About This Page
This is a clinician-written, evidence-based summary aligned to the USMLE Step 2 CK Content Outline. It is intended for medical students preparing for USMLE Step 2 CK. Management reflects current ACC/AHA, USPSTF, and APA guidelines. Always cross-reference with UpToDate, institutional protocols, and clinical judgment.
The Bottom Line
- Confounder = associated with exposure and outcome but not on the causal pathway; control by randomization, restriction, matching, stratification, or multivariable adjustment
- p-value = probability of observing data as extreme or more extreme if the null hypothesis is true; it is not the probability the null is true
- 95% CI excluding the null value is statistically significant at alpha 0.05; for RR/OR null = 1, for mean difference null = 0
- Power = 1 - beta; higher sample size, larger effect size, and higher event rate increase power
- NNT = 1 / ARR; NNH = 1 / absolute risk increase. Absolute risk matters more clinically than relative risk alone
Overview
This topic covers the conceptual language needed to interpret clinical research. Step 2 CK often asks what bias is present, how to reduce confounding, whether a confidence interval is significant, what happens to power when sample size changes, or how to calculate NNT. The safest approach is to identify the study design, define the exposure and outcome, determine whether the problem is random error, bias, or confounding, and then calculate absolute effects before being impressed by relative percentages.
Epidemiology
Medical evidence can be distorted by systematic error (bias), random error (imprecision), confounding, and selective reporting. Large sample sizes reduce random error but do not fix bias. Randomization reduces confounding in trials; careful design and statistical adjustment reduce but do not eliminate confounding in observational studies. Clinically meaningful evidence requires both internal validity and relevance to the patient population.
Concept Recognition
Symptoms
Selection bias: study groups differ because of how participants enter or remain in study
Recall bias: patients with disease remember exposures differently than controls
Lead-time bias: earlier diagnosis appears to improve survival time without changing death time
Length-time bias: screening preferentially detects slower-growing, less aggressive disease
Observer/measurement bias: outcome assessment differs because assessor knows exposure or treatment status
Publication bias: positive studies are more likely to be published
Signs
Confounding: crude association changes substantially after stratification or adjustment
Effect modification: effect genuinely differs across strata; do not adjust it away as error
Type I error alpha: false positive; rejecting a true null
Type II error beta: false negative; failing to reject a false null
Confidence interval width reflects precision; narrow CI = more precise estimate
Calculations and Statistical Interpretation
First-line
Absolute risk reductionARR = control event rate - treatment event rate. This is the basis for NNT
Relative risk reductionRRR = ARR / control event rate. Can sound impressive even when absolute benefit is small
Number needed to treatNNT = 1 / ARR using proportions, not percentages. Round up to the next whole patient
Number needed to harmNNH = 1 / absolute risk increase. Lower NNH means harm is more frequent
Second-line
p-valueProbability of observing results at least as extreme as those observed if the null hypothesis is true. Lower p-value suggests data are less compatible with null
Confidence intervalRange of plausible values for estimate. 95% CI excluding RR/OR 1 or mean difference 0 is statistically significant at p<0.05
PowerProbability of detecting a true effect if it exists = 1 - beta. Increase by increasing sample size, event rate, effect size, or alpha
Specialist
Multivariable regressionAdjusts for measured confounders but cannot correct unmeasured confounding or major bias
Stratified analysisCan reveal confounding or effect modification by comparing stratum-specific effect estimates
Reducing Bias and Interpreting Results
Standard epidemiology and evidence-based medicine principles used in USMLE Step 2 CK1
Control confounding
- Randomization: best method in trials for known and unknown confounders
- Restriction: include only one level of confounder, improving internal validity but reducing generalizability
- Matching: select controls similar to cases for confounders, common in case-control studies
- Stratification: analyze within levels of confounder
- Multivariable adjustment: statistical control for measured confounders
2
Reduce measurement and performance bias
- Blinding participants reduces placebo effects and behavior differences
- Blinding clinicians reduces differential co-interventions
- Blinding outcome assessors reduces observer bias
- Use objective outcomes and standardized protocols when possible
3
Interpret significance and precision
- Statistical significance does not prove clinical importance
- A very large study can detect trivial differences
- A small study may miss important effects because of low power
- Confidence interval gives effect size and precision; p-value alone does not
4
Report clinically useful effects
- Prefer absolute risk reduction and NNT alongside relative risk reduction
- Balance NNT against NNH, severity of outcome, patient preferences, and cost
- For screening, consider lead-time, length-time, and overdiagnosis bias
- Assess whether the study population resembles the patient in the question stem
Complications
- Overstating benefit: Relative risk reduction without absolute risk can exaggerate clinical impact
- False confidence: Small p-value can coexist with bias, confounding, or trivial effect size
- False negative studies: Low power may miss true effects
- Screening bias: Lead-time and length-time bias can make survival appear better without reducing mortality
- Residual confounding: Unmeasured or poorly measured confounders remain after adjustment
USMLE Step 2 CK Exam Tips
- 1NNT = 1 / ARR; always use decimals, then round up
- 2RR or OR confidence interval crossing 1 = not statistically significant
- 3Mean difference confidence interval crossing 0 = not statistically significant
- 4Power = 1 - beta. Increase sample size to increase power
- 5Type I error = false positive; Type II error = false negative
- 6Lead-time bias: survival after diagnosis improves but mortality does not
- 7Length-time bias: screening finds slow-growing disease
- 8Randomization reduces confounding; blinding reduces observer/performance bias
practicetest your knowledge on biostatistics — bias, confounding, and statistical conceptsApply what you've learnt with USMLE Step 2 CK-style questions from the iatroX Q-Bank — preventive medicine and beyond.
open q-bank