CODEX Digest - 4.9.26

1) Push notifications can be used for safe and timely communication of critical lab results in the emergency department.

Amin DP, Bonomo M, Chesebro L, et al. Jt Comm J Qual Patient Saf. Epub 2026 Mar 18.

Poor communication of test results can hinder diagnostic accuracy. This quality improvement initiative evaluates push notifications for critical results in one emergency department. Push notifications took longer to be acknowledged by clinicians compared to phone calls, and a smaller proportion were acknowledged within 30 minutes, although the intervention was time-saving for laboratory personnel. Push notifications may be an option for critical test result alerts, but will require further study to ensure patient safety is not compromised by delayed acknowledgements.

2) Anything but endo: diagnostic buck-passing in endometriosis diagnosis.

Dexter R, Kitts M, Welty H et al. Sociol Health Illn. 2026;48(1):e70142.

Patients with endometriosis often face prolonged diagnostic journeys marked by physical and emotional distress. This interview study introduces the concept of diagnostic buck-passing, describing how patients cycle through various specialists, contributing to diagnostic delay. The findings show that these struggles are not just mischance but reflect the systemic history of underfunding research on women’s health, which has left doctors with major gaps in knowledge about conditions like endometriosis.

3) An AI approach to differentiating lung squamous cell carcinoma from metastases of other origins.

Evans MG, Ribeiro JR, Maney T, et al. JAMA Netw Open. 2026;9(3):e260908.

AI shows potential for improving accuracy in pathology. This cross-sectional study explores the use of AI as a screening tool to identify misdiagnoses in lung squamous cell carcinoma. Using a multipronged AI-assisted approach, the study found a meaningful number of misdiagnoses. These AI-identified diagnostic changes may help guide disease prediction and therapy selection, underscoring AI’s clinical promise.

4) From tool to teammate in a randomized controlled trial of clinician-AI collaborative workflows for diagnosis.

Everett SS, Bunning BJ, Jain P, et al. npj Digital Med. Epub March 18, 2026.

AI performance shows promise as a tool to support diagnostic decision-making. This randomized controlled trial tests a custom system where the physician and AI give independent diagnostic opinions, then the AI combines them to highlight agreements and differences. Whether the AI went first or second, it improved the doctor’s accuracy, showing the potential of AI as a collaborative partner rather than a replacement for decision makers.

5) Evaluating the AI potential as a safety net for diagnosis: a novel benchmark of large language models in correcting diagnostic errors. (This is a preprint that has not gone through the peer review process).

Hassoon A, Peng X, Irimia R, et al. medRxiv. Epub 2026 Feb 24.

Diagnostic errors often happen early in care processes when a patient’s condition is most uncertain. This study tests whether 16 prominent AI models could spot and challenge an incorrect diagnosis. The top models caught about half of the errors, but performance varied widely and were easily influenced by non-clinical factors, showing that current AI models still have significant limitations.

6) Cognitive biases and collapse of prioritization accuracy under incongruent clinical data: a mixed-methods study of nursing diagnostic reasoning.

Milani A, Saiani L, Gandini S, et al. Diagnosis (Berl). Epub 2026 Mar 13.

Mistakes in prioritizing patient care during diagnostic tasks are critical but unappreciated errors. This mixed-methods Italian study finds that when nurses receive conflicting clinical information, their prioritization accuracy drops sharply—regardless of experience or education. This happens due to predictable cognitive biases, revealing a fundamental flaw in how humans process complex information under uncertainty.

7) Diagnostic outcomes among patients with positive multi-cancer early detection test results.

O’Donnell EK, Kauffman TL, Asnis S, et al. Cancer Res Commun. Epub 2026 Feb 20.

A collective approach to early disease detection may transform cancer care. This study evaluated a new blood test for early detection of multiple cancers. In a small patient group, the multi-cancer early detection (MCED) test accurately identified the cancer’s origin in about 90% of cases. MCED tests offer potential for wider screening and early diagnosis but face challenges like specialist access, false positives, and follow-up issues.

8) Eligibility and prognostic performance of smoking duration-based versus pack-year-based U.S. national lung cancer screening criteria across racial and ethnic groups.

Su CC, Ding VY, Ten Haaf K, et al. Ann Intern Med. 2026;179(2):196-206.

Screening guidelines can unintentionally acerbate racial disparities. This prospective study compares screening eligibility rules across diverse groups. Using a 30-year smoking history reduced racial disparities in who qualifies for lung cancer screening, especially for African Americans and Latinos. Risk-based approaches performed well overall but sometimes increased gaps.

9) Synergy of diagnosis coding between administrative claims and electronic health records of large patient populations across multiple healthcare organizations.

Xiong MC, Pittell H, Kitchen C, et al. JAMIA Open. 2026;9(2):ooag030.

Accurate diagnosis records are essential for research, billing, and patient care. This study compared diagnostic data in insurance claims and electronic health records for the same patients. Diagnostic information was often inconsistent across sources, and electronic records missed nearly half of all diagnoses, highlighting the need for better data completeness to support population health efforts.

10) Medical student experiences with ChatGPT: national cross-sectional study.

Xu AY, Speakman S, Piranio VS, et al. JMIR Formativ Res. 2026;10:e76838.

AI-driven chatbots are becoming widespread in medical education. This survey study finds that medical students use these tools for both studying and clinical practice—42.7% reported using ChatGPT to help generate possible diagnoses. As AI use grows, the authors suggest medical schools adapt their curricula to address both the opportunities and challenges AI presents to students.

11) An agentic system for rare disease diagnosis with traceable reasoning.

Zhao W, Wu C, Fan Y, et al. Nature. 2026;651(8106):775-784.

Rare diseases affect over 300 million people but are hard to diagnose due to limited clinical team familiarity and diverse symptoms. This Chinese study tested an AI system designed to assist with rare disease diagnosis through its use of information and reasoning process. Across nearly 3,000 rare diseases representing 14 medical specialties, the system had superior performance compared to existing methods on several benchmarks and high expert agreement on its reasoning chains.

12) Nursing Homes Inappropriately Diagnosed Residents with Schizophrenia to Mask the Misuse of Antipsychotic Drugs.

Office of Inspector General; Department of Health and Human Services: 2026.

Inappropriate diagnoses can lead to unnecessary treatment, waste, and patient harm. This review of 40 inspections reveals that some nursing home residents were incorrectly diagnosed with schizophrenia to boost antipsychotic drug quality scores. It highlights how financial and regulatory incentives influence medical decisions. The report recommends commitment to ongoing improvement projects, better data use, and increased communication with patients and families about antipsychotics to address the problem.