Bayesian Network Meta-Analysis With One or Two Continuous Outcomes Measured at Multiple Time Points Using Gaussian...

Area of Science:

• Biostatistics
• Medical Informatics
• Clinical Epidemiology

Background:

• Traditional network meta-analysis synthesizes single outcomes at one time point.
• Existing methods struggle to simultaneously analyze multiple continuous outcomes across various time points.
• Trials often report secondary outcomes and longitudinal data, which are underutilized.

Purpose of the Study:

• To develop novel Bayesian network meta-analysis models for continuous outcomes measured longitudinally.
• To simultaneously account for multiple outcomes and their correlations over time.
• To improve the synthesis of evidence from clinical trials with complex outcome reporting.

Main Methods:

• Development of two Bayesian network meta-analysis models incorporating Gaussian random walks with drift.
• The first model handles a single continuous outcome over multiple time points.
• The second model extends this to incorporate a second outcome using cointegration of random walks.

Main Results:

• Both proposed models provide unbiased estimates for treatment effects and drift parameters with reasonable coverage.
• The cointegration model offers slight precision gains in certain scenarios.
• The models outperform a previously used random walk model in synthesizing osteoarthritis trial data.

Conclusions:

• The proposed Bayesian models offer advanced tools for network meta-analysis of longitudinal continuous outcomes.
• These models effectively handle multiple outcomes and time-dependent correlations.
• They are valuable for synthesizing evidence in areas like knee and hip osteoarthritis treatment.