CODEX Digest 11.12.26

Older adults’ complex presentations make diagnostic decision support particularly challenging. This study developed and evaluated an AI-based system using collective intelligence and imitation learning to support primary care decisions. The approach trained on 707,598 encounters aimed to represent clinical practice and uncertainty without relying exclusively on narrow gold-standard labels. Its potential value requires further assessment of external validity, clinician interaction, missed diagnoses, and patient outcomes before routine deployment can be justified. For more from senior author Gary Weissman, watch his CODEX webinar "AI Diagnostic Decision Support for Older Adults in Primary Care."

Unnecessary diagnostic testing can expose patients to avoidable procedures, risks, and costs. This analysis synthesized 14 studies involving low-yield vascular diagnostics among 16.7 million patients in high-income countries. The results documented overuse in vascular diagnostic imaging and interventions, and all studies assessing costs linked overuse to increased healthcare spending. However, authors call for additional research before designing interventions to reduce inappropriate vascular testing.

Timely cancer diagnosis depends on access, recognition, and efficient follow-up. This cohort study examined 1,071 women diagnosed with breast cancer at age 40 or younger and identified presentation, diagnostic, and treatment delays. Lower income was associated with presentation delays, while younger age was associated with diagnostic delays. Delays were not associated with disease-free or overall survival in this cohort. The authors suggest that further work should investigate disparities and opportunities for earlier diagnosis.

Patients increasingly use AI chatbots to explore symptoms and health concerns, creating a need for practical safety guidance. This guide used evidence review, patient and expert input, and consensus to develop five steps supporting the safe use of AI for information generation: slow down, make questions clear, consider privacy, check facts, and talk to a person when needed.

Diagnostic AI performance may vary substantially between controlled case studies and real clinical encounters. This preprint evaluated LLM diagnostic multi-agent (team) performance against single-model responses using curated cases and emergency department notes. LLM teams improved differential diagnosis accuracy on curated benchmarks but performed worse on real emergency presentations. The findings underscore the need for clinical validation before deployment.

Diagnostic excellence depends on recognizing the contributions of all team members involved in patient care. This review draws on historical, legal, theoretical, and scientific evidence to examine registered nurses’ diagnostic roles. The authors argue that regulatory barriers, limited visibility in health information systems, and gaps in clinical-reasoning preparation constrained nursing contributions. They call for systematic inclusion of nurses in team-based diagnosis.For more from author Kelly Gleason, PhD, RN, FAAN watch her Editor's Pick Q&A. 

Respectful clinical documentation is essential to diagnostic excellence because records can shape subsequent clinicians’ judgments and patients’ willingness to seek care. In this cross-sectional survey of 500 adults, 17.1% of patient-portal users reported disrespectful or judgmental language in medical records. Approximately three-quarters anticipated adverse consequences for future care including diagnostic errors. The findings support person-centered documentation standards and clinician training.

Successful diagnostic AI integration depends on more than technical performance; it also affects trust, communication, and professional responsibility. This mixed-methods Belgian study surveyed 51 patients with localized prostate cancer and convened focus groups. Participants expressed mixed views about AI, citing concerns about reliability, understanding, and accountability. Authors propose that AI is changing the way trust is developed within the diagnostic encounter.

Sepsis diagnosis depends on interacting clinical, organizational, and system-level factors rather than a single decision. In this qualitative study involving 20 hospitals across four states and two health systems, researchers developed and empirically validated a multilevel Sepsis Diagnostic Excellence (SepsisDx) framework through semi- structured multidisciplinary interviews. The framework offers a system-focused basis for targeted improvement and future implementation research, although its application in other types of healthcare systems requires evaluation.

Diagnostic errors can harm patients and leave clinicians struggling without adequate organizational support. Drawing on empirical findings and relevant literature, this UK-led intervention-development study proposed three approaches for obstetric sonographers: a supportive learning environment, a manager support tool, and phased early intervention. The models emphasized compassionate responses, peer support, and learning rather than blame. Future implementation studies should assess clinician well-being and measurable patient-safety outcomes.

Strengthening clinical reasoning education is an important strategy for preventing diagnostic errors. This medical-school program evaluation describes a longitudinal, peer-assisted learning model in which senior medical students taught and mentored peers. Across six recruitment cycles, an emphasis on small-group and individualized teaching explored elements of diagnostic decision-making skills for both teachers and learners. This education model offers a scalable approach to expanding diagnostic reasoning practice and feedback.

Diagnostic excellence depends on clinicians’ ability to develop, test, and revise diagnostic hypotheses—not simply produce polished documentation. This commentary discussed “inference impersonation,” whereby ambient AI scribes imitate reasoning without a way to verify if clinicians independently reasoned through a case. The authors warn that the practice could weaken diagnostic judgment development, and conceal reasoning gaps, recommending independent trainee assessments and competency-based safeguards.