CODEX Digest 05.07.26

Patient safety report analysis can provide important avenues for error identification and improvement opportunities. This study explores using natural language processing (NLP) and machine learning (ML) to identify cardiovascular diagnostic errors from patient safety event reports. Findings demonstrate feasibility, bridging data science methods with patient safety needs. This approach can enhance surveillance of diagnostic errors, supporting quality improvement efforts in cardiovascular care.

In rural and wilderness settings, paramedics often manage patients during long transports with limited resources. This prospective study found that training paramedics to use point-of-care ultrasound (POCUS) was feasible, well accepted, and helped support patient care and transport decisions. Wider implementation and further research could clarify its effects on clinician performance, diagnostic accuracy, and patient outcomes in remote environments.

Diagnostic overshadowing occurs when a pre-existing condition influences the assessment of other diagnoses. This study examines how substance use disorder (SUD) affects the time to surgical intervention in patients with traumatic brain injuries, revealing that SUD leads to delays in care. The findings demonstrate that diagnostic overshadowing impacts surgical timeliness and is measurable.

Diagnostic errors are common in patients with abdominal pain. This study found that errors often stem from misinterpreting physical exams or imaging findings and difficulty distinguishing between conditions within the same organ system. Improving bedside assessment skills, using structured differential diagnosis approaches, and strengthening radiology support may help reduce errors and improve patient safety.

Canadian medical educators recognize the need for deliberate clinical reasoning (CR) instruction at the pre-clinical level, though integration into real-time education remains unclear. This focus group study finds that medical students gained diverse CR knowledge during their courses and applied it to diagnostics, history-taking, and clinical problem-solving. Broader stakeholder experiences captured largely aligned with curriculum goals but highlighted some difficulties. Results show CR can be integrated in undergraduate curricula, though challenges exist.

Large language models (LLMs) are increasingly promoted for clinical use, but their ability to support real-world clinical reasoning remains uncertain. This cross-sectional study found that while LLMs were often accurate in identifying final diagnoses, they performed poorly in generating differential diagnoses and managing uncertainty. The authors propose a multidimensional benchmark to better assess the safe clinical deployment of AI tools.

AI and machine learning are increasingly used in diagnostic imaging, impacting malpractice risk, legal standards, liability, and informed consent. While AI can boost accuracy and efficiency, its adoption raises potential negligence claims if clinicians misuse or fail to use validated tools. Liability is unclear due to shared responsibility among clinicians, organizations, and developers, and traditional frameworks lack clear accountability for AI-related errors. Clear guidelines on responsibility, decision-making, and validation are vital for patient safety and clinician protection.

Patients with undiagnosed cancer may first receive an alternative “interim” diagnosis, potentially delaying cancer detection. This study examines healthcare professionals’ views on when interim diagnoses occur, their impact on cancer diagnosis, and triggers for prompt review. Key factors included diagnostic uncertainty, non-specific symptoms, and remote consultations limiting information. While delays were common, safety-netting and team-based “fresh eyes” discussions enabled timely review.

Generative AI (GenAI) tools are increasingly used by patients and families seeking diagnoses of rare diseases. This preprint study explores their active use of GenAI for information, preparation, and advocacy, despite their concerns about reliability and safety. The results reveal the concealed behaviors associated with patient-driven GenAI use outside the clinical environment. Improved patient-centered design, education, and governance are necessary to ensure GenAI helps reduce—not worsen—inequities in rare disease care.

Perinatal depression (PND) is a common but serious complication. This single-system study examines racial, ethnic, and sociodemographic disparities in PND screening and diagnosis. Results show that historically marginalized groups are more likely to be screened but less likely to be diagnosed. These findings reveal critical inequities, underscoring the need for targeted interventions to ensure equitable care.

Radiation safety culture is increasingly recognized as essential to improving diagnostic imaging safety. This review highlights recent advances in implementing radiological protection practices, emphasizing safety reporting, learning from errors, regular performance review, and continuous improvement in imaging services to reduce radiation risks and prevent safety incidents.

As AI becomes more integrated into clinical care, clear communication with patients about AI findings will be critical. This Italian randomized survey study found that disclosing AI results that conflicted with a radiologist’s interpretation reduced patient's trust, increased their anxiety, and raised their interest in second opinions and legal action. Providing brief context for discordant AI findings may help maintain trust and reduce unnecessary concern.