Many apparent gaps in cholesterol management are not due to clinical oversight but are hidden in unstructured clinical notes. Analysis presented at the European Society of Cardiology Conference 2025 showed that a large language model can accurately extract external cholesterol values and documented reasons for statin non-use from free-text clinical notes, revealing the true context behind guideline-discordant care.

The model accurately extracted external low-density lipoprotein cholesterol values and reasons for statin non-use, including documented intolerance, physician decision, and patient refusal. Among patients lacking recent in-system lipid panels, 18.2% had external results identified. Including these data modestly improved lipid testing rates, though the proportion achieving target LDL-C remained unchanged. Accounting for statin intolerance increased overall statin use from 75.1% to 81.9% and high-intensity statin use from 45.5% to 53.6%.

This approach demonstrates that unstructured clinical data can reveal the true clinical context behind apparent care gaps. Large language models enable more precise quality assessment and targeted interventions to improve lipid management in patients with cardiovascular disease.