CODEX Digest 03.12.26

AI-based smartphone apps may help detect skin cancer, but real-world studies are limited. This Belgian study tests an app on 1,904 lesions (9.7% malignant) using different smartphones and conditions. The application was reasonably accurate, and dermatologist review reduced false positives but missed more cancers. However, only participants with fair skin tones were represented. Results stress the need for independent validation of AI tools in clinical settings.

Few studies have evaluated telemedicine's impact on cardiovascular diagnostics and care. This large health system study finds that during the COVID-19 pandemic, higher telemedicine use by clinicians was linked to less diagnostic testing and lower initiation of guideline-recommended therapy in heart failure patients.

Patients using LLMs to diagnose symptoms receive inconsistent results, posing safety risks. This study assesses ChatGPT Health’s triage accuracy using a wide range of clinician-authored scenarios. The results find that ChatGPT Health under-triaged emergencies in over 50% of cases, frequently recommending non-urgent care when immediate action was needed and failed to provide crisis resources in a suicide scenario. Across all scenarios, ChatGPT Health’s triage advice was more accurate when objective data (vital signs or lab results) were included in the vignettes. Dive into more insights from UCSF CODEX Director Sumant Ranji, MD, SFHM.

Disagreements between radiologists and AI in interpreting imaging studies raise concerns about diagnostic accuracy, medical decision-making, and patient safety. This review examines how differences in perceptual learning can lead to such discrepancies. While AI may result in earlier disease detection and non-invasive lesion analysis, it also presents risks like demographic and technical bias and lack of transparent reasoning.

Lung cancer screening with low-dose CT is underused in the US, partly due to incomplete smoking history records and limited time in primary care. This randomized study finds that adding a patient-facing tool to EHR workflows increased both identification of eligible patients and ordering of low-dose CTs compared to clinician tools alone, though the absolute increases were modest. More research is needed on intensive interventions to improve screening rates further.

Standardized handoff tools for diagnostic pauses may reduce information loss in complex settings. This multicenter qualitative study evaluates the ICU-PAUSE tool, a structured handoff program that helps clinicians summarize care, address unresolved issues, and discuss uncertainty. Results showed positive feedback on the pause tactic, value in the diagnostic uncertainty section, and that local adaptation aided implementation success.

Effective communication of urgent, unexpected diagnoses is vital for diagnostic excellence and patient safety. This article summarizes focus group findings from the American Society for Clinical Pathology Collaborative for Diagnostic Excellence. Insights from the 11 pathologists representing a variety of practice settings emphasize standardized definitions, protocols, and collaboration as ways to improve results communication. Future studies should evaluate these approaches and their effects on patient care.

Medical gaslighting contributes to misdiagnosis, delayed care, emotional distress, and mistrust in healthcare, especially among marginalized groups and women. Through two diagnostic case studies, this commentary highlights how bias and systemic issues worsen health disparities. The authors recommend clinician bias training, stronger patient advocacy, and policy reforms to rebuild trust and improve outcomes for vulnerable populations.

Clinical quality measures (CQMs) monitor various aspects of care, including diagnostics. This commentary outlines three scalable CQM projects that used natural language processing (NLP) to assess diagnostic performance for venous thromboembolism, pneumonia, and pulmonary embolism—conditions with non-specific symptoms and safety risks. These projects show strong potential for broader adoption, highlighting how NLP can enhance CQMs focused on diagnosis through effective integration.

Diagnostic errors are common in all specialties often due to gaps in clinical reasoning training during medical education. This North American study evaluates whether a classroom curriculum using virtual patients with automated scoring and deliberate practice improves diagnostic accuracy and reasoning. Results indicate that deliberate practice effectively trains clinical reasoning skills, reduces misdiagnosis rates, and is beneficial even for first-year medical students before they gain active clinical experience.

Conservative disease management may limit diagnostic accuracy. This retrospective single-center study evaluated the incidence of unexpected diagnoses in children who underwent appendectomy for presumed uncomplicated appendicitis and examined related outcomes. While few patients were found to have alternative conditions, they were significant, highlighting the importance of surgical specimen evaluation and the risk of missing other pathologies with non-operative treatment.

Autopsy-based studies help estimate cancer prevalence, including hidden and missed cases. This cohort study in Japan reveals a significant number of undiagnosed cancers, some with metastatic risk, emphasizing the importance of autopsies for cancer assessment and the need to improve early detection while limiting overdiagnosis.