Utility of glucose, lipid and kidney function trajectory measures for incident cardiovascular disease risk prediction for people living with type 2 diabetes: A case-study using Danish registry data

Background Cardiovascular disease (CVD) is an important complication of type 2 diabetes (T2D). Current CVD-prediction models use single baseline measurements and achieve moderate performance (C-indices ≈0.7). Repeated measurements available in modern healthcare registries may provide incremental predictive value. In this case study, we investigated whether HbA1c, LDL-cholesterol and eGFR trajectory measures might improve incident CVD-risk prediction. Methods We studied 83,326 people with T2D from Danish nationwide registers without CVD-history at baseline (1 st January 2015), who had ≥2 HbA1c, LDL-cholesterol and eGFR measurements between 2012-2014; their last measurement defined baseline. For each risk-factor we derived three paired trajectory measures across 2012-2014: mean & SD, median & IQR, and intercept & slope from growth models. A reference Cox-regression model for CVD-events (ICD-10 codes, until 31 st December 2020) included baseline age, sex, age-at-T2D-onset, HbA1c, LDL-cholesterol, HDL-cholesterol, eGFR, and medication use. Trajectory measures were added sequentially and jointly using an offset-procedure, computing hazard ratios (HRs), C-indices and continuous net reclassification index (NRI) with 95% CIs. Results Mean age was 65 (SD±12), and 48% were female. Over median 6 years 11,280 (13.5%) people experienced a CVD-event. Accounting for the reference model, trajectory measures of dispersion and change were associated with CVD-events (HRs ≈ 1.1 for HbA1c and eGFR; >1.4 for LDL-cholesterol), centrality measures were not. Adding trajectory measures produced minimal discrimination gains (C-index Δ +0.001-+0.003) but modest NRI improvements (+3-+9%). Conclusions Within-person dispersion or change measures for HbA1c, eGFR, and especially LDL-cholesterol, obtained from routine data, might moderately enhance incident CVD-risk prediction in people with T2D.