Recurrence of Gestational Diabetes Mellitus: A systematic review

We searched PubMed for studies published from 1965 until November 2006 using the search strategy “gestational diabetes AND recurrence.” We reviewed the abstracts and then reviewed the articles that met the selection criteria. We also reviewed the reference lists of review articles as well as articles ultimately included in our search and consulted with experts to complete the data search. The search was not restricted to English-language journals.

Entry criteria for the studies included 1) specified criteria for the diagnosis of GDM and 2) rates of recurrence in subsequent pregnancies. When the same study population was examined in several reports (18–20), only the study with the largest number of subjects was selected (6,7). Information abstracted from each report included year of publication, population characteristics including race/ethnicity if reported, diagnostic criteria for GDM, exclusion criteria, number of women with GDM, the average length of time between pregnancies if available, and recurrence rates of GDM and preexisting diabetes in future pregnancies.

In two reports, rates of recurrence of GDM were not calculated through the use of retrospective cohorts (11,16). Instead, by use of a case-control design, the risk of recurrent GDM pregnancy was calculated through an odds ratio (OR) comparing the odds of a previous GDM pregnancy in women with a current GDM pregnancy versus women with a normal current pregnancy. In these instances, the OR rather than the recurrence rate was recorded.

We also abstracted other factors associated with GDM recurrence, including characteristics of the index pregnancy (maternal age, BMI, weight gain during pregnancy, insulin use, glucose levels, gestational age at time of diagnosis, fetal weight or presence of macrosomia, parity, and breast-feeding) as well as characteristics of the subsequent pregnancy (BMI, weight gain during pregnancy, and number of months between pregnancies). No studies reported socioeconomic status. In a few reports, multivariate analysis was used, and the significance of these factors after multivariate adjustment was noted.

Initial searches yielded 45 titles for studies that examined GDM recurrence. By reviewing the abstracts, we excluded articles with no original data or articles that did not examine the recurrence of GDM or diabetes in future pregnancies. The majority of excluded articles focused on recurrence of perinatal complications rather than recurrence for GDM per se or did not report a rate or odds for recurrent GDM.

Overall, 13 studies met the inclusion criteria. Rates of GDM recurrence ranged between 30 and 84% (Table 1). Study cohorts were most commonly women with a history of GDM who had subsequent pregnancies, who received care for the subsequent pregnancy at the same center as the index pregnancy, and who were tested for GDM in the subsequent pregnancy. Therefore, the denominator excluded women who did not become pregnant again, thus reflecting risk in a subsequent pregnancy rather than absolute rates of GDM recurrence. The percentage of women who conceived versus the percentage of women who did not conceive after the index pregnancy was not consistently reported. Philipson and Super (14) noted that 11% of women with index GDM had another pregnancy at that particular center. Gaudier et al. (12) noted 17%, Moses (6) noted 21%, and Major et al. (14) noted 36%.

With two exceptions (8,14), the rate of postpartum diabetes screening after the index pregnancy and its possible effects on the diagnosis of GDM versus diabetes in the subsequent pregnancy were not described. One of these exceptions, a study by Major et al. (14), only included women who had a negative postpartum diabetes screening test after the index GDM pregnancy. The other, a subanalysis of a study by Farrell et al. (8), included only women who had GDM and became pregnant again within 1 year and were unable to complete postpartum diabetes screening after the index GDM pregnancy. Because the average length of time between pregnancies was usually not recorded, it was not possible to plot the incidence of recurrent GDM by length of time elapsed after the index pregnancy. However, it was possible to figure out an upper limit for the length of time between pregnancies when the study was conducted over a brief period and therefore to draw comparisons between rates of recurrent GDM and reported rates of diabetes outside of pregnancy over that time period.

Different testing criteria for GDM were noted across studies, with those of the 1979 National Diabetes Diagnostic Group (NDDG) being used most often (21). These guidelines recommended confirmation of GDM by administration of a 100-g glucose load accompanied by at least two values equaling or exceeding the following cut points: fasting glucose, 105 mg/dl; 1-h glucose, 190 mg/dl; 2-h glucose, 165 mg/dl; and 3-h glucose, 145 mg/dl. No studies used the lower cutoffs that have recently been endorsed (22). Other criteria included those described by the Australasian Diabetes in Pregnancy Society (ADIPS) (23), which recommended confirmation of GDM by administration of a 75-g glucose load and then at least one value equaling or exceeding the following cut points: fasting glucose, 100 mg/dl; and 2-h glucose, 144 mg/dl. These criteria were found to be the most sensitive compared with other testing criteria that also used a 75-g glucose load (24). The Japan Society of Obstetrics and Gynecology recommended a 75-g glucose load and then at least two values equaling or exceeding the following cutoffs: fasting glucose, 100 mg/dl; 1-h glucose, 180 mg/dl; and 2-h glucose, 150 mg/dl (25). Finally, other studies used criteria defined by the perinatal outcomes at that particular center. Unfortunately, the sensitivity and specificity of the 1979 NDDG criteria versus those of other criteria, as defined by perinatal outcomes, have not been compared due in part to the burden of testing that would be required (to compare directly the sensitivity of a strategy requiring a 100-g glucose load versus a 75-g glucose load versus a 50-g glucose load, a woman would need to undergo all three screening strategies).

Despite this variation in testing criteria, studies of non-Hispanic white (NHW) populations showed similar rates of recurrence of <40% (Table 1). The one study of NHW women using the 1979 NDDG criteria (5) had recurrence rates comparable to those of other studies examining NHW women using the ADIPS or center-specific criteria. Of note, in several cases the NHW makeup of the population was assumed rather than explicitly stated, and it was not stated how the rates of recurrent GDM varied by race within the study. However, studies including NHW women who had higher rates of recurrence either had a significant percentage of nonwhite women (9,10) or focused on women who became pregnant within the next year (8). In turn, these women may have had undiagnosed preexisting diabetes or an increased prevalence of other risk factors because of the short interval between pregnancies. In contrast, studies of predominantly minority populations, including African Americans, Latinas, and Asians, had recurrence rates of >50%. One of the Latina studies only included women who had undergone postpartum diabetes screening after the index pregnancy, so the recurrence rate of 69% was probably lowered compared with that in other studies (14).

The significant risk factors for recurrent GDM were inconsistent among studies (Table 2). No single factor was consistently associated with recurrence across studies. However, several of the studies in this review had relatively small cohorts, particularly for racial subanalysis, and lack of significance may be partially attributable to sample size and lack of variation within the particular study population (Table 2). Multivariate logistic regression was not used in most studies, so it is unclear how these factors might change when the interaction with other factors is taken into account. The examination of the significance of factors after adjustment would have been difficult anyway because of the different set of factors included in each study. Moses (6) found that although insulin use was not associated with recurrent GDM, it was associated with reduced odds of a subsequent pregnancy.

One study that was excluded from the primary review because the study population overlapped with that in another study showed a relationship between total fat intake and recurrence of GDM (18). The single study that examined breast-feeding did not find a significant association with recurrence rates (5). No studies examined physical fitness or activity between pregnancies. Other indicators of glucose tolerance, such as β-cell reserve, or risk factors for future diabetes in a mother, such as waist circumference and the presence of other cardiovascular risk factors, were not examined.

Two studies noted the risk of a future pregnancy being affected by preexisting diabetes. MacNeill et al. (5) noted that among 651 women with GDM, 2.4% were noted to have preexisting diabetes in a subsequent pregnancy, although postpartum diabetes screening rates were not reported. In a case-control study, women with GDM had an OR of 55 for a future preexisting diabetes pregnancy compared with women without diabetes, although absolute risks were not calculable (11).

In this systematic review, we found that the recurrence rate of glucose intolerance during subsequent pregnancies varied markedly across studies. The most consistent predictor of future recurrence appeared to be nonwhite race/ethnicity, with the caveat that the racial breakdowns within a study were not always explicit. Minority populations had a markedly higher recurrence of GDM than NHWs, even after preexisting diabetes before the subsequent pregnancy was taken into account (14). Other risk factors, such as maternal age, parity, BMI, oral glucose tolerance test levels, and insulin use inconsistently predicted development of recurrent GDM across studies.

The lower rates of recurrent GDM among NHWs compared with minority populations parallel the lower rates of future maternal diabetes among NHWs compared with minority populations. In a previous systematic review, we found that the rate of future maternal diabetes in women with GDM seemed to be lower among NHW women versus minority women, although direct comparisons were not possible because of the different testing criteria among studies (17). In that report, we also found that rates of future maternal diabetes were similar across different minority populations after adjustment for length of time after the index pregnancy and differing diagnostic criteria.

Although the average length of time between pregnancies was usually not specified by the studies in this report, the study periods sometimes placed a limit on the time between pregnancies. Studies conducted over a relatively short time frame would necessarily place a limit on the period between pregnancies. For these studies, the recurrence rates of GDM exceeded the rate of frank diabetes in other reports for that population within 5 years. Among minority populations, two studies on Latinas reported that the study period included pregnancies over a period of <4 years (13,14). Rates of recurrent GDM were ∼68%. In other reports of Latinas, the rate of maternal diabetes outside of pregnancy at 5 years ranged between 22% (26) and 45% (27). The former study excluded women who had an abnormal postpartum glucose tolerance test, which would lower the rate of diabetes at 5 years. Similarly, the rates of recurrent GDM at ∼35% in NHW women (7) exceeded the rates of frank diabetes in NHW women in other reports (20). Grant et al. (20) noted that NHW women with a history of GDM were tested for type 2 diabetes after delivery ∼40% of the time, and diabetes was subsequently diagnosed in 11% over a period of 12 years. Among Swedish women with a history of GDM, Efendic et al. (28) noted a cumulative incidence of type 2 diabetes of ∼26% at 3 years.

Studies within a particular racial/ethnic group were not possible in our review, although previous studies of incident GDM have suggested significant variation within Asians/Pacific Islanders, with subgroups such as Chinese having a significantly higher risk (29), and within NHWs, with women from Mediterranean countries having a higher risk than their northern European counterparts (30). It is unknown whether the risk for recurrent GDM varies within racial/ethnic group, once women have had glucose intolerance during the index pregnancy.

Although several screening strategies were used, the contribution to variation seemed relatively small. Among the four studies that showed recurrence rates of <40% among NHWs, four different screening strategies were used (4–7). However, when studies that used similar screening strategies and different racial/ethnic populations were performed, recurrence rates were different. This was the case with the study performed by Moses (6), who found recurrence rates of 35% using the ADIPS criteria, versus the study performed by Foster-Powell and Cheung (10), who found recurrence rates of 62% using the ADIPS criteria but whose population also included one-fifth minority women. When studies using similar screening strategies had different upper lengths of time between pregnancies, recurrence rates were also different. This was the case with the study performed by MacNeill et al. (5), who applied the 1979 NDDG criteria and found recurrence rates of 33%, versus the study performed by Farrell et al. (8), who also applied 1979 NDDG criteria but only among women who conceived again within the year of their index pregnancy and found recurrence rates of 84%. It is possible that the variation among studies due to different testing criteria was minimized because the GDM criteria for the index pregnancy and follow-up pregnancy are the same within a study, and therefore recurrence rates among studies were more comparable.

We also found that most studies did not specify postpartum diabetes screening rates between pregnancies or rates of future pregnancies affected by preexisting diabetes, and it is possible that a significant proportion of subsequent GDM pregnancies are actually affected by preexisting diabetes. As a result, the rate of GDM pregnancies might decrease with increased screening for detection of diabetes before a pregnancy, particularly for women who have had previous GDM. Alternatively, the rates of both GDM pregnancies and diabetes pregnancies could be fueled by obesity (31) and maternal age (32), and both might increase. Examination of both GDM and diabetes during pregnancy as linked phenomena would help distinguish which scenario is occurring, along with reports of postpartum diabetes screening rates.

The studies examined showed that the percentages of women with GDM who conceived again at the same center ranged between 11 and 36% (6,9,12,14). Explanations for the relatively low conception rates are speculative. It is possible that women who did not conceive again were discouraged from doing so by their diagnosis of GDM. Alternatively, it is possible that women who did not conceive at the center conceived at a different center. If the former were true, the low percentages suggest that the absolute risk of recurrent GDM for each individual woman with GDM may actually be much lower than the range listed in Table 1. Also, although this is a speculative statement, bias may have been introduced if the women who did not conceive again were different from women who conceived again with respect to factors that could also increase their risk for glucose intolerance, particularly education or psychosocial characteristics associated with family planning (33). Therefore, the overall recurrence rate in the study could reflect the recurrence in a particularly vulnerable group of women.

The high recurrence rates of GDM suggest several lines of translation and future research regarding screening. First, the inconsistency of risk factors for recurrent glucose intolerance suggests that these are not reliable and should not be used to select women for screening for recurrent GDM, postpartum diabetes, or treatment in subsequent pregnancies. Although speculative, it is also possible that the changing demographics of obesity may further undermine risk assessment in these women. For example, as the population becomes obese at earlier ages, women with GDM may have a history of more prolonged insulin resistance, rendering risk predictions based on age less accurate. Second, recommended postpartum screening after the index pregnancy needs to increase from its current suboptimal levels (34). Such screening may result in identification of preexisting diabetes; outcomes in subsequent pregnancies affected by frank diabetes may then be improved by preconception care as well as closer monitoring during the pregnancy (35). Identification of preexisting diabetes in subsequent pregnancies could also be accounted for in studies examining incident GDM. Along these lines, surveillance of women across their lifespan is needed to determine how rates of GDM and rates of preexisting diabetes in pregnancy equilibrate. More specifically, linkage of pregnancy databases to those databases monitoring glucose levels outside of pregnancy would provide better assessment of glucose intolerance during pregnancy. Given that screening for postpartum diabetes occurs at low rates, investigation of alternatives, including earlier screening during subsequent pregnancies, should be considered. Researchers comparing earlier versus later screening would need to weigh the benefit of treating preexisting type 2 diabetes and the presumed smaller proportion of women with preexisting type 2 diabetes versus the inconvenience of repeated testing later in pregnancy if the initial screen was negative, lack of evidence for treating subsequently identified early glucose intolerance, and the greater proportion of these women. Although no reports examined the effects of socioeconomic status, it is possible that these effects contributed to the racial/ethnic differences seen in recurrence rates, and further research is needed in this area.

High recurrence rates also suggest several opportunities for prevention research. The study of interventions on GDM recurrence rates and outcomes of subsequent preexisting diabetes pregnancies is needed. The Diabetes Prevention Program (DPP) showed that lifestyle and pharmacologic interventions were both successful in delaying diabetes among women with impaired glucose tolerance (36), and this finding also applied to the subset of women with a history of GDM (37). The Troglitazone in Prevention of Diabetes (TRIPOD) (38) and the Pioglitazone in Prevention of Diabetes (PIPOD) studies (39) demonstrated that pharmacological intervention may preserve β-cell function among women with a recent history of GDM. Evaluation of the effectiveness of interventions on the rates of GDM recurrence and outcomes of subsequent preexisting diabetes pregnancies would be useful in calculating the cost-effectiveness of such interventions. Such cost-effectiveness assessments should include women's absolute risk of recurrent GDM, i.e., recurrence rates of GDM that also account for the percentage of women who do not conceive again. Second, most pregnancies are unplanned, including those among women with glucose intolerance (33), and preconception counseling is suboptimal (40), but pharmacological interventions pose potential harm to the fetus. Therefore, investigation of the translation of pharmacological interventions to women with a history of GDM of childbearing age who do not engage in family planning is needed. Finally, the DPP found that tailoring of behavioral interventions by race/ethnicity was successful and may also be worth doing for women with GDM, given the greater risk in minorities (36). For women with recent GDM, further tailoring of interventions with the added goal of reducing the risk of a future glucose-intolerant pregnancy may provide added motivation. Women may be responsive to improving their fetus's health even more than their own, as suggested by the high rates of smoking cessation during pregnancy (41).

C.K. was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (K23DK071552).