ChatGPT Reduces Clinical Trial Screening Time from 40 Minutes to Under 3 Minutes, But Human Oversight Still Required
- ChatGPT-4 demonstrated superior performance over ChatGPT-3.5 in screening patients for clinical trials, achieving 84% accuracy with better sensitivity and specificity balance.
- Screening times were dramatically reduced from over 40 minutes per patient to 1.4-3.0 minutes with GPT-3.5 and 7.9-12.4 minutes with GPT-4, though costs ranged from $0.02-$0.27 per patient.
- Both AI models showed high specificity but low sensitivity in identifying eligible patients, with GPT-4 having only 16% median sensitivity despite 100% specificity.
- Researchers concluded that large language models should complement rather than replace manual chart reviews due to difficulties in identifying patients who meet all eligibility criteria.