CODEX Digest - 8.28.25

Deep learning for pediatric chest x-ray diagnosis: repurposing a commercial tool developed for adults.

Agarwal P, Rau A, Ngo H, et al. PLoS ONE. 2025;20(7):e0328295.  

AI tools specific to pediatric care support are limited. This single-center retrospective cohort study demonstrated the viability of the reuse of an adult AI chest radiograph system to diagnose children. The tool was trained on 958 pediatric radiographs and demonstrated high performance on the radiographs tested. Better results were achieved on chest images of older children. These results indicate that repurposing adult AI tools for children show promise as a stop-gap solution until pediatric AI tools are developed.  

Is a score enough? Pitfalls and solutions for AI severity scores.

Bernstein MH, van Assen M, Bruno MA, et al. Eur Radiol Exp. 2025;9(1):67.  

Test interpretation can be affected by human factors that degrade decision making. This commentary discusses the value of AI-generated severity scores on the diagnostic accuracy of radiologists. The authors review human limitations affecting severity score application to test review including the variabilities of people, systems, data, and perceptions. The authors suggest that providing context through false discovery rate and false omission rate transparency could alleviate these potential biases. 

Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study. (subscription required) 

Budzyń K, Romańczyk M, Kitala D, et al. Lancet Gastroenterol Hepatol. Epub 2025 Aug 12.

This retrospective, observational study randomly assigned colonoscopies to be conducted with and without AI tools for polyp detection at endoscopy centers in Poland. After three months of using the AI assistance, the authors found the unassisted adenoma detection rate fell significantly compared to before AI was introduced. The researchers suggest that the continuous exposure to AI may have caused overreliance on the tools and de-skilling among clinicians.  

**This is a recent Editor’s Pick. 

Surgeon-led point-of-care ultrasound-guided thoracic biopsy: a new paradigm in efficient diagnosis and resource-sparing care. 

Gupta V, Gowing SD, Pandya R, et al. J Thorac Cardiovasc Surg. 2025;170(2):369-378.  

Diagnostic rapidity and precision are elements of excellence. This prospective cohort study in Canada tested a surgeon-driven, outpatient clinic, bedside ultrasound-guided biopsy program to improve diagnostic timeliness for 160 patients with thoracic cancers. The researchers found the process to have diagnostic accuracy and reduced time to diagnosis; if implemented it may also free up diagnostic resources for other patients. 

Identifying diagnostic errors in the emergency department using trigger-based strategies.

Khalili M, Enayati M, Patel S, et al. BMJ Open Qual. 2025;14(3):e003389.  

EHR-based triggers are an established mechanism for flagging problems detracting from safe care. This consecutive cohort, retrospective observational design examined the performance of three pre-established triggers in one US emergency department to determine their accuracy in recognizing diagnostic errors. Chart review did not find many errors; detailed review found a very low amount of diagnostic errors. The authors suggest AI and other data-driven approaches could improve trigger tool performance and chart review strategies as a diagnostic error tracking mechanism in the emergency department.

Communicating a diagnosis of mild cognitive impairment or dementia. (subscription required) 

O’Brien K, Largent EA, Karlawish J. JAMA. Epub 2025 Jul 30. 

Patient and family-centered clinician communication of a difficult diagnosis can support trust and patient engagement in their care. This commentary outlines elements of an effective conversation with the patient and family explaining a diagnosis of cognitive disease. Key elements of communication excellence include the involvement of the patient, caregivers, and clinicians, personalized context of the discussion, and the psychological impact of the diagnosis and should be evident in the sharing of information and determination of next steps.      

Equity-driven diagnostic excellence framework: an upstream approach to minimize risk of diagnostic inequity. (subscription required) 

Simsam NH, Abuhamad R, Azzam K. Diagnosis (Berl). 2025;12(3):358-364.  

Inequalities are embedded across healthcare and require systematic approaches to reduce their impact. This commentary used a modified Delphi method model to conduct focus groups across clinical, patient, and administrative healthcare and evaluate a competency framework for Equity-Driven Diagnostic Excellence. The authors suggest next steps of implementing the framework into training and education and other practical applications including reporting patient outcomes and experiences and for use in AI-enabled patient simulations.  

Early detection of ICU-acquired infections using high-frequency electronic health record data

Varkila MRJ, Lancia G, van Smeden M, et al. BMC Med Inform Decis Mak. 2025;25(1):273.  

Hospital acquired infections are a persistent challenge in healthcare that, if recognized early, can be safely addressed. This single center ICU-based cohort study in Utrecht, Netherlands examined infectious disease cases over a seven-year period to develop a longitudinal prediction model that flagged patients at risk of developing nosocomial infection. Results show that the model was accurate in predicting ICU-acquired infection risk and performed slightly better when including vital sign data via a convolutional neural network.  

Artificial intelligence for reducing missed detection of adenomas and polyps in colonoscopy: a systematic review and meta-analysis. 

Wang S-Y, Gao J-C, Wu S-D. World J Gastroenterol. 2025;31(21):105753.  

Colorectal cancer is common yet treatable if diagnosed early using colonoscopy, and AI has the potential to improve this process. This systematic review and meta-analysis examine computer-assisted diagnostic (CAD) system accuracy focusing on the adenoma miss rate, which is more sensitive than the adenoma detection rate. The results indicate that CAD systems were able to reduce missed polyps and neoplasms and add to the literature supporting computer assisted colonoscopy for diagnostic accuracy.

Delayed diagnosis of aortic dissection: a scoping review. 

Xiao Y, Huang S, Zheng D, et al. Rev Cardiovasc Med. 2025;26(6):33487.  

Aortic dissection (AD) is challenging to diagnose, and delayed recognition can be fatal. This helpful review, which also includes studies from Chinese databases, summarizes research on factors contributing to diagnostic delays. Time to AD diagnosis was quite variable, from minutes to days. Demographics such as sex, geographic location, residence in distressed communities, and transfers from non-tertiary care hospitals were associated with longer times to diagnosis.  In addition, atypical symptoms or symptoms mimicking congestive heart failure or acute coronary syndrome were linked to longer diagnosis times.