Evaluating a large language model for identifying key comorbidities in discharge summaries of patients with obesity: a comparative validation study against ICD codes using MIMIC-IV
Authors' conclusion
The LLM showed promising concordance with ICD codes for T2DM, HTN, OSA, and HLD in obesity patient discharge summaries.
Study context
- Population
- Discharge summaries of adult patients with obesity, MIMIC-IV; mean age 59.7 years, 60.4% female341 unique patients
- Participants
- 350 admissions
- Comparison
- LLM identification of comorbidities vs ICD codes and manual review
- Outcome
- Performance (accuracy, precision, recall, F1-score) of LLM vs ICD codes for T2D, hypertension, hyperlipidaemia and obstructive sleep apnoea
Methodological detailsretrospective, validation study, MIMIC-IV data
- Retrospective
- Validation study
- MIMIC-IV data
Performance
As reported in the abstract
F1-score of LLM vs ICD codes for T2DM, HTN, HLD and OSA
LLM F1-scores, primary ICD-based analysis (range)
0.815 to 0.948
Further findings
LLM performance vs manual review (sensitivity analysis)
- Accuracy for T2DM
- 0.963
- Accuracy for Hypertension (HTN)
- 0.971
- Accuracy for Hyperlipidemia (HLD)
- 0.863
- Accuracy for Obstructive Sleep Apnea (OSA)
- 0.966
- F1-scores (range)
- 0.879 to 0.981
- Sensitivity analysis
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
Evaluating a large language model for identifying key comorbidities in discharge summaries of patients with obesity: a comparative validation study against ICD codes using MIMIC-IV. Int J Obes (Lond). Published 8 July 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →