CODEX Digest 10.15.26

Who is caring for whom may matter less than how clinicians interpret clinical information and diagnose. This randomized study included 1,027 primary-care physicians evaluating standardized cases in which patient sex and race were experimentally varied. Patient-physician sex concordance did not significantly affect diagnostic accuracy across five cases, although female physicians showed greater accuracy across several scenarios. The findings complicate assumptions about how demographic concordance influences diagnostic equity.

The diagnostic protential of AI can exceed the evidence supporting its clinical use. This narrative review provides a practical overview of AI in radiology before critically examining applications to cancer-associated thromboembolism. It discusses incidental detection, risk stratification, and treatment support while highlighting the limitations of AI such as data heterogeneity, limited prospective validation, generalizability problems, and workflow challenges.

Triage may be the first diagnostic decision as it articulates how urgently a patient needs attention. This evidence review provides a comparative global evaluation of emergency department (ED) triage systems rating patient acuity from most urgent (1) to least urgent (5). The results of this analysis find that established triage systems provide moderate performance that varies by outcome and setting for a selected set of outcomes. The authors support their use as prioritization tools rather than as a mechanism to determine diagnostic accuracy in the ED.

Generative AI may change not only what people decide, but how they decide. This literature review examines how AI-assisted decision-making interacts with human heuristics, analytical reasoning and cognitive biases, highlighting its impact in a variety of contexts impacting diagnostic work in medicine, management, education, and research. The authors consider potential benefits and risks of AI-supported reasoning, including improved efficiency but weakened cognitive reasoning.

History taking is an elemental exchange between patients and clinicians. This narrative review examines history taking as a core component of diagnostic safety, emphasizing communication, contextual understanding, and patient-centered information gathering. The authors discuss practical approaches for clinicians and patients and consider digital tools and AI as potential adjuncts rather than replacements for human interaction. The paper reinforces history taking as a diagnostic-excellence intervention.

As AI systems become more capable of supporting diagnosis, questions about when they can safely act independently become increasingly important. This study describes a hospital-based AI system that runs locally to support diagnosis and was evaluated on diagnostic performance using test cases across age groups and sex. The system flagged how confident it is in each diagnosis and triggered cases that need human attention. While promising, the results require prospective clinical validation to assure real-world performance.

Diagnostic reliability remains a fundamental challenge particularly in psychiatry. This international, cross-sectional examination of diagnostic reliability across 19 countries explores the consistency of 1,038 clinicians arriving at the same psychiatric diagnosis for nine cases. The results show diagnostic agreement to be modest with 66% overall concordance with reference diagnoses. The findings highlight persistent uncertainty in psychiatric classification and clinical interpretation.

Physician review of clinical reasoning is essential for evaluating medical AI but difficult to scale. This preprint introduces an LLM evaluator and physician-annotated benchmark. Despite limited examples, this study finds that conventional LLM judges often disagree with physicians, while the tested evaluator more closely matches physician scores across clinical-AI tasks, including stepwise diagnosis as case details arrived. The authors have made the code and data for the tools freely available.

Pulse oximeters can overestimate blood oxygen in Black patients compared with blood gas tests, potentially delaying diagnosis and treatment. In this prospective cohort study of 198 ICU patients with 709 paired measurements at two hospitals, Black race—but not objectively measured skin pigmentation—was associated with bias. The findings suggest additional physiological or device-related causes and warrant further study of inequities in pulse-oximeter use.

More evidence does not necessarily produce better diagnosis when responsibility for synthesizing it becomes fragmented. This historical and conceptual analysis traces how specialization, episodic care, reimbursement, documentation, and administrative systems may have weakened longitudinal diagnostic responsibility. The proposed concept of “structural diagnostic failure” shifts attention from individual cognitive mistakes toward organizational conditions that can prevent distributed information from becoming a coherent diagnostic account.

Earlier cancer detection can reduce mortality while also creating opportunities for overdiagnosis. This population-level ecological analysis used Global Burden of Disease 2021 data from 21 countries across six continents to examine lung cancer incidence and mortality trends and compared observed versus expected cases. Most countries showed parallel trends, but China showed a potential overdiagnosis signal among older women. Though the work is not a definitive finding of overdiagnosis, their findings do support careful risk-stratification of CT screening.

Discounting a patient’s experience can initiate diagnostic delay early in the care process. This commentary examines diagnostic delay through the concepts of epistemic injustice and US medical malpractice law, drawing on published evidence about communication, gender, and racial disparities. The author argues that dismissing credible patient testimony can contribute to delayed diagnosis and proposes recognition of epistemic injustices as legal injury within tort law.