Predict progression free survival and overall survival using objective response rate for anti-PD1/PDL1 therapy...

Area of Science:

• Oncology
• Clinical Trial Design
• Biostatistics

Background:

• Objective response rate (ORR) is a key early endpoint in oncology anti-PD1/PDL1 therapy trials, while progression-free survival (PFS) and overall survival (OS) are critical late-phase endpoints.
• Predicting late-phase outcomes from early-phase data can optimize trial design and probability of success (POS) evaluations.
• Existing literature has limited data on the association between ORR and survival outcomes (mPFS, mOS).

Purpose of the Study:

• To develop predictive models for median PFS (mPFS) and median OS (mOS) based on early-phase objective response rate (ORR) in solid tumor anti-PD1/PDL1 therapy development.
• To optimize late-phase clinical trial design and evaluate the probability of success (POS).

Main Methods:

• Constructed a comprehensive quantitative clinical trial landscape database (QLD) by integrating data from multiple sources.
• Developed a generalizable algorithm for systematic, accurate, and complete structured data extraction.
• Utilized a tree-based machine learning regression model incorporating tumor type, stage, line of therapy, and treatment class to predict mPFS and mOS from ORR.

Main Results:

• Identified over 150 late-phase clinical trials for predictive model development, significantly expanding upon existing literature.
• The developed tree-based model demonstrated robust predictions, accounting for heterogeneity and borrowing strength across trials.
• Cross-validation showed the model's predictive mean square error was competitive with advanced machine learning methods and superior to linear regression models.

Conclusions:

• The developed predictive model offers a robust and interpretable method for estimating late-phase survival outcomes from early-phase ORR data in oncology.
• This approach can significantly inform and optimize the design of late-phase clinical trials for anti-PD1/PDL1 therapies.
• The model provides a valuable tool for evaluating the probability of success in late-phase trials, particularly for combination therapies.