CODEX Digest 09.03.26

Diagnostic safety research often focuses on errors rather than how clinicians and systems successfully manage uncertainty. This ethnographic study examines pediatric outpatient diagnostic encounters involving children with medical complexity. Researchers identified resilient behaviors from patients/families, clinicians, and systems to navigate diagnostic complexity. The findings broaden diagnostic excellence toward understanding how teams anticipate, adapt to, and recover from diagnostic challenges.

Diagnostic excellence includes obtaining the right tests for the right patients while avoiding unnecessary testing. This multisite, observational audit examined 1,068 blood cultures across nine pediatric intensive care units. Only 6% were positive, and more than half were obtained despite a documented alternative source of symptoms. Findings support standardized thresholds and diagnostic stewardship to reduce potentially low-value blood cultures.

Diagnostic excellence requires balancing early detection with overuse, cost, and potential harms. This serial cross-sectional SEER-Medicare study examined breast cancer screening among women aged 67 years and older from 2009–2019. Screening patterns, costs, and estimated value varied by age, modality, and frequency, and ineffective screening costs accounted for over 50% of spending and over $1 billion. The results highlight the importance of aligning screening intensity with expected benefit.

Organized screening can improve timely colorectal cancer diagnosis, but real-world population benefit depends on sustained participation and follow-up. This observational, population-based cohort, Spanish study evaluated mortality after implementation of a fecal immunochemical test (FIT)-based screening program. Screening was associated with substantially lower colorectal cancer mortality, supporting organized screening as a population-level diagnostic pathway while underscoring the importance of program implementation and follow-up.

Reliable identification of clinical events from electronic medical records (EMR) is essential for measuring diagnostic quality and outcomes. This multisite, retrospective, validation study compares EMR retrieval approaches, including a Gemini V.1.5 powered LLM, for identifying cardiovascular events. The study demonstrates potential efficiency gains in extracting diagnostic information, while emphasizing that improved information retrieval does not itself establish improved clinical diagnosis or patient outcomes.

Diagnostic excellence requires equitable diagnosis, yet unfairness in diagnostic processes remains comparatively underexamined. This ethical analysis and narrative review conceptualizes diagnostic injustice across five dimensions. Factors contributing to diagnostic injustice include uncertainty, prejudice, and biomedical dominance. The framework positions equity as a fundamental dimension of diagnostic excellence rather than a separate quality objective.

AI assistance may improve diagnostic performance while simultaneously introducing automation bias. This experimental preprint study evaluates how physicians use large language models (LLM) for clinical diagnosis. LLM assistance increased diagnostic accuracy, but the 46 physicians testing a proprietary agentic retrieval-augmented LLM were more likely to accept incorrect recommendations when supported by apparently credible citations. Findings highlight a critical diagnostic-safety tradeoff: improved performance can coexist with false reliance, emphasizing calibration, oversight, and human factors.

Timely diagnosis of cancer in primary care may be suboptimal as patients often have non-specific clinical concerns. This secondary qualitative analysis examined UK patient perspectives on diagnostic delay following an initial non-cancer diagnosis, aka an "interim" diagnosis, before subsequent cancer diagnosis. Participants described consequences for trust, uncertainty, and lack of consideration for their concerns. The findings reinforce the importance of safety-netting, listening to patients, and examining diagnostic processes through the eyes of patients.

Successful implementation of AI in diagnosis requires patient trust as well as technical performance. This German focus-group study engaged 34 patients across seven groups to explore expectations and concerns about AI-assisted radiological diagnosis. The results indicate patients want evidence of the safety and added value AI would bring and reflect that a sociotechnical orientation to AI is necessary for full acceptance of its use. Concerns discussed include excessive clinician reliance on AI and the potential erosion of the physician-patient relationship.

AI-generated patient communications may improve efficiency but could introduce new communication and safety problems. This South Korean multi-reader study evaluates radiologists’ ability to identify deficiencies in LLM-generated communications accompanying radiology reports. The findings underscore the continuing need for expert review of AI-generated patient-facing information, which should be considered related to balancing AI rollout versus workforce needed for human review.

Diagnostic reasoning education must develop more than factual knowledge; clinicians must also learn effective diagnostic processes. This Australian qualitative educational study examines how 10 medical students engaged with a deliberate reflection "grid" during diagnostic reasoning instruction. The findings suggest reflection may influence process-oriented reasoning rather than simply increase knowledge. The study supports educational approaches that explicitly cultivate metacognition, diagnostic flexibility, and awareness of reasoning processes.

As AI becomes embedded in diagnostic workflows, health systems need measures that determine whether it actually improves diagnosis. This commentary describes the results of a multidisciplinary consensus-driven Action Incubator designed to develop actionable diagnostic-excellence measures for AI. Proposed domains include accuracy, timeliness, equity, uncertainty, and patient partnership—shifting evaluation beyond AI performance toward real-world diagnostic outcomes.