As AI systems approach and in some domains exceed human diagnostic performance, a critical gap in medical education is coming into focus.
Diagnosis, the cognitive task that has anchored clinical training for generations, is increasingly something AI can do, and in some cases do better than clinicians working alone. What remains distinctly human is deciding how to test and treat a particular patient in their specific context: translating evidence into action for an individual, with their values, their circumstances, and what matters most to them.
That cognitive work has a name: management reasoning. The process between a working diagnosis and enacting a management plan. It has received far less attention in medical education than diagnostic reasoning. That needs to change.
What is management reasoning in clinical practice?
Management reasoning is defined as "the process of making decisions about treatment, further testing, monitoring, follow-up visits, and allocation of limited resources" (Cook et al., JAMA 2018). It is the other domain of clinical reasoning, the part that begins where diagnosis ends (or sometimes happens in parallel).
The distinction matters more than it might initially appear:
- Diagnosis is a task of categorization: matching a patient's presentation to a known pattern and applying a label
- Management is a task of prioritization: selecting among multiple defensible options based on who the patient is, what they value, and what is feasible in their specific context
Same diagnosis, different management.
Why AI's diagnostic capability makes management reasoning more important
Recent evidence suggests AI has closed, and in some experiments surpassed, the diagnostic performance of human clinicians:
- Goh et al. (JAMA Network Open, 2024): an LLM alone scored 16 points higher than physicians using conventional resources. Giving physicians access to the LLM improved efficiency but did not improve diagnostic accuracy.
- Brodeur et al. (arXiv, 2025): superhuman LLM performance across six experiments involving hundreds of physicians. Median score of 97% on complex diagnostic cases versus 74% for clinicians alone.
This performance is not entirely surprising. Both expert cognition and LLM architecture rely on pattern matching across large volumes of prior cases: illness scripts for humans, statistical associations across training corpora for LLMs. The mechanisms differ; but the cognitive task is structurally similar.
Management reasoning is different. It requires:
- Direct communication with the patient and interprofessional team
- Ongoing, fluid decision-making rather than a fixed-point conclusion
- Navigation of competing factors: patient goals, comorbidities, cost, risk tolerance, institutional norms, and clinician uncertainty
What do learners understand about management reasoning, and where do they struggle?
A qualitative study using focus groups with 28 senior residents across four specialties at two academic medical centers identified three consistent themes (Parsons et al., Med Educ 2025; Parsons et al., Acad Med 2025):
- Linked but distinct. Trainees recognize that management is cognitively different from diagnosis: "something the patient can actually see or experience or feel." Arriving at the correct diagnosis does not automatically determine what to do next.
- Uncertain and risky. Unlike diagnosis, management decisions involve trade-offs with consequences that play out over time. Trainees found this intellectually and emotionally taxing: it "breaks down the idea that there's one right answer to the multiple-choice question."
- Context-specific. What is clinically appropriate depends on available services, time of day, institutional practices, and the specific patient. There is no transferable algorithm.
What trainees consistently lack is a structured framework for navigating this complexity: a shared language for articulating why one option is chosen over another.
How programs can structure teaching and assessment of management reasoning
The management script
The management script offers a practical teaching framework. Analogous to illness scripts for diagnosis, it is organized for action (Parsons, Wijesekera & Rencic, Acad Med 2020):
- Map all available options for a given diagnosis across categories: medications, procedures, monitoring, imaging, laboratory, and specialists
- Apply a structured set of influencing factors to select what is appropriate for this patient, spanning intervention characteristics, patient goals and values, case-specific variables, provider factors, and clinical setting constraints
Test and treatment thresholds
The threshold model (Pauker & Kassirer, NEJM 1980; Wijesekera, Parsons et al., Acad Med 2022) provides a framework for teaching calibration:
- Below the test threshold: the risk of testing exceeds its benefit — do not test
- Above the treatment threshold: the benefit of treatment clearly outweighs risk — treat
- Between them lies the zone of uncertainty, where most clinical care actually occurs and where influencing factors determine the plan
Teaching learners to name uncertainty explicitly and make decisions despite it, rather than defaulting to over-testing or deferring to a senior, is a core skill that receives insufficient formal attention.
A richer model of risk
Risk in management decisions operates across three simultaneous lenses (Renn, 1992):
- Technical: probability × magnitude of harm
- Psychological: risk as perceived, shaped by heuristics and individual tolerance
- Social: risk as socially constructed, shaped by relationships, team culture, and power dynamics
Shared decision making, reframed as risk calibration, is the process of reconciling those competing perspectives across patient, provider, and system (Elwyn et al. 2012/2017; Montori et al. 2023).
Frequently asked questions
What is the difference between diagnostic reasoning and management reasoning? Diagnostic reasoning is a task of categorization: matching a patient's presentation to a known pattern. Management reasoning is a task of prioritization, selecting among multiple defensible options based on the specific patient's context, values, and circumstances. Both are forms of clinical reasoning, but they involve different cognitive processes and require different educational approaches.
Why is management reasoning difficult to teach? Management reasoning is context-dependent, ongoing, and involves competing factors that cannot be resolved through pattern matching or guideline application alone. Research with senior residents shows trainees recognize its complexity but often lack a structured framework for navigating uncertainty, articulating their rationale, or adapting plans as clinical situations evolve.
How does AI affect the teaching of clinical reasoning? As AI systems demonstrate diagnostic performance at or above physician level, the educational case for investing in management reasoning becomes stronger. AI is well-suited for diagnosis for structural reasons; management requires contextual judgment, patient communication, and risk calibration that remains distinctly human.
When should management reasoning be introduced in training? Evidence from clerkship research suggests students encounter management-heavy situations from their first clinical rotations. Introducing the management script framework early, as a scaffold that trainees populate as their knowledge grows, allows the process to be practiced repeatedly before the content fully matures.
This post is part of DDx by Sketchy's ongoing webinar series on clinical reasoning and AI in medical education.
This session is the third in the series. If you missed the earlier sessions, both recordings are available on the DDx blog:
- Part 1: A cognitive framework for understanding AI's role in clinical reasoning education
- Part 2: Integrating generative AI into clinical reasoning education: What clinical educators need to know
The full recording of this session, including the Q&A with Dr. Parsons and a live demo of how DDx supports management reasoning teaching, is available here.
References
Brodeur PG et al. Superhuman performance of LLMs on complex clinical reasoning tasks. arXiv. 2025.
Cook DA, Sherbino J, Durning SJ. Management reasoning: beyond the diagnosis. JAMA. 2018.
Elwyn G, Durand M A, Song J, Aarts J, Barr P J, Berger Z et al. A three-talk model for shared decision making: multistage consultation process BMJ 2017; 359 :j4891 doi:10.1136/bmj.j4891
Elwyn G, Frosch D, Thomson R, Joseph-Williams N, Lloyd A, Kinnersley P, Cording E, Tomson D, Dodd C, Rollnick S, Edwards A, Barry M. Shared decision making: a model for clinical practice. J Gen Intern Med. 2012 Oct;27(10):1361-7. doi: 10.1007/s11606-012-2077-6. Epub 2012 May 23. PMID: 22618581; PMCID: PMC3445676.
Goh E, Gallo R, Hom J, et al. Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial. JAMA Netw Open. 2024;7(10):e2440969. doi:10.1001/jamanetworkopen.2024.40969
Montori VM, Ruissen MM, Hargraves IG, Brito JP, Kunneman M. Shared decision-making as a method of care. BMJ Evid Based Med. 2023 Aug;28(4):213-217. doi: 10.1136/bmjebm-2022-112068. Epub 2022 Dec 2. PMID: 36460328; PMCID: PMC10423463.
Parsons AS, Morris C, Bryan K, Durning SJ, van Mook WNKA, Ryan MS, Abdoler EA. How postgraduate medical trainees conceptualise management reasoning: A qualitative study. Med Educ. 2025 Dec 12. doi: 10.1111/medu.70123. Epub ahead of print. PMID: 41386648.
Parsons AS, Bryan K, Morris C, Durning SJ, van Mook WNKA, Ryan MS, Abdoler E. Learning to manage: a qualitative exploration of how graduate medical trainees develop management reasoning. Acad Med. 2026 Apr 1;101(4):411-421. doi: 10.1093/acamed/wvaf068. PMID: 41769763.
Parsons AS, Wijesekera TP, Rencic JJ. The Management Script: A Practical Tool for Teaching Management Reasoning. Acad Med. 2020 Aug;95(8):1179-1185. doi: 10.1097/ACM.0000000000003465. PMID: 32349018.
Pauker SG, Kassirer JP. The threshold approach to clinical decision making. N Engl J Med. 1980 May 15;302(20):1109-17. doi: 10.1056/NEJM198005153022003. PMID: 7366635.
Renn, Ortwin. Concepts of Risk: A Classification. University of Stuttgart, 1992.
Wijesekera TP, Parsons AS, Abdoler EA, Trowbridge RL, Durning SJ, Rencic JJ. Management Reasoning: A Toolbox for Educators. Acad Med. 2022 Nov 1;97(11):1724. doi: 10.1097/ACM.0000000000004796. Epub 2022 Jun 21. PMID: 35731592.
