Cohort Optimizer, a Software Solution That Uses Real-World Data and AI to Optimize Criteria of Oncology Trials
Bhandari S1, Mann J1, Hajela R2, Govil V2, Greatti Y1, Maguire R1, Powers J1, Southwick S1, Mueller J1
1ConcertAI, Cambridge, MA, USA, 2ConcertAI, Bengaluru, India
Background:Recent work demonstrates that many clinical trial criteria negatively affect enrollment without significantly improving the health of the cohort. We developed a software solution, Cohort Optimizer (CO) that uses Real-World Data (RWD) and AI to modify criteria to improve the size, overall survival, and diversity of the cohort. Unlike many proposed solutions which only eliminate criteria, CO optimizes the cohort by intelligently relaxing or tightening criteria.
Methods:CO allows the user to assign priorities to each characteristic of the cohort—size, overall survival, and diversity. CO estimates characteristics using ConcertAI’s RWD that includes 5.4 million individuals, gathered from oncology clinics throughout the US. After imputing missing patient characteristics, CO calculates the size, survival statistics, and diversity of the baseline cohort and candidate cohorts. To efficiently search through the high-dimensional space of criteria and arrive at a recommendation, CO uses an evolutionary algorithm and a custom fitness function. As validation, we implemented CO on 15 oncological clinical trials, translated into the Observational Medical Outcomes Partnership Common Data Model. For all trials, we prioritized total enrollment over survival and diversity, while ensuring that the overall survival of the recommended cohort be no less than 95% of the survival of the baseline cohort.
Results:On average, our modifications increased patient counts by 75.64% compared to the baseline, without significant sacrifices in either survival or diversity. Much of the changes in these cohorts were attributable to relaxation (not outright elimination) of the ECOG criterion.
Conclusion:Our method indicates that one can increase patient enrollment without compromising the overall health of the cohort. We demonstrate the value of RWD in estimating how changes to criteria affect the characteristics of its associated cohort. Given representative data, we believe our solution can be applied in clinical trials beyond oncology.
Conference/Value in Health Info
Methodological & Statistical Research, Study Approaches
Artificial Intelligence, Machine Learning, Predictive Analytics, Clinical Trials, Electronic Medical & Health Records
No Additional Disease & Conditions/Specialized Treatment Areas