Regimen-stratified Bayesian model-based meta-analysis of targeted biologics in generalised myasthenia gravis: efficacy comparison, mechanism-level class contrast, and endpoint sensitivity with model-informed trial design.

Nine targeted biologics spanning four mechanisms have reached phase III development for generalised myasthenia gravis, yet no cross-programme comparison accounts for the distinct continuous versus cyclical dosing patterns or integrates both patient-reported (MG-ADL) and clinician-assessed (QMG) scales. We performed a regimen-stratified Bayesian model-based meta-analysis (MBMA) to compare efficacy, evaluate mechanism-level class differences, and inform trial design. Longitudinal MG-ADL and QMG data were extracted from nine placebo-controlled phase III trials. Six hierarchical nonlinear Emax models were fitted with inverse-variance weighting. Drug-specific maximum efficacy, net benefit over placebo, week 26 predictions, time to the minimum clinically important difference (MCID), cumulative efficacy, ranking probabilities, and mechanism-level class contrasts were derived. Robustness was assessed via model comparison, prior sensitivity, and Leave-one-study-out analyses on both scales. Endpoint-sensitivity simulation and a treatment-selection framework were appended. The study was registered with PROSPERO: CRD420261288549. On MG-ADL, zilucoplan led with the highest Emax (4.8 points; 95% CrI: 4.2-5.4) and largest net benefit (2.5; 2.0-3.1). On QMG, batoclimab led (Emax 5.9; 4.4-9.5; net benefit 4.1; 2.7-7.7). Efgartigimod achieved the fastest onset on both scales (≤ 1.6 weeks). At the mechanism class level, neonatal Fc receptor (FcRn) inhibitors demonstrated decisive superiority over complement inhibitors on QMG (mean Emax 5.28 vs 3.99; posterior probability 99.7%) while no class difference was observed on MG-ADL (52.5%), identifying a scale-dependent mechanism-efficacy dissociation. The regimen-stratified architecture was decisively favoured over a pooled model (Δ Expected log pointwise predictive density [ELPD] =  + 19.8). Drug rankings were invariant across prior specifications and Leave-one-study-out permutations on both scales (maximum rank shift: 1 position). For rozanolixizumab on MG-ADL, snapshot week 26 endpoints were underpowered (65%); integrated endpoints recovered power to 94-95%. In the treatment-selection framework, batoclimab held the highest composite score in all eight priority profiles, while the identity of the runner-up varied with clinical priority. This analysis identifies a novel scale-dependent mechanism-level class effect - FcRn inhibitors outperform complement blockade on objective motor endpoints with near-certainty, while both classes improve patient-reported function equally. Across the multi dimensional treatment-selection framework integrating efficacy, safety, tolerability, and convenience, batoclimab emerged as the highest-ranked agent in all eight clinical priority profiles. These findings provide a pharmacodynamic evidence base for mechanism-guided treatment individualisation and endpoint optimisation in generalised myasthenia gravis.