Describe the problem before naming it
A patient says, “I'm dizzy.” That word could describe spinning, faintness, imbalance, or another experience. Ask about onset, duration, triggers, associated symptoms, and functional effects before adopting a diagnosis from the problem list. Decide early whether the patient is stable enough for an office evaluation.
A working diagnosis is the explanation currently guiding care. It can be reasonable without being certain. Separate observations from interpretation: “lightheaded when standing after a medication change” is an observation; “medication-induced hypotension” is a possible explanation that needs supporting assessment.
The referral says “anxiety”
A patient referred for anxiety describes new breathlessness when climbing one flight of stairs. The referral label does not establish the cause of the breathlessness.
Clarify onset and progression, symptoms at rest, associated chest discomfort or fainting, medication changes, and relevant cardiopulmonary risks. Assess vital signs and examination findings, then determine urgency and the needed workup. Anxiety may be relevant, but the new limitation should be evaluated on its own evidence.
A concise problem summary helps you choose the next question. Compare “short of breath” with “new, progressive exertional breathlessness over two weeks in a previously active adult.” Add the findings that change your differential, including relevant negatives. Do not fill the summary with normal findings that do not affect the decision.
Build a useful differential
Rank the possibilities by how well they fit and what you need to do about them. Include common causes that fit, serious causes for which delay matters, and alternatives suggested by a discordant finding.
Pretest probability is the estimated chance of a particular disease before the next test. It depends on the population and the patient's history, examination, exposures, and prior findings. A disease common in a specialist referral clinic may be uncommon among unselected primary care patients.
| Question | What it changes |
|---|---|
| What best fits the course and findings? | The leading explanation and initial management. |
| What serious alternative remains plausible? | The urgency and the evidence needed to exclude it sufficiently. |
| Which finding does the leading diagnosis fail to explain? | A reason to broaden or revise the differential. |
| What question or examination would separate the main possibilities? | The next useful information to obtain. |
Probability and consequence are different. A relatively unlikely condition may merit urgent investigation if a missed diagnosis would be dangerous. That does not mean investigating every imaginable catastrophe: the possibility must be grounded in the patient's presentation and an appropriate clinical pathway.
A probability estimate need not pretend to more precision than the evidence supports. When no suitable validated estimate exists, state the reasons for your assessment and what remains uncertain. If the decision changes across a plausible range of risk, obtain better information or choose a plan that accounts for that uncertainty.
A sore calf after a longer walk
An adult reports unilateral calf pain after increasing their walking distance. A muscle injury is plausible. Before settling on it, ask about swelling, previous clots, immobility, surgery, cancer, and chest or breathing symptoms; examine both legs.
The important next finding is whether there is objective unilateral swelling or tenderness along the deep veins. “It started after exercise” supplies a possible explanation, but does not settle which pathway fits. Continue this case in the decision-rule example.
Understand test performance
A test result is evidence about a diagnosis. To interpret it, distinguish performance among people whose disease status is known from the probability of disease in the person who just received a result.
| Term | Question it answers |
|---|---|
| Sensitivity | Among people who have the disease, what proportion test positive? |
| Specificity | Among people who do not have the disease, what proportion test negative? |
| False negative | The test is negative despite disease being present. |
| False positive | The test is positive despite disease being absent. |
| Positive predictive value | Among people with positive results, what proportion have the disease? |
| Negative predictive value | Among people with negative results, what proportion do not have the disease? |
A test with 90% sensitivity does not give a patient with a positive result a 90% probability of disease. Predictive values depend on how common the disease was before testing as well as the test's performance.
Performance also depends on how and when the test is used. Sampling, disease stage, prior treatment, the chosen cutoff, and the population studied can affect accuracy. Before applying a published estimate, check whether those circumstances resemble your patient's situation.
AHRQ: probability-based diagnosis.
Calculate what a result means
A positive result is not a diagnosis
Imagine testing 1,000 people, of whom 100 have a disease. The invented test has 90% sensitivity and 95% specificity. Assume everyone is tested under the conditions for which those estimates apply.
| Result | Disease present: 100 | Disease absent: 900 |
|---|---|---|
| Positive | 90 true positives | 45 false positives |
| Negative | 10 false negatives | 855 true negatives |
There are 135 positive results: 90 are true positives. The probability of disease after a positive result is 90 ÷ 135, or about 67%. Among 865 negative results, 10 are false negatives: the remaining probability of disease after a negative result is about 1.2%.
Whether either result is enough to act depends on the disease and the proposed action. A 1.2% residual probability may be acceptable for one decision and unacceptable for another. If the hypothetical disease could cause irreversible harm before reassessment and confirmation is low risk, further testing may be worthwhile. If waiting would not worsen the outcome and confirmation requires a hazardous procedure, planned observation may be preferable. Those are different assumptions about consequences, even with identical arithmetic.
Change the starting risk; keep the same test
| Probability before testing | Disease probability after a positive | Disease probability after a negative |
|---|---|---|
| 1% | 15.4% | 0.1% |
| 10% | 66.7% | 1.2% |
| 50% | 94.7% | 9.5% |
At low starting probability, false positives can outnumber true positives. At high starting probability, a negative result can leave substantial risk. This is why the same test cannot be interpreted identically in every patient.
Calculating probability with likelihood ratios
A likelihood ratio (LR) describes how much a result changes the odds of disease. For a binary test, LR+ = sensitivity ÷ (1 − specificity); LR− = (1 − sensitivity) ÷ specificity.
For this test, LR+ is 18 and LR− is about 0.105. Convert probability to odds using p ÷ (1 − p), multiply by the appropriate LR, then convert back using odds ÷ (1 + odds).
At a 10% starting probability, odds are 0.10 ÷ 0.90. After a positive result, multiply by 18 to get odds of 2; probability is 2 ÷ 3, or 66.7%. Use a validated calculator when the arithmetic distracts from clinical assessment, and verify its inputs.
Do not repeatedly multiply likelihood ratios for related tests as if they supplied independent information. Two measurements affected by the same biological or sampling problem may add much less evidence than that calculation implies.
Decide whether testing will help
Before ordering, describe what you would do after a positive, negative, or inconclusive result. A test may help by changing treatment, deciding the care setting, directing further evaluation, or resolving an uncertainty that matters to the patient. If none of its likely results would change the plan, reconsider why it is needed.
A testing threshold is the level of concern at which the expected value of more information justifies testing. A treatment threshold is the point at which the expected benefit of treatment justifies acting. These are not universal percentages. They depend on the consequences of missed disease, the harms of testing and treatment, available alternatives, and patient preferences.
A sore throat with clear viral features
An otherwise well adult has two days of mild sore throat, cough, and rhinorrhea. They are drinking normally and have no airway concern or concerning examination finding.
CDC guidance advises against testing for group A streptococcal pharyngitis when clear viral symptoms are present. In this presentation, a strep test is unlikely to improve management and can create confusion if it detects carriage. Explain the clinical assessment, offer appropriate symptom care, and give instructions for an unexpected course.
Without clear viral features, examination alone does not reliably separate viral from streptococcal pharyngitis; the testing decision changes. A negative rapid antigen test also has different follow-up rules in children and adults. The patient's age and the actual assay matter.
CDC clinical guidance for group A streptococcal pharyngitis; CDC explanation of carriers and testing.
Broad panels create findings to interpret
A laboratory reference interval often contains the central 95% of values in a reference population. A value outside it is not necessarily disease, and a value inside it does not exclude every clinically important condition. A reference interval and a disease-specific decision threshold serve different purposes.
In a simplified model of 20 independent measurements, each with a 95% chance of lying inside its reference interval, the chance of at least one outside result is 1 − 0.9520, about 64%. Real laboratory values are often correlated and reference populations may differ from the patient; this is an illustration of multiple testing, not a prediction for a particular panel.
Choose tests that address the clinical question, then interpret the magnitude, trend, and context of each finding. An incidental borderline result can lead to repeat testing, imaging, or a procedure whose harms outweigh the value of the original low-yield test.
National Library of Medicine: reference ranges and test results.
Use clinical decision rules correctly
A clinical decision rule combines defined findings to estimate risk or select a diagnostic pathway. Check the intended population, inclusion and exclusion criteria, exact variables, assay requirements, and the action attached to the result. A score validated for one setting is not automatically valid in another.
The calf examination changes the next test
The patient is a stable, nonpregnant adult with a first suspected DVT and no chest or breathing symptoms. Calf circumference is 3.5 cm greater than on the asymptomatic side. There is localized tenderness along the deep venous system. You assess every item in the two-level DVT Wells score: all other positive criteria are absent, and no alternative diagnosis is at least as likely.
Swelling contributes 1 point and deep venous tenderness 1: 2 points, the DVT-likely branch. In this NICE pathway, seek a proximal leg vein ultrasound result within four hours if possible; a D-dimer test alone cannot finish this evaluation. In this case, the result returns within that interval and confirms proximal DVT, so the next decision is treatment assessment.
If that timing cannot be met, use the delayed-imaging branch: D-dimer testing, interim therapeutic anticoagulation with assessment of contraindications, and ultrasound within 24 hours. Negative imaging also needs the pathway’s next steps, which depend on the D-dimer result. Follow the complete guidance. The score selects a pathway; it does not diagnose DVT. Pregnancy, instability, or suspected PE requires an appropriate separate assessment.
Worked example uses NICE NG158, two-level DVT Wells score and diagnostic pathway, a UK guideline. See also the American Society of Hematology VTE diagnosis guideline; use the complete pathway supported by your local service.
When results conflict with the presentation, check whether the test was appropriate, timely, and technically adequate. Reassess the original probability and seek additional evaluation when residual concern warrants it.
Make observation a specific plan
Observation is appropriate when immediate testing or treatment is not needed and the expected course can be followed safely. It requires a working explanation, a time for reassessment, symptoms that should trigger earlier care, and a way for the patient to obtain that care.
“Come back if worse” leaves too much undefined
For the adult with a mild viral-appearing sore throat, a plan might explain the likely cause, expected improvement over the next few days, and when persistent symptoms need reassessment. Tell the patient which changes require urgent evaluation, such as difficulty breathing or inability to swallow fluids, and how to seek it.
If a patient cannot reliably communicate or return, adjust the plan: arrange a scheduled contact, involve an authorized support person, or use a different setting when needed. The medical finding has not changed, but the safety of waiting may have.
Communicate uncertainty directly: “The findings today fit a viral infection. If the course changes, we need to reconsider that explanation.” This gives the patient a reason to report new information without implying that every alternative has been excluded.
For ordered tests, identify who will review the result and act. For referrals, track whether the consultation occurred and whether its conclusions answered the question. The inbox primer describes how to distinguish reviewed results from completed follow-up.
Revisit the working diagnosis
Return visits, unexpected deterioration, persistent symptoms, and discordant results should prompt another look at the explanation. Read the original reasoning, but obtain a current history and examination rather than copying its conclusion.
| Error | Question to ask |
|---|---|
| Anchoring: holding too tightly to the first explanation | “If this patient were new to me, what would I consider today?” |
| Premature closure: ending the search before the evidence is sufficient | “Which important finding remains unexplained?” |
| Confirmation bias: favoring evidence that supports the current view | “What finding would make me change my mind, and have I looked for it?” |
| Diagnosis carried forward without verification | “Was this condition established, suspected, or simply copied?” |
Part of the story improves
A patient with dizziness and exertional breathlessness is found to be taking a duplicated blood-pressure prescription. After the regimen is corrected, the dizziness improves but breathlessness persists.
The response supports a medication contribution to dizziness. It does not explain the remaining breathlessness. Reassess that symptom's course, associated findings, and urgency, and investigate accordingly. Two processes can coexist; treatment response is evidence, not proof that the entire presentation had one cause.
The medication review is described in resolving changes after discharge.
When reviewing an eventual missed diagnosis, separate what was knowable at the time from what became clear later. Examine whether important information was missed, whether the chosen pathway fit the patient, and whether follow-up could have detected the changing course. A poor outcome alone does not establish that the initial judgment was unreasonable.
AHRQ diagnostic-safety resources cover clinical reasoning, communication, and system failures.
Cases to work through
You ordered a broad panel and one value is just outside the reference interval. What comes before assigning a diagnosis?
Review the magnitude, prior values, clinical context, medications, sample issues, and whether a disease-specific threshold applies. Decide whether confirmation or further evaluation is useful. A borderline flag alone does not establish a disease.
A decision rule gives a low-risk result, but the patient falls outside the population in which it was validated.
The score does not establish the promised level of safety for that patient. Use an appropriate pathway or obtain further assessment. Document why the rule was not applied rather than treating its numerical output as authoritative.
The working diagnosis is reasonable, but the patient has no reliable way to obtain planned follow-up. Is observation still the same decision?
No. Determine whether a scheduled call, coordinated visit, authorized support person, or different care setting can make the plan safe. A follow-up-dependent decision must account for whether follow-up can actually happen.