Study guide
This chapter is educational content only and does not guarantee any exam outcome. Biostatistics and epidemiology carry meaningful weight on Step 3, and questions typically embed a calculation or interpretation task inside a clinical scenario rather than asking for a bare definition. This chapter builds the core toolkit: recognizing study designs and their biases, working with diagnostic test statistics, interpreting measures of association and treatment effect, and applying evidence to an individual patient's care.
Study Design and Bias Recognition
Step 3 expects an examinee to identify the type of study described in a short passage and to recognize its characteristic strengths and vulnerabilities. A randomized controlled trial assigns participants to intervention or control by chance, which is the strongest design for establishing causation because randomization tends to balance both known and unknown confounders between groups. A cohort study follows groups defined by exposure status forward in time to observe outcomes, useful for studying multiple outcomes from one exposure but vulnerable to confounding since exposure is not randomly assigned. A case-control study works backward, starting with people who have an outcome and people who do not, then looking at prior exposures; this design is efficient for rare diseases but is particularly prone to recall bias, since people with the outcome may remember past exposures differently than those without it. Cross-sectional studies capture exposure and outcome at a single point in time, useful for estimating prevalence but unable to establish which came first. Common biases tested include selection bias, where the study sample is not representative of the target population; recall bias, most relevant to case-control designs; and confounding, where a third variable is associated with both the exposure and the outcome and distorts the apparent relationship. Lead-time bias and length-time bias specifically distort apparent survival benefits in screening studies and deserve separate attention, since they can make a screening test appear to improve survival when it has only shifted the point of diagnosis earlier or preferentially detected slower-growing disease.
Diagnostic Test Performance and Predictive Values
Sensitivity is the proportion of people with disease who test positive, calculated as true positives divided by the sum of true positives and false negatives; a highly sensitive test is useful for ruling out disease when negative, since it misses few true cases. Specificity is the proportion of people without disease who test negative, calculated as true negatives divided by the sum of true negatives and false positives; a highly specific test is useful for ruling in disease when positive, since it rarely flags people who do not have the condition. Positive predictive value is the probability that a person with a positive test actually has the disease, and negative predictive value is the probability that a person with a negative test truly does not; unlike sensitivity and specificity, both predictive values shift with the prevalence of disease in the population being tested, so the same test performs differently in a high-prevalence referral population than in a low-prevalence screening population. Likelihood ratios combine sensitivity and specificity into a single figure that describes how much a given test result shifts the probability of disease from the pre-test level to the post-test level; a positive likelihood ratio well above one increases confidence in disease, while a negative likelihood ratio well below one decreases it. Step 3 vignettes often supply a two-by-two table or raw counts and expect the examinee to calculate one or more of these measures and then use the result to make a clinical decision, such as whether a positive result truly changes management in a patient with a very low pre-test probability.
Measures of Association and Treatment Effect
Relative risk compares the probability of an outcome in an exposed group to the probability in an unexposed group and is appropriately calculated from cohort studies and randomized trials, where the incidence of disease can be directly measured. The odds ratio compares the odds of exposure among cases to the odds of exposure among controls and is the appropriate measure for case-control studies, where true incidence cannot be calculated directly; when a disease is rare, the odds ratio closely approximates the relative risk. Absolute risk reduction is the arithmetic difference in outcome rates between a control group and a treatment group, and the number needed to treat is the reciprocal of that difference, representing how many patients must receive a treatment for one additional patient to benefit. The number needed to harm follows the same logic applied to an adverse outcome, representing how many patients must be exposed for one additional patient to be harmed. These figures matter clinically because a treatment can have an impressive relative risk reduction while still providing a modest absolute benefit if the underlying event rate is low, and Step 3 vignettes frequently test whether an examinee can recognize this distinction when a patient or a colleague conflates a relative benefit with a large absolute benefit.
Survival Analysis, Prognosis, and Population Health
Survival analysis techniques, most notably the Kaplan-Meier method, are used to estimate the probability that patients remain event-free over time when some individuals are followed for different lengths of time or are lost to follow-up before an outcome occurs, a situation called censoring. A Kaplan-Meier curve that drops sharply at a particular time point suggests a cluster of events occurring around that point, and comparing two curves for separate patient groups is a common way trials illustrate a difference in prognosis or treatment effect. Step 3 vignettes may ask an examinee to interpret such a curve to estimate a patient's prognosis or to judge whether an observed difference between two curves appears clinically meaningful. Population health content extends this reasoning to public health surveillance, including how incidence and prevalence are defined and used differently, how outbreaks are investigated, and how screening programs are evaluated at a population level rather than for a single patient. Applying evidence-based medicine to an individual patient requires integrating the best available evidence, such as a trial's reported relative and absolute effects, with that specific patient's values, comorbidities, and pre-test probability of disease, since a therapy proven beneficial on average in a trial population may still be inappropriate for a particular patient whose characteristics differ substantially from the study sample.
Key terms
- Randomized controlled trial
- — A study design in which participants are randomly assigned to intervention or control, minimizing confounding and supporting causal inference.
- Cohort study
- — An observational study that follows groups defined by exposure status forward in time to compare outcome rates.
- Case-control study
- — An observational study that compares prior exposures between people with an outcome and people without it, efficient for rare diseases but prone to recall bias.
- Sensitivity
- — The proportion of people with disease who correctly test positive; a highly sensitive negative test helps rule out disease.
- Specificity
- — The proportion of people without disease who correctly test negative; a highly specific positive test helps rule in disease.
- Predictive value
- — The probability that a test result correctly reflects a person's true disease status; it varies with disease prevalence, unlike sensitivity and specificity.
- Likelihood ratio
- — A figure combining sensitivity and specificity that indicates how much a test result shifts the probability of disease from pre-test to post-test.
- Relative risk
- — The ratio of outcome probability in an exposed group compared to an unexposed group, calculated from cohort studies or trials.
- Odds ratio
- — The ratio of exposure odds between cases and controls, the appropriate association measure for case-control studies.
- Number needed to treat
- — The number of patients who must receive a treatment for one additional patient to experience benefit, the reciprocal of absolute risk reduction.
- Kaplan-Meier analysis
- — A statistical method for estimating event-free survival over time that accounts for participants censored before an outcome occurs.
- Lead-time bias
- — A distortion in which earlier detection by screening appears to lengthen survival without actually delaying death.
Exam tips
- When a vignette names a study design, immediately consider its most characteristic bias, since the question often hinges on that vulnerability.
- If the question gives prevalence and asks about predictive value, remember predictive values change with prevalence while sensitivity and specificity do not.
- Distinguish odds ratio from relative risk by checking the study design first: case-control studies use odds ratios, cohort studies and trials use relative risk.
- Do not equate a large relative risk reduction with a large absolute benefit; check the baseline event rate before judging clinical significance.
- When interpreting a Kaplan-Meier curve, look for where curves separate or drop sharply, and remember censoring means loss to follow-up, not treatment failure.