Subgroup discovery in non-inferiority trials

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Machine Learning in Healthcare

Background:

• Subgroup analysis is vital for assessing treatment effect heterogeneity, particularly in non-inferiority (NI) trials where overall similarity may mask interactions.
• Traditional methods for subgroup analysis in NI trials are limited, with growing interest in advanced machine learning techniques for post-hoc discovery.

Purpose of the Study:

• To evaluate the performance and practical application of machine learning methods for subgroup identification in non-inferiority trials.
• To adapt and assess the Virtual Twin algorithm, combining random forest and classification/regression trees, for the NI setting.

Main Methods:

• Conducted extensive simulation studies to examine the Virtual Twin method's performance under various NI trial conditions.
• Developed decision rules for selecting final subgroups identified by the Virtual Twin algorithm.
• Applied the method to real-world data from a non-inferiority trial comparing two acupuncture treatments for chronic musculoskeletal pain.

Main Results:

• The Virtual Twin method demonstrated utility in identifying relevant subgroups within the non-inferiority trial context.
• Simulation studies provided insights into the algorithm's performance across different NI trial scenarios.
• Decision rules were devised to guide the selection of meaningful subgroups.

Conclusions:

• Machine learning approaches, specifically the Virtual Twin method, offer a flexible and powerful tool for subgroup discovery in non-inferiority trials.
• This research addresses a gap in systematically exploring advanced methods for subgroup analysis in NI settings.
• The findings support the application of the Virtual Twin method for uncovering treatment effect heterogeneity in clinical trials.