CODEX Digest 04.30.26

Physicians' adoption of AI is vital for its safe, effective implementation. This nationwide survey reveals an increase from 65 to over 75% of physicians having a positive view of AI for healthcare, particularly for diagnosis. The survey also reviews physician comfort with patient AI use.

Evidence-based screening tools for delirium in older adults are available but often underused. This ethnographic study found incomplete screenings and identified barriers such as staff prioritization of other duties, limited awareness of the tool's value, and language barriers. Facilitators included input from care partners on baseline mental status and nurses introducing the screening to boost patient participation.

For patients, delayed diagnosis carries deep emotional and psychological costs, eroding trust and future care engagement. This interview study of patients who experieced a delayed diagnosis reveals three key themes: feeling dismissed, frustration and self-doubt; relief and validation upon finally receiving a diagnosis; and lasting mistrust in healthcare. These harms span ages and conditions, underscoring the need for patient-centered, validating communication and partnership to improve diagnostic safety.

Diagnosing stroke in emergency department patients presenting mainly with dizziness is difficult. This retrospective study estimates stroke-related disability or mortality after such visits, finding these outcomes are quite rare when patients are discharged home. The authors address challenges in accurate stroke diagnosis during initial assessments and suggest decision support tools to help identify strokes in dizziness cases.

Patient reporting is crucial for flagging problems to inform diagnostic accuracy improvement efforts. This report describes the development and evaluation of PROMOTE-Dx, a survey-based patient-reported measure for assessing timely cancer diagnosis. Healthcare organizations can use PROMOTE-Dx to evaluate and enhance the timeliness of cancer diagnoses.

People of color face higher rates of delayed or incorrect dermatologic diagnoses, worsening health inequalities. This UK mixed-methods study of podiatry students found low confidence in diagnosing dermatological conditions, especially on dark skin tones. It highlights educational gaps in podiatry education, addressing a critical research need at the intersection of training and health equity.

Delayed venous thromboembolism (VTE) diagnosis is an impactful harm in hospitals, but identifying cases is difficult and feedback is limited. This retrospective real-world study shows that a LLM-based electronic trigger outperforms traditional electronic triggers in detecting VTE delays; although positive predictive value was only 48% and observed incidence of VTE Dx delay was quite low.

Illness scripts are one element of teaching clinical reasoning, and AI chatbots may improve illness script quality more effectively than conventional teaching. This Thai study compared chatbot-enabled feedback and script co-construction with traditional instruction, finding that AI-enhanced personalized feedback within a structured framework boosts medical students’ diagnostic reasoning skill development.

Multi-cancer detection tests (MCDs) may help reduce cancer deaths by enabling early diagnosis. This focus group study finds that participants valued early detection and screening convenience with a single blood test but were concerned about limited clinical validation and risks from false positives, overdiagnosis, and overtreatment. Of importance, the implementation of MCD testing at the NHS identifed safety and feasibility issues, showing the importance of more large-scale trials and clinical validation for any future implementation.

Low dose CT (LDCT) is an important screening method, but many lung cancer patients face diagnostic uncertainty after a positive initial screen. This UK study investigates potential missed diagnostic opportunities (PMDO), defined as no lung cancer diagnosis following a positive LDCT, later followed by another positive annual LDCT that confirms the disease. Over 37% of patients diagnosed with lung cancer and considered at risk for PMDO experienced at least one PMDO, highlighting an important area of focus for timely lung CA diagnosis.

AI-driven diagnostic support systems can boost accuracy and efficiency, but opaque, usage-based pricing hinders adoption. This European interview study explores ways to make pricing clearer and more predictable, considering technical and organizational factors. Ten themes reveal resistance to usage-based pricing and a desire for transparency. The authors offer design principles and recommendations for pricing models that support procurement, reimbursement, and scaling.

Evaluating the suitability of diagnostic tests and devices for children is fundamental for their safe use. This review explores whether radiology AI devices including software approved in Europe specify pediatric use, revealing many do not. The authors recommend a standardized safety indicator to help clinicians choose appropriate devices.