Accounting for time-varying exposures and covariates in the relationship between obesity and diabetes: analysis using parametric g-formula

Background Previous studies investigating the association between obesity and diabetes often did not consider the role of time-varying covariates affected by previous obesity status. This study quantified the association between obesity and diabetes using parametric g-formula. Methods We included 8924 participants without diabetes from the Korean Genome and Epidemiology Study—Ansan and Ansung study(2001–2002)—with up to the seventh biennial follow-up data from 2015 to 2016. Obesity status was categorised as normal (body mass index (BMI) <23.5 kg/m 2 ), overweight (23.5–24.9 kg/m 2 ), obese 1 (25.0–27.4 kg/m 2 ) and obese 2 (≥27.5 kg/m 2 ). Hazard ratios (HRs) comparing baseline or time-varying obesity status were estimated using Cox models, whereas risk ratio (RR) was estimated using g-formula. Results The Cox model for baseline obesity status demonstrated an increased risk of diabetes in overweight (HR 1.85; 95% CI=1.48–2.31), obese 1 (2.40; 1.97–2.93) and obese 2 (3.65; 2.98–4.47) statuses than that in normal weight status. Obesity as a time-varying exposure with time-varying covariates had HRs of 1.31 (1.07–1.60), 1.55 (1.29–1.86) and 2.58 (2.14–3.12) for overweight, obese 1 and obese 2 statuses. Parametric g-formula comparing if everyone had been in each obesity category versus normal over 15 years showed increased associations of RRs of 1.37 (1.34–1.40), 1.78 (1.76–1.80) and 2.42 (2.34–2.50). Conclusions Higher BMI classification category was associated with increased risk of diabetes after accounting for time-varying covariates using g-formula. The results from g-formula were smaller than when considering baseline obesity status only but comparable with the results from time-varying Cox model.