Improving Clinical Trial Enrollment for Smartphone-Based AI Data Collection: A Methodological Analysis of Nudge-Based Interventions

Abstract

The development of impactful medical AI technology hinges on access to high-quality training data, yet insufficient patient enrollment in clinical studies is often a limiting factor. This study examines the methodological challenges we faced in assembling a training dataset from 256,600 PCR-tested patients in clinical studies conducted in Colombia, Japan, Pakistan, and India to build a smartphone app that uses AI analysis of breathing and cough sounds to detect COVID-19. We analyzed the effectiveness and cost implications of three different data collection methods. We found that robocalls were the most efficient, enrolling 247,450 patients at USD 0.04 per call and an 8.5% success rate. Face-to-face methods achieved high enrollment rates (on average 45% to 61%) but were resource-intensive. Flyers were largely ineffective. Based on our findings and leveraging behavioral economics, we propose a framework of targeted nudge strategies to improve patient enrollment in future smartphone-based clinical studies.