Insulin resistance, indexed by the estimated glucose disposal rate (eGDR), may evolve over time, yet its longitudinal patterns and relevance for cardiovascular prevention remain unclear. Using data from a nationwide cohort of adults without diabetes, we derived eGDR from routinely collected variables and classified participants into data-driven trajectories across two follow-up visits. We then applied longitudinal targeted maximum likelihood estimation (LTMLE) with Super Learner to estimate the association between eGDR trajectories and incident cardiovascular disease (CVD) within a causal-inference framework, while accounting for time-varying confounding. Compared with the low-stable trajectory, participants in the high-stable and increasing trajectories had significantly lower risks of incident CVD. Restricted cubic splines confirmed an inverse, nonlinear relationship between eGDR and CVD. Benchmarking against the China-PAR model showed that adding eGDR trajectories significantly improved discrimination (AUC 0.554 → 0.609), reclassification (IDI 0.0153; continuous NRI 0.364, both P < 0.001), and net clinical benefit across threshold probabilities of 10–25% on decision curve analysis, whereas TyG trajectories conferred no meaningful incremental value. Findings were consistent in sensitivity and subgroup analyses and were replicated in an independent UK cohort. These results suggest that longitudinal monitoring of eGDR, derived from routine clinical measurements, may help identify adults at elevated cardiometabolic risk before diabetes onset and support early prevention strategies.