Strengths and Limitations
This analysis has several strengths and limitations. Strengths include the use of a rich set of covariates from the prenatal and early postpartum period, including glucose metabolism values at 2 days postpartum, robust methods including the use of a predictive modelling algorithm to identify important predictors in the presence of a small number of events and several collinear variables, internal validation with k-fold cross validation, and calibration assessment of the final model.22 In addition, we used multiple imputation to address missing covariate data, which is often ubiquitous in routinely collected clinical data.36 Limitations include the fact that only 68% of women from the parent study (n=300) had glucose metabolism assessed at 1 year postpartum and were included in the present analysis, raising the possibility of selection bias. However, a previous analysis of these data indicated that on average, women who returned at 1 year postpartum were similar to women who did not across measured covariates, with the exception of being more likely to be privately insured.14 Women in the study were followed for only 1 year postpartum, which is too short to evaluate type 2 diabetes as an outcome. Our analysis also included information on race/ethnicity as a potential predictor of impaired glucose tolerance. Type 2 diabetes is well known to have a genetic component37,38. However, race and ethnicity are not reliable indicators of genetic differences.39 In this context, race/ethnicity likely represent social rather than genetic differences. We note the lack of additional covariates related to social stratification (socioeconomic status, healthcare access, etc.) as a limitation. Finally, in our calibration analysis we observed some evidence of model overfitting in women in the highest two quartiles of predicted risk, which may reflect the small sample size in our cohort and indicates the need to validate these findings in larger cohorts of women with GDM.