Prediction of drug approvals in oncology using explainable artificial intelligence

Area of Science:

• Oncology
• Biostatistics
• Drug Development

Background:

• Oncology drug development faces lower clinical trial success rates compared to other areas.
• Existing machine learning models for drug approval prediction lack oncology-specific trial design details.
• This deficiency hinders effective portfolio optimization in oncology drug development.

Purpose of the Study:

• To develop a machine learning model incorporating oncology-specific trial design information for predicting drug approval.
• To enhance the accuracy of drug approval predictions in the oncology pipeline.
• To support strategic planning in oncology drug development.

Main Methods:

• Developed a machine learning model integrating oncology-specific clinical trial design features.
• Utilized data on trial duration, patient enrollment, and sponsor product launches.
• Validated model performance using prospective data.

Main Results:

• Achieved high predictive accuracy with an area under the receiver operating characteristic curve (AUC) ranging from 0.86 to 0.98.
• Identified key predictive features including trial duration, number of enrolled patients, and sponsor's product launch history.
• Demonstrated high accuracy in prospective validation.

Conclusions:

• The developed machine learning model effectively predicts oncology drug approval by incorporating trial design specifics.
• The model provides valuable estimates of approval probabilities for drugs in the oncology pipeline.
• Findings support improved strategic planning and resource allocation in oncology drug development.