GPT-4 Improves Clinical Guideline Readability With Caveats

In a 2026 study, researchers applied a GPT-4–based pipeline to 20 NHS Injectable Medicines Guide (IMG) guidelines to improve readability while preserving content. BERT scores showed high semantic similarity (0.88–0.96), and SMOG readability improved modestly (mean difference 0.32; P=.02), but pharmacists identified omissions or changes in 20% of subsections, so authors caution against unverified automated edits.
Key Points
- 1Reports GPT-4 pipeline yields high semantic similarity across guidelines (BERT 0.88–0.96) in 20 IMG documents
- 2Highlights modest readability gains: SMOG improved by mean 0.32 (95% CI 0.10–0.55; P=.02) relative to original
- 3Warns of safety risk: pharmacists found omissions/additions/meaning changes in 20% of subsections requiring checks
Scoring Rationale
Peer-reviewed, quantitative evaluation demonstrating readability gains and high semantic similarity; limited by clinically relevant omissions requiring human verification.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
