who returned for postpartum testing within a specified time period. Four studies collected data prospectively.⁹⁸–¹⁰¹ These four studies recruited patients with a history of gestational diabetes for screening for type 2 diabetes. Seven studies used the same OGTT results, but applied different diagnostic threshold criteria. One study¹⁰⁰ independently performed the FBG and OGTT as two separate tests for comparison (see Appendix F, Evidence Table 27).

As shown in Evidence Table 27, the population in two studies^{98 100} was more than 50 percent Caucasian. One study,⁹⁵ performed in the United Arab Emigrates, had mostly Arab (80 percent) subjects. One study¹⁰¹ was performed in the United Kingdom and included participants from three racial/ethnic groups: European, South Asian, and Afro-Caribbean. Three studies,^{94 97 99} including the study performed in the United States,⁹⁷ did not report the racial composition of their study populations. The cohort included in the study by Reichelt et al.⁹⁹ was part of a Brazilian Cohort Study, which previously reported high representation from non-white populations.¹⁰²

The majority of the studies screened for type 2 diabetes within 1 year of delivery.⁹⁴–^{97 100} Two studies^{98 101} reported wide ranges of postpartum testing intervals, from 1 to 86 months and from 6 to 72 months, respectively. Only one study⁹⁹ conducted late screening of all subjects (between 4 and 8 years postpartum).

Overview of studies evaluating comparison and reference tests for type 2 diabetes. Our review yielded three general comparisons: (1) two different diagnostic threshold values applied to the 75-gm OGTT (the WHO 1985 criterion versus the WHO 1999 criterion); (2) FBG level greater than 7.0 mmol/L (126 mg/dL) (ADA 1997) and the 75-gm OGTT (WHO 1999); and (3) FBG greater than 7.0 mmol/L (126 mg/dL) (ADA 1997) and the 75-gm OGTT (WHO 1985).

For each eligible study, two of our investigators abstracted data serially to create a two-by-two table for each comparison test. The two-by-two tables contained data for the number of true positives (TP), false positives (FP), false negatives (FN), and true negatives (TN). We then calculated the sensitivity [# TP/(# TP + # FN)] and specificity [(# TN/ (# TN + # FP)] for each comparison test using the structured two-by-two tables. Since some cells included zero, standard errors and confidence intervals were calculated by means of the exact binomial formula using Stata command “cii” (Intercooled, version 8.2, StataCorp, College Station, TX).¹⁰³ An example of our calculation of the sensitivity, specificity, and standard errors is shown in Table 7, using the study by Costa et al.¹⁰⁰

Key Question 1

What is the evidence for the risks and benefits of oral diabetes agents (e.g., second-generation sulfonylureas and metformin), as compared to all types of insulin, for both the mother and neonate in the treatment of women with gestational diabetes?

Relatively few studies have examined the effect of oral agents or insulin analogues, as compared to insulin, on a number of significant maternal and neonatal outcomes in women with gestational diabetes. Only three RCTs assessing the efficacy of glyburide and insulin met our inclusion criteria,^{32 33 37} and only two maternal outcomes were evaluated in more than one RCT: cesarean delivery and maternal glycemic control. There was little difference in maternal outcomes between those treated with glyburide and those receiving insulin. In the largest trial (n = 404) comparing glyburide and insulin, 49 percent of the women on insulin underwent cesarean delivery, as compared to 46 percent of those on glyburide.³² A second trial³⁷ reported no difference in cesarean delivery rates for 51 women on glyburide, insulin, or acarbose. Three trials found no statistically significant differences in glucose control between women treated with insulin and those receiving glyburide.^{32 32 33} There was one study that considered pre-eclampsia, two studies that included maternal weight, and two studies that included information on maternal hypoglycemia. There were no available data with regard to perineal tears, operative vaginal delivery, or postpartum hemorrhage. Because of the small number of RCTs and the lack of consistency in the maternal outcomes measured across studies, we graded the overall strength of evidence as very low.

Only four neonatal outcomes were evaluated by more than one RCT: birth weight, LGA, macrosomia, and neonatal hypoglycemia. We conducted a meta-analysis of three RCTs with a total of 478 pregnancies. There was a lower average infant birth weight in the insulin group as compared to the glyburide group (-93 gm; 95 percent CI: -119 to 5). This difference was not statistically significant and is unlikely to have substantial clinical influence. We were unable to draw any definitive conclusions regarding neonatal hypoglycemia, given the limited data available. Langer et al.³² reported no significant difference between glyburide and insulin in the percentage of infants with hypoglycemia (9 percent versus 6 percent, p = 0.25), but Bertini et al.³⁷ reported a higher percentage of infants with hypoglycemia in the glyburide group than in the insulin or acarbose groups (33 percent compared to 4 percent and 5 percent, respectively; p = 0.006). Several of our neonatal outcomes of interest were not included in the RCTs reviewed. Therefore, we were unable to draw conclusions about anoxia, birth trauma, respiratory distress syndrome, or shoulder dystocia.

We extended our review of the literature on insulin and glyburide to include four observational studies.⁴⁵–⁴⁸ None of the observational studies were strong enough to justify a modification of the conclusions drawn from the RCTs.

Two RCTs compared the effect of insulin lispro and regular insulin on maternal and neonatal outcomes in women with gestational diabetes.^{34 36} We concluded that there was little difference in maternal glucose control (glycosylated hemoglobin or 1-hr glucose levels) between the women treated with insulin lispro and those treated with regular insulin. Neither Jovanoic³⁶ nor Mecacci³⁴ reported significant differences in mean infant birth weight between the insulin lispro and regular insulin groups. We concluded that no evidence exists to suggest that neonatal outcomes differ between those treated with regular insulin and those receiving insulin lispro. The limited number of trials, limited sample size, and paucity of information on neonatal outcomes made it difficult to draw any firm conclusions.

There was insufficient evidence to draw meaningful conclusions about the effect of long-acting versus short-acting insulin, twice-daily versus four-times-daily use of regular insulin, or diet alone versus diet plus insulin. In one study comparing long-acting to short-acting insulin, there was a higher percentage of infants with macrosomia in the long-acting insulin group than in the short-acting insulin group.³¹ Limited data from one RCT³⁵ suggested that twice-daily insulin may be associated with worse neonatal outcomes (neonatal hypoglycemia, macrosomia, LGA, and SGA) than four-times-daily use of insulin. We found no evidence to suggest a difference in maternal outcomes between twice-daily and four-times-daily use of regular insulin. In the study by Thompson,³⁰ women were randomized to diet alone or diet plus a fixed insulin regimen that included 20 units of NPH insulin and 10 units of regular insulin. There was no reported difference in maternal glucose control or the proportion of women undergoing cesarean delivery in the two groups. In terms of neonatal outcomes, infant birth weight was higher in the diet-alone group than in the diet and insulin group. Similarly, there was a higher proportion of infants with macrosomia in the diet-alone group. These findings must be viewed with caution because the overall strength of the evidence for diet compared to insulin and dietary management was very low.

We did not identify any studies that compared metformin with other diabetes medications in women with gestational diabetes. Also, we found no evidence regarding maternal or neonatal outcomes as related to the level of glucose at the initiation of a medication.

Key Question 2

What is the evidence that elective cesarean delivery or the choice of timing of induction in gestational diabetes results in beneficial or harmful maternal and neonatal outcomes?

There is little evidence on the effect of gestational age or EFW on the timing of labor induction or performance of elective cesarean delivery in women with gestational diabetes. The findings from one experimental study⁵⁵ suggested that active induction of labor at 38 weeks of gestation reduced infant birth weight (3,672 gm versus 3,446 gm; p < 0.01) and rates of macrosomia (27 percent versus 15 percent; p = 0.05) when compared to expectant management, with no concomitant increase in the rate of cesarean delivery (25 percent in the active induction group versus 31 percent in the expectant management group; p = 0.43). While these results suggested that maternal outcomes might be better in women who undergo elective induction, we were unable to draw firm conclusions based on this one trial.

Observational studies⁵²–^{54 56}–⁵⁹ provided some additional evidence of a reduction in macrosomia and shoulder dystocia with elective labor induction or cesarean delivery, when compared to expectant management. For example, in the study by Conway,⁵³ women with diabetes underwent ultrasonographic estimates of fetal weight between 37 and 38 weeks of gestation. Women whose EFW was greater than or equal to 4,250 gm underwent cesarean delivery; those whose EFW was estimated at less than 4,250 gm but LGA (defined as ≥ 90^th percentile for the gestational age in their population) underwent labor induction. Fewer infants were macrosomic (weighing 4,000 gm or more) in the group undergoing elective cesarean or labor induction than in the expectant management group (8.9 percent versus 11.6 percent; p = 0.04). In addition, the incidence of shoulder dystocia was higher in the expectant management group (OR = 1.9, 95 percent CI: 1.0 to 3.5) than in the group undergoing elective cesarean or labor induction. The overall strength of evidence on this comparison was graded as very low. Only one of the observational studies adjusted for potential confounders,⁵⁶ so any measures of association may have been biased. Second, there may have been selection bias in the recruitment of women to participate in the studies. Third, there was substantial heterogeneity in terms of the comparison groups, length of followup, and outcome measures included in the analysis. Fourth, the four primary observational studies were conducted over a wide timeframe. It would be difficult to adequately adjust for changes in practice patterns and treatment modalities that occurred over the long time periods of the studies.

Key Question 3

What risk factors, including but not limited to family history, physical activity, pre-pregnancy weight, and gestational weight gain, are associated with short-term and long-term development of type 2 diabetes following a pregnancy with gestational diabetes?

Several factors were associated with the development of type 2 diabetes in women with previous gestational diabetes. Anthropometric measures before, during, and after pregnancy were found to be positively associated with the development of type 2 diabetes in 10 of 11 cohort studies. Waist circumference and BMI were the strongest anthropometric measures associated with type 2 diabetes in gestational diabetic women. Early gestational age at diagnosis of gestational diabetes (primarily less than 24 weeks) and use of insulin versus diet for glucose control were key pregnancy-related clinical factors that were positively associated with type 2 diabetes. Physiologic measures, including FBG and 2-hr plasma glucose levels during the diagnostic OGTT, were also associated with development of type 2 diabetes. Higher blood glucose following a screening 50-gm GCT, prior gestational diabetes, and OGTT area under the curve during both the antepartum and postpartum periods were positively associated with development of type 2 diabetes, but the strength of the associations was not consistent across studies. There is conflicting data on progesterone-only contraceptive use and the risk for developing type 2 diabetes. Elevated postpartum homocysteine levels were positively associated with type 2 diabetes in one study. Surprisingly, there were no studies of lifestyle factors in women with gestational diabetes that met our review criteria.

After a review of the available evidence, we concluded that the strongest epidemiological risk factors were anthropometric measures prior to pregnancy and during both the antepartum and postpartum periods. Taking into consideration the quantity, quality, and consistency of the studies evaluating the association of risk factors for type 2 diabetes following a pregnancy with gestational diabetes, we graded the strength of the evidence as very low. While there was substantial consistency in the direction of association across studies for many of the risk factors, there was considerable variation in the covariates adjusted for in multivariate models across studies.

Key Question 4

What are the performance characteristics (sensitivity, specificity, and reproducibility) of tests for diagnosing type 2 diabetes after pregnancy in patients with a history of gestational diabetes? Are there differences in the performance characteristics of the test results based on subgroup analysis?

Several studies have pointed to poor physician compliance with postpartum glucose screening for type 2 diabetes among women with a history of gestational diabetes.^{19 20} We reviewed the available studies of the diagnostic accuracy of screening for type 2 diabetes in this population. We identified 8 studies and 10 evaluations of screening tests, with three types of comparisons:

• 1

  two different diagnostic fasting value thresholds applied to the 75-gm OGTT (the WHO 1985 criteria compared to the WHO 1999 criteria);

• 2

  single FBG level greater than 7.0 mmol/L (126 mg/dL) (ADA 1997) compared to the 75-gm OGTT (WHO 1999); and

• 3

  single FBG greater than 7.0 mmol/L (126 mg/dL) (ADA 1997) compared to the 75-gm OGTT (WHO 1985).

For the first comparison, we concluded that there was acceptable specificity (98 percent) for the OGTT using either a FBG value greater than 7.0 mmol/L (126 mg/dL) or greater than 7.8 mmol/L (140 mg/dL). For the second comparison, we were unable to draw meaningful conclusions. The sensitivities for a single FBG greater than 7.0 mmol/L (126 mg/dL), as compared to a complete OGTT using the same FBG threshold, ranged from 46 to 89 percent in the three studies. For the third comparison, there were five studies, which reported a high specificity of the FBG greater than 7.0 mmol/L (126 mg/dL). However, there was a wide range of sensitivity, from 14 to 100 percent.

The six studies⁹⁵–^{98 100 101} that used an FBG threshold greater than 7.8 mmol/L (140 mg/dL) in the reference test may be obsolete, since current guidelines recommend an FBG greater than 7.0 mmol/L (126mg/dL)^{17 93 105} The wide variation in the reported sensitivities for studies that compared the OGTT as the reference test to a single FBG greater than 7.0 mmol/L (126 mg/dL) may reflect differences in the study samples' risk for type 2 diabetes, based on heterogeneity of study design and population. The overall strength of evidence was very low because of the high loss-to-followup rates (22 to 82 percent) for studies using clinic convenience samples.