Start with the question being asked
The official Step 3 content outline currently lists biostatistics, epidemiology, population health and medical-literature interpretation at 11–13% of MCQ content; USMLE notes that ranges may change. These three exercises cover only selected basics within that broader area.
- Identify the outcome, population and follow-up period.
- Write the numerator and denominator before substituting numbers.
- Say the result in ordinary words, including its limits.
Sensitivity, specificity, PPV and NPV
Assume an appropriate reference standard establishes whether the fictional condition is present. TP means true positive, FP false positive, FN false negative and TN true negative.
| Measure | Formula | Starting group |
|---|---|---|
| Sensitivity | TP / (TP + FN) | People who have the condition |
| Specificity | TN / (TN + FP) | People without the condition |
| Positive predictive value (PPV) | TP / (TP + FP) | People with positive test results |
| Negative predictive value (NPV) | TN / (TN + FN) | People with negative test results |
Multiply each proportion by 100 to express a percentage. These are FDA diagnostic-performance definitions. Estimates depend on the study population and reference standard; they are not guarantees about an individual patient. If condition status is not established appropriately, agreement with another test alone does not establish these accuracy measures.
ORIGINAL EXAMPLE 1 · DIAGNOSTIC DENOMINATORS
What does a positive result mean?
A fictional community study tests 1,000 volunteers for condition Z. Every volunteer also receives the reference-standard assessment. The complete results are below. Among volunteers whose test is positive, what proportion actually has condition Z?
| Test result | Condition present | Condition absent | Total |
|---|---|---|---|
| Positive | 80 (TP) | 180 (FP) | 260 |
| Negative | 20 (FN) | 720 (TN) | 740 |
| Total | 100 | 900 | 1,000 |
- 80.0%
- 20.0%
- 30.8%
- 97.3%
- 10.0%
Show answer and worked explanation
Answer C: 30.8%. The question starts with positive results, so use all 260 positive results as the denominator. Of these, 80 are true positives.
PPV = 80 / (80 + 180) = 80 / 260 ≈ 30.8%
Sensitivity is 80/100 = 80%; specificity is 720/900 = 80%; NPV is 720/740 ≈ 97.3%. The 20% value is the false-negative proportion among people with the condition. The 10% value is this study population’s prevalence. Each distractor answers a different question.
In natural frequencies: about 31 of every 100 positive results in this fictional study are true positives. That is not the same statement as detecting 80 of every 100 people who have the condition.
Why the base rate changes predictive value
With sensitivity and specificity held at 80%, a condition that is present in 10% of 1,000 people gives the table above. If it is present in only 1%, the expected counts become TP 8, FN 2, FP 198 and TN 792. PPV is then 8/206 ≈ 3.9%. The many people without the condition can generate more false positives than true positives.
This is a controlled arithmetic comparison, not a claim that test sensitivity and specificity remain fixed across real settings. CDC’s field epidemiology guidance specifically calls for considering prevalence when interpreting predictive values. Do not transfer the PPV from an enriched study sample directly to a different population.
Try the 2×2 calculator
Change the four counts to explore the denominators. Use fictional whole-number counts, not patient information. The calculator runs only in your browser, sends no entries anywhere and uses no storage. Worked answers above remain available without JavaScript.
Relative risk, absolute reduction and NNT
For a binary adverse outcome, first calculate the risk in each arm over the same follow-up period. A risk is events divided by participants at risk; it is not an incidence rate based on person-time. CDC’s archived methods lesson defines RR as a ratio of risks.
| Measure | Calculation |
|---|---|
| Risk ratio (RR) | Treatment risk / control risk |
| Absolute risk reduction (ARR) | Control risk − treatment risk |
| Relative risk reduction (RRR) | ARR / control risk = 1 − RR |
| NNT for an additional beneficial outcome | 1 / ARR, using a proportion such as 0.05, not the whole number 5 |
NNT refers to an average additional benefit compared with a specified alternative over a stated time. Round a noninteger NNT upward. If ARR is zero, there is no finite NNT for benefit; if treatment increases the adverse outcome, describe harm instead of reporting a negative “benefit.” These interpretation and rounding points follow the Cochrane methods handbook.
ORIGINAL EXAMPLE 2 · TRIAL EFFECT SIZE
Halving relative risk: how large is the absolute benefit?
A fictional randomized trial compares a new discharge-support program with usual support. All 400 participants complete 12 months of follow-up. An unplanned readmission occurs in 10 of 200 participants assigned to the new program and 20 of 200 assigned to usual support. What are the point estimates for RR, ARR and the NNT to prevent one additional participant having a readmission over 12 months?
- RR 2.0; ARR 5 percentage points; NNT 20
- RR 0.5; ARR 5 percentage points; NNT 20
- RR 0.5; ARR 50 percentage points; NNT 2
- RR 0.5; ARR 5 percentage points; NNT 5
- RR 1.0; ARR 0; no finite NNT for benefit
Show answer and worked explanation
Answer B. Treatment risk is 10/200 = 0.05 = 5%; control risk is 20/200 = 0.10 = 10%.
RR = 0.05 / 0.10 = 0.5
ARR = 0.10 − 0.05 = 0.05 = 5 percentage points
RRR = 0.05 / 0.10 = 0.50 = 50%
NNT = 1 / 0.05 = 20 over 12 months
Relative risk is halved, but the absolute change is 5 fewer participants with the event per 100 assigned to the program. Reversing the RR produces choice A. Treating a relative reduction as an absolute 50-point benefit produces C. Choice D mistakes the whole-number percentage for the NNT; E discards the observed difference.
These are estimates from one fictional trial. NNT 20 does not guarantee exactly one prevented event in every group of 20 people, establish statistical significance or tell us the balance of all benefits and harms. A confidence interval and study-quality assessment would be needed to judge uncertainty. Do not extend this 12-month estimate to a lifetime or use it as advice about an actual program.
Keep the units visible: 10% down to 5% is a 5-percentage-point absolute reduction and a 50% relative reduction. Both are true; they answer different questions.
A p-value does not establish equivalence
For a specified test and model, a p-value describes how incompatible the observed data are with the null hypothesis. More precisely, under that null and its assumptions, it is the probability of a test statistic at least as extreme as the one observed. It is not the probability that the null hypothesis is true, an effect size or the chance that a treatment will work for a patient.
A result above a prespecified significance threshold means the test did not detect a statistically significant difference. It does not prove equality. CDC notes that small samples can miss meaningful differences; FDA/ICH E9 separately explains why equivalence needs an appropriate design and justified margins. Read the confidence interval, effect size, study design and possible bias together.
ORIGINAL EXAMPLE 3 · INTERPRETING UNCERTAINTY
What can a nonsignificant comparison establish?
A separate fictional superiority trial compares two follow-up strategies. The estimated difference in adverse-event risk, strategy A minus strategy B, is −2 percentage points, with a 95% confidence interval from −6 to +2 percentage points. A two-sided test reports p = 0.33; the prespecified significance threshold is 0.05. No equivalence margin was specified. Which conclusion is best supported?
- The strategies have been shown to be equivalent because p exceeds 0.05.
- There is a 33% probability that the null hypothesis is true.
- Strategy A will benefit 33% of individual participants.
- No statistically significant difference was detected; equivalence has not been established.
- Every clinically important benefit of strategy A has been excluded.
Show answer and worked explanation
Answer D. The interval includes no difference and estimates in both directions. At the stated threshold, the superiority test is nonsignificant. Equivalence was not tested with prespecified acceptable margins.
Choice A confuses failure to detect a difference with evidence of equivalence. B turns a probability calculated under a hypothesis into a probability about that hypothesis. C mistakes a p-value for a patient-level treatment effect. E is unsupported: the interval still includes a 6-percentage-point reduction, and the prompt does not define what difference would be clinically important.
In real evidence review, a point estimate and interval should be interpreted with the outcome’s importance, precision, missing data, bias and applicability. The toy p-value is stipulated for this exercise; it is not calculated from the earlier trial.
A short recall checklist
- Positive test? Start with all positive results if asked for PPV.
- Condition present? Start with all people who have the condition if asked for sensitivity.
- Trial benefit? State both arm risks, the comparator and the follow-up before calculating ARR or NNT.
- Uncertainty? Interpret the interval and design; p above 0.05 is not proof of equivalence.
These examples do not cover all study designs, likelihood ratios, confounding, survival analysis, clustered data or medical-literature questions. The calculator supplies only point estimates from a simple fully classified 2×2 table.
Methods and official resources
Links and the relevant sections were checked on . Older foundational methods are identified by date; availability today does not make them newly issued guidance. None of these sources supplied the fictional questions or datasets above.
- USMLE: Step 3 Exam Content ↗Current examination outline: the broad content range, not validation of these exercises.
- FDA: Reporting results from studies evaluating diagnostic tests ↗Diagnostic definitions and reference-standard requirements. Published March 2007; checked at its current official URL.
- CDC Field Epidemiology Manual: Epidemiology–laboratory collaborations ↗Why sensitivity, specificity and prevalence must be considered together. Chapter dated August 2024.
- CDC: Measures of association (archived methods lesson) ↗Risk-ratio definition. This is an explicitly archived 2012 mathematical lesson, not new clinical guidance.
- Cochrane Handbook: Interpreting results ↗Sections 15.4.1–15.4.4: relative/absolute effects, time-specific NNT and rounding convention.
- CDC/NCHS: Statistical significance ↗Sample size and the distinction between statistical and public-health importance. Reviewed July 2024.
- FDA/ICH E9: Statistical Principles for Clinical Trials (PDF) ↗Section 3.3.2: equivalence/noninferiority and prespecified margins. Foundational 1998 guidance; not a 2026 guideline.
FreeUSMLE is independent and is not affiliated with or endorsed by USMLE, NBME, CDC, FDA or Cochrane. See the Medical & AI notice and Sources & copyright policy. No examination result is guaranteed.