High-Yield Biostatistics Formulas for the USMLE
Biostatistics on the USMLE rewards a handful of formulas that recur across diagnostic-test, epidemiology, and clinical-trial questions. The organizing framework is to know which measures describe a test (sensitivity, specificity, likelihood ratios — fixed properties), which depend on the population (PPV, NPV — vary with prevalence), which measure association (relative risk, odds ratio), which quantify treatment benefit (ARR, RRR, NNT), and which describe disease frequency (prevalence, incidence). Master both the formula and the setting in which each is valid.
Diagnostic Test Characteristics (Test Properties)
These are intrinsic properties of the test and do not change with prevalence. Sensitivity = TP / (TP + FN), the proportion of diseased correctly identified. Specificity = TN / (TN + FP), the proportion of non-diseased correctly identified. Memory aids: a highly Sensitive test, when Negative, rules OUT disease (SnNOut); a highly Specific test, when Positive, rules IN disease (SpPin). Screening tests prioritize high sensitivity so as not to miss disease; confirmatory tests prioritize high specificity so as not to misdiagnose.
Predictive Values (Population-Dependent)
Predictive values answer the patient's real question and depend on prevalence. PPV = TP / (TP + FP): given a positive test, the probability of disease. NPV = TN / (TN + FN): given a negative test, the probability of no disease. Because they depend on prevalence, they vary by population. When prevalence is LOW, even highly specific tests generate many false positives, so PPV falls (most positives are false positives) while NPV rises (a negative is very reassuring). When prevalence is HIGH, PPV rises and NPV falls. With a 90% sensitive, 85% specific test at 1% prevalence, PPV is only about 5% — the reason we don't screen everyone for rare diseases.
Likelihood Ratios (Prevalence-Independent)
Likelihood ratios express how much a test result shifts the probability of disease and, unlike PPV, apply across populations because they are independent of prevalence. Positive LR = Sensitivity / (1 − Specificity): how much more likely a positive test is in someone with disease versus without. Negative LR = (1 − Sensitivity) / Specificity: how much more likely a negative test is in someone with disease versus without. Interpretation: LR+ >10 is strong evidence for disease, 5–10 moderate, 2–5 weak, and 1 means the test is useless; LR− <0.1 is strong evidence against disease and 0.1–0.2 moderate. They combine with pre-test odds: post-test odds = pre-test odds × LR (where odds = probability / (1 − probability)).
Measures of Association (RR vs. OR)
Relative Risk = Risk in exposed / Risk in unexposed, and is used in cohort studies where you can measure incidence. Odds Ratio = (a × d) / (b × c), used in case-control studies, where you select on outcome and therefore cannot calculate incidence or relative risk. Case-control designs are efficient for rare diseases because they recruit cases directly, avoiding the need to follow a huge at-risk cohort for years to accumulate enough incident cases. Separately, when a disease is rare the OR mathematically approximates the RR, which is what justifies using the OR as a valid estimate of the RR that cannot be measured directly in these studies. Be precise: an OR of 9.3 means the odds (not the incidence or relative risk) of disease are 9.3 times higher in the exposed group. Attributable Risk = Risk in exposed − Risk in unexposed, the excess risk due to the exposure.
Treatment Benefit (ARR, RRR, NNT)
These translate trial results into meaningful benefit. Absolute Risk Reduction = Risk in control − Risk in treatment. Relative Risk Reduction = ARR / Risk in control. Number Needed to Treat = 1 / ARR, the number of patients you must treat to prevent one outcome. Example: a drug lowering heart attack risk from 4% to 3% gives ARR = 1%, RRR = 25%, and NNT = 100. Marketing touts the impressive-sounding RRR, but the honest counseling figure is the ARR — always communicate absolute numbers to patients.
Measures of Disease Frequency
Prevalence = Cases / Total population (all existing cases at a point). Incidence = New cases / At-risk population over time (new cases only). The water-bucket analogy: incidence is water flowing in, prevalence is water in the bucket, and deaths/cures flow out — so Prevalence = Incidence × Duration. High incidence with low prevalence signals a rapidly fatal or rapidly cured disease (acute MI, many cancers); high prevalence with low incidence signals a long-lasting disease (HIV on treatment, diabetes).
High-yield
- Sensitivity and specificity are fixed properties of the test; PPV and NPV depend on prevalence.
- SnNOut: a sensitive test that is negative rules OUT; SpPin: a specific test that is positive rules IN.
- Low prevalence → low PPV even for an excellent test (many false positives); high prevalence → low NPV.
- Likelihood ratios are prevalence-independent and can be applied across populations for bedside reasoning.
- LR+ >10 = strong evidence for disease; LR− <0.1 = strong evidence against; LR = 1 means a useless test.
- Post-test odds = pre-test odds × likelihood ratio.
- Relative risk comes from cohort studies; odds ratio comes from case-control studies.
- Case-control studies are efficient for rare diseases because they recruit cases directly rather than following a large cohort for years; separately, when disease is rare the OR approximates the RR, validating OR as an RR estimate.
- NNT = 1/ARR; always counsel patients with absolute (ARR) rather than relative (RRR) figures.
- Prevalence = Incidence × Duration.
Pitfalls
- Confusing OR with RR or incidence in a case-control study — you cannot calculate incidence or relative risk when selecting on outcome.
- Reporting an odds ratio as a risk or incidence ratio; an OR of 9.3 means 9.3× higher odds, not 9.3× the incidence.
- Conflating why case-control studies are efficient for rare diseases (direct recruitment of cases, no need to follow a huge cohort) with the separate rare-disease assumption that makes OR ≈ RR (a validity justification, not an efficiency rationale).
- Assuming PPV is a property of the test — it changes with prevalence, so a great test can still yield mostly false positives in a low-prevalence population.
- Being impressed by a large RRR while ignoring a small ARR (e.g., 25% RRR but only 1% ARR, NNT of 100).
- Using survival from diagnosis to judge screening — lead-time bias makes screening appear beneficial; measure mortality instead.
- Mixing up sensitivity (TP/(TP+FN)) with specificity (TN/(TN+FP)) or swapping their denominators.
- Forgetting that likelihood ratios, unlike predictive values, stay constant across populations.
Clinical pearls
- A negative highly sensitive test is your best tool to rule out disease.
- At 1% prevalence, a 90/85 test yields a PPV near 5% — most positives are false alarms.
- OR ≈ RR only when the disease is rare — this validates OR as an RR estimate, not why case-control studies are chosen.
- Case-control studies win for rare diseases because you can enroll cases directly instead of waiting for them to accrue in a cohort.
- Convert to odds before applying a likelihood ratio: post-test odds = pre-test odds × LR.
- 'This drug lowers your risk from 4 in 100 to 3 in 100' beats '25% reduction' every time.
- High incidence, low prevalence = rapidly fatal or rapidly cured disease.
Frequently asked
Why does PPV fall when disease prevalence is low even if the test is excellent?
Because in a low-prevalence population there are far more disease-free people, so even a small false-positive rate generates many false positives that outnumber the true positives. At 1% prevalence a 90% sensitive, 85% specific test gives a PPV of only about 5%.
When do I use relative risk versus odds ratio?
Use relative risk (Risk exposed / Risk unexposed) in cohort studies where incidence can be measured. Use the odds ratio ((a×d)/(b×c)) in case-control studies, where selection is on outcome and incidence cannot be calculated. For rare diseases the OR approximates the RR, which validates using it as an RR estimate.
Why are case-control studies efficient for rare diseases, and how does that differ from the OR ≈ RR rule?
Case-control studies are efficient for rare diseases because they let you recruit a sufficient number of cases directly, avoiding the need to follow an enormous at-risk cohort for years to accumulate enough incident cases as a cohort study would require. That efficiency is about study design and feasibility. The OR ≈ RR approximation is a separate statistical property — when the disease is rare, the odds ratio mathematically approaches the relative risk, which justifies interpreting the OR as an estimate of the RR you cannot directly measure. Don't conflate the two: one explains why the design is chosen, the other explains why the resulting number is a valid RR proxy.
How do I interpret an odds ratio of 9.3 in a case-control study of smoking and lung cancer?
It means smokers have 9.3 times the odds of lung cancer compared with non-smokers. It does not mean 9.3 times the incidence or a relative risk of 9.3, because case-control studies cannot yield incidence or relative risk (though for a rare disease the OR closely estimates the RR).
What is the difference between ARR and RRR, and why does it matter?
ARR = Risk in control − Risk in treatment (the absolute benefit), while RRR = ARR / Risk in control (the proportional benefit). A drug lowering risk from 4% to 3% has an ARR of 1% but an RRR of 25%. RRR sounds larger, so always counsel patients with absolute numbers.
Why are likelihood ratios preferred over predictive values for bedside reasoning?
Likelihood ratios are calculated from sensitivity and specificity and are independent of prevalence, so they apply across different populations. Predictive values change with prevalence, making LRs more portable for individual clinical decisions.
How does lead-time bias make screening look better than it is?
Screening detects disease earlier, so survival measured from diagnosis appears longer even if the time of death is unchanged. The fix is to measure mortality (death rate) rather than survival from diagnosis.
What does the relationship Prevalence = Incidence × Duration tell me?
Prevalence reflects both how many new cases arise (incidence) and how long people remain cases (duration). High incidence with low prevalence indicates a rapidly fatal or rapidly cured disease; high prevalence with low incidence indicates a long-lasting disease like treated HIV or diabetes.
Turn this into reasoning you can use on exam day — practice High-Yield Biostatistics Formulas on branching cases where your decisions shape the patient.