Large language models generate diagnostic likelihood ratios with low mean bias but wide dispersion.

Paul Chong¹, Shuhan He^2,3, Kian Samadian², Amal Mohamed², Boyu Peng⁴, Emma Chua⁵, Cory Rohlfsen⁶, Brian W Locke^7,8

¹School of Osteopathic Medicine, Campbell University, Lillington, NC, USA.

²Department of Emergency Medicine, Massachusetts General Hospital, Boston, MA, USA.

³Lab of Computer Science, Harvard Medical School, Boston, MA, USA.

⁴MSDA Program, MGH Institute of Health Professions, Boston, MA, USA.

⁵Department of Natural Sciences, Pasadena City College, Pasadena, CA, USA.

⁶Department of Medicine, University of Nebraska Medical Center, Omaha, NE, USA.

⁷Department of Pulmonary and Critical Care, Intermountain Medical Center, Murray, UT, USA. brian.locke@imail.org.

⁸Shock Trauma Intensive Care Unit, Intermountain Medical Center, 5121 Cottonwood Street, Murray, UT, 84107, USA. brian.locke@imail.org.

PubMed DOI

Abstract

Accurate, context-appropriate likelihood ratios (LRs) are needed for Bayesian diagnosis, but empirical LRs are sparse because diagnostic accuracy studies are costly and context-dependent. We evaluated whether large language models (LLMs) can generate LR outputs that align with literature-reported values. We compared LR outputs generated by three OpenAI models (GPT-4o, o3, and GPT-5) with all liter…

Create a free account to keep reading

Free members get 10 full research views every month across publications, clinical trials, FDA clearances, adverse events, and NIH grants. No credit card required.

Create Free Account Sign In

Want unlimited research access? See Pro plans

Data Accuracy Notice: Research intelligence on Health AI Central is aggregated from public sources (PubMed, ClinicalTrials.gov, FDA, NIH, CMS, and others) and refreshed nightly. Classifications and derived metrics are produced by automated methods described in our Methodology. We recommend verifying critical data points against the primary sources before making decisions.