Identifying PSC: ICD-10 code vs LASSO model vs LLM

What is known

• The ICD-10 code alone had a c-statistic of 0.87, sensitivity 87.6%, and PPV 68.8%. ¹
• The LASSO model (incorporating ICD-10 code, elevated alkaline phosphatase, ursodiol use, IBD, and history of cholangitis) had a c-statistic of 0.92, sensitivity 87.4%, and PPV 70.7%. ¹
• The LLM had a c-statistic of 0.77, sensitivity 91.7%, and PPV 51.0%, inferior discrimination compared to either the ICD-10 code or the prediction model. ¹
• The authors concluded the LASSO model's performance was not meaningfully improved over the ICD-10 code alone, and the code alone was sufficient for identifying PSC. ¹

What is unknown / caveats

• Single retrospective single-center study (UCSF), internal validation only
• Population enriched with likely-PSC patients, so PPV may be inflated and NPV could not be calculated
• LASSO model variables limited to those available in the electronic health record
• The 0.92 vs 0.87 difference in c-statistic is statistically higher but the authors judged it not clinically meaningful.

References

1. Wang M, Dao M, Delk M, Fenton C, Li MY, Rubin JB, Ge J, Lai JC, Li M. Identification of primary sclerosing cholangitis: ICD-10 code validation and comparison with a large language model approach. Res Sq. 2026. PMID: 41960311.

For educational use only, not a substitute for clinical judgement or medical advice. Consult qualified clinicians for diagnosis and treatment decisions. This tool may contain errors; use the flag button below to report inaccurate or harmful content.

Draft, generated by the IBDology RAG and not yet clinician-reviewed. Answers are grounded in the retrieved literature listed above; a high faithfulness score means the answer matches its sources, not that the sources are correct.