Rajpurkar et al. (2022) — including Eric Topol — published a follow-up assessment of AI in healthcare, finding both progress and persistent gaps:
Progress
- Over 500 AI/ML-enabled medical devices cleared by the FDA by 2022
- AI-powered COVID-19 tools (diagnosis, drug repurposing, protein structure prediction) demonstrated rapid response capability
- AlphaFold's protein structure predictions opened new avenues for drug discovery
Persistent Gaps
- Dataset bias: Most AI models trained on data from wealthy nations, performing poorly in diverse populations
- Lack of prospective validation: The vast majority of AI studies are retrospective; few test performance in real clinical settings
- Regulatory lag: The regulatory framework hasn't kept pace with the speed of AI development
- Implementation barriers: Even validated AI tools face resistance from clinicians and integration challenges
Source
Rajpurkar, P. et al. (2022). Nature Medicine, 28(1), 31-38.