Influence of Gestational Diabetes Mellitus on Diabetes Risk and Glycemic Control in a Retrospective Population-Based Cohort

Gestational diabetes mellitus (GDM), a complication of glucose intolerance with first recognition in pregnancy, is strongly linked with later-life type 2 diabetes (1–3). Alarmingly, rates of GDM and type 2 diabetes continue to rise. Between 2016 and 2020, the rate of GDM increased nationally by 30% from 6.0 to 7.8 per 100 births, while the rate of diagnosed diabetes rose 6% over a similar period (4,5). Both conditions are marked by persistent racial/ethnic disparities and are leading risk factors for cardiovascular disease and premature mortality (6,7). Uncontrolled diabetes may further increase the risk of cardiovascular comorbidity and underlie later-life disparities attributed to cardiovascular disease (8,9). Despite this, existing studies of the influence of GDM have focused on type 2 diabetes incidence (1–3). In contrast, comparatively few studies have examined the influence of GDM on glycemic control, once diagnosed.

An understanding of how race/ethnicity and GDM interact to influence diabetes risk and control also remains incomplete. While some evidence suggests that associations between GDM and type 2 diabetes risk differ by race/ethnicity, clarity of the understanding is limited by several factors (10–12), including lack of consideration of multiple GDM pregnancies over the life course, potential bias in who receives diabetes screening, lack of examination of differences by race/ethnicity or inadequate sample sizes for subgroup analysis, and use of broad racial/ethnic categorizations (1–3,5). Furthermore, while some retrospective evidence has shown that well-controlled glycemia during pregnancy reduces both the risk of GDM recurrence and type 2 diabetes, differences in reversion from diabetes onset to normoglycemia by race/ethnicity warrant examination (13,14). The effects of systemic racism, which impede access to care, promote poor nutrition and sedentary lifestyle, and contribute to biological “weathering” because of the accumulated effects of stress, may underlie racial/ethnic differences in diabetes risk and control (15,16). Such pathways can induce chronic inflammation and higher levels of oxidative stress, which may facilitate pathophysiological differences in the metabolic characteristics that lead to type 2 diabetes following GDM (15,17). To inform effective prevention and treatment strategies in the postpartum period, further elucidation of how GDM history and race/ethnicity interact to influence diabetes risk and control is critical.

To address these gaps, we constructed a diverse, population-based cohort of postpartum women in New York City (NYC) with the creation of the A1C in Pregnancy and Postpartum Linkage for Equity (APPLE) cohort. Our main objective was to estimate racial/ethnic group differences in diabetes risk and control, once diagnosed, by GDM status.

The APPLE retrospective cohort was constructed by linking birth certificate data for all births in NYC from 2009 through 2011 with NYC A1C Registry data from 2009 to 2017 (Supplementary Table 1). NYC has mandated laboratory reporting of A1C results since 2005 (18). Hospital discharge data from the Statewide Planning and Research Cooperative System (SPARCS) were subsequently linked with birth record data for the same period to strengthen measures for pregnancy-related comorbidities, including GDM and prepregnancy and gestational hypertensive disorders. SPARCS data contain billing codes for all discharge diagnoses and procedures for hospitalizations in New York State. The Data Matching Unit of the Bureau of Vital Statistics in the NYC Department of Health and Mental Hygiene performed data linkages using a probabilistic matching algorithm using identifying patient information.

Women with known diabetes before the start of the second trimester were excluded (ICD-9 codes 250.X, 362.01, 362.02, 363.04–363.07, and 366.41 or birth certificate indication of prepregnancy diabetes). The date of the second trimester was calculated using gestational age via the birth record (19). Of the cohort who remained at risk to develop type 2 diabetes (n = 336,276), 56% of the birth cohort received an A1C test. An additional 4,685 unique mothers (1.4%) had A1C records that indicated incident diabetes during postpartum observation (Supplementary Fig. 1). Glucose control was measured among those with incident type 2 diabetes who had at least one additional A1C test result following the second confirmatory A1C test for diagnosis (n = 4,430). Prior to data linkage, institutional review board approval was obtained by the NYC Department of Health and Mental Hygiene, the New York State Department of Health, and the Icahn School of Medicine at Mount Sinai. Note, data were not available on participant gender. We use the term women to apply to any person who is pregnant or who has delivered a child.

Incident diabetes was defined as the second date of an A1C test with a value of ≥6.5% (18). This algorithm does not distinguish between type 1 and 2 diabetes. We henceforth refer to the risk of diabetes in reported results. However, we expected >95% of cases to be type 2 diabetes, as this is the most common form of diabetes among adults (20). Glucose control was defined as an A1C value <7.0% following postpartum diabetes diagnosis (14). To distinguish from the resolution of GDM and diabetes onset, study observation began at 12 weeks postpartum for all analyses. Survival time was calculated as the difference between the date at 12 weeks postpartum and diabetes diagnosis (in years). Among those with postpartum-onset diabetes, survival time was calculated as the difference from diabetes diagnosis and the date of the first A1C value <7.0%.

We classified GDM at the baseline or subsequent pregnancy using birth record data or as indicated on hospital discharge records (ICD-9 codes 648.01–648.04). This method produces high sensitivity with few false-negative results (21). Race/ethnicity was self-reported via the birth certificate and categorized as Asian, non-Hispanic Black, Hispanic, non-Hispanic White, and other/unknown. We further subdivided Asian-specific estimates into East/Central and South/Southeast Asian (22) given that these are the two largest prevalence Asian subgroups in NYC and evidence of a higher type 2 diabetes prevalence among South/Southeast Asian people (12). We consider race/ethnicity to be a social construct, representing group social relations and intergenerational experiences of racism (23).

We obtained maternal sociodemographic characteristics and health behaviors from the birth certificate, including age (5-year age-group), nativity (foreign born vs. U.S. born), educational attainment (less than a college degree vs. college graduate or higher), public or no insurance versus private, whether antenatal care was received during the first trimester, and alcohol, tobacco (prior 3 months), or drug use during pregnancy (yes/no). Maternal health history, including parity (0, 1, 2, or ≥3 prior births), multiple gestation pregnancy, and prepregnancy BMI (measured as kg/m²) were obtained from the birth record.

Adverse birth outcomes were included as markers of underlying poor health, which manifest during pregnancy and have the potential to influence the risk for both GDM and type 2 diabetes (24), including macrosomia (birth weight >4,500 g) identified on the birth certificate, preterm delivery calculated using gestational age via the birth record (19), and shoulder dystocia identified using hospital diagnosis codes (ICD-9 codes 660.4, 600.41, and 600.43). We also adjusted for both preexisting hypertension and hypertensive disorders of pregnancy given the common co-occurrence and similar etiology between diabetes and hypertension (25). To maximize measure sensitivity (26), we used the following definition for preexisting/chronic hypertension (ICD-9 codes 401.X–405.X and 642.0X–642.2X or chronic hypertension indicated on the birth certificate). Total eclampsia (ICD-9 codes 642.4X–642.6X and 642.7X) or eclampsia indicated on the birth record and gestational hypertension (ICD-9 code 642.3X or birth record indication) were combined into a hypertensive disorders of pregnancy category given moderate correlation (r = 0.62).

We estimated race-specific rates of diabetes and glycemic control by GDM status over 9 years of follow-up. Model estimation was first performed using the Kaplan-Meier method, using the log-rank test for differences in the survival function between groups (27). Next, as women without GDM during their baseline pregnancy could develop the condition in a later pregnancy, time-varying exposure Cox regression models were used to estimate the hazard of developing diabetes by GDM status, adjusting for the above-defined confounders. Survival models with a time-varying exposure are not considered proportional hazards models and do not require this assumption (28). Among women who developed diabetes after 12 weeks postpartum, we assessed time to glycemic control by GDM status using the same methods. An adjusted hazard ratio (aHR) >1 implies that GDM facilitates glycemic control, and an aHR <1 implies that GDM reduces its likelihood. We assessed effect measure modification to test whether the association between GDM and diabetes outcomes differed by race/ethnicity, using tests for interaction (see conceptual model in Supplementary Fig. 2). Given evidence of interaction (P < 0.001) in both models of diabetes risk and glycemic control, we stratified all models by race/ethnicity. To assess changes in the magnitude of effects over time, we also stratified models by postpartum time interval (0–1 years, 1.1–3 years, 3.1–5 years, 5.1–9 years). In post hoc analysis, differences in average A1C levels at diagnosis were examined by GDM status and race/ethnicity using two-sample t tests. Analyses were performed using SAS Enterprise version 7.15 (SAS Institute, Cary, NC) statistical software.

Given that we do not have information on whether pregnancies prior to the start of study observation were complicated by GDM and to further assess the potential effect on future GDM pregnancies on diabetes risk, we conducted several sensitivity analyses, which included restricting the sample to 1) no previous live births, 2) no previous live births and no subsequent pregnancies, 3) GDM as assigned at baseline pregnancy, and 4) multiparity with GDM only at a pregnancy subsequent to baseline.

Systematic differences in postpartum diabetes screening or loss to follow-up such that A1C results are not captured within the A1C Registry may bias estimated associations between GDM and diabetes risk and control. Representation in the A1C registry by GDM status and race/ethnicity is reported in Supplementary Table 1 and Supplementary Fig. 3. We adjusted for differential representation in the A1C Registry and loss to follow-up by GDM and race/ethnicity using probabilistic bias analysis (see Supplementary Appendix 2 for a detailed description).

A total of 336,276 women with linked vital statistics and SPARCS records gave birth between 2009 and 2011 and had no previous diabetes diagnosis. Of these, 4,685 (1.4%) had a diabetes diagnosis by the end of the 9-year follow-up. Characteristics of the birth cohort by representation in the A1C Registry are shown in Supplementary Tables 1 and 2. The cumulative incidence of diabetes was 11.8% (95% CI 11.4, 12.3) and 0.6% (95% CI 0.62, 0.67) among women with and without GDM at their baseline pregnancy, respectively. The overall prevalence of GDM at baseline was 6.7% (95% CI 6.6, 6.8). A small proportion (2.5%, n = 8,418) of women without GDM at baseline pregnancy had GDM at a future pregnancy, raising the cumulative prevalence to 7.9% (95% CI 7.8, 8.0). At baseline, GDM was highest among South/Southeast Asian (19.8%) and lowest among White (4.2%) women (Table 1). Of those with incident diabetes, 70.4% (95% CI 69.0, 71.7) achieved glycemic control during at least one follow-up observation. A slightly lower proportion of women with GDM (69.3%; 95% CI 67.5, 71.1) than without GDM achieved glycemic control (71.8%; 95% CI 69.8, 73.9). Glycemic control following diabetes was most likely among East/Central Asian (81.9%) and South/Southeast Asian (78.3%) women, followed by Black (70.6%), White (68.3%), and Hispanic (67.1%) women.

Figure 1A and B illustrate the Kaplan-Meier plot of cumulative risk of diabetes among postpartum women stratified by GDM status and race/ethnicity. The aHR for the risk of GDM on developing diabetes was 11.5 (95% CI 10.8, 12.3) (Table 2). Tests for interaction between GDM and race/ethnicity in the adjusted model showed evidence of effect measure modification (P < 0.001) (Supplementary Fig. 4), with differences in the risk of diabetes between South/Southeast Asian women (aHR 7.7; 95% CI 6.1, 9.6) and White women (aHR 12.5; 95% CI 10.0, 15.6) or Hispanic women (aHR 12.2; 95% CI 10.9, 13.5). The risk of GDM on diabetes was similar among Black (aHR 10.3; 95% CI 9.2, 11.5), Hispanic (aHR 12.2; 95% CI 10.9, 13.5), and White (aHR 12.5; 95% CI 10.0, 15.6) women. Racial/ethnic differences in diabetes risk were attenuated in sensitivity analyses (Table 3), although South/Southeast Asian women maintained a lower risk than Hispanic and White women. Main effects also persisted in a sensitivity analysis among primiparous women and those with no future pregnancies in the study observation period (Supplementary Table 3). In post hoc analysis, we found no mean differences in A1C levels at the time of diabetes diagnosis by GDM status overall. Mean A1C level at the time of diabetes diagnosis did, however, differ by race/ethnicity. South/Southeast Asian women with or without GDM had significantly lower A1C levels than Black, Hispanic, and White women. Black and Hispanic women with GDM had significantly higher A1C levels than Asian women (Supplementary Table 4).

The cumulative risk of glycemic control among respondents with postpartum-onset diabetes by GDM status and race/ethnicity is shown in Fig. 1C and D. The unadjusted average time to glycemic control was 2.0 (SD 2.1) years and 1.6 (SD 1.7) years among those with versus without GDM at baseline pregnancy, respectively. In adjusted analyses, GDM reduced the likelihood of attaining glycemic control (aHR 0.85; 95% CI 0.81, 0.93) (Table 2). When stratified by postpartum time interval, GDM was associated with the least likelihood of attaining glycemic control in the first 12 weeks to 1 year postpartum (aHR 0.88; 95% CI 0.80, 0.97) and had a nonsignificant influence at all later postpartum intervals. There was evidence of an interaction between GDM and race/ethnicity on time to glucose control (P < 0.001) (Supplementary Fig. 5). In stratified models, GDM was associated with a negative influence on glycemic control among Black women (aHR 0.77; 95% CI 0.68, 0.88) and Hispanic women (aHR 0.84; 95% CI 0.74, 0.95). GDM had a nonsignificant negative influence on glycemic control among White (aHR 0.90; 95% CI 0.70, 1.17) and South/Southeast Asian (aHR 0.96; 95% CI 0.76, 1.23) women and a nonsignificant positive influence among East/Central Asian women (aHR 1.45; 95% CI 0.95, 2.20). Adjustment for loss to follow-up by GDM status and race/ethnicity showed little influence on effect estimates. In sensitivity analyses (Supplementary Table 3), main effects also persisted among primiparous women and those who had no subsequent pregnancy to baseline.

In this population-representative cohort of postpartum women (the APPLE cohort), we found that GDM across multiple pregnancies was associated with >11 times the risk of diabetes. To a lesser extent, this association also varied by race/ethnicity. Most notably, women with the highest GDM incidence (South/Southeast Asian) were characterized with the lowest postpartum diabetes risk. Critically, among those with postpartum-onset diabetes, we found that a history of GDM was associated with a lower likelihood of glycemic control. In particular, Black and Hispanic women with GDM experienced a longer time to achieve glycemic control compared with those without.

Our findings confirm prior estimates of diabetes risk attributed to GDM while building on limited existing evidence of racial/ethnic differences in the influence of GDM across multiple pregnancies (1–3). Three prior meta-analyses compared type 2 diabetes risk following GDM among White, non-White, or mixed ethnicity subgroups and documented either no association or effects in mixed directions (1–3). True measures of association, however, may be obscured by use of broad racial/ethnic characterizations. Similar to our findings among South/Southeast Asian women, two prior observational cohort studies found a disproportionate burden of GDM among Asians but a smaller incremental effect of GDM on diabetes risk (11,12). In contrast to an earlier health plan cohort study, we document a slightly lower, although substantially elevated, effect of GDM on diabetes risk among Black women (10). Plausible mechanisms for discrepancies across studies are differences in the distribution of social and structural factors that contribute to pathophysiological differences leading to the development of diabetes (16). For example, progressive deterioration of β-cell function following GDM has been shown to be critical to the development of type 2 diabetes (29). Chronic inflammation and oxidative stress induced by an obesogenic environment and the cumulative effects of structural racism and disadvantage leading to biological weathering have both been shown to contribute to β-cell damage (15,30). These mechanisms warrant further investigation and may represent one pathway through which racial/ethnic disparities in diabetes pathogenesis manifest (17,31).

A major contribution of this study is new evidence on the influence of GDM on time to glycemic control among women with postpartum-onset diabetes, which to our knowledge has not been previously investigated. The finding that, on average, those with GDM have a lower rate of glycemic control following diabetes onset (aHR 0.85; 95% CI 0.79, 0.92) is consistent with previous evidence of elevated later-life cardiometabolic risk following GDM (13,14). The results also give weight to the importance of achieving glycemic control in the 1st year following diabetes onset. Early control has been associated with lower risk of future diabetes-related complications and mortality, even after adjusting for glycemic control in later life (32). The implications of early glycemic control underscore that the first 12 weeks following pregnancy is a critical transition period, where timely intervention and monitoring can lessen the severity of, if not prevent, diabetes (33). In vivo and in vitro studies have suggested that with short duration type 2 diabetes, β-cell failure is potentially reversible in response to weight loss and other lifestyle changes (34). In NYC, residential resources are highly segregated by race/ethnicity (35). While further study is warranted, it may be that greater neighborhood disadvantage, in part, hinders health-promoting lifestyle changes and contributes to the lower likelihood of glycemic control following GDM (29).

Finally, we found that the 1st year postpartum also corresponds with the largest negative influence of GDM on glycemic control among women with incident diabetes, which is consistent with our finding that the rate of diabetes onset was highest in the first 12 weeks to 1st year postpartum, which is supported in prior literature (2,36). It is possible that the high rate of transition in the first 12 weeks to 1 year postpartum could reflect undiagnosed diabetes prior to pregnancy and underscores the potential importance of first trimester screening. Excess risk of the 1st year postpartum may also, in part, reflect the key challenges of this period (37). Missed opportunities (e.g., lack of access to care, low rates of provider visits, lack of screening by pregnancy history among women who do see a provider) likely compound racial/ethnic disparities in type 2 diabetes onset and severity (37). Policies such as extended Medicaid postpartum coverage, care coordination between obstetric and primary care, and provider education on the importance of obstetric history taking are essential in facilitating diabetes awareness and early glycemic control (38,39).

This study has several strengths, including the use of a longitudinal population-based cohort with 9 years of follow-up and high diversity by racial/ethnic subgroup, which was 70% non-White. This cohort is unique in that it includes repeated A1C measures in a mandatory, citywide A1C registry. However, it is possible that some individuals are not captured in the A1C Registry (e.g., relocation from NYC, care outside of NYC). Furthermore, data on diet, exercise, family history of type 2 diabetes, pharmaceuticals, and lifestyle behaviors were not available. These variables may account for some diabetes risk associated with GDM in our analysis and should be further explored in future analyses. We note that diabetes diagnosis and glycemic control are also tied to disparities in access to screening services and treatment. While we adjusted for socioeconomic and access-to-care indicators to limit confounding, residual bias may remain. Furthermore, while we used hospital diagnosis codes to strengthen measures of clinical comorbidity where possible, all measures are subject to misclassification. Notably, while the majority of incident cases are likely type 2 diabetes (20), it was not possible to distinguish between type 1 and type 2 diabetes using A1C Registry data. Bias analysis results are based on race-specific estimates of the rate of undiagnosed diabetes reported in Supplementary Tables 7 and 8 (5). Furthermore, we classified the distribution of missed diabetes cases according to GDM status by the empirical risk of diabetes in each exposure group (Supplementary Tables 7 and 8). Bias in health care screening and loss to follow-up are common threats in longitudinal health cohorts, and this study models how to consider these biases and their potential magnitude of influence on study conclusions.

In conclusion, using a diverse, population-based cohort of postpartum women in NYC, the APPLE cohort, we demonstrated that the influence of GDM on postpartum diabetes risk was elevated among all racial/ethnic groups, although least so among groups with the highest GDM incidence. We found that GDM reduces the likelihood of early glycemic control, a stage of disease comparatively less explored in relation to GDM, particularly among Black and Hispanic women. Greater understanding of how GDM influences various stages of diabetes progression and among which groups this diagnosis confers the highest risk are critical steps in elucidating the mechanism through which life course cardiometabolic disparities manifest. A GDM diagnosis is a powerful early warning signal for interventions to improve the immediate risk environment and alter the course and speed of disease progression in years to come.

Acknowledgments. The authors thank Katherine Aherns (Muskie School of Public Service, University of Southern Maine) for guidance on generating probabilistic bias analysis estimates.

Funding. This study was funded by National Institutes of Health grants R21DK122266 and R01DK134725.

The National Institutes of Health had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.J.M. analyzed the data and wrote the first draft of the manuscript. S.H.L. advised on the analysis and reviewed and edited the manuscript. M.H. and E.A.H. advised on the study design and reviewed and edited the manuscript. J.K. and H.T.C. conducted the data linkage and reviewed the manuscript. V.L.M. contributed to the discussion and reviewed and edited the manuscript. L.V. advised on the analysis, contributed to the discussion, and reviewed and edited the manuscript. B.T. provided the data, advised on the study design, and reviewed and edited the manuscript. F.H. provided project management and reviewed the manuscript. G.V.W. and A.L. reviewed and edited the manuscript. T.J. conceptualized and oversaw the study, contributed to the discussion, and reviewed and edited the manuscript. K.J.M. and T.J. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in poster form at the Society for Epidemiologic Research Conference, Chicago, IL, 14–17 June 2022, and the Society for Pediatric Epidemiologic Research, Chicago, IL, 13–14 June 2022.