A Safety-Oriented Hypothetico-Deductive Framework for AI-Assisted Differential Diagnosis

Jul 16, 2026

Upload by Rowan

Diagnostic error poses a serious threat to patient safety, yet many large language model systems handle diagnosis as a single-step prediction without systematic safeguards against missed high-risk alternatives or rigorous verification. The framework AegisDx applies hypothetico-deductive clinical reasoning by coordinating role-specific LLM components, structured intermediate outputs, evidence-retrieval interfaces, and verification gates. It generates broad differentials, enforces explicit “must-not-miss” screening, verifies against grounded medical evidence, and outputs actionable next steps, improving safety and transparency.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages26

Reading Time1 h 6 min

Introduction

Diagnostic error and patient safety

LLMs in medical differential diagnosis

Translational gap in current systems

Proposed safety-oriented framework

arXiv :2607 .08038v 1 [ cs .AI] 9 Jul 2026 A safety-oriented hypothetico-deductive framework for AI-assisted differential diagnosis Fan Ma 1,∗, Mauro Giuffrè2,∗, Donald Wright3 , Kent McCann3 , Mark Iscoe 1,3 , Lingfei Qian 1 , Mingyang Jiang4 , Chi Wing Ng 1 , Na Hong 1 , Huan He 1 , Cathy Shyr5 , Qingyu Chen 1 , Lee Schwamm 1,6 , Lucila Ohno-Machado 1 and Hua Xu 1,† 1 Department of Biomedical Informatics and Data Science, Yale School of Medicine, Yale University, New Haven, USA 2 Department of Medical, Surgical and Health Sciences, Università degli Studi di Trieste, Trieste, Italy 3 Department of Emergency Medicine, Yale School of Medicine, Yale University, New Haven, USA 4 Applied Mathematics and Computer Science, Vanderbilt University, Nashville, USA 5 Department of Biomedical Informatics, Vanderbilt University, Nashville, USA 6 Department of Neurology, Yale School of Medicine, Yale University, New Haven, USA ∗ Equal contributions †Corresponding author Hua Xu: [hua.xu@yale.edu](hua.xu@yale.edu) Diagnostic error is a major threat to patient safety, yet current large language model (LLM) systems often treat diagnosis as a one-shot prediction task, lacking systematic safeguards against missed high-risk alternatives or rigorous verification of their reasoning. Here, we present AegisDx, a safety-oriented diagnostic reasoning framework for hypothetico-deductive clinical reasoning rather than a conventional free-form model call or agent-only pipeline. AegisDx coordinates specialized LLM components through role-specific contracts, structured intermediate outputs, evidence-retrieval interfaces, and verification gates to generate broad differential diagnoses, enforce explicit screening for dangerous “must-not-miss” conditions, verify reasoning against grounded medical evidence (e.g., PubMed and clinical guidelines), and structure actionable next diagnostic and management steps. We evaluated AegisDx across three distinct layers. First, on literature-derived case reports from two general medical journals—The New England Journal of Medicine (NEJM) and the Journal of the American Medical Association (JAMA)—using the final diagnoses reported in the source articles as the reference standard, AegisDx consistently outperformed the matched standalone LLM. With GPT-oss-120B as the shared backbone, Top-3 diagnostic accuracy was 59.9% versus 52. 1% on JAMA cases (+7.8 percentage points) and 62.7% versus 51.4% on NEJM cases (+11.3 percentage points) . Second, on cases from Annals of Emergency Medicine (Annals of EM), Top-3 accuracy against the final diagnoses reported in the source articles was 85.7% for AegisDx versus 68.6% for the standalone LLM (+17.1 percentage points) . Against physician-consensus must-not-miss diagnosis sets, AegisDx captured at least one such condition among its top three diagnoses in 78.0% of cases versus 52.0% for the standalone LLM (+26.0 percentage points) . Third, in a blinded physician evaluation of 43 real-world emergency department (ED) clinical notes from the Yale New Haven Health System (YNHHS) compared against GPT-5, AegisDx improved the physician-rated composite safety score from 4.31 to 4.55 on a 5-point scale (adjusted p = 2 .1 × 10 −4), with specific qualitative gains in“must-not-miss” condition identification and reasoning safety. Our findings suggest that engineering diagnostic artificial intelligence (AI) as a safety-oriented diagnostic reasoning framework, rather than optimizing raw predictive accuracy alone, can provide a safer, more transparent, and clinically meaningful layer of bedside decision support for acute care workflows. 1 Introduction Diagnostic error remains one of the most consequential and unresolved patient-safety challenges in modern medicine [1, 2 , 3] . Across all healthcare settings, missed, delayed, or incorrect diagnoses contribute substantially to preventable morbidity, mortality, and unsustainable downstream resource utilization [4, 5 , 6] . In a natio

FAQs

What problem does AegisDx address in AI-assisted diagnosis?

It targets diagnostic error by addressing the lack of safeguards in many LLM systems, specifically omissions of dangerous alternatives and insufficient verification of reasoning against evidence.

How does AegisDx produce and validate differential diagnoses?

It coordinates specialized LLM components using role-specific contracts, structured intermediate outputs, evidence-retrieval interfaces, and verification gates to generate broad differentials and check “must-not-miss” conditions.

What evaluation evidence supports the safety benefits of AegisDx?

Across literature-derived journal cases and emergency department real-world notes, AegisDx improved top-3 diagnostic accuracy, increased capture of must-not-miss conditions, and raised physician-rated safety scores versus a standalone LLM.

Jul 16, 2026

Upload by Rowan