CODEX Digest - 12.18.25

1)  Unravelling the rise in thyroid cancer incidence and addressing overdiagnosis.(subscription required)

Chen DW, Haymart MR. Nat Rev Endocrinol. Epub 2025 Sep 3. 

Increases in the global thyroid cancer rate have been ascribed to overdiagnosis of nonaggressive disease and some increase in advanced-stage thyroid cancer. This review discusses trends, clinical and environmental risk factors, and strategies to reduce overdiagnosis and overtreatment at the community level. 

2) To what extent should doctors communicate diagnostic uncertainty with their patients? An empirical ethics vignette study.

Cox C, Hatfield T, Parry M, et al. J Med Ethics. 2025;51(11):754-765.  

Diagnostic uncertainty is common, yet research on how it is shared with patients is limited. This study investigates patient preferences and the impact of clinician candor regarding diagnostic uncertainty. Greater communication of uncertainty is recommended; withholding or inconsistently sharing this information can deprive patients and families of knowledge, potentially worsen inequalities, and challenge ethical practice. 

3) What Is implementation science: and why it matters for bridging the artificial intelligence innovation-to-application gap in medical imaging. (subscription required) 

Fayaz-Bakhsh A, Tania J, Lutfi SL, et al. PET Clin. 2026;21(1):1-16. 

While AI technology is advancing quickly, adoption faces organizational and practical barriers. This commentary highlights how implementation science can help integrate AI into imaging workflows. Recommendations include early and collaborative interdisciplinary stakeholder engagement to infuse AI into imaging practice.  

4) From Hindsight Bias to Machine Bias: Dr. Laura Zwaan on Learning from Mistakes.

NEJM AI Grand Rounds. November 18, 2025.  

Cognitive science helps explain decision making, biases, errors, and their influence on diagnostic excellence. This New England Journal podcast covers cognitive concepts relevant to diagnostic excellence, discusses how AI adopts human biases, and stresses the need for transparency. It also introduces "machine psychology," and argues that understanding human reasoning is vital for the development of reliable AI. 

5) Strategies to reinvigorate the bedside clinical encounter. (subscription required) 

Garibaldi BT, Russell SW. N Engl J Med. 2025;393(21):2142-2150. 

Inadequate history taking and physical exam mistakes are primary detractors from diagnostic excellence, and lack of time at the bedside contributes to this problem. This commentary shares evidence-based strategies, practical guidance, and case study content to help medical educators strengthen learner bedside clinical skills to support effective diagnosis, treatment, and patient communication.   

6) To recommend or not to recommend: designing and evaluating AI-enabled decision support for time-critical medical events. 

Mastrianni A, Kim MS, Sullivan TM, et al. Proc ACM Hum-Comput Interact. 2025;9(7):10.1145/3757512.   

AI decision-support aids clinicians in emergency decision making with limited data. In an online experiment, clinicians who received AI recommendations made more accurate diagnoses than those without, but information synthesis alone did not improve results. Key barriers included balancing accuracy with speed and disagreements over recommendations. 

7) Advancing Diagnostic Excellence in Rural Areas: A Workshop.

National Academies of Sciences, Engineering, and Medicine. October 25, 2025.   

Rural communities harbor barriers to timely diagnosis that contribute to inequitable care. This set of workshop videos reviews current challenges and potential improvements for rural diagnostics. Key topics include delivering patient-centered care, leveraging technology, and adapting strategies for diagnostic excellence to other underserved regions.   

8)  A framework for defining diagnostically challenging conditions identifiable through electronic algorithms.(subscription required)

Olson APJ, Sloane J, Zimolzak A, et al. Diagnosis (Berl). Epub 2025 Oct 27.  

Triggers can help target priorities for diagnostic improvement work. This study compares using specific high-risk diagnoses (diagnosis-based triggers) against traditional quality assurance methods. For appendicitis and neurologic hemorrhage, lookback review found more diagnostic errors from prior ED visits which quality assurance processes missed.  

9) A dangerous retreat: defunding AHRQ and the threat to diagnostic safety. 

Schulson LB, Fischer MA, Streed CG. J Gen Intern Med. Epub 2025 Nov 17.  

Public, federally-funded entities play a vital role in improving healthcare by providing funding, sharing free tools, and raising awareness. This commentary discusses the impact the loss of AHRQ could have across the field of diagnostic excellence. The piece calls for medical, research, and healthcare organizations to reach out to policymakers to share the importance of AHRQ’s continued investment in diagnostic safety work.  

10) Differences in contributing factors to diagnostic errors between physicians and allied health professionals: a nationwide analysis in Japan. 

Shimizu I, Shikino K, Harada Y, et al. BMJ Open Qual. 2025;14(4):e003419.  

Most diagnostic error research centers on physicians, but allied health professionals (AHPs) such as nurses, also affect patient safety. This study reviews national adverse-event reports and compares diagnostic errors by physicians and AHPs. By analyzing expert-reviewed data, it highlights AHPs' role in diagnostic errors and underscores the need for targeted strategies to improve diagnosis. 

11) Artificial intelligence chain-of-thought reasoning in nuanced medical scenarios: mitigation of cognitive biases through model intransigence. (subscription required) 

Wang J, Redelmeier DA. BMJ Qual Saf. Epub 2025 Nov 24. 

Large language models (LLMs) are used in clinical decision making but can reinforce cognitive biases seen in physicians. This study evaluated whether chain-of-thought LLMs reduced these biases.  However, certain biases persisted regardless of the LLM and deviated substantially from the average response of practicing physicians. Clinicians should remain vigilant when interpreting LLM outputs for complex medical decisions. 

12) Age-based screening for lung cancer surveillance in the US.

Yang HC, Chang A, Visa M, et al. JAMA Netw Open. 2025;8(11):e2546222.   

Lung cancer is increasingly found in people who have never smoked, highlighting limitations in the US Preventive Services Task Force’s (USPSTF) current screening criteria. This study finds that USPSTF guidelines exclude about two-thirds of patients—mainly women and never-smokers. The age-based screening model tested here improves detection, is six times more cost-effective, and improves fairness compared to the existing program.