Psychiatric Morbidity in Women With Previous Gestational Diabetes Mellitus: A Nationwide Register-Based Cohort Study

Gestational diabetes mellitus (GDM), a common pregnancy complication, is characterized by hyperglycemia resulting from a combination of physiological pregnancy-induced insulin resistance and insufficient insulin response. GDM risk factors include advanced maternal age, overweight/obesity, polycystic ovary syndrome (PCOS), certain ethnicities, higher parity, and previous GDM (1). GDM is associated with an increased risk of complications related to pregnancy, delivery, and the immediate postpartum period (1–3). The risk increases with increasing blood glucose levels (4); therefore, additional perinatal care and examinations are recommended for women diagnosed with GDM, including counseling on diet and exercise and introduction of insulin treatment, if necessary (1,3). After delivery, glucose metabolism generally normalizes (1), but women with GDM have an increased long-term risk of developing manifest diabetes (mainly type 2 diabetes) (5), as well as metabolic morbidity (6,7) and cardiovascular disease (7,8), compared with women without previous GDM.

In addition to physical morbidity, women with previous GDM may have an increased risk of developing psychiatric morbidity later in life; however, the existing evidence is limited. Two recent meta-analyses found an increased risk of postpartum depression in women with previous GDM (9,10); however, evidence regarding incident anxiety disorders is scarce and provides conflicting results (11). Only two studies have examined the risk of depression beyond the 1-year postpartum period (12,13); moreover, psychiatric disorders beyond depression and anxiety have been investigated in only one other study using a composite score of psychiatric morbidity (12). Therefore, population-based data on the long-term risk of overall and subcategorized psychiatric morbidity in women with previous GDM are lacking and will provide novel insight. The development of subsequent diabetes is more common in women with versus without GDM; in addition, diabetes is a well-established risk factor of depression (14). Therefore, GDM and incident psychiatric morbidity may be linked through subsequent diabetes development; however, the role of subsequent diabetes in the association between GDM and psychiatric morbidity is unknown. Finally, psychiatric disorders are associated with diabetes development (15) and diabetes complications (16) and can adversely affect the mother-child bond and child development (17); therefore, it is important to clarify the risk of psychiatric disorders after GDM while additionally considering subsequent diabetes development.

The overall aim of this study was to examine long-term psychiatric morbidity in women with previous GDM based on data from national registers including the complete population of delivering women in Denmark from 1997 to 2018. The research objectives were 1) to compare the risk of incident psychiatric morbidity in women with and without previous GDM, 2) to investigate the role of subsequent diabetes development in the association between previous GDM and incident psychiatric morbidity, and 3) to quantify the potential mediating effect of subsequent diabetes in the abovementioned association.

This nationwide register-based cohort study included data from registers provided by Statistics Denmark and the Danish Health Data Authority. In Denmark, data on a large range of sociodemographic, economic, and health care variables, including redemptions of prescribed medications, are prospectively collected and stored in national registers, data from which can be linked at the individual level (18–20). Data were extracted from the Danish Medical Birth Register, which contains data on all pregnancies and deliveries in Denmark (21). The cohort was enriched with data on inpatient, outpatient, and emergency department contacts at psychiatric and non-psychiatric hospitals from the Danish National Patient Registry (22,23). The Danish National Prescription Register provided data on redemptions of prescribed psychopharmacological medications (24), and demographic and socioeconomic data were extracted from relevant registers (25–27).

The inclusion criterion was delivery during the study period from 1 January 1997 to 31 December 2018. Women were considered as the study unit, and the index date was defined as the date of conception in the index pregnancy (i.e., the first pregnancy during the study period). Women with preexisting diabetes and/or preexisting psychiatric morbidity at or up to 2 years before the index date were excluded. The 2-year time frame was adopted to ensure identical criteria for all women (study data were available from 1995 onward; study period commenced in 1997). Descriptions of diagnosis codes and medications are listed in Supplementary Table 1. Women with missing data on the a priori–defined demographic/socioeconomic covariates included in the main adjusted analyses were excluded.

GDM exposure, based on diagnosis code, was defined as one or more pregnancies with GDM during the study period. To address potential change in exposure status in case of women with additional pregnancies during the study period, GDM was considered a time-varying exposure; thus, correct classification of each pregnancy was possible. In Denmark, screening for GDM is based on an individual risk factor assessment (28).

Psychiatric morbidity was based on selected psychiatric diagnosis codes and/or two or more redemptions of prescribed psychopharmacological medications from 6 weeks after delivery (Supplementary Table 1). The primary outcome was depression (composite of diagnosis codes of depressive disorders and/or two or more redemptions of antidepressants). Secondary outcomes were any psychiatric diagnosis (composite of diagnosis codes of depressive disorders, anxiety disorders, bipolar disorders, psychotic disorders, and/or postpartum psychiatric disease) and use of any psychopharmacological medication (composite of two or more redemptions within each group of psychopharmacological medications [antidepressants, antipsychotics, and anxiolytics]). Additionally, to address each of the subcategories in the outcome composites, data on the subcategories were analyzed separately. Over time, women may have received several different psychiatric diagnoses/medications, and each woman may therefore have been represented in more than one outcome category.

Follow-up and risk time contribution commenced 6 weeks after delivery (marking the end of the immediate postpartum period) and continued until outcome occurrence, death, emigration, or 31 December 2018, whichever occurred first. The risk time spent during any subsequent pregnancy was extracted from the total follow-up period. Because GDM was a time-varying exposure, categorization of subsequent risk time as either exposed or unexposed was subject to potential change. However, once GDM occurred, all subsequent risk time was considered exposed.

On the basis of existing literature, covariates from the index pregnancy were a priori identified as potential confounders or intermediate factors. Table 1 and Supplementary Table 1 present details on categorizations and definitions. The potential confounders were selected demographic/socioeconomic covariates: maternal age, parity, preexisting comorbidity (29), ethnicity, marital status, income level, education level, occupation, and calendar year of delivery. The potential intermediate covariates of the association between GDM and incident psychiatric morbidity were selected pregnancy complications: preeclampsia, gestational hypertension, preterm delivery, induced labor, cesarean section, stillbirth, and offspring malformations.

Comparisons of baseline characteristics between women with and without GDM were performed using the Wilcoxon rank-sum test and χ² test. Cox regression models were used to estimate crude and adjusted hazard ratios (aHRs), including 95% CIs, using women without GDM as reference. Because each woman could contribute with more than one pregnancy during the study period, repeated measurements from the same woman were taken into account by clustering on each woman in the model specification. The assumption of proportional hazards was tested, and interaction terms were included in the final adjusted model in case of nonproportional hazards, preceded by a likelihood ratio test for model fit.

Missing data regarding pregestational BMI and smoking were expected, because these data were not registered until late 2003 and 1997, respectively. Therefore, these variables were not included in the main analyses but handled in separate sensitivity analyses. For missing data regarding gestational age (<2%), imputation was performed using the mean value.

Subsequent diabetes development was defined as diabetes diagnosis code and/or two or more redemptions of antidiabetic agents after delivery and before psychiatric morbidity outcome occurrence (Supplementary Table 1). To address the role of subsequent diabetes in future psychiatric morbidity risk according to GDM history, previous GDM and subsequent diabetes development were included as interaction terms in an additional analysis of the Cox regression models, including a likelihood ratio test for model fit. Therefore, risk estimates were produced for the study population when stratified into four mutually exclusive groups: 1) women without previous GDM and no subsequent diabetes (no GDM and no diabetes), 2) women with previous GDM and no subsequent diabetes (GDM and no diabetes), 3) women without previous GDM who subsequently developed diabetes (no GDM and then diabetes), and 4) women with previous GDM who subsequently developed diabetes (GDM and then diabetes).

Additionally, mediation analyses were performed to quantify the potential mediating effect of subsequent diabetes development after pregnancy and before outcome occurrence in the association between GDM and psychiatric morbidity. These analyses were based on a counterfactual framework (30), according to which the total effect of an exposure (i.e., GDM) on outcome incidence (i.e., psychiatric morbidity) can be decomposed into components, one of which resembles the mediated effect by the mediator included within the model (i.e., subsequent diabetes). The aHRs were computed by first not including subsequent diabetes and then including subsequent diabetes in the Cox regression models. Thereafter, quantification of the proportion mediated effect was estimated for the outcomes, resulting in significant associations in the analyses both excluding and including subsequent diabetes. The proportion mediated effect was reported as proportion with corresponding 95% CI obtained using bootstrapping with 100 replicates.

All statistical analyses were performed using STATA 17 software (StataCorp, College Station, TX). P values <0.05 were considered statistically significant.

A series of sensitivity analyses was conducted. In order to minimize the risk of misclassifying GDM exposure, we restricted the study population to women with 1) no previous deliveries before study entry and 2) GDM diagnosis code after gestational week 20 and no diagnosis code for diabetes registered during the same pregnancy. To investigate whether potential bias was introduced as a result of imputation for missing data on gestational age at delivery, women with missing data on this covariate were excluded. Furthermore, women with missing data on the selected confounders were included in the study population, using missing categories for the missing data. Subsequently, confounder adjustment was expanded by individually including covariates from the index pregnancy in the model: pregestational BMI, smoking, multiple pregnancy, preexisting PCOS, and preexisting metformin treatment. Finally, women with selected pregnancy complications during the index pregnancy were excluded.

No ethical approval was required (assessment by the Regional Committees on Health Research Ethics for Southern Denmark [20192000-27]). However, the study was approved by the Danish Data Protection Agency (19/11440).

The data set is not publicly available because of Danish legislation concerning personalized data. Access to the data set has been authorized restrictively to the authors by the Danish National Health Data Authority; other researchers are eligible to apply for access.

In total, 758,978 women delivered from 1997 to 2018. Women with preexisting diabetes (n = 3,323), preexisting psychiatric morbidity (n = 50,207), or missing data on the selected covariates (n = 45,431) were excluded. The resulting study population consisted of 660,017 women (Supplementary Fig. 2). Of these, 20,663 (3.1%) were registered with GDM in one or more delivery during the study period. Follow-up periods ranged from 0 to 21.9 years, with medians differing from 9.5 (interquartile range 4.0–16.4) to 12.4 years (interquartile range 5.6–18.0) depending on outcome category. Subsequent diabetes development occurred in 17.3% and 1.4% of women with and without GDM, respectively. Women were censored from the Cox regression analyses if psychiatric morbidity occurred during the time period from the index date to the beginning of follow-up (i.e., 6 weeks postpartum), because they were then not considered to be at risk during the follow-up period. This reduced the total number of women eligible for these analyses to 656,783.

Clinical, demographic, and obstetric characteristics in the index pregnancy for women with and without GDM are listed in Table 1. The two groups differed significantly in terms of almost all parameters. Compared with women without GDM, a greater proportion of women with GDM were age >30 years, overweight/obese, primiparous, of non-Danish ethnicity, nonsmokers, and from the low- or high-income tertile and had a lower educational level; moreover, they were more often in the undereducation or early retirement category. Pregnancy and labor complications were more prevalent in women with GDM.

Table 2 shows associations between previous GDM and incident psychiatric morbidity. Previous GDM was associated with a significantly increased risk of depression, any psychiatric diagnosis, and use of any psychopharmacological medication, with aHRs ranging from 1.20 (95% CI 1.13–1.27) to 1.22 (95% CI 1.18–1.27). The findings were similar for the subcategory outcomes depressive disorders, anxiety disorders, antidepressant medication use, and antipsychotic medication use, with aHRs ranging from 1.14 (95% CI 1.05–1.25) to 1.32 (95% CI 1.22–1.42). Risk of incident diagnosis of psychotic disorder, bipolar disorder, substance use disorder, postpartum psychiatric disease, and anxiolytic medication use was similar in women with and without previous GDM.

Generally, sensitivity analyses produced risk estimates equaling those of the main analyses (data not shown). Minor differences were found when women with previous deliveries before the study period were excluded. When pregestational BMI was included in the adjusted analyses (n = 342,572), risk estimates were lower, although the differences remained statistically significant (Supplementary Table 3). After excluding women with selected pregnancy complications, risk estimates remained relatively unchanged; however, risk of antipsychotic medication use lost significance.

Figure 1 depicts the risk of incident psychiatric morbidity according to previous GDM, considering subsequent diabetes development. Overall, risk of psychiatric morbidity was higher in women who subsequently developed diabetes compared with women without subsequent diabetes. Specifically in women without subsequent diabetes, GDM was associated with significantly increased risk of psychiatric morbidity, with aHRs ranging from 1.15 (95% CI 1.08–1.23) to 1.18 (95% CI 1.13–1.23). This pattern of increased risk related to GDM history was not similarly seen in women who subsequently developed diabetes; however, risk estimates were higher in women with versus without subsequent diabetes.

Results of the mediation analyses are presented in Table 3. Subsequent diabetes development significantly mediated the association between previous GDM and incident depression by 35.7% (95% CI 27.3–44.2). Similarly, diabetes was found to have a mediating effect on the incidences of all other psychiatric outcome categories that were significantly associated with previous GDM; the mediation analyses produced estimated proportions of mediation ranging from 35.0% (95% CI 11.1–59.0) to 42.2% (95% CI 27.5–57.0).

This large nationwide cohort study provides novel insight, because it is the first study to examine the risk of a variety of psychiatric disorders after GDM beyond the 1-year postpartum period. Overall, significant associations between previous GDM and incident psychiatric morbidity were observed. Subsequent diabetes development played a significant role in future psychiatric morbidity risk, although it only partly explained the associations between GDM and psychiatric morbidity.

Our study showed that GDM was associated with increased risk of incident depression. This is in line with the findings of studies focusing on the risk of depression after GDM up to 1 year postpartum (9). However, our results contrast with a study from the U.S. (N = 18,109), which reported a similar confounder-adjusted risk of depression in women with and without GDM after ∼4 years of follow-up (12). This difference may be explained by the longer follow-up period and the larger study population in our study.

Our study also demonstrated a clear and not previously reported increased risk of developing anxiety disorders in women with previous GDM compared with women without a GDM history. This finding differs from those of previous studies (12,31,32), probably because of the use of self-reported anxiety symptoms (31,32), shorter follow-up periods, and smaller study populations in the previous studies (12,31,32).

To our knowledge, only one other large-scale study has examined risk of psychiatric disorders other than depression and anxiety after GDM; the study investigated a composite score of mental health disorders, excluding depression and anxiety. It found a similar psychiatric morbidity risk among women with and without previous GDM (12). In our study, we reported data on type-specific subcategories of psychiatric morbidity outcomes and found that the risk of being diagnosed with bipolar disorders, psychotic disorders, substance use disorders, and postpartum psychiatric disease was comparable in women with and without previous GDM. The difference in the risk of requiring anxiolytic medication according to previous GDM was also insignificant. Thus, our study showed that the increased psychiatric morbidity after GDM was constituted by diagnosis of depressive and anxiety disorders as well as use of antidepressant and antipsychotic medications, whereas the other investigated psychiatric outcomes seemed to be unrelated to GDM history.

This study provides unprecedented results regarding the role of subsequent diabetes development in the psychiatric morbidity risk in relation to GDM history. We found that whether or not a woman developed diabetes subsequently constituted a significant role in the association between previous GDM and the future risk of psychiatric morbidity. In women without subsequent diabetes, previous GDM was associated with modestly increased risk of psychiatric morbidity. Women with subsequent diabetes had a much higher risk of developing psychiatric disorders compared with women without subsequent diabetes; however, history of GDM did not seem to substantially affect future risk of psychiatric morbidity in women with subsequent diabetes, because risk estimates were relatively similar regardless of previous GDM. These findings require further exploration.

Our mediation analyses showed that subsequent diabetes development contributed substantially to risk of incident psychiatric morbidity. The associations between GDM and psychiatric morbidity were mediated by subsequent diabetes by ∼40%, implying that the associations were only partly mediated by subsequent diabetes; thus, a direct association between GDM and incident psychiatric morbidity remained. This pattern of mediation was observed consistently for all outcomes that were significantly associated with previous GDM. In studies among individuals with type 2 diabetes, both shared and causal physiological and psychological mechanisms have been proposed to explain the link between diabetes and depression (33). Next to incident diabetes, mechanisms linking GDM to psychiatric disorders could be negative self-perceptions in connection with GDM diagnosis (34), awareness of increased diabetes risk, and increased parenting stress resulting from more adverse outcomes observed among offspring of women with GDM (35,36). More research is warranted to clarify the link between GDM and incident psychiatric disorders. To our knowledge, the mediating role of diabetes has not previously been quantified with regard to incident psychiatric morbidity after GDM; however, a study found that type 2 diabetes mediated the association between GDM and incident cardiovascular disease by 23% (37). Thus, subsequent diabetes seems to play a profound mediating role in future morbidity after GDM but does not exclusively account for it. Prevention of subsequent diabetes after GDM could potentially reduce not only diabetes but also risk of other morbidity types.

A major strength of this study was the large study population size and, specifically, that it included the complete population of delivering women in Denmark during a 22-year period, minimizing the risk of selection bias. Furthermore, this approach facilitated an unprecedented population-based insight into long-term mental health implications beyond the postpartum period, because we addressed previously unexplored conditions.

GDM exposure and psychiatric morbidity outcomes were based on register data that were considered valid and reliable for health research (22,38). Although prior research has shown that a GDM diagnosis based solely on diagnosis codes is valid, this strategy potentially underestimates future health consequences of GDM because of misclassification (38). By considering GDM as a time-varying exposure, we optimized insight into the implications of GDM, because the strategy facilitated full exposure history in each woman during the entire study period. Thus, exposure status was not limited to a selected/random pregnancy. Outcome categories, including depressive and anxiety disorders, were analyzed as distinct conditions, although they may co-occur and present similar traits. Both diagnosis codes and medication use were addressed, which constituted a valuable strategy, because it also enabled investigation into the proportion of women experiencing psychiatric conditions without assigned hospital diagnosis codes. The latter group received psychopharmacological medication prescribed exclusively from the primary health care sector, which accounts for most patient cases with mild to moderate mental illness (22). Through this strategy, we obtained more clinically relevant and extensive insight into the mental health implications of GDM. Requiring at least two redemptions further qualified and validated the use of redemptions as indicative of morbidity. Potential confounding and influence of pregnancy/labor complications were addressed, and sensitivity analyses were performed to test the robustness of the findings.

Our study had some limitations. Because the scope was to investigate incident psychiatric morbidity, we excluded women with preexisting psychiatric morbidity within 2 years before the index date. However, some women may have been diagnosed/medicated before the 2-year period and therefore not correctly excluded from the study. This would result in a misinterpretation of an incident outcome that was rather a recurrent condition. Additionally, psychiatric illness may increase GDM risk (39). Psychiatric illnesses often co-occur, and depression is known to be recurrent (40); therefore, the GDM-related risk with regard to psychiatric morbidity might have been overestimated. Moreover, use of prescribed psychopharmacological medication only resembled a proxy for psychiatric morbidity, because these medications have heterogeneous and also nonpsychiatric indications; this is an important limitation. Regarding GDM exposure, this may have been underestimated as a result of selective rather than universal screening in Denmark. In addition, GDM exposure may have been misclassified, because some women classified as non-GDM may have experienced GDM pregnancies after the study period. However, this would have generated more conservative results. Another important imitation of this study design is the definition of subsequent diabetes, which was defined as incidence of diagnosis codes and/or antidiabetic medication. Therefore, we were not able to address unknown/undiagnosed diabetes or explore whether the mediation by subsequent diabetes reflected biological/physiological and/or psychological mechanisms. Additionally, in the mediation analysis, some women diagnosed with GDM may have had undiagnosed type 2 diabetes instead. Data on potential confounders were derived from the index pregnancy; such data might not be representative in subsequent pregnancies or in later life. Furthermore, some data were self-reported, with potential limited validity. Because obesity and depression are linked, the inability to adjust for BMI beyond pregestational BMI at the index pregnancy is a noteworthy limitation. Furthermore, it was beyond the scope of this study to explore the incidence of additional comorbidities developing after GDM, which might influence future psychiatric morbidity risk. Residual confounding also remained a limitation. Other statistical approaches could have been chosen, including a multiple outcome approach addressing correlations between outcomes, which could have further elucidated the interdependence between psychiatric outcomes. However, in a large study population like ours, such an approach would be computationally infeasible, and the interpretation of results would be too complex. Finally, surveillance bias might be a concern; women with GDM were recommended to undergo follow-up with a general practitioner at an interval of 1–3 years (28). Such women might be more prone to be diagnosed/medicated, which may have resulted in overestimation of the incidence of psychiatric morbidity in women with versus without previous GDM. The generalizability of the study findings is potentially restricted. The Danish health care system provides free access to most services, and the Danish population is relatively homogenous with regard to race/ethnicity. Therefore, our findings may not necessarily be applicable to countries with different health care systems and racial/ethnic profiles.

In conclusion, we found that previous GDM was significantly associated with an overall higher risk of incident psychiatric morbidity (especially depression and anxiety); however, other psychiatric outcomes were not related to GDM history. Subsequent diabetes development played an important role in the association between previous GDM and future psychiatric morbidity risk, but it only partly explained the association. The findings highlight the importance of special attention to the long-term mental health of women with previous GDM, irrespective of subsequent diabetes. Further research is required to address the development of strategies for prevention, detection, and management of psychiatric morbidity in women with previous GDM.

Acknowledgments. The authors would like to acknowledge Sören Möller, biostatistician at Research Unit OPEN, Department of Clinical Research, University of Southern Denmark, Odense, Denmark, for assistance with mediation analyses and additional statistical assistance and Tanja Gram Petersen, epidemiologist and data manager at Research Unit OPEN, Department of Clinical Research, University of Southern Denmark, Odense, Denmark, for assistance with data management and data analyses. This manuscript was edited for 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. performed data management and statistical analyses and wrote the first and final manuscript drafts. M.H.C., M.S.A., and D.M.J. conceptualized and initiated the study. All authors contributed to the study design, contributed substantially to data interpretation and critical revision of the manuscript, and read and approved for publication the final manuscript. D.M.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.