CODEX Digest - 10.23.25

1) Physical harms in colorectal cancer screening: an overview of the reporting in systematic reviews and randomised controlled trials. 

Bie AKL, Handberg Juul Martiny F, Jauernik CP, et al. PLoS ONE. 2025;20(9):e0331104.  

Gaps of reported harms in the literature minimize the evidentiary quality of research exploring care problems. This analysis found inconsistencies in colorectal screening program harm reporting in randomized controlled trials and systematic reviews. The results illustrate concerns on the value of safety conclusions drawn by these studies. 

2) Artificial intelligence and computer-aided diagnosis in diagnostic decisions: 5 questions for medical informatics and human-computer interface research (subscription required) 

Brunyé TT, Mitroff, SR, Emore JG. JAMIA. Epub 2025 Oct 17.  

This paper proposes five questions to guide research and development on AI-based clinical decision support systems. These questions are grounded in the existing evidence of where diagnostic errors are being made and aimed at ensuring effective integration into real-world clinical workflows taking into account cognitive science alongside medical informatics and human-computer interface design.   

3) Applying ChatGPT to plan and create a realistic collection of virtual patients for clinical reasoning training.

Fąferek J, Kononowicz AA, Bogutska N, et al. BMC Med Educ. 2025;25(1):1277.  

Clinical reasoning tools are costly to produce. This study compares five virtual patient cases made by ChatGPT compared to virtual patients made by humans. This paper showed the GPT generated virtual patients were not adequate in quality or representativeness and describes what would be needed to serve as a support tool for developing diagnostic education materials. 

4) Delayed venous thromboembolism diagnosis and mortality risk. 

Kang M-J, Schreiber R, Baris VK, et al. JAMA Netw Open. 2025;8(9):e2533928.  

Venous thromboembolisms (VTE) or blood clots are a priority in diagnosis. This study uses EHR systems to analyze VTE diagnostic delays and determine their connection with mortality to inform quality improvement efforts. Contributors to missed diagnosis were looked at in the context of clinician, patient, and other factors. 

5) Digital quality measure of potentially avoidable emergency presentations among patients with colorectal cancer. (subscription required) 

Khalaf N, Ali B, Zimolzak A, et al. BMJ Qual Saf. Epub 2025 Sep 30. 

Identifying potentially avoidable emergent cancer presentations may inform diagnostic improvement efforts. This study proposes an electronic quality measure to efficiently flag cases of emergency presentations of colon cancer, so organizations can prevent diagnostic delays contributing to increased harm from disease. 

6) Evaluation of prostate cancer pathology reports generated by ChatGPT to enhance patient comprehension.

Khanmammadova N, Gevorkyan R, Afyouni AS, et al. Sci Rep. 2025;15(1):34459.  

Effectively communicating test results to patients supports shared decision making. This study explores the viability and accuracy of AI generated patient-focused pathology and biopsy reports. Both surgeons and patients found the summaries to be easy to read and useful in understanding the diagnosis.   

7) Facilitators and barriers to the timely diagnosis and treatment of melanoma in Latino persons. (subscription required)

Miller KA, Marìn P, Aguero R, et al. JAMA Dermatol. Epub 2025 Oct 1.

Delays in skin cancer diagnosis contribute to more advanced disease and mortality. This interview study summarizes the insights of 20 Latino patients to uncover systemic factors contributing to their delayed melanoma diagnosis including language and specialist access barriers. 

8) The rise in early-onset cancer in the US population--more apparent than real. (subscription required) 

Patel VR, Adamson AS, Welch HG. JAMA Intern Med. Epub 2025 Sep 29. 

Early-onset cancer incidence is increasing, yet there is a lack of clarity as to why and what this really means. This study explores these questions in the eight fastest growing cancers in younger adults. It highlights which cancers also have higher mortality while others have no mortality change, signaling this may be an issue of overdiagnosis and increased diagnostic inquiry. 

9) Parental involvement in paediatric patient safety incidents in general practice: a cross-sectional study.

Purchase T, McFadzean IJ, Donovan L, et al. Br J Gen Pract. 2025;75(758):e607-e617.   

Parent advocacy can help to identify unsafe situations in healthcare. The study finds that parents were instrumental in addressing diagnostic delay, mistakes, and assessments, amongst the other safety foci reviewed. The research acknowledges that parents can also contribute to safety incidents but suggests that healthcare builds on the evidence of parent engagement to improve safety.

10) Roadmap to success: blueprint for enterprise-wide deployment of a point-of-care ultrasound platform, inclusive of governance, policy, education, credentialing, and quality assurance (parts 1 and 2). 

Waldman D, Doughton J, Pino C. J Clin Imaging Sci. 2024;14:35; 2025;15:28.  

Point of case ultrasound (POCUS) can increase diagnostic excellence at the bedside by quickly providing real-time imaging information for questions about physical conditions in a cost-effective manner. This implementation commentary describes steps taken in an academic medical institution to deploy POCUS and discusses nursing education, image management, value, and credentialing activities associated with the initiative. 

11) Relationship between diagnostic accuracy and self-confidence among medical students when using Google search: a mixed-method study.

Yanagita Y, Ishizuka K, Yokokawa D, Shikino K. PLoS ONE. 2025;20(9):e0332918.  

Some may poke fun at “Dr. Google”, but a simple Google search may be a valuable clinical diagnostic support tool. This Japanese study explores how fifth-year medical students used Google as a decision-support tool and assessed its effect on their diagnostic accuracy and confidence. Results, while of limited generalizability, show that Google improves students’ clinical reasoning and educational outcomes. 

12) Enhancing clinician trust in AI diagnostics: a dynamic framework for confidence calibration and transparency. 

Yu Y, Gomez-Cabello CA, Haider SA, et al. Diagnostics (Basel). 2025;15(17):2204.  

Transparency in LLM development is essential for building physician trust in AI. This proof-of-concept study proposes a means to integrate confidence prediction from AI into recommendations. The AI generated diagnoses corresponded with the “true” diagnosis much more often when rated with high confidence vs. low confidence. This is a promising way to incorporate some transparency into AI diagnostic output.