Comment on “Associations between gestational diabetes and cardiovascular disease largely operate independently of postpartum causal pathways: A population‐based cohort study in England”

Astbury et al. present a large, methodologically sophisticated analysis linking gestational diabetes mellitus (GDM) and later cardiovascular disease using linked primary care, hospital episode and mortality records.1 Their causal four-way decomposition showing a substantial controlled direct effect after accounting for postpartum diagnoses is important for framing pregnancy as a sentinel moment for long-term cardiometabolic risk. First, to evaluate heterogeneity, prespecify how categorical covariates will be operationalised and the statistical approach to test effect modification. We recommend categorising self-reported ethnicity into White, South Asian (Indian/Pakistani/Bangladeshi/Other Asian), Black (African/Caribbean), Chinese, and Other, using White as the reference, and retaining Townsend deprivation in quintiles; effect modification should be assessed using interaction terms in Cox models and by reporting stratified mediation decomposition (e.g., med4way or counterfactual mediation models within major ethnic and deprivation strata) with presentation of subgroup Controlled Direct Effects and Pure Indirect Effects and their confidence intervals.2 These prespecified choices will aid interpretation and avoid post-hoc subgroup selection bias. Second, the proposal to incorporate continuous biomarkers deserves clarification about data availability and analytic strategy. Where routinely recorded laboratory measures exist within the QResearch linkage (e.g., glycated haemoglobin [HbA1c], routine lipid panels and serial blood pressure readings), time-updated quantitative mediators or cumulative exposure metrics (area under the curve or trajectory modelling) could be incorporated to characterise mediator intensity and duration; if these measures are incomplete, we recommend stating this explicitly and framing the request as a priority for future biomarker-rich cohorts or nested substudies. Modelling mediator dose and temporality would help distinguish persistent metabolic disturbance from discrete incident diagnoses.3 Third, to make the translational recommendations more actionable, briefly address feasibility, priority populations and cost implications. Implementation should prioritise women from South Asian and Black ethnic groups and those in the most deprived Townsend quintiles given their disproportionate burden; pragmatic pilots of electronic health record flags and integrated cardio-obstetric pathways can be evaluated using stepped-wedge designs with builtin cost-effectiveness and equity metrics to inform scale-up. Early piloting will clarify resource needs and facilitate allocation where benefit is greatest. We have adjusted tone to avoid subjective expressions and focused recommendations on analyses and pragmatic next steps. Prioritising heterogeneity analyses, time-updated mediator modelling, and pragmatic implementation pilots will accelerate translation of these findings into equitable prevention strategies for women after GDM. The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/dom.70402. Not applicable, as no data were generated or analysed in this study.