CODEX Digest - 6.11.26

1) Possible diagnostic error in cervical artery dissection: analysis of STOP-CAD study.

Bakradze E, Shu L, Yaghi S, et al. J Am Heart Assoc. 2026;15(10):e046408.

Cervical artery dissection (CeAD), a common cause of stroke in young adults, is often misdiagnosed because symptoms are nonspecific. This multicenter international study finds that one in six patients had a possible diagnostic error, especially younger patients, those with migraines, and those presenting with headache and diffuse symptoms.

2) Use of large language models by academic hospitalists: results of a multicenter survey. (This is a preprint that has not yet gone through the peer review process). 

Bressman E, Auerbach A, Keniston A, et al. medRxiv. Epub 2026 May 29.

The use of LLMs in support of clinical practice is being adopted across healthcare, though not as a part of everyday care delivery. This cross-sectional survey finds that 67% of academic hospital medicine faculty across eight institutions use LLMs to support diagnostic reasoning and treatment decisions. However, most are still not using it daily in their practice due to workflow, accuracy, and patient privacy concerns.

*UCSF CODEX’s director, Sumant Ranji, MD, and project manager, Caroline Jens, MPH, are authors for this publication.

3) A bumpy ride: how patients experience diagnostic uncertainty and anxiety.

Dahm MR, Dahm AE. J Gen Intern Med. Epub 2026 May 26.

Discussing diagnostic uncertainty with patients and families is challenging, but clinicians must be transparent about what is not yet known while remaining sensitive to its impact. This commentary from an uncertainty researcher and their child examines the emotional toll of diagnostic uncertainty on patients and families. The authors suggest that understanding patients’ experiences of diagnostic anxiety and illness severity can help tailor information and care, build trust, and support shared decision-making.

4) Equity evaluation of an intervention to increase colorectal cancer screening at Federally Qualified Health Centers.

Ganguly AP, O’Leary MC, Stradtman LR, et al. Ann Fam Med. 2026;24(3):231-234.

Federally Qualified Health Centers (FQHCs) are a key source of care for racially and ethnically minoritized populations, yet colorectal cancer screening rates are 30 percentage points lower than in the general US population. In this secondary analysis, a mailed fecal immunochemical test (FIT) outreach program plus navigation to colonoscopy after abnormal results improved screening similarly across racial and ethnic groups compared to usual care. The findings suggest such organized outreach could raise screening rates without widening disparities, an important consideration for FQHCs.

5) Core + Clusters: a practical clinical reasoning-based toolkit for revitalizing physical diagnosis training.

Gowda D, Silvestri RC, Blatt B, et al. Clin Teach. 2026;23(3):e70416.

Physical examination is foundational to diagnosis, yet clinical reasoning is not consistently taught as part of the skill. This article describes an innovated approach to medical teaching history which includes a "core" physical exam and "clusters" of exam features that correspond with diagnostic reasoning, with curricular examples and lessons from more than 10 years across several schools.

6) Frequent ambulatory care visits predict harmful diagnostic errors in high-risk hospitalized patients: a retrospective cohort study.

Goyal A, Konieczny K, Plombon S, et al. J Gen Intern Med. Epub May 22.

Diagnostic errors threaten hospital patient safety. In this single-center retrospective cohort study, having two or more ambulatory care visits in the 14 days before admission was associated with about twice the odds of harmful diagnostic error in high-risk hospitalized patients, suggesting a useful signal for proactive diagnostic safety surveillance.

7) Gaslighting in the context of diagnostic safety: a concept analysis.

Hermosilla AL. J Healthc Risk Manag. Epub 2026 May 24.

Gaslighting is an underexplored form of team dysfunction that can undermine diagnostic safety. This review describes gaslighting in healthcare teams as dismissiveness, undermining, and inducing doubt in colleagues’ clinical judgment. Contributing factors include hierarchy, workplace stress, and poor communication, consequences include emotional distress, loss of trust, and greater risk of diagnostic error. The author argues that organizations should promote psychological safety, open dialogue, and professional respect to mitigate gaslighting in diagnostic teams.

8) Applying natural language processing and large language models to clinical notes for phenotyping and diagnosing rare diseases: a systematic review.

Kim S, Zhou Y, Guo Y, et al. J Am Med Inform Assoc. 2026;33(6):1225-1235.

Patients with rare diseases often experience substantial diagnostic delays. This systematic review examines how NLPs and LLMs are being used to analyze unstructured clinical notes, where key diagnostic information is frequently documented. Of 27 studies, a few showed increased genetic testing and identification of a condition; but most studies relied on retrospective, single site studies, and low validity.

9) Cognitive mechanisms underlying anchoring bias in diagnosis: a randomized controlled experiment.

Mamede S, Schmidt HG, Specian Junior FC, et al. Diagnosis (Berl). Epub 2026 May 14.

Anchoring bias happens when physicians fail to adjust diagnostic decisions despite conflicting information. This Brazilian eye-tracking study of 32 medical residents examined differences in attention when distracting features were placed in different locations in case vignettes. Accuracy of diagnosis did not differ based on attention to the distracting features, but accurate physicians remembered more critical diagnostic information and less distracting features. This implies that the accurate clinicians have different cognitive processing (retaining key data and discarding the distractors) than those who were inaccurate although they viewed the information similarly.

10) Cancer diagnostic delay rates associated with a population-based screening trial evaluating a cell-free DNA multicancer early detection test.

Mann S, Nascimento de Lima P, et al. JAMA. Epub 2026 May 30.

Novel innovations in cancer diagnostic care at a population level may have unintended "spillover" side effects. This study compared cancer diagnostic delays across NHS regions participating in the large multicancer Galleri screening trial with those that were not. Participating regions experienced a small but statistically significant increase in diagnostic delays during the trial’s first six months in part possibly due to increased cancer referral patterns, highlighting the importance of evaluating system-level consequences alongside potential benefits of new screening technologies.

11) AI in point-of-care imaging for clinical decision support: systematic review of diagnostic accuracy, task-shifting, and explainability.

Wadie P, Zakher B, Elgazzar K, et al. JMIR AI. 2026;5:e80928.

AI integration is a promising approach for expanding service access in settings with limited access to specialists. This systematic review from settings examines the use of AI to support clinical decision-making with point-of-care imaging in global health settings, such as colposcopy, retinopathy screening, and DVT diagnosis. The findings suggest that AI may achieve strong diagnostic performance and move practitioners from nonspecialist to specialist-level performance. However, important evidence gaps remain in understanding its true potential, including limited evaluation of patient outcomes, explainability, regulatory readiness, and performance across diverse clinical and geographic settings.

12) Impact of using artificial intelligence as a second reader in breast screening including arbitration.

Warren LM, Venton J, Young KC, et al. Nat Cancer. 2026;7(3):507-521.

Artificial intelligence may help address workforce pressures in breast cancer screening by supporting mammography interpretation and review. In this retrospective study of 50,000 women from two NHS screening centers, researchers evaluated AI as a replacement for one radiologist in a standard double-reading workflow, including specialist arbitration. AI-assisted screening was noninferior to the traditional approach while reducing overall reading workload, suggesting a potential path to maintaining screening quality with greater efficiency.