Internal Medicine · Clinical Reasoning · Article / Review
Clinical reasoning is not a checklist — it is a process of updating probabilities
Lede: Good clinical reasoning is the disciplined conversion of incomplete information into an evolving probability map—not the mechanical completion of a differential diagnosis.
Why it matters
Internal medicine is practiced under uncertainty. The clinician rarely begins with a complete dataset; instead, each history item, examination finding, laboratory result, image, and response to treatment should modify the relative probability of competing diagnoses. Expert reasoning therefore depends on how well we represent the problem, activate relevant illness scripts, interpret evidence in context, and reopen the case when new data stop fitting the favored explanation.
1. Start by compressing the case
A useful problem representation is not a shorter history. It is a selective synthesis of the features that change diagnostic probability: age and risk context, tempo, severity, localization, key positive findings, and discriminating negatives. Semantic qualifiers such as acute versus chronic, focal versus diffuse, inflammatory versus noninflammatory, or exertional versus resting transform raw observations into diagnostically useful information.
2. Use illness scripts to organize—not replace—thinking
Illness scripts are structured knowledge networks that link predisposing conditions, pathophysiologic mechanisms, and expected clinical consequences. They allow clinicians to recognize patterns quickly, but they are also used to compare near-neighbor diagnoses and identify mismatches. The value of a script is not that it makes diagnosis automatic; it makes comparison more efficient.
3. Think in pretest probabilities
Before ordering a test, ask what you believe before seeing the result. That pretest probability is shaped by prevalence, patient-specific risk, setting, and the clinical pattern. A positive result in a low-probability patient may mean something very different from the same result in a high-probability patient.
4. Tests should update probability—not merely generate data
Sensitivity and specificity describe test performance, but likelihood ratios are especially useful for bedside reasoning because they indicate how much a particular result should shift diagnostic odds. The practical question is whether that shift crosses a threshold that changes the next action: stop testing, obtain more evidence, or treat.
5. Search for disconfirming evidence
A working diagnosis should explain the important findings with minimal contradiction. When new information does not fit, do not simply add exceptions to protect the original hypothesis. Ask what diagnosis would better account for the discordant feature. This is one of the most useful safeguards against premature closure.
6. Reassess after every major piece of new information
Clinical reasoning is iterative. A patient who looked like uncomplicated pneumonia at presentation may require a different frame after persistent hypotension, a new murmur, an unexpected eosinophilia, or failure to respond as predicted. A diagnosis is not a label attached once; it is a hypothesis that survives repeated attempts to falsify it.
The takeaway
Clinical reasoning is the repeated cycle of representing the problem, estimating probability, testing the hypothesis, and updating when new evidence arrives. The goal is not to generate the longest differential—it is to make the next decision more accurate.
