CODEX Digest - 8.7.25

On context specificity and management reasoning: moving beyond diagnosis. (subscription required) 

Boyle JG, Walters MR, Burton FM, et al. Diagnosis (Berl). 2025;12(2):217-222. 

Context specificity occurs when a physician sees two patients with the same symptoms, history, and exam findings, yet comes to different diagnostic conclusions given physician, patient, or environmental conditions that influence management reasoning and decision making. This randomized controlled crossover study in Scotland tests 20 medical students on two encounters with identical clinical information, except one encounter contains contextual factors such as an earlier suggestion of COVID-19. Results show that contextual factors can affect both diagnostic and treatment choices.  

Impact of AI-assisted diagnosis on American patients’ trust in and intention to seek help from health care professionals: randomized, web-based survey experiment.

Chen C, Cui Z. J Med Internet Res. 2025;27:e66083. 

AI-support of diagnosis is a promising application of AI, but patient reception of AI-driven conclusions is underexplored. This study surveyed participants on willingness to seek care and trust in a doctor based on disclosed AI use in four theoretical cases. Results show patient trust in physicians and willingness to seek care was highest in cases where it was explicitly stated they did not use AI and declined when clinicians disclosed using AI. This pattern was not related to partisanship or other demographics and resonates with other findings that patients seek transparency regarding AI use and AI use can worsen medical mistrust.  

Diagnosis and symptom assessment in telepsychiatry vs. face-to-face settings: a systematic review and meta-analysis.

Fujikawa M, Hagi K, Kinoshita S, et al. Psychiatry Clin Neurosci. Epub 2025 Jun 25.   

Video telemedicine is an established option for patient consultation appreciated by both patients and providers. This meta-analysis explores its use in mental healthcare as a tool to identify psychiatric conditions. The authors found telepsychiatry diagnoses to track appropriately with in-person assessments highlighting the value of its use in this specialty. The review was conducted without language restrictions and included studies conducted in Japan, Australia, Iran, and more.  

The performance of machine learning for predicting the recurrent stroke: a systematic review and meta-analysis on 24,350 patients. (subscription required) 

Habibi MA, Rashidi F, Mehrtabar E, et al. Acta Neurol Belg. 2025;125(3):609-624. 

Stroke is one of the “Big Three” serious missed conditions that detract from systemic diagnosis excellence. This review looks at how well machine learning (ML) algorithms recognize stroke patients who are at risk of having another stroke. ML methods including logistic regression and artificial neural networks were found to be effective in predicting repeat strokes. However, the methodological differences between studies call for more uniform research to make these tools more reliable. 

Time to de-implementation of low-value cancer screening practices: a narrative review. 

LeLaurin JH, Pluta K, Norton WE, et al. BMJ Qual Saf. 2025;34(8):547-555.  

Value-based cancer screening programs support diagnostic excellence. This review highlights it took 16 years to de-implement by 50% cervical cancer screening for women over 65, and other grade D cancer screening modes have not yet been de-implemented. Measuring and de-implementing unnecessary care is essential to improve patient care, lower unnecessary invasive testing and costs, and make diagnosis more efficient. 

The Leng Review: An Independent Review of the Physician Associate and Anaesthesia Associate Roles: Final Report.

Leng G, ed. Crown Copyright; 2025.  

This government-commissioned mixed-methods review of advanced practice provider services in the UK’s NHS touches on safety, access, and cost concerns. Further analysis found that UK physicians and patients were less confident in physician associates (PA) or anesthesia associates (AA) taking a primary role in initial diagnosis. The results point to tension with addressing workforce burdens and interdisciplinary care with unclear evidence regarding which roles should provide a diagnosis.  

Delayed hypertension diagnosis and its association with cardiovascular treatment and outcomes. 

Lu Y, Brush JE Jr, Kim C, et al. JAMA Netw Open. 2025;8(7):e2520498.  

The timeliness of a hypertension diagnosis launches effective treatment and care. This cohort study of 311,743 adults from one regional health system explored EHR data to determine cardiovascular outcomes for patients with delayed diagnosis and treatment initiation. The study found delayed diagnosis of hypertension to be common and highly connected to medication therapy delays and patient harm, particularly for patients whose diagnosis was delayed more than one year after a recorded blood pressure elevation. Greater delays were also associated with patient demographics with significant differences found in non-Hispanic Black patients.  

Utility of the US Preventive Services Task Force for preeclampsia risk assessment and aspirin prophylaxis. 

McElrath TF, Jeyabalan A, Khodursky A, et al. JAMA Netw Open. 2025;8(7):e2521792.

Preeclampsia causes significant preventable perinatal mortality; recent guidelines have utilized risk stratification to give aspirin prophylaxis. This cohort study sought to evaluate the clinical validity of established USPSTF criteria on identifying risk for developing preeclampsia and for allocating aspirin prophylaxis. The researchers found that the risk levels did seem to match low, moderate, and high incidence of preeclampsia, but not all patients who were eligible for prophylaxis received aspirin. This study also implies that the much-criticized risk factor focused on race-based medicine for Black patients was not clinically valid. This study highlights that “moderate” risk factors will require further synthesis and that “high” risk patients are undertreated for prophylaxis. 

The impact of large language models on diagnostic reasoning among LLM-trained physicians: a randomized clinical trial. 

Qazi IA, Ali A, Khawaja AU, et al. medRxiv. Epub 2025 Jul 21.  

LLMs could improve clinical decision making, but they sometimes provide inaccurate answers. This non-peer reviewed preprint shares the results of an exploration into the value of AI training for physicians in the use of LLMs. Across multiple medical institutions in Pakistan, 58 physicians trained to use LLMs had dramatically greater diagnostic reasoning scores over clinicians using conventional resources such as PubMed.  

Clinical decision support for septic shock in the emergency department: a cluster randomized trial.(subscription required) 

Scott HF, Sevick CJ, Colborn KL, et al. Pediatrics. 2025;156(1):e2024069478.  

Early recognition and treatment of sepsis is crucial for preventing septic shock. This study examines how EHR-embedded clinical decision support can improve the identification of pediatric patients suspected of having sepsis that are at risk to develop septic shock. While the system design was rated favorably by clinicians, its use did not reduce patients with risk factors for sepsis to experience septic shock, likely because both control and usual care arms had almost identical rates of goal-directed antibiotic and fluid therapy.