Gestational diabetes, a risk factor for postpartum type 2 diabetes mellitus (T2DM), threatens women’s health [1–3]. The prevalence of obesity, a known risk factor for gestational diabetes, has increased, and a further increase in the incidence of gestational diabetes and postpartum T2DM is expected [4–6]. The incidence of diabetes in Asians, especially in age groups younger than 40 years, has escalated [4,7,8], and the onset of diabetes at a young age increases the chances of long-term complications and decreased quality of life [9]. Therefore, increased interest in women with gestational diabetes who have a high risk of diabetes is essential to prevent diabetes, particularly in young women.

Previous studies have reported postpartum T2DM incidence rates of 2.6% to 70% in women with gestational diabetes [10–12]. The prevalence of gestational diabetes and incidence of diabetes can differ by race and geographic region, and Asian women are known to have a higher risk of developing postpartum T2DM [1,13,14]. In a recent meta-analysis of T2DM after gestational diabetes, the highest incidence rate (45.96/1,000 person-years) was in Asian women, whereas the incidence rates in four Korean studies ranged from 10.51 to 78.33 per 1,000 person-years [1]. Additionally, the postpartum T2DM incidence rate differs according to the follow-up period, with a relatively higher risk in studies with longer observation periods [3]. Hospital-based cohort studies are limited because of biased recruitment, limited subjects, and drop-outs during follow-up. These may overestimate the incidence of postpartum T2DM since women with pregestational diabetes may be included [11,15,16]. Therefore, it is necessary to conduct a population-based study with a large representative sample of women with gestational diabetes, excluding pregestational diabetes and a long observation period.

Postpartum screening tests and healthy lifestyles are recommended to prevent and manage postpartum T2DM in women with gestational diabetes [17–19]. However, the recommended performance level of health behaviors is often not achieved [20]. The key obstacle for why women with gestational diabetes have suboptimal health behavior performance is because they do not perceive themselves to be at an increased risk of postpartum T2DM [21]. Therefore, identifying individualized risk levels according to individual risk factors is an important first step in the prevention and management of T2DM. When identifying risk factors, the timing of diabetes onset is essential. The risk and risk factors differ in the early and late (greater than 1 year follow-up) postpartum periods [15,22], and defects in beta-cell function, partly explained by genetic predisposition, are more remarkable in early converters [15]. Risk factors of postpartum T2DM across the perinatal period include pre-pregnancy obesity, gestational diabetes diagnostic test results during pregnancy, postpartum metabolism tests results, and obesity [22–27]. Given the potential for prevention strategy and usefulness of postpartum counseling for women with gestational diabetes, it is necessary to focus on late-onset postpartum T2DM and modifiable factors in the postpartum period. Therefore, more comprehensive risk factor assessments, including postpartum fasting glucose level, body mass index (BMI), healthy behaviors, type of breastfeeding, as well as known pre-pregnancy and pregnancy risk factors of T2DM 1-year postpartum, are needed.

In this study, the long-term incidence of postpartum T2DM in postpartum women, including those with gestational diabetes, was assessed using data from the National Health Information Database (NHID) of the National Health Insurance Service (NHIS), South Korea. This study aimed to assess the incidence of diabetes according to the presence of gestational diabetes and comorbid obesity and to identify risk factors, including postpartum factors, for the incidence of T2DM 1 year after delivery.

The cumulative incidence of postpartum T2DM in women diagnosed with gestational diabetes during a follow-up period of up to 14 years (median, 5.9) was 21.3%. This is similar to the cumulative incidence rate of 18.9% reported by Feig et al. [10], using Canadian population-based data for up to 9 years (median, 5.4); 21,823 of 633,449 women who delivered had gestational diabetes. The cumulative incidence rate observed in our study was also higher than the 8.4% reported in an 8-year survey study, using population-based data in Korea, of 2,962 women with gestational diabetes who delivered in 2004 [34]. In another 5-year study of women with gestational diabetes in Korea, a 40% incidence rate of postpartum diabetes was predicted, based on the results of hospital-based prospective cohort studies [11]. This incidence rate was remarkably greater than that in our study. It was observed that the incidence rate tends to be lower in population-based studies with a long follow-up period for a large number of subjects than in hospital-based studies with a short follow-up period for a small number of subjects [1].

The incidence rate of postpartum T2DM was slightly higher in the early postpartum period and increased at a constant rate. The cumulative incidence rate in the first year after delivery was 2.2%. Our finding is lower than the incidence rate of 10% to 15% observed in other Korean hospital-based cohort studies [11,15,35,36], which may not have excluded pregestational diabetes. Failure to exclude pregestational diabetes can lead to an overestimation of the incidence rate in the early postpartum period. By strictly excluding pregestational diabetes in our study, a relatively lower risk of diabetes was observed.

In women with gestational diabetes, a HR of 2.59 to 2.78 was observed after adjusting for other variables. The HR observed in our study is lower than the HR of 3.5 to 4.5 reported in cohort studies on Korean women [11,16], and the relative risk ratios of 7.43 and 7.76 in meta-analysis studies, which included cohort studies on women from various regions or races [2,3]. The studies included in these meta-analyses are highly heterogeneous, such that the relative risk reported in the studies varied from 1.48 to 86.41 [2,3]; thus, generalizing the results to specific regions or populations would be limited. The results of our study differ from those of previous studies that reported that Asian women with gestational diabetes have a higher risk of developing postpartum T2DM in general [13,14]. This finding suggests that it is necessary to assess the incidence risk through population-based studies, considering the differences in specific regions and races.

Fasting glucose level and BMI were identified as the main postpartum risk factors for the incidence of T2DM 1 year after delivery; these may be important for a prevention strategy for T2DM, as they can be monitored and controlled. In this study, the risk of diabetes significantly increased as the fasting glucose level increased, even if the level was below 126 mg/dL, the diagnostic criterion for diabetes. Similarly, in two other prospective studies that followed women with gestational diabetes for 4 and 7 years, respectively, the groups with and without diabetes showed significant differences in fasting glucose at the postpartum baseline, even within the normal range (100.8±14.4 mg/dL vs. 90.0±9.0 mg/dL, P<0.05; 97.9±2.5 vs. 86.0±0.9, P<0.05) [37,38]. Tura et al. [38] reported that the HR for each SD increase in fasting glucose was 2.99 (95% CI, 1.96 to 4.57). These findings suggest that the fasting glucose level can be used as an indicator of diabetes risk within the normal range before reaching the diagnostic level. Additionally, we observed that women with both obesity and gestational diabetes had a much higher risk of developing postpartum T2DM, indicating that they require more attention to prevent diabetes. BMI at postpartum was a significant risk factor for postpartum T2DM in this study, and the HR per 1 SD for postpartum BMI was 2.19 (95% CI, 1.54 to 3.13) in another study [38]. Previous studies reported that pre-pregnancy, pregnancy, and postpartum weight were significant risk factors for postpartum T2DM [26], and Cho et al. [39] reported that postpartum obesity parameters, such as waist circumference, body fat weight, BMI, and weight, significantly increased the risk. Longitudinal studies showed that postpartum weight gain or BMI change increased the risk even further after adjusting for postpartum weight or BMI [40,41]. In a study of 418 Korean women, Moon et al. [40] reported that the incidence of diabetes increased to 8.6%, 12.6%, and 16.9% (P=0.039) along three groups categorized by postpartum BMI change (the lowest tertile [mean −1.8 kg/m²], the middle tertile [mean −0.2 kg/m²], and the highest tertile [mean 1.6 kg/m²], respectively) [40]. Women in the highest tertile had an approximately 2-fold increase in risk compared to those in the lowest tertile after adjusting for initial postpartum BMI and age [40]. Bao et al. [41] reported through a study with over 18 years of follow-up that there was a 27% increase in the risk of T2DM for every 5 kg increase in weight after gestational diabetes (adjusted HR, 1.27; 95% CI, 1.04 to 1.54). Even the HR was elevated to 43.19 (95% CI, 13.60 to 137.11) among women who had a ≥30 kg/m² BMI at baseline and ≥5 kg weight gain after gestational diabetes as compared to women with a <25 kg/m² BMI and <5 kg weight gain [41]. Therefore, this evidence suggests that postpartum weight management may be an effective prevention strategy that can lower the risk of postpartum T2DM.

This study had issues related to the identification of gestational diabetes and postpartum T2DM subjects. First, it was difficult to confirm the method and diagnostic criteria used for the diagnosis of gestational diabetes because the NHID data do not include oral glucose tolerance test (OGTT) results. Referring to Korean ‘Treatment Guidelines for Diabetes’ during the study period [42], universal screening at 24 to 28 weeks for gestational diabetes and two-step approach (if over 140 mg/dL [over 130 mg/dL in case of high risk] in 50 g OGTT, and then 100 g OGTT) were recommended and diagnostic criteria were changed from the National Diabetes Data Group (NDDG) criteria (100 g OGTT, at least 2 of the following: fasting ≥105 mg/dL, 1-hour ≥190 mg/dL, 2-hour ≥165 mg/dL, and 3-hour ≥145 mg/dL) to the Carpenter-Coustan (C-C) criteria (100 g OGTT, at least 2 of the following: fasting ≥95 mg/dL, 1-hour ≥180 mg/dL, 2-hour ≥155 mg/dL, and 3-hour ≥140 mg/dL) in the 3rd guideline of 2007. Since the 4th guideline of 2011, one-step approach with the International Association of the Diabetes and Pregnancy Study Group (IADPSG) criteria (75 g OGTT, at least 1 of the following: fasting ≥92 mg/dL, 1-hour ≥180 mg/dL, and 2-hour ≥153 mg/dL) was also recommended in addition to a previous two-step approach. The prevalence of gestational diabetes is affected by the applied diagnostic criteria. The prevalence of gestational diabetes increased by 50% to 60% when the C-C criteria were applied instead of the NDDG criteria [43,44], and it was twice as high when the one-step approach with IADPSG criteria was applied compared to the two-step approach with the C-C criteria [45]. Therefore, it is necessary to consider applied diagnostic criteria and their impacts on the prevalence of gestational diabetes and further incidence of postpartum T2DM. To consider the impact of changes of diagnostic criteria over time and multiple recommended criteria, our study compared the incidence risk by matching subjects with the same year of delivery. Nonetheless, careful interpretation of our results is needed due to the lack of information on applied criteria to each subject for the diagnosis of gestational diabetes. Second, verification of T2DM by diagnosis or drug codes may be different from the real incidence. Studies for validation of identification of diabetes using diagnosis codes showed that the sensitivity and specificity changed according to the operational definition [46,47]. More complex criteria for the identification of diabetes than more simplified ones were related to relatively lower sensitivity and higher specificity [46,47]. The level of sensitivity and specificity could affect the incidence estimation: low sensitivity leads to underestimation through false negatives, and low specificity leads to overestimation through false positives. Thus, an appropriate condition of an operational definition is required in consideration of the sensitivity and specificity of the tradeoff relationship. Identification of diabetes through two or more relevant diabetes ICD codes, as one criterion in this study, showed fair validity (sensitivity 88.4% to 92.3%, specificity 96.9% to 99%, positive predictive value [PPV] 77.2% to 94%) [47–49]. Lipscombe et al. [47] recommended diagnosis code ‘AND’ drug/fee code combination (“one hospitalization or physician claim for diabetes AND either one prescription for an anti-diabetic medication or one physician claim with a diabetes-specific fee code” [sensitivity 84.2%, specificity 99.2%, PPV 92.5%]) as the best definition than diagnosis code ‘OR’ drug code combination (“one hospitalization or 2 physician claims for diabetes OR one prescription for an anti-diabetic medication within 2 years” [sensitivity 90.7%, specificity 97.5%, PPV 98.9%]). Although there may be concerns about the possibility of overestimation due to the ‘OR’ combination used in this study, it is assumed that overestimation was not serious enough to concern because the incidence rate in our study was lower than that of the clinical studies and similar to that of other population-based studies [10].

This study minimized the effects of recruitment bias and drop-outs during follow-up and long-term follow-up by using a large-scale nationally representative sample of female subjects. Additionally, this study tried to overcome the limitations of heterogeneous or unconfirmed regions or races observed in other studies by using homogenous Korean national data (above 95% estimated as Korean). Moreover, subjects with pregestational diabetes were strictly excluded by confirming that there was no diagnosis of diabetes before inclusion in the study (using data until index pregnancy since 2002, which ranged from 1 to 13 years depending on the subject’s inclusion date). The control group was established by matching the gestational diabetes group. The effects of confounding variables were adjusted to accurately assess the risk of postpartum T2DM. Furthermore, comprehensive factors were included in the assessment of risk factors during the postpartum period to help develop strategies that can prevent postpartum T2DM.

There were some limitations to this study. As a retrospective cohort study, we could not preplan subjects’ identification and data measurements and had limited control of them [50]. First, there are concerns about accuracy and validity in the identification of subjects through diagnosis or drug codes in the NHID. For example, there may be inaccuracy while entering the diagnosis code, and the medication may be prescribed for purposes other than for diabetes treatment. For these reasons, in this study, postpartum diabetes was defined based on diagnosis codes that were entered twice over different treatment days. For the hypoglycemic agent metformin, considering the possibility that it could be used for other purposes, postpartum diabetes was only defined when the drug and the diabetes diagnosis codes existed together. However, discrepancies between recorded codes and actual subjects’ medical diagnoses could still exist. Second, due to the incompleteness of existing data, there is a limit to the availability of important variables. For example, clinical information on disease severity such as OGTT values at diagnosis, treatment and control of glucose levels during pregnancy, and recurrence of gestational diabetes was not available. Moreover, hemoglobin A1c (HbA1c) of 6.5% (48 mmol/mol) or higher may be used to diagnose diabetes [32,33]; however, HbA1c was not one of the health screening items of the NHID and could not be used in our study.

This population-based cohort study assessed the risk and risk factors for postpartum T2DM in women with gestational diabetes. The findings of this study may be used to increase the risk awareness of subjects and provide motivation for prevention actions by notifying the subjects with more accurate individualized risks. Additionally, results of health examinations may also be used to identify groups with a high risk of diabetes in clinical settings and increase the likelihood of active intervention or early diagnosis.