Clinical reasoning is the thinking process clinicians use to gather patient information, form a working diagnosis, and decide on a course of action. There is no single universally accepted definition, but the core idea is consistent: a clinician takes what they know about medicine, combines it with what they observe about a patient, builds a mental picture of the problem, and refines that picture as new information comes in. This cycle of gathering, interpreting, and revising repeats until the clinician is confident enough to act.
It sounds straightforward, but clinical reasoning is one of the most complex cognitive tasks in professional life. When it works well, patients get accurate diagnoses and effective treatment. When it breaks down, the consequences can be serious: an estimated 10% to 15% of diagnostic errors lead to adverse patient outcomes, and roughly 75% of those errors trace back to cognitive mistakes in the reasoning process itself.
How the Reasoning Cycle Works
The most widely referenced model of clinical reasoning is called the hypothetico-deductive method, and it follows four steps. First, the clinician collects initial cues: symptoms the patient describes, vital signs, visible findings, medical history. Second, those cues trigger one or more preliminary hypotheses about what might be going on. Third, the clinician interprets additional information in light of those hypotheses, refining or narrowing them. Fourth, the hypotheses are evaluated against all available evidence and either confirmed or ruled out.
In practice, this isn’t a neat linear process. A clinician might loop back to gather more information after a lab result surprises them, or shift hypotheses entirely when a patient mentions a detail that changes the picture. The cycle can repeat multiple times in a single encounter. What matters is that each pass through the loop sharpens the mental representation of the patient’s problem until it’s clear enough to support a decision.
Two Thinking Systems at Work
Clinical reasoning draws on two distinct modes of thinking, often called System 1 and System 2. System 1 is fast, intuitive, and largely unconscious. It’s the “gut feeling” that fires when a pattern looks familiar, and it works by matching the current situation to similar cases stored in memory. System 2 is slow, deliberate, and analytical. It involves consciously working through evidence, weighing possibilities, and applying logical rules.
Neither system is inherently better. System 1 is essential in time-pressured situations like emergency medicine, where a clinician may need to act in seconds. It also becomes more accurate with experience, because complex reasoning operations gradually migrate from System 2 to System 1 as a clinician sees more cases and builds stronger pattern recognition. An experienced emergency physician recognizing a heart attack from a patient’s appearance and a few key symptoms is using System 1 effectively.
System 2 takes over when situations are unfamiliar, complex, or high-stakes with time to think. It’s more cognitively demanding and slower, but it’s better suited to problems where the answer isn’t obvious. The risk with System 1 is that it can misfire when a clinician is fatigued, hungry, stressed, or emotionally distracted. The risk with System 2 is that it’s not foolproof either, and it can be derailed by the same cognitive biases that affect all human thinking.
How Expertise Changes the Process
One of the clearest differences between novice and experienced clinicians is the mental structures they bring to a case. Experienced clinicians develop what researchers call “illness scripts,” which are stored mental templates for diseases. Each script contains the underlying mechanism of a disease, its typical signs and symptoms, the conditions that predispose someone to it (age, genetics, lifestyle), and how it’s typically managed.
When an experienced clinician encounters a patient, they rapidly match the presentation against these scripts. This is why a seasoned physician can often generate an accurate hypothesis within minutes. Medical students, by contrast, haven’t built a library of illness scripts yet. They tend to work through case information more thoroughly and systematically because they lack the shortcuts that come from having seen hundreds of similar patients. The transition from novice to expert is largely a story of building, refining, and reorganizing these mental templates through clinical experience.
Experts also perceive clinical information differently. They literally “see” things that novices miss, not because their eyes are better, but because their knowledge allows them to focus attention on what’s relevant and filter out what isn’t. This perceptual advantage is a hallmark of expertise across many fields, from chess to radiology.
Cognitive Biases That Derail Reasoning
Clinical reasoning is vulnerable to the same cognitive biases that affect all human decision-making, and several show up repeatedly in medical settings:
- Anchoring bias: Locking onto the first piece of information and letting it disproportionately shape the diagnosis, even when later evidence points elsewhere.
- Confirmation bias: Favoring information that supports a hypothesis you’ve already formed while downplaying evidence that contradicts it.
- Search-satisficing bias: Stopping the search for a diagnosis too early because the current explanation seems “good enough,” potentially missing a second condition or the correct one.
- Framing bias: Being influenced by how information is presented rather than what it actually contains. The same data framed positively versus negatively can lead to different conclusions.
- Blind spot bias: Believing you’re less susceptible to bias than your colleagues, which makes you less likely to check your own reasoning.
These biases don’t reflect carelessness. They’re built into how human cognition works, and even highly skilled clinicians fall prey to them, particularly under time pressure or cognitive load.
Strategies for Sharper Reasoning
The most effective countermeasure to cognitive bias is metacognition: thinking about your own thinking. One practical technique is called decision justification, where a clinician pauses to explicitly articulate why they believe a particular diagnosis is correct. This interrupts the automatic flow of System 1 thinking and forces a shift to more analytical processing. Research shows it reduces cognitive bias by prompting clinicians to re-evaluate initial hypotheses and reconsider available evidence rather than running on pattern recognition alone.
Decision justification works because it creates a moment of friction in the reasoning process. Instead of moving smoothly from pattern to conclusion, the clinician has to defend the conclusion to themselves. This simple act of self-explanation exposes weak links in the logic, surfaces assumptions that might be wrong, and increases awareness of which reasoning strategy is being used.
Other approaches include deliberately considering alternative diagnoses (sometimes called a “diagnostic time-out”), seeking disconfirming evidence for a working hypothesis, and being especially cautious when conditions known to impair System 1 are present, such as fatigue, hunger, or emotional distress.
How Clinical Reasoning Is Assessed
Testing clinical reasoning is harder than testing factual knowledge because reasoning involves judgment under uncertainty, not just recall. One tool designed for this is the Script Concordance Test, which presents a clinical scenario, states a diagnostic hypothesis, then introduces a new piece of information and asks how that information changes the likelihood of the hypothesis. Responses are scored on a five-point scale, from “much less likely” to “much more likely,” and compared against the responses of expert clinicians.
What makes this test distinctive is that it targets reasoning in ambiguous situations where there’s no single right answer. It measures whether a trainee’s thinking aligns with how experienced clinicians process uncertain information. Studies consistently show that more experienced clinicians score higher, confirming that the test captures something real about how reasoning develops with practice.
Where AI Fits In
Recent benchmarking has tested how well large language models perform on clinical reasoning tasks compared to human clinicians. In a 2025 study published in NEJM AI, ten AI models were compared against over 1,000 medical students, 193 residents, and 300 attending physicians using Script Concordance Tests. The best-performing model, OpenAI’s o3, scored 67.8%, matching or exceeding student performance on several exams but falling short of senior residents and attending physicians.
The more revealing finding was in how the AI models reasoned. All models showed systematic overconfidence, overusing extreme responses and rarely selecting the neutral option that indicates new information shouldn’t change a hypothesis. In other words, the models struggled with a skill that’s central to good clinical reasoning: knowing when evidence is irrelevant and your current thinking should stay put. Models specifically optimized for step-by-step logical reasoning actually performed worse on this dimension, suggesting that the kind of flexible, uncertainty-tolerant thinking that experienced clinicians develop doesn’t emerge naturally from current AI architectures.