CODEX Digest 07.30.26

Safety II shifts focus on learning from failure to success. This multi-institutional qualitative study of 36 individuals (34 healthcare/academic personell and 2 patient advocates) named structured yet adaptive reasoning, collaborative teamwork, and systems capacity as key ingredients in diagnostic excellence in hospital medicine. However, paticipants were challenged to systematically identify "Safety II' cases.

Inappropriate resource use, such as emergency room for primary care needs, can contribute to fragmentation and diagnostic delays. This case study describes a single-center Epic-integrated AI triage tool implementation in New York which combines AI with an "expert authored rules engine." The effort completed 22,000 sessions with 88%–96% concordance with blinded physician review and reportedly zero safety adverse events.—demonstrating safe, usable, clinically reliable navigation. The authors provide implementation recommendations based on their experience.

Diagnostic excellence requires proactive policy and regulatory frameworks that maintain public trust while leveraging AI expansion into diagnostic pathways. This qualitative systematic review outlines ethical considerations for integrating AI into medical screening. To ensure patient safety, the literature emphasizes the necessity of human oversight, transparent decision-making, and robust liability frameworks. Primary findings highlight the need to implement clear ethical frameworks for AI implementation within health policy.

Low-value lab testing can result in patient harm, resource waste, and environmental cost. This interview study on Canada’s "Using Labs Wisely" collaborative shows that national benchmarking, peer learning, and structured QI effectively reduce overuse. The initiative fostered institutional engagement and lab-led stewardship, yet data extraction and resource constraints persisted. Addressing these infrastructure gaps is essential for scaling sustainable, multidisciplinary diagnostic excellence efforts nationwide.

Cognitive bias and conclusions drawn from a patient’s appearance can affect medical students’ diagnostic decision making related to pain detection and treatment decisions. This Italian study finds that a brief evidence-based educational intervention to raise student awareness of implicit racial bias improved perception of pain-related cues, showing potential to support more equitable pain recognition and management. However, future studies should investigate whether the effects of the training have long-term impact.

Direct test result release, prior to discussion with a clinician, risks patient confusion and distress without structured support. This national study reveals a communication gap: only 66% of patients understood the results accessed via a patient portal, and just 28% could choose delivery methods to manage that communication which was associated with digital literacy. Result release processes can do better to incorporate patient preferences and account for digital health inequities.

Over-reliance on routine diagnostic scripts can lead to reasoning limits and premature closure in uncommon, evolving, or socially problematic cases. This commentary describes the "Pattern Disruption Intervention" a multimodal training approach which inserts novel elements to routine cases to expand flexibility in student diagnostic reasoning, improve tolerance of uncertainty, clarify diagnostic risk, and support effective follow-up planning. Of note, the report shows some feasibility without clinical outcomes.

Real-world emergency triage can be challenging due to time pressures and incomplete information. In this German preprint study of more than 16,000 emergency department visits, LLM triage performance as a diagnostic and care prioritization decision did not acheive significant agreement with human assessment. Most of the 16 models tested were overconfident, inconsistent, and weak at routing patients, underscoring the deficits of current AI models for appropriate clinical triage.

Non-medical community contacts—librarians, hairdressers, home aides—are uniquely positioned to identify early cognitive concerns in their older customers. This Dutch survey study finds these individuals often feel responsible to act when suspecting dementia, with about half of participants reporting they acted on their concern. Enhancing diagnostic timeliness requires equipping these observers with clear communication tools and referral guidance, though broader sampling is needed to validate findings across diverse settings.

Gen AI may be able to support mentorship in clinical reasoning for medical learners when its use is guided by established educational theory. This commentary presents a conceptual framework for integrating gen AI as a real-time feedback and instruction tool that supports psychologically safe clinical reasoning skill development. The theories anchoring the discussion were selected as they address a distinct challenge in providing remediation to stressed medical learners

Autopsy remains an underutilized tool to support learning from gaps in diagnostic decisions. This 10-year Spanish retrospective observational review of 284 autopsies found major discrepancies in 30% of cases—predominantly missed malignancies, infections, and cardiovascular diagnoses. While mortality review committees could harness these insights for improvement, the authors highlight that persistently low autopsy volumes and absent feedback loops constrain their influence on reducing diagnostic error.

Overuse of imaging degrades diagnostic excellence as it results in unnecessary costs, radiation exposure, and patient stress. This randomized trial examines whether the use of a diagnostic rule set could safely reduce CT imaging for patients with active cancer and suspected pulmonary embolism. The results find that the algorithm was as safe in detecting clinically significant clots as routine CT scanning while avoiding unneeded exams. These findings suggest evidence-based diagnostic pathways can reduce unnecessary imaging without compromising patient safety in this high-risk population.