Step 1 topics → Public Health Sciences → Risk and Effect Measures (OR, RR, ARR, NNT)

Risk and Effect Measures (OR, RR, ARR, NNT)

USMLE Step 1 trap: Confuses when OR is a valid approximation of RR versus when it overestimates effect size. OR approximates RR only when the outcome is rare (<10%); for common outcomes, OR overestimates the magnitude of association compared to RR.

Risk and effect measures are the quantitative language of clinical research, and USMLE Step 1 expects you to move fluently between them — not just define them, but apply them to passage data and interpret what they mean clinically. The core measures are OR (odds ratio), RR (relative risk), ARR (absolute risk reduction), RRR (relative risk reduction), NNT (number needed to treat), and NNH (number needed to harm). Each one answers a slightly different question, and the exam exploits the fact that students often mix them up or apply them in the wrong study design context.

The trickiest part isn't the math — it's knowing which measure is appropriate for which study design, and understanding that the same data can look wildly different depending on whether you're reporting RRR versus ARR. USMLE Step 1 loves to give you a drug trial where RRR sounds impressive (say, 50%) but ARR is tiny (0.2%), making NNT enormous. Students who don't distinguish relative from absolute measures will be fooled every time. The exam also tests whether you know when OR is a valid stand-in for RR — and it isn't always.

Another common trap: students calculate NNT using RRR instead of ARR. The formula is NNT = 1/ARR, full stop. Getting the denominator wrong produces a completely different number and a wrong answer. Understanding why each formula works — not just memorizing it — is what separates students who ace the biostatistics questions from those who guess. This page will walk you through every angle the exam tests and the exact misconceptions that sink students.

From real student decks
70%have cards covering this topic
51%have mature cards
1.0median lapses in 14 days

Common misconceptions

Common mistake
Wrong: OR and RR are interchangeable in all study designs.
Right: OR approximates RR only when the outcome is rare (<10%); for common outcomes, OR overestimates the magnitude of association compared to RR.
OR and RR are mathematically distinct: OR divides the odds of an outcome in exposed vs. unexposed, while RR divides the probabilities. When the outcome is rare (incidence <10%), odds and probability are nearly identical, so OR ≈ RR — this is called the rare disease assumption. But for common outcomes, odds inflate compared to probabilities, so OR will be farther from 1.0 than RR is, making the association look stronger than it really is. Bottom line: always ask how common the outcome is before treating OR as a proxy for RR.
Common mistake
Wrong: A large relative risk reduction (RRR) always means a treatment is clinically important.
Right: RRR can be large even when ARR is tiny if baseline risk is very low; ARR and NNT better reflect absolute clinical benefit.
RRR is calculated as ARR divided by the control event rate, so it's entirely dependent on baseline risk. If the control group has a 0.4% event rate and the treatment group has a 0.2% rate, RRR = 50% — sounds dramatic — but ARR = 0.2% and NNT = 500. You'd need to treat 500 patients to prevent one event. RRR strips out the baseline context, which is why pharmaceutical marketing loves it and why USMLE Step 1 tests whether you can see through it. Always anchor RRR to ARR and NNT.
Common mistake
Wrong: NNT is calculated from RRR rather than ARR.
Right: NNT = 1/ARR; it requires the absolute risk reduction, not the relative risk reduction.
NNT answers the question: how many patients must be treated to prevent one bad outcome? The answer comes from the absolute difference in event rates — ARR — not the relative one. NNT = 1/ARR. If you plug in RRR instead, you get a number that has no valid clinical interpretation and is almost always much smaller (falsely optimistic). Remember: ARR already accounts for baseline risk, which is why it's the right denominator. If a question gives you both ARR and RRR, use ARR for NNT without hesitation.
Common mistake
Wrong: OR is the appropriate effect measure to report from a cohort study.
Right: Cohort studies can directly calculate incidence in both groups, so RR is the preferred effect measure; OR is used in case-control studies where incidence cannot be calculated.
In a cohort study, you follow exposed and unexposed groups forward in time, so you directly observe how many people in each group develop the outcome — that's incidence. Since you have incidence in both groups, you can calculate RR directly and it's the preferred, most interpretable measure. OR is used in case-control studies because you start with cases and controls (not exposed and unexposed), so you can't calculate incidence and therefore can't compute RR. Reporting an OR from a cohort study isn't wrong mathematically, but it's inappropriate and potentially misleading — the exam tests whether you know the difference.
Free Deck audit

See if your Anki deck covers this topic.

Upload your deck →
Guided session

Stuck on this? An AI tutor that probes your understanding.

Start a session →

What the exam tests

  1. Know the formulas for OR and RR, understand which study designs use each, and identify the specific condition (rare disease/outcome, <10% prevalence) under which OR closely approximates RR.
  2. Distinguish ARR from RRR, and explain why a large RRR can coexist with a clinically meaningless ARR — the exam will present trial data and ask you to identify which measure better reflects real-world benefit.
  3. Calculate NNT and NNH directly from a table or passage: NNT = 1/ARR, NNH = 1/absolute risk increase — and interpret what the number means for a single patient.
  4. Define hazard ratio, recognize that it is used in survival analyses (e.g., Kaplan-Meier studies), and understand its key assumption that the hazard ratio remains constant over time (proportional hazards).
  5. Given a complete clinical trial result, translate between RR, RRR, ARR, and NNT to determine the most clinically meaningful way to present a treatment effect.

Can you avoid these mistakes?

A cohort study of smokers vs. non-smokers finds that 30% of smokers and 10% of non-smokers develop disease X over 10 years. Calculate the RR. Would it be appropriate to report an OR here instead, and would it closely approximate the RR? Why or why not?
A clinical trial shows a new statin reduces cardiovascular events from 2% to 1% over 5 years. What is the ARR, RRR, and NNT? A colleague says 'the drug cuts risk in half' — is that statement misleading? How would you respond?
A student calculates NNT for a new antihypertensive by dividing 1 by the RRR of 40% and gets an NNT of 2.5. What error did they make, and what additional piece of information do you need to calculate NNT correctly?
A case-control study of a rare cancer (prevalence 0.5%) reports an OR of 4.2 for exposure to a certain chemical. A classmate says the RR must be much lower than 4.2. Are they correct? What principle applies here, and how would your answer change if the outcome had a 40% prevalence?

Related topics

Case-Control Studies Cohort Studies Observational Study Designs Randomized Controlled Trials

See how your Anki deck covers this topic.

Upload your deck for a free audit →