Overview
Clinical trials are the gold standard for testing new cancer treatments, but they're expensive, time-consuming, and require large numbers of patients. A promising new approach uses real-world patient data to build "digital twin" models—artificial intelligence (AI) systems that predict how individual patients would respond to treatments they didn't actually receive. If validated, this could help design more efficient trials with smaller sample sizes while maintaining statistical power.
Researchers used Flatiron Health’s US Non–Small Cell Lung Cancer (NSCLC) Panoramic Database, inclusive of more than 336,000 patients with NSCLC, to develop machine learning models using real-world data from patients with advanced NSCLCtreated with first-line chemotherapy between 2011-2016. They then tested whether these models could accurately predict outcomes for patients in two major clinical trials (IMpower131 and 132). The models were validated by comparing their predictions to actual trial results.
The models successfully predicted overall survival with high accuracy, and mean absolute overall survival differences less than 5%. Based on these findings, researchers estimated that digital twins could reduce trial sample sizes by 9-21% while maintaining statistical power.
Why this matters
This research demonstrates that real-world data can be used to accurately predict treatment responses using AI, potentially enabling faster, more efficient clinical trials. This breakthrough could accelerate the development and approval of new cancer treatments, getting life-saving therapies to patients more quickly while reducing the burden on trial participants.