Machine learning approach for detection of MACE events within clinical trial data.

Area of Science:

• Clinical research methodology
• Health informatics
• Machine learning applications in medicine

Background:

• Randomized controlled trials (RCTs) are standard but may not reflect real-world medicine impact.
• Real-world data (RWD) offers broader patient populations and care settings.
• A key challenge in RWD is the absence of algorithmic outcome identification.

Purpose of the Study:

• To develop machine learning models for identifying Major Adverse Cardiovascular Events (MACE) within clinical trial data (CTD).
• To assess the feasibility of using these models to analyze outcomes in both CTD and RWD.
• To demonstrate the potential for reducing resources in RCTs and supporting regulatory submissions for pragmatic trials.

Main Methods:

• Anonymized CTD was used to develop features for identifying MACE.
• Three random forest models were trained to detect components of 3-point MACE.
• Model performance was evaluated using recall and precision metrics.

Main Results:

• Developed models demonstrated viability for identifying clinical outcomes in prospective trials.
• Models achieved recall of 0.72 (0.07) and precision of 0.68 (0.12).
• The study presented a cost-benefit analysis for deploying these models in clinical settings.

Conclusions:

• Advanced algorithms can effectively identify clinical outcomes in prospective trials.
• Deploying these machine learning models can potentially decrease the resources needed for RCTs.
• Extending these models to RWD could facilitate regulatory approval of pragmatic clinical trials.