Step 1 topics → Public Health Sciences → Likelihood Ratios

Likelihood Ratios

USMLE Step 1 trap: Confuses LRs as prevalence-dependent when they are actually test-intrinsic. Likelihood ratios are derived solely from sensitivity and specificity and are independent of prevalence, making them portable across populations.

Likelihood ratios quantify how much a test result shifts your probability of disease, and USMLE Step 1 tests them both as formula recall and as clinical reasoning. LR+ tells you how much more likely a positive result is in someone with disease versus without; LR- tells you how much less likely a negative result is in disease versus without. Step 1 expects you to know the math and know what the numbers mean at the bedside. The formulas are LR+ = sensitivity / (1 − specificity) and LR- = (1 − sensitivity) / specificity. That's it. Everything else flows from those two.

The tricky part is that students conflate LRs with PPV and NPV, which are prevalence-dependent. LRs are not. They're derived entirely from sensitivity and specificity, so they travel with the test, not with the population. A test with LR+ of 15 has LR+ of 15 whether you're in a high-prevalence ICU or a low-prevalence outpatient clinic. USMLE Step 1 will sometimes set up a scenario in a new population and ask you which diagnostic measure stays the same — LRs do, PPV and NPV don't. That distinction is a high-yield trap.

The other place students struggle is Bayesian updating. You can't just multiply LR × pretest probability. You have to convert pretest probability to pretest odds (odds = p / (1 − p)), multiply by the LR to get post-test odds, then convert back to probability. It feels like extra steps, but the exam will occasionally walk you through a vignette that requires this logic. Know the threshold numbers cold: LR+ > 10 or LR- < 0.1 are clinically significant shifts; LRs between 0.5 and 2 are essentially noise.

From real student decks
35%have cards covering this topic
17%have mature cards

A gap in most decks — fewer than half of students in our cohort have cards covering this topic.

Common misconceptions

Common mistake
Wrong: Likelihood ratios change with disease prevalence, like PPV and NPV do.
Right: Likelihood ratios are derived solely from sensitivity and specificity and are independent of prevalence, making them portable across populations.
LRs are calculated purely from sensitivity and specificity — two properties of the test itself, not of the population being tested. PPV and NPV change with prevalence because they depend on how many true positives and true negatives exist in that specific population. LRs sidestep this entirely by comparing the probability of a result in diseased versus non-diseased individuals, a ratio that doesn't shift when prevalence shifts. This is precisely why LRs are the preferred metric when applying research data from one population to a different clinical setting.
Common mistake
Wrong: LR+ and LR- both increase post-test probability when elevated.
Right: LR+ >1 increases post-test probability of disease, while LR- <1 decreases it; a very low LR- is what rules out disease.
The key insight is that LR+ and LR- work in opposite directions. LR+ > 1 means a positive result is more common in disease than in non-disease, so it raises your post-test probability — the higher above 1, the better it rules in disease. LR- < 1 means a negative result is less common in disease than in non-disease, so it lowers your post-test probability — the closer to 0, the better it rules out disease. Students get tripped up thinking a 'high' LR- is good; it's actually a very low LR- (like 0.05) that gives you confidence in ruling out.
Common mistake
Gap: Unaware of the LR magnitude thresholds that define clinically meaningful diagnostic impact
LR+ >10 or LR- <0.1 generate large, clinically significant shifts in post-test probability; LRs between 0.5 and 2 are clinically negligible.
Not all statistically defined LRs are clinically useful. The widely used Jaeschke thresholds give you a practical filter: LR+ > 10 produces a large, clinically meaningful increase in post-test probability, while LR- < 0.1 produces a large, meaningful decrease. Values in the moderate range (LR+ 5–10, LR- 0.1–0.2) give moderate shifts. Values between 0.5 and 2 barely move the needle and generally shouldn't change your clinical decision. Knowing these cutoffs lets you quickly judge whether a test is diagnostically powerful or just generating paperwork.
Common mistake
Gap: Does not know the pretest odds × LR = post-test odds formula for Bayesian updating
Bayesian updating with LRs requires converting pretest probability to pretest odds, multiplying by the LR, then converting back to post-test probability.
The formula post-test probability = pretest probability × LR is wrong — you can't multiply probabilities by LRs directly. The correct process has three steps: first, convert pretest probability to odds using odds = p / (1 − p); second, multiply those pretest odds by the LR to get post-test odds; third, convert post-test odds back to probability using p = odds / (1 + odds). This works because LRs are ratios of probabilities in a likelihood framework, and odds-form is the natural unit for that multiplication. Skipping the conversion gives you a nonsensical answer outside the 0–1 range when pretest probability is high.
Guided session

Stuck on this? An AI tutor that probes your understanding and catches where your reasoning breaks.

Start a session →
Free Deck audit

Already run Anki? See if your deck covers this topic.

Upload your deck →

What the exam tests

  1. Calculate LR+ and LR- from sensitivity and specificity, and correctly identify which direction each ratio shifts post-test disease probability.
  2. Interpret the clinical significance of a given LR value — recognizing that LR+ > 10 and LR- < 0.1 produce meaningful diagnostic shifts, while values near 1 do not.
  3. Explain why likelihood ratios remain constant across populations with different disease prevalence, unlike PPV and NPV.
  4. Apply Bayesian pretest-to-posttest updating using the formula: post-test odds = pretest odds × LR, with correct conversion between probability and odds.

Can you avoid these mistakes?

A test has sensitivity of 90% and specificity of 80%. Calculate LR+ and LR-. Which one would you use to rule in disease, and which to rule out? What do the magnitudes tell you about clinical utility?
A researcher derives LR+ = 12 for a new cardiac biomarker using a hospital cohort where disease prevalence is 40%. A colleague wants to apply this LR+ in a primary care setting where prevalence is 5%. Is the LR+ still valid? What would change and what would stay the same?
A patient has a pretest probability of 30% for pulmonary embolism. A V/Q scan returns a result with LR+ = 6. What is the post-test probability? Walk through the full odds-based calculation.
You see two tests described in a vignette: Test A has LR+ = 1.8 and Test B has LR- = 0.08. A colleague says Test A is better because its LR is higher. Are they right? Which test is more clinically useful and why?

Related topics

Sensitivity, Specificity, and Cutoff / ROC Observational Study Designs Case-Control Studies Cohort Studies Randomized Controlled Trials

Find out what you actually understand about this topic — a tutor that probes your reasoning and catches where it breaks.

Start a session →

Already run Anki? Get a free deck audit →