Biostatistics and Epidemiology on Step 1: Formulas, Designs and Bias

Scope and time budget

Fourteen questions is less than one full block, and the 12 subtopics fit in roughly three to four focused sessions. Biostatistics works best when you do it early and often: a 20-minute session twice a week for the first weeks, then a few items in every mixed block. Some biostatistics also hides in other rows (a screening question in the cancer content, a study-design question in a pharmacology vignette). The 4-6% official range belongs to the stand-alone row, and the same skills support the practice-based learning competency, which the outline gives 4-6% of the exam.

What is tested

• Test characteristics. Sensitivity, specificity, predictive values and likelihood ratios.
• Study design. Cohort, case-control, cross-sectional and randomized trials.
• Bias and confounding. Selection and information bias, confounding and effect modification.
• Risk measures. Relative risk, odds ratio, absolute risk reduction and number needed to treat.
• Inference. Hypothesis testing, p-values, confidence intervals and power, and choosing among t-tests, ANOVA, chi-square and regression.
• Descriptive statistics. Distributions, central tendency and standard deviation.
• Disease frequency. Incidence, prevalence and mortality measures.
• Screening and prevention. Screening principles, lead-time and length-time bias, levels of prevention and public health surveillance.
• Research and quality. Clinical trial phases, research ethics, quality improvement and patient safety methods.

Anchor facts and formulas

Learn these as a one-page sheet and rewrite it from memory until it is automatic.

• Study designs and their measures. Cohort studies give relative risk, case-control studies give odds ratios, cross-sectional studies give prevalence, and randomized trials give the strongest evidence of causation.
• Errors. Type I error (alpha) is a false positive, usually set at 0.05. Type II error (beta) is a false negative. A 95% confidence interval for a ratio that includes 1 is not statistically significant.
• Screening biases. Lead-time bias makes survival look longer after earlier detection even if death comes at the same time. Length-time bias over-represents slow-growing disease because it is detected more often.
• US context. Screening and prevention recommendations in the United States come from bodies such as the USPSTF and the CDC. The exam tests the logic, not the age cutoffs.

Three typical traps

1. Reading the 2 x 2 table the wrong way round. The most common error is flipping rows and columns, so sensitivity becomes predictive value. Always label the table with disease status across the top and test result down the side, then write the formula next to the labels before inserting numbers.
2. Using the wrong measure for the design. A case-control study cannot give incidence or relative risk directly, so the answer is an odds ratio. A cross-sectional study gives prevalence. Name the design first, then the measure.
3. Confusing statistical and clinical significance. A very small p-value in a large study can be a tiny effect, and a wide confidence interval crossing 1 in a small study can hide a large effect. Questions ask which statement is supported by the data. Read the interval, not just the p-value.

How to revise this system

• Memorize the 2 x 2 table. Draw it and derive every formula from it. A table is faster than recalling five separate formulas.
• Use a decision tree for tests. Question type: comparing means (t-test for two groups, ANOVA for more), comparing proportions (chi-square), relating two continuous variables (correlation or regression). Draw it once and review it weekly.
• Do calculations by estimation. A calculator is available, but most answers differ widely enough that rounding is enough. Practice rounding to two significant figures.
• Tie each bias to a one-line example. Selection bias, recall bias, observer bias, lead-time bias and confounding. An example you can picture is easier to apply than a definition.
• Revisit weekly. Do 5 to 10 questions in each mixed block. The study plan guide starts the row in week 1.

Worked illustration

A quick illustration of why the table matters. Suppose a screening test is applied to 1,000 people, and 100 have the disease. The test is positive in 90 of them and negative in 10 (so sensitivity is 90/100 = 90%). Among the 900 without the disease, it is negative in 810 and positive in 90 (so specificity is 810/900 = 90%). The positive predictive value is 90/(90 + 90) = 50%. If the same test is used in a population where prevalence is much lower, the sensitivity and specificity do not change, but the positive predictive value falls. This is why a stem that changes the population changes the answer. The numbers here are illustrative, not from any exam.

What the questions look like

Items typically give a table, a short trial or study description, or a graph, and then ask for a calculation, the best description of a bias, the study design, the most appropriate test or an interpretation of a confidence interval. Our original practice bank follows the same format with five options, and it does not copy official or recalled items. See the ethics and communication page for the other small rule-based row, the Step 1 hub for all systems and weights, and the high-yield topics guide for the formula list alongside other cross-system facts.