Synthetic target trial emulation and predictive modeling of amylin-pathway therapies for obesity and type 2 diabetes.

INTRODUCTION: Amylin-pathway therapies represent a novel therapeutic class for obesity and type 2 diabetes, however head-to-head comparative data and long-term outcome predictions remain limited. We conducted target trial emulation and computational predictive modeling aiming to predict future trial outcomes and comparative effectiveness across the amylin-pathway development program. METHODS: Following PRISMA 2020 and TARGET framework guidelines, we search in the current literature for eligible trials and extracted data from seven randomized controlled trials (N = 5,786 participants) of amylin-pathway therapies published up to September 2025. We reconstructed high-precision synthetic individual patient data (IPD) and developed computational models for virtual head-to-head comparisons, dose-response optimization, longitudinal trajectory prediction, and trial simulation. Network meta-analysis integrated evidence across CagriSema, cagrilintide, and amycretin formulations. RESULTS: Synthetic IPD reconstruction achieved >99 % fidelity to source trials, validated through leave-trial-out cross-validation (efficacy RMSE: 2.9 % points, calibration slope: 0.61; discontinuation RMSE: 0.18, slope: 1.08). Virtual head-to-head modeling confirmed CagriSema superiority over amycretin subcutaneous at matched timepoints (posterior probability >0.95). Dose-response modeling identified optimal amycretin exposures (ED80: 8.88 mg subcutaneous, 95 % CI: 7.12-11.08), with benefit-risk frontier analysis delineating a therapeutic window at 10-20 mg balancing efficacy plateau against tolerability thresholds (GI-AE <75 %, discontinuation <20 %). Longitudinal kinetics showed plateau timing at 52-68 weeks for obesity outcomes and 24-32 weeks for glycemic endpoints. Heterogeneity analysis revealed complete resolution for GI adverse events (I_DL = 0 %, τ = 0) and moderate variation for discontinuation (I_DL = 13 %, τ = 0.03) after logit-scale correction with proper within-arm variance weighting. Machine learning models predicted treatment response with 82-87 % accuracy using baseline characteristics. CONCLUSIONS: Synthetic target trial emulation with structured validation (leave-trial-out, posterior predictive checks, simulation-based calibration) demonstrated promising evidence for amylin-pathway development optimization. Benefit-risk frontier analysis identified an optimal 10-20 mg subcutaneous therapeutic window, and heterogeneity quantification through maximum a posteriori (MAP) predictive interval provides design-ready estimates for confirmatory trials requiring around 800-1,200 participants per arm for 90 % power.