CODEX Digest - 10.16.25

1) A comparison of blood culture diagnostic stewardship across three emergency departments in a healthcare network. (subscription required) 

Ramos J, Theophanous R, Gettler E, et al. Am J Emerg Med. 2025;93:135-139. 

Blood culture (BCx) diagnostic stewardship helps reduce unnecessary treatments, false positives, and improves patient outcomes and resource use. This study compares the effectiveness of diagnostic stewardship strategies on BCx rates in three emergency departments (ED). Direct audit and feedback mechanisms temporarily reduced BCx use but this effect was not sustained after the intervention ended.  

2) Causes of myocardial infarction in younger patients: troponin-elevation in persons ≤65 years old in Olmsted County. (subscription required) 

Raphael CE, Sandoval Y, Beachey JD, et al. J Am Coll Cardiol. 2025;86(12):877-888. 

Misdiagnosis of heart conditions in women can lead to delayed treatment and mortality. This study examined records of 2​,​780 patients under the age of 65 experiencing​ myocardial infarction​ ​(​MI​)​ events. This retrospective analysis found over 1/3 of MIs were among women and had much different etiologies than men. This paper pushes us to incorporate sex related differences with heart disease diagnoses.   

3) Assessing artificial intelligence as a diagnostic support tool for surgical admissions in the emergency department. (subscription required) 

Rice J, Ó’Briain E, Kilkenny CJ, et al. J Surg Educ. 2025;82(10):103676. 

AI​ is increasingly being considered as a tool to support frontline care. This retrospective study examined how well ChatGPT-4o diagnosed 100 patients presenting with surgical admissions to a Dublin ED​ and compared​​​​ its accuracy to that of early-stage surgical trainees on call. The ​diagnostic ​accuracy of the AI and junior trainee were comparable, but management was less concordant. The paper shows examples of where both learners and AI may make diagnostic mistakes.  

4) Can ChatGPT identify diagnostic uncertainty upon admission?  (subscription required) 

Rodriguez JA, Leeson M, Plombon S, et al. J Gen Intern Med. Epub 2025 Sep 5. 

Early diagnostic errors in hospitalized patients often result from incomplete initial assessments and are signaled by uncertainty in clinical notes. This comparative study explores the use of AI to identify diagnostic uncertainty in real time through the review of admission notes to facilitate harm prevention during or after hospitalization. 

5) Does knowledge of clinical case outcome influence supervisor evaluation of resident clinical reasoning?

van Sassen C, Mamede S, van den Broek W, et al. Acad Med. 2025;100(10):1194-1202.  

Clinical reasoning feedback helps develop diagnostic decision-making skills. This Dutch single-center analysis at a teaching hospital used clinical cases with unclear diagnoses to assess the impact of positive or negative clinical outcomes on supervisors’ review. Negative outcomes did provoke worse scores and more negative and specific feedback. Educators can learn how to address outcome bias in their diagnostic teaching methods.   

6) Critical thinking for 21st-century medicine--moving beyond illness scripts. (subscription required) 

Schwartzstein RM, Iyer AA. JAMA. Epub 2025 Sep 25. 

Clinical reasoning education is challenging. This provocative commentary proposes that as AI will adopt “pattern recognition” in assessment tools, medical education should move away from “pattern recognition” and towards analytic pathophysiological thinking and critical thinking, which may remain a uniquely human value. 

7) Learning from what goes right: a safety-II framework for improving diagnosis at the point of care. (subscription required)

Shimizu T. Diagnosis (Berl). Epub 2025 Jul 17. 

Safety I and ​S​afety II are inverse perspectives representing important approaches to inform work toward diagnostic excellence. This commentary argues that the application of Safety II strategies would benefit from improved application in healthcare. The author shares a five-part protocol that emphasizes routine reflection, the use of diagraming​,​ and teamwork ​to ​learn from diagnostic cases that go right. 

8) Disparities in diagnosis and management of chronic overlapping pain conditions in an observational study of Medicare beneficiaries. 

Smith MA, Knecht TPE, Irwin MN, et al. J Pain. 2025;34:105490.  

Pain conditions such as fibromyalgia, endometriosis​,​ and lower back pain can be difficult to diagnose. This cohort study documented racial disparities in diagnosis of chronic pain conditions that reduced effective treatment. The data found that Black and Asian Medicare patients were less likely to receive a diagnosis, yet more likely to be given a potential harmful opioid prescription for pain. 

9) The STARD-AI reporting guideline for diagnostic accuracy studies using artificial intelligence.

Sounderajah V, Guni A, Liu X, et al for the STARD-AI Steering Committee. Nat Med. Epub 2025 Sep 15.  

Inaccurate or missing diagnostic test study data can reduce effective application of results and bias output. The Standards for Reporting Diagnostic Accuracy (STARD) program guides the communication of diagnostic test accuracy evidence to combat unintended misinformation. This commentary discusses work to create STARD guidance to support reporting of AI-developed test studies. Recommended reporting elements include how AI algorithmic bias was evaluated and what comprised dataset practices. 

10) Missed injuries in trauma care: an analysis of mechanisms and prevention of one of the surgeon’s worst nightmares.

Vieira LF, Grover A, Parreira JG, et al.  Injury. 2025;56(8):112600.  

The chaotic nature of the ​ED​ can contribute to care mistakes. This systematic review explores ​five ​areas of trauma associated with missed injuries: trauma characteristics, injury-specific factors, diagnostic limitations, patient-related challenges, and human factors. Factors such as polytrauma, injury severity, clinician fatigue​,​ and cognitive biases were associated with missed injuries. Trauma team collaboration with subsequent reevaluations was one recommended approach to improve diagnosis in trauma care.

11) Peer perceptions of clinicians using generative AI in medical decision-making.

Yang H, Dai T, Mathioudakis N, et al. NPJ Digital Med. 2025;8(1):530.  

Little is known how physicians perceive colleagues who use AI in real time within healthcare practice. This study asked clinicians to rate vignettes featuring different uses of generative AI and found that colleagues rated other clinicians using AI as a primary tool much lower than no AI used at all, but in the middle if AI was mentioned as a “verification” tool.  Clinician use of AI may have downstream consequences on peer trust and teamwork. 

12) Understanding clinical decision support failures in pediatric intensive care units via applied systems safety engineering and human factors problem analysis: insights from the DISCOVER learning lab.

Zackoff M, Graciela A, Collins K, et al.  J Patient Saf. 2025;21(7Supp):S21-S28.  

Pediatric intensive care units (PICUs) often support patients in challenging conditions that contribute to diagnostic mistakes. This team applied safety engineering and human factors methods, including virtual reality systems, to identify workflow conditions that serve as barriers to the use of clinical decision support in the PICU.