Comparative Effectiveness in Cardiometabolic Diseases

Abstract

Cardiometabolic disease -- the clinical cluster encompassing type 2 diabetes, hypertension, dyslipidaemia, and cardiovascular disease -- now accounts for over 40% of premature European mortality, yet treatment guidelines are built predominantly on randomised trial populations that exclude patients with multiple concurrent conditions. Comparative effectiveness research offers a route to evidence that more closely matches clinical reality, but its validity depends heavily on the analytical approach used to address treatment selection confounding. We evaluated five comparative effectiveness frameworks applied to 108,642 cardiometabolic patients from six Baltic and Central European national health registries -- Tallinn, Vienna, Helsinki, Warsaw, Riga, and Vilnius -- between 2015 and 2023. Frameworks spanned passive claims-based analysis through ML-enhanced doubly robust estimation with target trial emulation. Performance was benchmarked using our Cardiometabolic Comparative Effectiveness Score (CCES), integrating causal validity, population representativeness, data completeness, clinical translation utility, and cross-registry replication consistency. The target trial emulation with ML confounding control achieved the highest CCES (0.927), generating GLP-1 agonist cardiovascular benefit estimates within 3.2% of landmark trial values while extending findings to populations with eGFR below 45 and BMI above 40 who were excluded from those trials. SGLT2 inhibitor heart failure benefit was confirmed across all five frameworks but heterogeneous treatment effects revealed substantially attenuated benefit in patients with preserved ejection fraction below median, a finding invisible to average-effect analyses.