Comment on “The association between COVID ‐19 and incident gestational diabetes: A population‐based case–control study of the National Health Insurance Research Database in Taiwan”

Dear Editor, We read with interest the population-based case–control study by Lin et al.1 evaluating the association between SARS-CoV-2 infection and incident gestational diabetes mellitus (GDM) using Taiwan's National Health Insurance Research Database (NHIRD). By linking nationwide insurance claims with birth registry and COVID-19 testing data, the authors assembled a large pregnancy cohort and applied landmark and sensitivity analyses to address time-related bias. This study provides important population-level evidence on COVID-19-related metabolic outcomes in pregnancy, particularly in an East Asian setting with high vaccine uptake. Several aspects of the analysis merit consideration when interpreting both the reported null association between SARS-CoV-2 infection and GDM and the observed inverse association with vaccination. The finding that COVID-19 vaccination was associated with a lower risk of GDM (adjusted odds ratio 0.90; 95% CI 0.87–0.93) is difficult to reconcile with current biological understanding. Vaccination is unlikely to directly influence glucose metabolism or insulin resistance during pregnancy in the absence of infection, consistent with systematic reviews showing no clear evidence that COVID-19 vaccines exert a direct effect on glycemic outcomes in noninfected populations2. In observational settings, vaccination status often reflects differences in health awareness, healthcare utilization, and socioeconomic circumstances, all of which are strongly associated with GDM risk. Although insurance premium–based income was included as a covariate, this measure may not adequately capture individual or household socioeconomic position within Taiwan's National Health Insurance system. The presence of an apparent vaccine “protective effect” therefore suggests that residual confounding due to healthy user bias may remain and may also affect estimates of the association between infection and GDM. Adjustment for prepregnancy adiposity also appears limited. Diagnostic codes for morbid obesity identify only a small proportion of women with overweight or obesity because of their low sensitivity in claims data. Given the strong association between body mass index and GDM, as well as its relationship with COVID-19 susceptibility and detection, incomplete control of this confounder could materially bias effect estimates. Such bias does not necessarily exaggerate risk and may, under certain behavioral or epidemic conditions, attenuate or obscure a true association. Finally, the analysis combines infections occurring across different epidemic phases. The study period spans multiple SARS-CoV-2 variants with distinct virulence profiles, whereas most infections in Taiwan occurred during the Omicron wave, which has been characterized by a high proportion of asymptomatic infections3. Pooling infections across calendar time may dilute variant-specific effects and bias estimates toward the null. Analyses stratified by epidemic period could help distinguish between a true absence of association and averaging across heterogeneous exposures. In this context, the null association between COVID-19 infection and GDM, together with the apparent protective association with vaccination, is best interpreted as specific to a highly vaccinated, Omicron-dominant population rather than as definitive evidence against a biological link between SARS-CoV-2 infection and gestational metabolic risk. We commend the authors for their rigorous use of national data resources and encourage further work addressing residual confounding and temporal heterogeneity to strengthen causal interpretation. No funding was received for this study. The author declares no conflict of interest. Approval of the research protocol: N/A. Informed consent: N/A. Registry and the registration no. of the study/trial: N/A. Animal studies: N/A. Lidong Gao: writing—original draft; writing—review and editing. Not applicable, as this is a commentary on a published study; no new data were generated or analyzed.