General Internal Medicine and Clinical Reasoning
This domain tests whether you can turn an undifferentiated presentation into a ranked differential, then into a defensible next step. The exam rewards probabilistic thinking and disciplined reassessment, not encyclopedic recall of rare syndromes.
Anchor on pretest probability
Every test result means nothing without a pretest probability. Estimate it from prevalence, age, sex and the illness script, then ask how far the test moves you. A likelihood ratio above 10 (or below 0.1) usually changes management; ratios between 0.5 and 2 usually do not.
- Sensitivity rules out when negative (SnNout); specificity rules in when positive (SpPin).
- Do not order a test whose result will not change what you do.
- Tests in series raise specificity and lower sensitivity; in parallel the reverse.
Structure the reasoning
Use a systems screen (VINDICATE) to avoid premature closure, and a diagnostic timeout when the working diagnosis is not evolving as expected. Review every medication as a potential cause; polypharmacy and prescribing cascades are classic exam material.
Worked example
A 60-year-old with pleuritic pain has a Wells score of 1.5 (low pretest probability, roughly 5%). D-dimer sensitivity is about 95% with specificity 40%, so the negative likelihood ratio is approximately 0.125. A negative result moves probability to about 0.6% - below the testing threshold - so pulmonary embolism is safely excluded without CT. In a high pretest probability patient the same test cannot exclude disease; go straight to imaging.
Traps that cost marks
- Treating a highly sensitive test as a rule-in test.
- Applying a validated score to a population it was not derived in.
- Ignoring comorbidities that change the risk-benefit balance.
- Failing to state the management consequence of the test you propose.