CODEX Digest - 12.4.25

1) Performance of large language models on the acute coronary syndrome guidelines using retrieval-augmented generation. (subscription required)

​Alexandrou M, Kumar S, Mahtani AU, et al. JACC Cardiovasc Interv. 2025;18(20):2458-2467.

Improving cardiac symptom diagnosis can reduce patient harm and anxiety. This study explores methods of infusing specialty-focused cardiac guideline content into LLM workflows to enhance timeliness, improve accuracy, and reduce hallucinations. The strategy enhances the potential of LLMs to provide decision making assistance that aligns with universal care recommendations. 

2) Comparing clinical decision support systems for improving follow-up of abnormal cervical cancer screening test results. (subscription required)

Atlas SJ, Burdick TE, Wright A, et al.J Biomed Inform. 2025;170:104908.

Test follow-up is crucial yet error prone in diagnosis. This study finds that clinical decision support systems (CDSS) combined with patient outreach improved follow-up of abnormal cervical cancer screening, while CDSS alone did not. Future CDSS may benefit from open-source tools created through public-private partnerships. 

3) Evidence categories in systematic assessment of cancer overdiagnosis.

Barchuk A, Nordlund NK, Halme ALE, et al. BMJ Evid Based Med. 2025;30(5):333-339.

Diagnosing tumors that would never cause harm creates unnecessary stress and waste. This review summarizes key evidence on cancer overdiagnosis from four sources: (1) registry data, (2) prevalence studies, (3) diagnostic impact on incidence and mortality, and (4) effects of care advances on mortality. It outlines each evidence type’s strengths and limits, offers cancer-specific examples, and shows how combining sources of evidence clarifies overdiagnosis. 

4) Diagnostic Research.

Booth G, Nelson P, eds. AMA J Ethics; 2025;27(10):e713-e767.

The Undiagnosed Diseases Network (UDN) offers insight into how precision medicine, genetics, and new diagnostics shape care for all patients. This special issue uses UDN experience to explore the ethical and equity duties of clinicians, researchers, and policymakers in rare and undiagnosed diseases, emphasizing ethical, compassionate, patient-centered care even when diagnoses remain uncertain. 

5) Digital detection of dementia in primary care: A randomized clinical trial.

Boustani MA, Ben Miled Z, Owora AH, et al. JAMA Netw Open. 2025;8(11):e2542222.

Detecting Alzheimer disease and related dementias (ADRD) can be difficult in primary care settings. This study examines the use of a patient reporting tool along with a machine learning method that analyses EHR data to identify ADRD in clinics. The combined approach was found to be scalable for early detection of ADRD in primary care environments. 

6) Evaluation of four learning collaboratives for improving diagnostic excellence in radiology.

Holdsworth LM, Mui HZ, Tomkins KG, et al. Learn Health Sys. 2025;9(4):e70035. 

Learning collaboratives are used in healthcare to promote practice improvement through collaboration among clinical teams from different organizations. This assessment examined factors associated with the effectiveness of diagnostic excellence learning collaboratives and identified social system aspects such as leader credibility, engagement, and guidance as influential elements. 

7) Patterns and implications of missed injuries on computed tomography imaging in older blunt trauma patients.

Kishawi SK, Mahajan A, Nahmias J, et al for the EAST SAMMI MCT Group. Surgery. 2025;185:109524.  

Missed diagnoses in older adults can lead to greater morbidity. This study found that over 8% of older blunt trauma patients initially received suboptimal CT imaging that missed an injury. Clinicians should be particularly vigilant with patients who are transferred, need consults, or have language barriers, as these factors are linked to lower imaging quality. 

8) Transformative impact of deep learning and machine learning in oncology: a comprehensive review of AI-based approaches for early detection, diagnosis and therapeutics across different cancer types. (subscription required)

Kour T, Raina JK, Gondhi NK, et al. Arch Computat Methods Eng. Epub 2025 Sep 29.

Early detection of disease is a promising role for advanced computer systems. This review analyzes deep learning (DL) applications in oncology, highlighting both their promise and challenges like data availability, model generalization, and diagnostic consistency. It summarizes commonly used DL techniques, compares tool performance, and outlines future research directions with high impact potential to improve cancer diagnostic timeliness. 

9) Emotions and clinical reasoning in medical education and clinical practice: a scoping review. 

​Merkebu J, Y Soh M, Loncharich M, et al. Acad Med. 2025;100(11):e80-e90.

Clinical reasoning in medicine is strongly affected by emotions, yet their impact is less studied than cognitive factors. This review highlights that emotions influence all aspects of patient care, including diagnosis, and actively shape clinical decisions. The findings suggest clinicians would benefit from training in emotional awareness and regulation to improve clinical reasoning. 

10) How accurate is point-of-care ultrasound for detecting paediatric appendicitis? A systematic review and meta-analysis. 

​Miller B, McCreary D, Rees J. Arch Dis Child. Epub 2025 Oct 17. (subscription required)

Rapid diagnosis of pediatric appendicitis can be difficult. This review examines the accuracy of point-of-care ultrasound (POCUS) for this purpose. The results suggest POCUS is an effective diagnostic tool due to its moderately high sensitivity, but it should not be solely relied on to rule out appendicitis. 

11) Leveraging implementation science to address diagnostic disparities and promote equity in healthcare. (subscription required)

​Schulson L, Drainoni M-L, Austad K. Jt Comm J Qual Patient Saf. Epub 2025 Oct 1.

Diagnostic errors cause preventable harm and worsen health inequities. Improving accuracy is essential for safety and fairness. Applying implementation science with an equity focus helps ensure diagnostic improvements are effective, sustainable, and reduce disparities in care. 

12) Medical imaging and pediatric and adolescent hematologic cancer risk. 

Smith-Bindman R, Alber SA, Kwan ML, et al. N Engl J Med. 2025;393(13):1269-1278.

Imaging overuse may lead to waste, increased workload, and patient harm. This study finds that a small but significant increase in hematologic cancer risk was linked to medical imaging radiation exposure in children and adolescents. The results, which highlight the risk for radiation-related cancer in young people, call for improvements in pediatric imaging decisions.