A recursive partitioning approach for subgroup identification in individual patient data meta-analysis

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Health Services Research

Background:

• Subgroup analyses in clinical trials are crucial for identifying patient groups with differential treatment effects.
• Traditional statistical tests for treatment-by-subgroup interactions often lack sufficient statistical power.
• Individual patient data (IPD) meta-analyses offer enhanced power but conventional subgroup analysis methods have limitations, such as investigating covariates one at a time.

Purpose of the Study:

• To investigate and extend statistical methods for identifying patient subgroups with substantial treatment effects, specifically focusing on treatment-by-subgroup interactions.
• To address the limitations of conventional subgroup analyses in both single trial and IPD meta-analysis settings.
• To develop an exploratory statistical approach suitable for IPD meta-analyses.

Main Methods:

• Employed tree-based methods, specifically extending the SIDES (Subgroup Identification Based on Effect Size) algorithm.
• Incorporated fixed-effects and random-effects models within the SIDES framework to manage between-trial variation in IPD meta-analyses.
• Assessed the performance of the extended IPD-SIDES method through simulation studies and application to a low back pain dataset.

Main Results:

• Simulation studies demonstrated that the extended IPD-SIDES method effectively detects patient subgroups with significant treatment effects, particularly when substantial between-trial heterogeneity is present.
• Application of the IPD-SIDES method to low back pain data successfully identified subgroups exhibiting an enhanced treatment response.
• The method proved adept at navigating the covariate space to define subgroups based on multiple patient characteristics.

Conclusions:

• The proposed extension of the SIDES method provides a robust exploratory statistical approach for subgroup analyses in IPD meta-analyses.
• This methodology is applicable across various research disciplines requiring subgroup identification within pooled individual patient data.
• The findings highlight the utility of advanced tree-based methods for uncovering complex treatment-effect modifiers.