CODEX Digest - 6.26.25

Variation in the use of primary care-led investigations prior to a cancer diagnosis: analysis of the National Cancer Diagnosis Audit. 

Akter N, Lyratzopoulos G, Swann R, et al. BMJ Qual Saf. 2025;34(6):367-376. 

Variation in standard processes can reduce quality and reliability. This study reviewed test use, such as imaging, to determine suspected cancer by general practitioners before referring patients to a specialist, as documented in a United Kingdom national cancer care audit. The study shows wide variations in cancer workup which may contribute to diagnostic disparities and need for gold standards. The results underscore the need for improvements to address overtesting and undertesting that could affect diagnosis.

AI-driven healthcare: fairness in AI healthcare: a survey.

Chinta SV, Wang Z, Palikhe A, et al. PLOS Digit Health. 2025;4(5):e0000864

Concerns regarding the biased nature of artificial intelligence system data and algorithms damper the optimism of AI to enhance diagnosis and other aspects of health care. This review highlights the impact of AI in a variety of specialties. The authors discuss weak points across the machine learning data processing continuum and the effects these biases have on diagnosis, resource use, and other segments of health care processes. Mitigation strategies include synthetically augmenting underrepresented groups in datasets and incentivizing the collection of more inclusive datasets.  

Interventions to improve timely cancer diagnosis: an integrative review.

Graber ML, Winters BD, Matin R, et al. Diagnosis (Berl). 2025;12(2):153-162. 

Cancer is one of the “big three” areas of concern in the quest for diagnostic excellence. This review summarizes evidence on misdiagnosis of breast, lung, and colorectal cancer in out-patient care. The authors discuss improved follow-up communication, patient engagement, and electronic health record systems as targets to enhance safety.

Safety and efficacy of digital check-in and triage kiosks in emergency departments: systematic review.

Lammila-Escalera E, Greenfield G, Aldakhil R, et al. J Med Internet Res. 2025;27:e69528. 

Emergency departments are complex triaging and diagnosis environments due to patient demand, flow uncertainties, and staffing challenges. This systematic review examines the utility of digital kiosks to gather needed data from patients and improve timeliness of identifying high-acuity patients. In reviewing five studies, the authors found while digital kiosks can effectively identify high-acuity patients, they were less efficient operationally and had high over-triage rates and poor concordance with nurse-assigned triage scores.  Improvements in the evidence base are needed for the concept to be implemented across emergency rooms.

Large language models and text embeddings for detecting depression and suicide in patient narratives. 

Lho SK, Park SC, Lee H, et al. JAMA Netw Open. 2025;8:e2511922.   

Suicide of a patient is a sentinel event, and efforts to improve diagnosis of depression could be useful in limiting its occurrence. This cross-sectional study used large language models and text-embedded models to detect depression and suicide risk in self-concept narratives of psychiatric patients​ ​​​from a medical center in South Korea​. While illustrating potential for improvement, the innovation will require further testing to confirm its safe use in active care. 

Diagnostic stewardship of blood cultures in the pediatric ICU using machine learning. 

Martin B, DeWitt PE, Payan M, et al. Hosp Pediatr. Epub 2025 May 22.  

Diagnostic stewardship affects the ordering, handling, and reporting of diagnostic tests to improve process reliability, reduce costs, and unnecessary testing that may contribute to patient harm and discomfort. This retrospective cohort study reports on the use of pediatric intensive care unit data to train a machine learning classifier in reducing blood culture test use for low-risk children. The authors found the system accurately predicted >25% negative blood cultures showing potential to reduce over testing in this patient population. 

​The Epidemiology of Missed and Delayed Medical Diagnosis: Implications for Health Equity and Public Health.

Mundt KA, Salvador D, Wyatt R, eds. Front Public Health. 2024-2025. 

This international article collection provides insights into incidence patterns and causes of diagnostic error. It examines advancements and methods for understanding diagnostic error with an emphasis on health equity and public health. The topics covered include diagnostic communication in pediatric cancer care, sociodemographic disparities, and race use in medical algorithms. 

Characterizing personal clinical cognitive uncertainty and its association with clinical judgment.

Simard-Sauriol P, Wassef A, Peters E, et al. J Eval Clin Pract. 2025;31(4):e70124. 

Individual and organizational understanding of and patience with uncertainty can affect patient outcomes and clinician wellbeing. This study explores the relationships between clinical uncertainty, judgement, and knowledge. The authors found learners able to assess their level of uncertainty appropriately in clinical situations and their awareness of uncertainty increased with training and experience. The findings inform educational efforts to enhance behaviors associated with uncertainty to improve clinical judgement. 

The impact of definitions of disease on overdiagnosis.

Tikkinen KAO, Halme ALE, Guyatt GH, et al. JAMA Intern Med. Epub 2025 June 9.  

Overdiagnosis can result in waste and potential patient psychological distress and physical harm. This commentary discusses how overmedicalization, commercialization and clinical definition evolution contributes to overdiagnosis and highlights both physician and researcher roles in mitigating the problem.  

Diagnostic scope: the AI can’t see what the mind doesn’t know.

Weissman GE, Zwaan L, Bell SK. Diagnosis (Berl). 2025;12(2):189-196. 

The distinct clinical setting experience on a defined set of diagnoses is referred to as diagnostic scope. This commentary describes how diagnostic scope can be used by system developers to train artificial intelligence for use at one organization. The authors highlight the importance of the accurate development of diagnostic scope and its evaluation at the local level to ensure AI supports safe diagnosis and treatment at that setting.