Using supervised machine learning classifiers to estimate likelihood of participating in clinical trials of a de-identified version of ResearchMatch

Summary

This study analysed ResearchMatch registry data (841,377 instances, 20 features including demographics, geography, medical conditions and visit history) using seven machine learning and deep learning classifiers to identify characteristics associated with clinical trial participation intent. A deep learning convolutional neural network model outperformed traditional classifiers with an AUC of 0.8105, demonstrating meaningful correlations between predictor variables and trial participation likelihood. The findings suggest machine learning approaches have promise for identifying and stratifying individuals more likely to participate when offered specific clinical trial opportunities.

UK applicability

This methodology could be adapted by UK clinical research networks and NHS trial recruitment systems to improve enrolment targeting, though applicability would depend on access to comparable de-identified registry data and consideration of UK-specific demographic and health service factors.

Key measures

Area under the curve (AUC); classifier performance comparison across Logistic Regression, Decision Tree, Gaussian Naïve Bayes, K-Nearest Neighbor, Adaboost, Random Forest, and Convolutional Neural Network models

Outcomes reported

The study identified predictive characteristics of individuals more likely to express interest in participating in clinical trials using supervised machine learning and deep learning approaches. Performance was evaluated using area under the curve (AUC) metrics across multiple classifier models.

Topic tags