Presentation (CTI)

OBJECTIVES: Network Meta-Regression (NMR) is a valuable instrument in synthesizing evidence across multiple studies while considering patient-level and study-level covariates. This study presents multi-level NMR approach using an individual patient data (IPD) as well as aggregate data (AgD) to assess the treatment effects, including count data outcomes.
METHODS: Multilevel NMR was performed in R with the ‘multinma’ package, IPD and AgD from multiple studies were included, with focus on the count data. Patient-level covariates were scaled and included in the model while treatment effects were estimated using fixed, as well as random effects model with a logit link function and binomial likelihood. Random effects models incorporated varying half-normal priors and model evaluation was based on DIC (Deviance Information Criterion), residual deviance, and effective number of parameters.
RESULTS: Both fixed and random effects models provided estimates of relative treatment effects as Rate Ratio with 95% credible intervals for comparisons between treatments A, B, and C. Fixed-effects models yielded precise and statistically significant estimates while random-effects models accounting for heterogeneity, resulted in wider credible intervals, reflecting increased uncertainty and capturing variability across studies. Sensitivity analyses with different heterogeneity prior determined the impact of prior specifications on the results.
CONCLUSIONS: This study reveals the feasibility of multi-level NMR in integrating IPD and AgD for evaluating treatment effects. The 'multinma' R package is a powerful tool for conducting ML-NMR in comparative effectiveness research. While the methodology offers several advantages, it is crucial to acknowledge that the results may be subject to certain limitations.