Bayesian one-step IPD network meta-analysis of time-to-event data using Royston-Parmar models

Area of Science:

• Biostatistics
• Medical Statistics
• Clinical Trials Methodology

Background:

• Network meta-analysis (NMA) is established for continuous and binary outcomes but less so for time-to-event data.
• Traditional Cox proportional hazard (PH) models may be inadequate for oncology due to long follow-up and time-dependent treatment effects.
• Bayesian NMA is gaining traction, but Cox model fitting is computationally intensive.

Purpose of the Study:

• To develop a flexible, computationally practical Bayesian NMA model for individual participant data (IPD) survival analysis.
• To address limitations of Cox PH models in oncology settings with time-dependent effects.
• To provide a robust method for analyzing complex survival data in network meta-analyses.

Main Methods:

• Developed an IPD Royston-Parmar Bayesian NMA model for overall survival, utilizing data from 37 cervical cancer trials.
• Incorporated a treatment-ln(time) interaction to test for proportional hazards (PH).
• Provided WinBUGS code for model implementation and assessed direct/indirect evidence consistency and heterogeneity.

Main Results:

• The Royston-Parmar model accommodates time-dependent effects, offering flexibility beyond standard PH models.
• The developed model is computationally practical for large IPD datasets.
• Methods for testing PH, consistency, and heterogeneity were demonstrated.

Conclusions:

• The Royston-Parmar Bayesian NMA approach offers a flexible and computationally efficient method for analyzing IPD survival data in oncology.
• This approach readily extends to incorporate complexities like non-PH effects.
• It provides a valuable tool for ranking treatments and understanding survival outcomes in network meta-analyses.