One-step parametric network meta-analysis models using the exact likelihood that allow for time-varying treatment...

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Pharmacoeconomics

Background:

• Network meta-analysis (NMA) is crucial in oncology for comparing multiple treatments.
• Time-to-event (TTE) data analysis in NMA often violates the proportional hazard (PH) assumption.
• Existing NMA methods for time-varying effects lack accuracy and are complex to implement.

Purpose of the Study:

• To introduce a novel one-step fully Bayesian parametric individual patient data (IPD)-NMA model.
• To enable accurate TTE data analysis with time-varying treatment effects without the PH assumption.
• To offer a flexible and implementable alternative to existing NMA methods.

Main Methods:

• Developed a one-step fully Bayesian parametric IPD-NMA model.
• Utilized exact likelihood for TTE data, accommodating time-varying treatment effects.
• Incorporated Weibull, Gompertz, log-normal, log-logistic, gamma, and generalized gamma distributions for fixed or random effects.

Main Results:

• The one-step model was applied to a network of advanced melanoma RCTs.
• Results were compared to a traditional two-step approach, demonstrating comparable or improved accuracy.
• A simulation study confirmed the one-step method's advantages over the two-step approach.

Conclusions:

• The proposed one-step IPD-NMA model provides a flexible and accurate approach for TTE data analysis in oncology.
• It simplifies model selection and allows for the inclusion of novel distributions like generalized gamma.
• This method enhances the reliability of evidence synthesis for treatment comparisons in clinical trials.