CODEX Digest - 2.26.26

1) Integrating AI scribes into medical education: guardrails for preserving clinical reasoning. 

Abernethy J, Shah A, Chen B, et al. J Gen Intern Med. Epub 2026 Feb 2.

AI scribes transcribe ambient audio into organized records and should lessen clerical tasks and allow clinicians to focus more on patients. This pilot project identifies shortcomings in AI scribe content, which among others, negatively impact diagnostic prioritization, differential diagnosis specificity, and uncertainty documentation. The authors suggest seven best practices to ensure AI scribes promote thoughtful clinical practice in learners and encourage educators to help guide effective AI use.

2) Artificial intelligence and radiologists in pancreatic cancer detection using standard of care CT scans (PANORAMA): an international, paired, non-inferiority, confirmatory, observational study.

Alves N, Schuurmans M, Rutkowski D, et al for the PANORAMA consortium. Lancet Oncol. 2026;27(1):116-124.

AI can enhance diagnostic accuracy, but direct comparisons with radiologists are limited. This study reviews cohorts from the Netherlands and US, finding that AI significantly outperformed radiologists in detecting pancreatic ductal adenocarcinoma on routine CT scans, suggesting potential for earlier cancer detection and better patient outcomes.

3) Reliability of LLMs as medical assistants for the general public: a randomized preregistered study.

Bean AM, Payne RE, Parsons G, et al. Nat Med. Epub 2026 Feb 9.

LLMs can accurately diagnose a broad range of conditions in simulated studies where they are provided with a clinical summary of patient information, but this study tested how well LLMs performed when patients inputted their symptoms. The researchers found that while the LLM provided an accurate differential diagnosis and triage advice on its own, the accuracy dropped dramatically when humans users provided the information.

Dive into more insights on this paper from CODEX Director Sumant Ranji, MD, SFHM.

4) Improving patient understanding of radiology reports using generative artificial intelligence: a vignette study of 2000 US adults.

Chen AH, Rudin RS, Levine DM, et al. J Am Med Inform Assoc. 2026;33(2):326-335.

While patients appreciate access to medical reports online, complex terminology may hinder its usefulness. This vignette study finds that AI simplification of radiology reports into plain language boosted patient comprehension and confidence, despite some trust issues and 60% still preferred the original versus AI-generated reports

5) Diagnostic accuracy of severe acute respiratory infection definitions in hospitalized children: a systematic review and meta-analysis.

Hersi L, Kant T, Kaziev CL, et al. AMA Netw Open. 2025;8(12):e2550298.

Inconsistent measurement hampers disease detection, surveillance, and public health readiness. This evidence analysis of 13 studies finds that the World Health Organization (WHO) 2014 severe acute respiratory infection (SARI) case definitions have lower sensitivity but higher specificity in children of younger ages, indicating SARI-based diagnostic accuracy may underestimate pediatric disease burden particularly for infants.

6) Association between underdiagnosed and overdiagnosed chronic obstructive pulmonary disease with all-cause mortality.

Huang S, Liang J, Deng Z, et al. J Thorac Dis. 2025;17(11):9329-9341.

Chronic obstructive pulmonary disease (COPD) is often misdiagnosed, but little is known about its long-term outcomes. This retrospective Chinese study using US survey data finds that patients with both underdiagnosis and overdiagnosis of COPD had higher disease burden and mortality. The authors recommend spirometry as a standard screening approach that supports diagnostic accuracy.

7) Evidence of availability bias in diagnostic reasoning following real clinical encounters: a cross-sectional survey of physicians.

Moura EC de, Goldszmidt R, Andrade EB. BMJ Qual Saf. Epub 2026 Feb 2.

Diagnostic errors are a major issue in healthcare, and cognitive biases contribute to their occurrence. This Brazilian survey shows that experience and specialization influence practitioner vulnerability to availability bias. The findings underscore how knowledge and feedback help reduce bias, highlighting the importance of expertise built over time through real-life exposure to clinical challenges.

8) How does socio-economic disadvantage influence the timeliness of lung cancer diagnosis? A systematic review and synthesis of published qualitative studies.

Sanghrajka A, Sharp L, Rowlands G. Public Health. 2025;249:105975.

Lung cancer survival and mortality rates vary by social status. This qualitative systematic review examines qualitative evidence on how low socio-economic status contributes to delays in lung cancer diagnosis in high-income countries, focusing on societal, psychological, and systemic barriers affecting timely diagnosis.

​​​​​​

9) "My confidence in the diagnosis is low, but I am confident in what needs to happen for the patient": a mixed-methods study exploring diagnostic confidence and its impact on physician well-being.

Sloane J, Matin R, Shahid U, et al. Diagnosis (Berl). Epub 2026 Jan 26.

Diagnostic confidence is understudied. This mixed-methods research finds that physicians often maintain strong confidence in their diagnostic strategies even amid uncertainty, and higher confidence correlates with improved well-being.

10) Diagnostic tests for ovarian cancer in premenopausal women with non-specific symptoms (ROCkeTS): prospective, multicentre, cohort study.

Sundar S, Agarwal R, Scandrett K, et al. BMJ. 2026;392:e083912.

Missing ovarian cancer can seriously harm patients. In this UK study of premenopausal women with symptoms or abnormal tests, researchers compared several ways to estimate cancer risk. The commonly used method missed many cancers. The IOTA ADNEX model found far more cases, though with more false alarms, and was recommended for practice.

11) Estimating the number of incorrect tuberculosis diagnoses in low- and middle-income countries.

van Lieshout Titan A, Dodd PJ, Cohen T, et al. Nat Med. Epub 2026 Jan 7.

Tuberculosis (TB) is the world’s deadliest infectious disease. Using data reported to the World Health Organization from 111 countries, this study finds that many people tested for TB are diagnosed incorrectly. In 2023, about 2 million people may have been told they had TB when they did not, while about 1 million with TB were missed. Improving accuracy will likely require better bacteriological tests and less trust in symptoms assessment alone.

12) A nurse-led study to investigate factors influencing patients’ perception of diagnostic quality in the emergency department.

Wood EB, Baidoo S, Babiera E, et al. J Emerg Nurs. Epub 2026 Feb 5.

Patient views on safety reveal important aspects of diagnostic performance. This single-center cross-sectional study found generally positive perceptions but identified key disparities by gender, age, and emergency department stay length, suggesting improved communication is needed for nurses to address patient understanding and safety.