Kidney Disease in Women With Previous Gestational Diabetes Mellitus: A Nationwide Register-Based Cohort Study

Insulin resistance (IR) increases as a physiological response during normal pregnancy, and insulin production increased accordingly to maintain normoglycemia (1). Gestational diabetes mellitus (GDM) develops when the insulin response is insufficient due to preexisting, underlying IR in combination with impaired β-cell function (2,3). The metabolic dysfunction may be subclinical or unknown and, hence, first disclosed by a GDM diagnosis (3).

GDM treatment includes lifestyle interventions and, in some cases, medication—primarily insulin—to reduce the risk of obstetric complications related to hyperglycemia (e.g., preeclampsia, cesarean section, preterm delivery) (4).

GDM is associated with an increased risk later in life for diabetes (5), hypertension (6,7), cardiovascular disease (CVD) (7,8), metabolic disease (7,9), and psychiatric morbidity (10). Diabetes and hypertension predispose to the development of kidney disease (11,12); therefore, it seems plausible to speculate about a link between GDM and kidney disease. However, relatively few studies have investigated this. A systematic review and meta-analysis from 2020 reported an association between GDM and chronic kidney disease (CKD) among Black but not White women (13). Recent cohort studies from Sweden, Australia, and Canada have shown increased risk of CKD and end-stage kidney disease (14,15) and renal dialysis (16). However, these findings were only significant in women who had developed diabetes subsequently. A Danish clinical study found increased estimated glomerular filtration rate in women with previous GDM (17) equal to glomerular hyperfiltration seen in prediabetes and prehypertension (18).

IR is evident in people with GDM (3) and CKD (19) and may be the underlying mechanism linking GDM and CKD. Insulin treatment for GDM may be regarded as a proxy of a more severe metabolic dysfunction of underlying IR and β-cell insufficiency. In a small Hungarian study, researchers used insulin treatment as a proxy for GDM severity and reported that increasing severity of GDM was associated with increased vascular dysfunction 3 years postpartum (20). Theoretically, women with severe metabolic dysfunction during pregnancy (expressed as insulin-treated GDM) may constitute a specific high-risk group with a substantially higher risk of CKD.

Kidney disease is a major global health problem that significantly affects quality of life, morbidity, and mortality (21). An investigation of the possible link with previous GDM, therefore, offers a valuable contribution to the understanding of a substantial health concern. Risk factors for kidney disease include preeclampsia (22), preterm delivery (23), diabetes (11), and hypertension (12,24). These conditions are more prevalent in women with GDM. The mediating effects of subsequent development of diabetes and hypertension in the association between GDM and kidney disease have not been investigated and quantified, to our knowledge, and thus represent gaps in the existing literature clinically relevant to address. Additionally, the risk of acute kidney disease after GDM has not been examined, as far as we are aware. An explorative investigation into this will provide fuller insight into kidney disease after GDM.

Therefore, our overall aim was to investigate the long-term risk of kidney disease after GDM in a Danish population of women giving birth between 1997 and 2018. The objectives were to investigate 1) the association between GDM and incident kidney disease, 2) the mediating effects of subsequent diabetes and hypertension, and 3) the impact of increasing severity of metabolic dysfunction during pregnancy on the risk of incident kidney disease.

This nationwide cohort study was based on data from national registers containing information on all residents in Denmark and stored at an individual level. We used the Danish Medical Birth Register with data on all births in Denmark since 1973 (25) and the National Patient Registry with data on all ICD-10 diagnosis codes from hospital contacts (outpatient, inpatient, and emergency) since 1995 (26). Data on prescribed medication came from the Danish National Prescription Register (27), whereas demographic and socioeconomic data were collected from other relevant registers (28–30). For diagnosis codes, definitions, and categories, see Supplementary Table 1.

The study population consisted of women who gave birth in Denmark between 1 January 1997 and 31 December 2018. The exclusion criteria were preexisting diabetes, kidney disease, CVD, autoimmune or immune disease, and coagulopathies or hemoglobinopathies at or within 2 years before the index date (defined as the conception date in the first pregnancy during the study period). The flowchart is illustrated in Supplementary Fig. 1. Each woman was represented by at least pregnancy during the study period. Women with preexisting hypertension were omitted from the study population in a sensitivity analysis.

GDM exposure was based on diagnosis codes. In Denmark, screening for GDM is selective and based on risk factors (namely, previous GDM, glucosuria, BMI ≥27 [calculated as kg/m²], family history of diabetes, previous delivery of a neonate weighing ≥4,500 g, polycystic ovary syndrome [PCOS], multiple pregnancy). Screening is performed using a 2-h 75-g oral glucose tolerance test during gestational weeks 24–28. Previous GDM, glucosuria, and/or presence of two or more risk factors warrant additional screening in gestational weeks 10–20. GDM was diagnosed at a 2-h glucose level of ≥9 mmol/L (venous plasma or capillary blood) or ≥10 mmol/L (capillary plasma) (31). Because GDM exposure may vary from pregnancy to pregnancy of each of the included women, GDM was handled as a time-varying exposure to ensure correct classification of each pregnancy and subsequent risk time.

The outcomes were composites of CKD and acute kidney disease, including different subcategories based on diagnosis codes (Supplementary Table 1). Each woman may be represented in more than one outcome category because different outcomes may develop over time.

Follow-up commenced 6 weeks after index delivery, which was considered the end of the immediate postpartum period. Contribution of risk time continued until outcome, emigration, death, or end of study period, whichever came first. In case of additional pregnancies during the study period, time from conception date to 6 weeks postpartum was excluded from the total risk time. Risk time was categorized as exposed or unexposed depending on GDM exposure status in the preceding pregnancy (i.e., this categorization was subject to change). However, if GDM was diagnosed, all subsequent risk time was considered as exposed.

Data on potential confounders were derived from the index pregnancy and included maternal age, parity, preexisting comorbidities (including a comorbidity score based on the Charlson Comorbidity Index) (32), ethnicity, marital status, income, education, occupation, and calendar year of delivery. Furthermore, data regarding pregnancy-related hypertensive disorders and preterm delivery were collected. A large amount of missing data for pregestational BMI and smoking during pregnancy was expected because registration of these data did not commence until late 2003 and 1997, respectively (25). Therefore, these covariates were not included in the main analysis although they were included in sensitivity analyses. We generated the following proxy for increasing severity of metabolic dysfunction during pregnancy, based on GDM diagnosis and insulin treatment: no GDM; GDM with no insulin treatment; and GDM with insulin treatment. The latter category represented the most severe metabolic dysfunction.

Subsequent development of diabetes and of hypertension after GDM but before development of kidney disease was considered a mediator in the association between GDM and kidney disease. Definition of the mediators was based on diagnosis codes and/or prescribed medication (Supplementary Table 1).

Baseline characteristics were compared using the Wilcoxon rank sum test for continuous variables and the χ² test for categorical variables. Using Cox regression models, we estimated incidence rates with corresponding 95% CIs and crude and adjusted hazard ratios (aHRs) for incident kidney disease. By clustering on each woman, the model specification accounted for the potential repeated measurements in case of multiple pregnancies during the study period for each woman. The proportional hazards assumption was tested by the Schoenfeld residuals. In case of nonproportionality of the covariates, we included interaction terms between the covariates and the time period in the final adjusted model. We imputed the mean value of gestational age at delivery for the missing data (<2%).

The mediating effects of subsequent development of diabetes and hypertension in the association between GDM and incident kidney disease were investigated through mediation analyses in accordance with the method suggested in a study by Yu et al. (6), following the approach originally suggested by VanderWeele (33). In this approach, the total effect (TE) of an association is decomposed into a controlled direct effect (CDE) obtained by adjusting for the potential mediator and a proportion eliminated effect (PE) by this adjustment. Because of the hazard ratio (HR) scale being estimated on the log-hazard scale in the Cox regression model, this results in the equation HR_TE = HR_PE × HR_CDE, corresponding to the equation HR_PE = HR_TE/HR_CDE used to obtain our estimate for HR_PE. From this, we determined the PE itself by PE = (HR_TE − HR_CDE)/(HR_TE − 1), following Yu et al. (6). In this process, HR could be interpreted as an approximation of a relative risk in the case of reasonably rare outcomes (34) and, hence, using the mediation analysis approach for relative risk suggested by VanderWeele (33). The proportion mediated effect was then calculated from the TE and the PE. The potential mediators were included as constants (e.g., not time-dependent covariates). Bootstrapping with 100 replicates was used to obtain the corresponding 95% CIs for the proportions.

The impact of increasing severity of metabolic dysfunction stratified according to subsequent development of diabetes and hypertension before development of kidney disease was investigated by including the proxy variable as interaction term with subsequent diabetes or hypertension, respectively, in the adjusted Cox regression models.

Statistical analyses were performed using Stata 17 software (StataCorp, College Station, TX). A P value <0.05 was considered statistically significant.

A series of sensitivity analyses was performed as described in Supplementary Material. Briefly, we applied further restrictions for the GDM exposure and the study population (e.g., excluding women with preexisting hypertension). Finally, we performed adjustments for additional confounders (e.g., BMI, pregnancy-related hypertensive disorders) (Supplementary Table 4).

According to Danish law, ethical approval was not required. This study was approved by the Danish Data Protection Agency (journal no. 19/11440).

Because Danish legislation restricts access to individual-level data, the data set is accessible only after special authorization granted by the Danish National Health Data Authority.

During the study period, 758,978 women were registered with at least one birth. Because of the exclusion criteria, data on 61,356 women were excluded from this study (Supplementary Fig. 1). Thus, the study population comprised 697,622 women, including 23,710 (3.4%) with GDM in at least one pregnancy. Furthermore, 12.8% of women with GDM received insulin treatment. Women diagnosed with kidney disease from the index date to 6 weeks postpartum were not considered at risk during follow-up and, therefore, were censored from the Cox regression models (i.e., these analyses included 694,178 women). The follow-up period was 0–21.9 years; the median was 11.9 years (interquartile range 5.4, 17.8) for CKD and 12.0 years (interquartile range 5.4, 17.8) for acute kidney disease. The cumulative incidence of subsequent diabetes in women with and without GDM was 19.9% and 2.0%, respectively, whereas the cumulative incidence of subsequent hypertension was 15.7% and 10.9%, respectively.

Women with GDM had significant differences in multiple characteristics compared with women without GDM (Table 1). Compared with women without GDM, those with GDM had higher pregestational BMI (median 27.2 kg/m² vs. 22.8 kg/m²), were more likely to have preexisting health-related morbidities, and were less likely to be of Danish ethnicity (79.4% vs. 88.1%). Additionally, women with GDM had lower income levels (32.9% vs. 31.4%) as well as higher income levels (35.2% vs. 34.3%), lower educational levels (24.3% vs. 19.3%), and lower employment rates (68.5% vs. 72.0%) than women without GDM. Moreover, obstetric complications occurred more often in women with GDM.

Women with GDM had significantly higher risk of CKD compared with women without GDM (aHR 1.92; 95% CI 1.67–2.21) (Table 2). This risk was constituted by a higher risk within the CKD subcategories: glomerular and proteinuric disease, unspecified CKD, and diabetic kidney disease. The risk of acute kidney disease was unrelated to GDM history; hence, results regarding acute kidney disease are not reported in the other tables and figures. The results of the sensitivity analyses are described in Supplementary Material.

The test of the proportional hazards assumption in the model test of the Cox regression analysis indicated a potential risk of violation regarding the association between exposure and outcomes. Hence, we computed a Kaplan-Meier curve that indicated 2 years after pregnancy as a relevant cutoff of time periods in which to investigate potential differences in aHRs. A Cox regression analysis was performed by stratifying the risk time into two periods and computing the aHRs for each of these periods (Supplementary Table 5). This showed that the aHRs were insignificant 0–2 years after pregnancy (except for diabetic kidney disease), whereas from 2 years and beyond, the aHRs and significance levels corresponded to those of the main analysis reporting the overall aHRs during the complete follow-up period.

As shown in Table 3, CKD after GDM was significantly mediated by subsequent diabetes and hypertension by 75.7% (95% CI 61.8–89.6) and 30.3% (95% CI 25.2–35.4), respectively. Significant mediation was evident regarding the CKD subcategories glomerular and proteinuric disease, and diabetic kidney disease only. Subsequent diabetes and hypertension mediated the association between GDM and glomerular and proteinuric disease by 35.8% and 37.5%, respectively, whereas the association between GDM and diabetic kidney disease was mediated by 96.3% and 31.5%, respectively.

Figure 1 illustrates the risk of CKD according to severity of metabolic dysfunction, stratified for subsequent diabetes and hypertension. The figure shows increasing risk with increasing metabolic dysfunction severity independent of diabetes and hypertension.

This nationwide cohort study based on register data of ∼700,000 women with almost 22 years of follow-up showed that GDM was associated with increased risk of CKD regardless of subsequent development of diabetes and hypertension. The association was, however, partially mediated by subsequent diabetes and hypertension. Moreover, the CKD risk increased with increasing severity of metabolic dysfunction during pregnancy.

We found that women with GDM had an almost twofold higher risk of CKD compared with women without GDM, whereas the risk of acute kidney disease was unrelated to GDM history. The significantly elevated CKD risk was observed from 2 years after pregnancy and beyond, and was evident despite the cohort being relatively young. The largest study to date exploring CKD after GDM is a Swedish register-based study including data on >1.1 million women (15). Although the outcome definitions varied slightly, the CKD risk was comparable in the Swedish study and in our study (aHR 1.81 vs. 1.92, respectively). An Australian cohort study found increased risk of CKD and end-stage kidney disease (5- and 10-fold, respectively) in Aboriginal women with previous GDM (14). The risk was substantially higher than in our cohort, and this might be related to more deprived health-related circumstances among Aboriginal women (14). The study populations in our study and the Swedish study (15) comprised primarily White women, thus indicating a significant association between GDM and CKD in White women specifically. This finding contrasts the conclusion in the systematic review by Barrett et al. stating that an association was only evident in Black women (13).

A cohort study from Israel including ∼100,000 women found a threefold higher risk of hypertensive kidney disease after GDM (35). The Swedish study reported a 2.5-fold increased risk of hypertensive kidney disease after GDM (15). Surprisingly, we did not observe any association between GDM and hypertensive kidney disease. Our previous study (7) demonstrated a twofold higher risk of incident hypertension (1.5-fold [95% CI 1.4–1.6] after adjustment for pregestational BMI) in women with previous GDM, and we expected this to be reflected in an increased risk of hypertensive kidney disease. The divergent findings between our study and the studies from Israel (35) and Sweden (15) are difficult to explain because all three studies excluded women with preexisting risk factors, had long follow-up periods, adjusted for confounders, and used hospital diagnosis codes. However, different coding practices and differences in GDM screening, definitions, and prevalence may explain the discrepancies.

A prospective cohort study from the United States including 37,716 women (1.5% with GDM) found higher rates of microalbuminuria (10.0% vs. 7.7%) and CKD (adjusted odds ratio 1.54; 95% CI 1.16–2.05) in women with previous GDM compared with normoglycemic women (36). We did not investigate microalbuminuria, because these data are not available in the registers. Obesity may seriously confound associations between GDM and CKD. However, after adjustment for pregestational BMI in our sensitivity analysis, GDM maintained the significant association with CKD (aHR 1.30; 95% CI 1.02–1.65) (Supplementary Material). We believe this finding is novel.

Importantly, our study is the first, to our knowledge, to show an association between GDM and CKD not mediated by subsequent development of diabetes and hypertension prior to CKD. The significant associations between GDM and CKD reported in previous studies were restricted to women with subsequent diabetes development (14–16), and the role of subsequent hypertension has not been addressed previously. Our finding contrasts with previous findings despite these being based on large study populations and the use of hospital data (14–16). The divergence may be explained by our analytical strategy of allowing additional pregnancies by each woman during the study period. This assured detailed insight into GDM exposure status throughout the period. The Swedish study applied a similar strategy (15); however, the GDM prevalence was lower in their study compared with our study (1.4% vs. 3.4%). This might explain why the researchers in Sweden did not find GDM significantly associated with CKD in the absence of subsequent diabetes.

Our mediation analyses showed that subsequent development of diabetes and hypertension significantly mediated the risk of CKD after GDM. The quantifications of the mediated effects provide unprecedented and clinically relevant insights regarding the risk of CKD in women with previous GDM. As expected, subsequent diabetes mediated the association between GDM and diabetic kidney disease almost completely (96.3%), which was reassuring with regard to the validity of this mediation analysis. The minor deviation from complete mediation may be explained by coding discrepancies. Diabetes and hypertension in women with previous GDM mediated the other subcategories of CKD by ∼31–37%. This suggests that prevention of diabetes and hypertension after GDM may reduce the development of CKD in women with previous GDM. Previous studies have reported that diabetes likewise acts as a significant mediator of the risk of CVD (6) and psychiatric morbidity (10) after GDM.

The potential biological mechanisms behind an association between GDM and CKD may be composed of multiple factors preceding pregnancy and persisting postpartum. These include subclinical inflammatory processes, IR, β-cell insufficiency, and vascular dysfunction influenced by genetics, epigenetics, obesity, lifestyle, and environmental factors (4,35,36).

The exploratory investigation of acute kidney disease after GDM contributes novel knowledge. Not surprisingly, we found no association.

To our knowledge, this is the first study investigating the impact of increasing severity of metabolic dysfunction during pregnancy on the risk of kidney disease. Our data showed a proportional increase in the CKD risk with increasing severity of metabolic dysfunction. This was evident regardless of subsequent diabetes and hypertension. Hence, our results suggest that severity of metabolic dysfunction during pregnancy may be a potential prognostic factor for the CKD risk. Thus, inclusion of GDM history and treatment modality as an indicator of metabolic dysfunction severity seems clinically relevant in the individual assessment of CKD risk.

A major strength of this study is the large study population of all women giving birth in Denmark over a 22-year period. The study comprised the total eligible population, thus minimizing selection bias and information bias. Generally, data from the Danish registers are of good quality for health research (25–27,37). A validation study of the GDM diagnosis code found a sensitivity of 76% and concluded that, given that this misclassification is nondifferential, future health consequences may be underestimated in studies based solely on register data like ours (38). For kidney disease outcomes, the diagnosis codes for CKD have a high positive predictive value (39). For the investigation of mediation by subsequent diabetes and hypertension, the strategy of combining diagnosis codes and prescribed medications in their definitions increased the clinical relevance of the findings. The strategy facilitated identification of women with hospital contacts for diabetes and hypertension (i.e., expectedly, the more severe conditions but, importantly, it also identified women only receiving treatment in the primary care sector, that is, expectedly, the less severe conditions). Finally, several potential confounders were addressed in the adjusted analyses and the robustness of the results was confirmed by comprehensive sensitivity analyses.

Limitations include the selective GDM screening, because GDM may be undiagnosed and misclassified as non-GDM, thereby underestimating the prevalence. However, this would expectedly yield more conservative results. Also, the Danish GDM guidelines were revised in 2003 (31). However, the changes in screening strategy and diagnostic criteria were diminutive and, hence, considered to be without any substantial impact on the characteristics of the study population over time. Furthermore, undiagnosed diabetes may be misclassified as GDM. Embedded in the study design is the premise that only conditions included in the registers were accounted for (i.e., unknown or undiagnosed conditions remained so in this study). Moreover, the mediation analyses estimated one overall estimate of PE. As we observed possible changes in the TE of GDM on outcomes over time, this proportion may vary over time as well. In this case, our estimate would only be an estimate of the average mediation effect over time, not necessarily the mediation effect at each specific time point. Furthermore, in our mediation analyses, we assumed no interaction between potential mediator and exposure, nor between mediator and adjustment covariates in the models, as well as no unmeasured confounding between any pair of exposure, mediator, and outcome. Also, GDM and insulin treatment were perceived as proxies for increasing severity of metabolic dysfunction; however, despite being a feasible assumption, it may not capture this concept adequately. Additionally, the prevalence of insulin treatment was lower than expected from our clinical practice. This may be an underestimation potentially explained by suboptimal coding practice and hospital practice with a patient’s first insulin pens being given to them without prescription.

Regarding incidence of outcome, we excluded women with preexisting kidney disease at or up to 2 years prior to the index date. Hence, diagnosis before that did not lead to exclusion and, in case of kidney disease during follow-up, this would wrongly be considered an incident outcome. Similarly, basing kidney disease only on hospital diagnosis codes expectedly infers an underestimation of outcome incidence; women with CKD that manifested merely as microalbuminuria or reduced estimated glomerular filtration rate >60 mL/min/1.73 m² may not be referred to hospital and thus may be missed. Detection bias is potentially an important limitation because women with GDM are recommended for long-term follow-up examinations (4). Detection of outcome, therefore, may be more likely in these women. However, the impact of this bias was expectedly minor due to relatively low participation rates over time (40). Residual confounding also posed a limitation and, similarly, data on the included confounders and covariates were derived from the index pregnancy, thus not taking into account that some confounders/covariates may change and hence differ in subsequent pregnancies or later in life. Finally, the generalizability of our findings is potentially restricted to populations similar to the Danish with relatively homogeneous ethnic composition and where health care services, to a great extent, are provided for free.

In conclusion, we found a significant association between GDM and incident CKD. Subsequent diabetes and hypertension mediated this partially, yet even in the absence of these conditions, the risk persisted. Our data suggest that women with severe metabolic dysfunction during pregnancy constitute a high-risk group regarding future CKD. More research is warranted to determine the optimal, long-term clinical management after GDM, especially development of health care recommendations reaching beyond the postpartum period to decrease the risk of CKD after GDM.

Acknowledgments. The authors acknowledge Tanja Gram Petersen, epidemiologist and data manager at Research Unit OPEN, Department of Clinical Research, University of Southern Denmark, for data management assistance. This manuscript was edited for correct use of English language by Charlesworth Author Services.

Funding. This work received financial support via a research grant from the Danish Diabetes Academy, which is funded by the Novo Nordisk Foundation (grant NNF17SA0031406) and via research grants from the University of Southern Denmark and the Region of Southern Denmark.

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

Author Contributions. M.H.C., C.B., and D.M.J conceptualized and initiated the study. M.H.C. performed data management and wrote the first and final manuscript drafts. M.H.C. and S.M. performed statistical analyses. All authors contributed substantially to the study design, data interpretation, and critical revision of the manuscript, and read and approved the final manuscript for publication. D.M.J. is the guarantor of this work, had full access to all the data in this study, and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This work was presented in part in a poster at the Diabetic Pregnancy Study Group annual meeting, Poznan, Poland, 7–10 September 2023, and in an oral presentation at the European Association for the Study of Diabetes annual meeting, Hamburg, Germany, 2–6 October 2023.