CODEX Digest - 4.23.26

1) A national survey of radiographer and assistant practitioner's experiences of incidents, investigation processes and safety culture.

Adamson HK, Chaka B, Marsh E, et al. Radiography (Lond). 2026;32(3):103376.

Safety culture is the foundation of learning from what goes wrong to drive improvement. This study examines radiographers’ and assistant practitioners’ (AP) views on safety culture in UK clinical practice. The responses highlight the need for improved incident management, accessible sharing of lessons learned, and enhanced engagement of radiographers and AP in incident reviews and policy revisions as an element of safe diagnosis and learning from error.

2) Evaluating the impact of age on prostate cancer overdiagnosis using long-term follow-up from a randomised trial. (This is a preprint that has not yet gone through the peer review process). 

Brentnall AR, Rebolj M, Sasieni P, et al. medRxiv. Epub 2026 Jan 28.

Prostate cancer overdiagnosis significantly affects quality of life and healthcare costs. This UK trial shows overdiagnosis increases with age due to competing mortality, while rates remain low for younger men. Policies with high screening rates in older men should be reconsidered to find other opportunistic testing.

3) Do mildly ill patients really need an offline doctor? Deciphering health choices from artificial intelligence diagnostic design and trust dynamics.

Chen A, Lu Y. J Manage Inform Syst. 2026;43(1):237-272.

Discussions abound about whether AI can replace clinicians. This study investigates whether AI can replace clinicians by assessing its impact on offline healthcare use among mildly ill patients in China. Through two experiments, the authors show that AI's design influence types of trust in clinicians and examine how medical fear affects comfort with AI for mild-illness diagnoses.

4) Achieving equitable care for racial minority patients with a lung cancer screening program.

Griffith A, Pratt CG, Whitrock JN, et al. Ann Thorac Surg. 2026;121(4):830-837.

Racial minority lung cancer patients often present with advanced disease, undergo fewer resections, and have poorer survival, which could be due to screening disparities. This cohort study assesses the impact of low-dose computed tomography (LDCT) on diagnosis and treatment in Black men, showing that structured LDCT programs improve screening access and facilitate timely care. The findings underscore the need to optimize screening for equitable outcomes.

5) Understanding clinician edits to ambient AI draft notes: a feasibility analysis using large language models. (This is a preprint that has not yet gone through the peer review process). 

Guo Y, Zhou Y, Hu D, et al. medRxiv. Epub 2026 Mar 2

Reviewing clinician edits to ambient AI-drafted notes can help assess AI reliability. This pre-print study tests if LLMs can learn from clinician revisions. Medication and symptom models performed well at categizing clinician edits, but diagnosis, orders/tests/procedures, and social history edits were less precise. Prompt engineering worked helped categorize straightforward edits, but complex cases need triage, with human review of labeled edits.

6) Teaching diagnosis in a fragmented healthcare system.

Jerjes W, Majeed A. Postgrad Med J. Epub 2026 Mar 20.

Diagnostic processes can be challenged outside traditional educational settings. In outpatient and community care, teams often diagnose over time with limited information and unreliable follow-up. Safe practice requires assessing reliability, communicating uncertainty, setting safety limits, and clearly assigning responsibility. This commentary calls for educators to focus on these skills and encourage system improvements for diagnostic safety.

7) Public risk perception, institutional AI adoption, and diagnostic safety: an exploratory cross-level analysis using a tracer condition approach.

Li Z, Lyu M, Huang J. Digit Health. 2026;12:20552076261425376.

Safe AI integration in healthcare means tackling both social concerns and clinical dependence on algorithms. This study analyzes Brazilian data to examine how social pressure influences technology strategies, measures the link between AI usage and diagnostic errors, and highlights the value of tracer conditions as tools for evaluating AI safety. The findings reveal underlying sociotechnical factors shaping medical AI adoption.

8) Characterizing diagnostic disclosure communication strategies and challenges in dementia care: a qualitative study of interprofessional clinicians across three systems.

Paladino J, Chavez Granados H, Bernstein Sideman A, et al. Dementia. 2026;25(2):376-400.

Disclosing a new diagnosis of dementia is difficult for clinicians and often falls short of patient and caregiver needs. This interview study identifies communication challenges, as well as skills clinicians use to build trust, tailor messages, and support emotional coping during diagnosis. The findings suggest that disclosure should focus on building a supportive environment rather than simply sharing medical facts.

9) The Elusive Body Patients, Doctors, and the Diagnosis Crisis.

Sifferlin A. Penguin Group; 2026

It is well known that patients struggle with undiagnosed health conditions and that shortcomings in the American healthcare system contribute to diagnostic delay. This book examines obstacles to accurate diagnosis, highlights patient and physician perspectives on the diagnostic journey, and explores what the search for answers reveals about systemic factors degrading diagnostic excellence.

10) Diagnostic timing and ovarian cancer survival.

Soppe SE, Kuo T-M, Bae-Jump VL, et al. JAMA Netw Open. 2026;9(3):e262434.

Earlier diagnosis of ovarian cancer doesn’t always improve survival. In this study, patients diagnosed quickly often had worse outcomes—likely because their disease was already severe (a “wait time paradox”). But longer delays were also linked to poorer survival. Simple, linear analyses can miss this pattern, which may explain why research finds no clear benefit to earlier diagnosis and could unintentionally discourage efforts to detect this hard-to-find cancer sooner.

*First author Sarah Soppe was awarded the People's Choice Award for Best of the Abstracts at DEX25 for her work addressing challenging yet crucial questions in cancer diagnosis. 

11) Closing the loop: a custom artificial intelligence agent to improve detection of radiologist follow-up recommendations.

Treacher A, Moran B, Case M, et al. NEJM Catal. 2026;7(3).

Inadequate follow-up and incomplete tests contribute to diagnostic failures. This study examines a custom AI system designed to help the digital health team reliably identify patients needing follow-up and integrate findings into patient outreach. The AI agent improved detection of missed diagnoses in radiologist notes, patient outreach, and scheduling. By streamlining follow-up identification, the tool increased the chances of patients receiving timely care to improve outcomes in busy clinical environments.

12) Economic evaluation of a digital symptom checker for endometriosis using a Markov decision process model.

Xu Y, Prentice C, Torres-Rueda S, et al. NPJ Digital Med. 2026;9(1):128.

Digital symptom checkers (SCs) are increasingly used for early symptom recognition, but their cost-effectiveness is not well established. This study evaluates a digital SC for endometriosis, showing it can reduce diagnostic delays, improve outcomes, and save costs across various scenarios. These results suggest digital SCs may support timely and efficient care.