4B-03 - Leveraging Genetic Evidence to Predict Clinical Trial Success: A Novel Approach to Enhancing Patient Outcomes

Background: Drug development is an inherently time-consuming and costly process with low overall success rates. Thus, innovative strategies to identify and prioritize promising therapeutic candidates are essential for streamlining drug discovery and ultimately enhancing clinical outcomes.

Objectives: To develop and validate a machine learning model that leverages genetic evidence of drug targets to predict clinical trial success, and to test the hypothesis that incorporating genetic data enhances predictive accuracy compared to traditional methods.

Methods: We collected data from 45,641 Phase III clinical trials registered on ClinicalTrials.gov. Genetic evidence of drug targets—including summary-data-based Mendelian randomization (SMR), DNA mutations, RNA differential expression, and protein differential expression—associated with their respective therapeutic indications was used as the exposure variable. The primary outcome was defined as trial success or failure based on whether the primary endpoint was achieved. A supervised machine learning model was developed to predict clinical trial outcomes, with performance evaluated using cross-validation and independent test sets via accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC).

Results: Incorporating summary-data-based Mendelian randomization (SMR) evidence significantly enhances clinical trial success rates, with supported trials demonstrating a 43% higher likelihood of success (OR=1.43, 95%CI:1.34-1.53, _P_ < 0.001). Our random forest prediction model integrating SMR-related biomarkers achieved robust performance: sensitivity 0.703, specificity 0.527, and positive predictive value 0.764. The model showed superior discriminative power compared to conventional approaches, providing a validated framework for prioritizing therapeutic candidates with genetic validation.

Conclusions: The machine learning model incorporating genetic evidence achieved high accuracy in predicting clinical trial success. These findings indicate that integrating genetic data into predictive models may improve candidate prioritization in drug development, although further validation is necessary to confirm its impact on patient outcomes.