Biostatistics on the USMLE is a small set of formulas asked in a large number of disguises. The arithmetic is almost always easy. What costs marks is choosing the wrong formula for the question, or putting the numbers in the wrong cell of the table before any calculating starts. Learn each formula as the answer to a specific question, set the table up the same way every time, and this becomes one of the most reliable areas of the exam.
Build the table the same way, every time
Nearly every diagnostic and risk question reduces to a two-by-two table. Use one fixed layout and never vary it:
- Columns are the truth: disease present on the left, disease absent on the right.
- Rows are the test or exposure: positive or exposed on top, negative or unexposed below.
- The cells are then a (true positive), b (false positive), c (false negative) and d (true negative).
Vignettes often present the numbers in a different order, or give totals and expect you to subtract. Draw the table on your noteboard before reading the answer options. Students who skip this step make most of their errors here, not in the division that follows.
Test characteristics follow the columns and rows
Four formulas describe how a diagnostic test performs.
Sensitivity = a / (a + c). Of the people who have the disease, the share the test catches. It reads down the disease-present column.
Specificity = d / (b + d). Of the people without the disease, the share the test correctly clears. It reads down the disease-absent column.
Positive predictive value = a / (a + b). Of the people who test positive, the share who truly have the disease. It reads across the top row.
Negative predictive value = d / (c + d). Of the people who test negative, the share who are truly free of it. It reads across the bottom row.
The distinction that the exam tests most often is this: sensitivity and specificity belong to the test, while predictive values depend on how common the disease is in the population being tested. Move the same test to a population with higher prevalence and the positive predictive value rises while the negative predictive value falls. If a question changes the population, it is almost always asking about predictive values.
Likelihood ratios combine both properties
Positive likelihood ratio = sensitivity / (1 - specificity).
Negative likelihood ratio = (1 - sensitivity) / specificity.
A positive likelihood ratio well above one means a positive result substantially raises the probability of disease; a negative ratio close to zero means a negative result substantially lowers it. Like sensitivity and specificity, likelihood ratios do not depend on prevalence, which is why they are useful for comparing tests.
Cut-offs move sensitivity and specificity in opposite directions
Many questions describe a test with a numerical threshold and ask what happens if the threshold moves. Lowering the cut-off for a positive result catches more true cases, so sensitivity rises, but it also labels more healthy people positive, so specificity falls. Raising the cut-off does the reverse. False negatives and false positives trade off against each other, and the question is usually asking you to notice which one the clinical setting cares more about.
Risk measures depend on the study design
Relative risk = [a / (a + b)] / [c / (c + d)]. The risk in the exposed group divided by the risk in the unexposed group. Use it for cohort studies and randomised trials, where you know how many people in each group developed the outcome.
Odds ratio = (a × d) / (b × c). Use it for case-control studies, where researchers chose people by outcome, so true risks cannot be calculated. When the outcome is rare, the odds ratio approximates the relative risk.
Attributable risk = risk in the exposed minus risk in the unexposed. It answers how much of the outcome in the exposed group is due to the exposure.
If the vignette describes recruiting people with and without a disease and looking back at exposures, the answer involves an odds ratio. If it follows exposed and unexposed people forward, it involves relative risk.
Treatment effects need absolute numbers
Absolute risk reduction = risk in the control group minus risk in the treatment group.
Relative risk reduction = absolute risk reduction / risk in the control group.
Number needed to treat = 1 / absolute risk reduction. Number needed to harm is the same calculation using the absolute increase in an adverse outcome.
The trap is trying to calculate the number needed to treat from a relative figure. A drug that halves a rare outcome and one that halves a common outcome have the same relative risk reduction and very different numbers needed to treat. If a question gives only relative figures, look for the baseline risk elsewhere in the stem.
Incidence, prevalence and the errors of a study
Incidence counts new cases over a period in the population at risk. Prevalence counts all existing cases at a point in time. In a stable situation, prevalence is roughly incidence multiplied by average disease duration, which explains a favourite question: a treatment that keeps people alive longer without curing them raises prevalence.
Type I error is concluding there is an effect when there is none; its probability is alpha. Type II error is missing a real effect; its probability is beta, and power is one minus beta. Larger samples increase power. A confidence interval for a ratio that includes one, or for a difference that includes zero, means the result is not statistically significant.
Practising so it sticks
Reading the formulas is not enough. Do biostatistics questions in mixed blocks rather than in a separate burst, because the skill being tested is recognising which formula a disguised question wants. When you miss one, write down whether the error was the table, the formula choice or the arithmetic. A bank that reports results by discipline makes it easier to see whether biostatistics is a genuine weak area or a few unlucky misses; QBankly reports that way, as do the established banks. Over a few weeks, the errors usually concentrate in one place, and that is the habit to fix.
Frequently asked questions
Which biostatistics formulas do I need to memorise for Step 1?
Sensitivity, specificity, both predictive values, relative risk, odds ratio, absolute and relative risk reduction, and number needed to treat cover most calculation questions. Likelihood ratios and the relationship between incidence and prevalence are worth knowing as well. Memorise them as questions they answer, not as strings of letters.
When do I use relative risk and when do I use an odds ratio?
Use relative risk for cohort studies and randomised trials, where you know how many people in each exposure group developed the outcome. Use an odds ratio for case-control studies, where participants were chosen by outcome and true risks cannot be calculated. For rare outcomes, the two give similar values.
How do I remember sensitivity and specificity?
Sensitivity is about people who have the disease, and specificity is about people who do not. A highly sensitive test, when negative, helps rule a disease out, and a highly specific test, when positive, helps rule it in. Tie each to its column of the two-by-two table and the formulas follow.