Learn
High-Yield Biostatistics and Epidemiology for Medical Exams
High-yield biostatistics revision: sensitivity and specificity, predictive values, NNT, relative risk, odds ratio, study designs, bias, p-values and exam traps.
Biostatistics questions are among the most predictable in any licensing exam. They use a small set of formulas and concepts, and the same traps come up repeatedly. With a few hours of focused practice, they become reliable marks. This guide is exam revision.
The 2ร2 table
Almost every diagnostic test question can be solved by drawing this table first.
| Disease present | Disease absent | |
|---|---|---|
| Test positive | True positive (a) | False positive (b) |
| Test negative | False negative (c) | True negative (d) |
| Measure | Formula | Meaning |
|---|---|---|
| Sensitivity | a / (a + c) | Proportion of diseased who test positive |
| Specificity | d / (b + d) | Proportion of non-diseased who test negative |
| Positive predictive value (PPV) | a / (a + b) | Probability of disease if the test is positive |
| Negative predictive value (NPV) | d / (c + d) | Probability of no disease if the test is negative |
| Positive likelihood ratio | Sensitivity / (1 โ specificity) | How much a positive result raises the odds |
| Negative likelihood ratio | (1 โ sensitivity) / specificity | How much a negative result lowers the odds |
| Prevalence | (a + c) / total | Proportion with disease at a point in time |
Memory aids: a highly Sensitive test, when Negative, rules out (SnNout). A highly Specific test, when Positive, rules in (SpPin).
Worked example
1,000 people are tested; 100 have the disease. The test detects 90 of them and is falsely positive in 45 of the 900 without disease.
- Sensitivity = 90/100 = 90%
- Specificity = 855/900 = 95%
- PPV = 90/(90+45) = 67%
- NPV = 855/(855+10) = 98.8%
Prevalence and cut-offs
- Sensitivity and specificity do not change with prevalence; predictive values do.
- Higher prevalence: PPV up, NPV down.
- Lowering a test's cut-off for positivity increases sensitivity and decreases specificity (more true and false positives).
Measures of risk and treatment effect
| Outcome | No outcome | |
|---|---|---|
| Exposed / treated | a | b |
| Unexposed / control | c | d |
| Measure | Formula |
|---|---|
| Risk in exposed (EER) | a / (a + b) |
| Risk in unexposed (CER) | c / (c + d) |
| Relative risk (RR) | EER / CER |
| Absolute risk reduction (ARR) | CER โ EER |
| Relative risk reduction (RRR) | ARR / CER |
| Number needed to treat (NNT) | 1 / ARR |
| Number needed to harm (NNH) | 1 / absolute risk increase |
| Odds ratio (OR) | (a ร d) / (b ร c) |
| Attributable risk | Risk in exposed โ risk in unexposed |
Worked example: stroke occurs in 8% of controls and 6% of treated patients. ARR = 2% (0.02), NNT = 50, RR = 0.75, RRR = 25%.
Trap: a relative risk reduction can sound impressive when the absolute benefit is small. Exams like to ask which number best communicates benefit to a patient (usually absolute risk or NNT).
Incidence and prevalence
- Incidence: new cases over a period in a population at risk.
- Prevalence: all existing cases at a point or period.
- Prevalence โ incidence ร duration (for stable conditions). A new treatment that prolongs life without curing raises prevalence; a cure or rapidly fatal disease lowers it.
Study designs
| Design | Starts with | Measure | Strengths | Weaknesses |
|---|---|---|---|---|
| Randomised controlled trial | Random allocation to intervention | RR, ARR, NNT | Controls known and unknown confounders | Cost, ethics, generalisability |
| Cohort (prospective or retrospective) | Exposure | RR, incidence | Multiple outcomes, temporal sequence | Slow for rare outcomes, loss to follow-up |
| Case-control | Outcome | Odds ratio | Rare diseases, quick | Recall and selection bias |
| Cross-sectional | A point in time | Prevalence | Quick, cheap | No temporal sequence |
| Ecological | Population-level data | Correlation | Hypothesis-generating | Ecological fallacy |
| Systematic review / meta-analysis | Multiple studies | Pooled estimate | Highest level when studies are good | Publication bias, heterogeneity |
Trial terminology
- Intention-to-treat analysis: analyse participants in the groups to which they were randomised; preserves randomisation.
- Per-protocol analysis: only those who completed treatment as allocated; can overstate benefit.
- Blinding: reduces performance and detection bias.
- Allocation concealment: prevents selection bias at enrolment.
- Phases of drug trials: phase I (safety, usually healthy volunteers), phase II (efficacy and dose in patients), phase III (comparison with standard care in large groups), phase IV (post-marketing surveillance).
Bias and confounding
| Bias | Description | Classic setting |
|---|---|---|
| Selection bias | Study group not representative | Hospital-based controls |
| Recall bias | Differential memory of exposure | Case-control studies |
| Lead-time bias | Earlier diagnosis appears to prolong survival without changing outcome | Screening programmes |
| Length-time bias | Screening detects slower-growing disease more often | Cancer screening |
| Observer / detection bias | Knowledge of exposure influences assessment | Unblinded outcome assessment |
| Attrition bias | Differential loss to follow-up | Cohort studies and trials |
| Publication bias | Positive results more likely to be published | Meta-analyses (funnel plot asymmetry) |
| Hawthorne effect | Behaviour changes because people know they are observed | Practice audits |
Confounding: a third variable associated with both exposure and outcome (for example, smoking confounding an association between coffee and lung cancer). Control by randomisation, restriction, matching, stratification or multivariable analysis.
Hypothesis testing
| Term | Meaning |
|---|---|
| Null hypothesis | No difference or association |
| p-value | Probability of results at least this extreme if the null hypothesis were true |
| Type I error (alpha) | Rejecting a true null hypothesis (false positive), usually set at 0.05 |
| Type II error (beta) | Failing to reject a false null hypothesis (false negative) |
| Power | 1 โ beta; probability of detecting a true effect; increases with sample size and effect size |
| 95% confidence interval | Range compatible with the data; if it includes the null value (0 for differences, 1 for ratios), the result is not statistically significant at the 5% level |
Trap: a statistically significant result is not necessarily clinically important, and a non-significant result in a small study does not prove there is no effect.
Choosing a statistical test
| Data | Comparison | Test |
|---|---|---|
| Continuous, normally distributed | Two groups | t-test (paired or unpaired) |
| Continuous, normal | Three or more groups | ANOVA |
| Continuous, not normal | Two groups | Mann-Whitney U (unpaired), Wilcoxon signed-rank (paired) |
| Categorical | Proportions | Chi-squared (Fisher's exact for small numbers) |
| Two continuous variables | Association | Pearson (normal) or Spearman (non-normal) correlation |
| Time to event | Survival | Kaplan-Meier, log-rank test, Cox regression |
Screening
Classic screening criteria (often attributed to Wilson and Jungner, adopted by WHO) include: an important health problem, a recognisable early stage, an acceptable and accurate test, an effective treatment with benefit from early detection, and a cost that is balanced against benefit. Screening tests favour high sensitivity; confirmatory tests favour high specificity.
Distributions
- Normal distribution: mean = median = mode; about 68% within 1 standard deviation, 95% within about 2, 99.7% within 3.
- Positive (right) skew: mean > median > mode.
- Negative (left) skew: mean < median < mode.
- Standard error of the mean = standard deviation / โn; used for confidence intervals.
Common exam traps
- Saying sensitivity changes with prevalence (it does not; predictive values do).
- Calculating relative risk from a case-control study.
- Using relative risk reduction instead of absolute risk reduction to describe benefit.
- Mistaking lead-time bias for improved survival.
- Concluding "no effect" from a small underpowered study.
- Forgetting that a confidence interval for a ratio crossing 1 means non-significance.
How to revise
Draw the 2ร2 table for every question, even simple ones, so the process becomes automatic. Do short daily sets of calculation questions. MediFlow Exam Prep includes biostatistics questions with worked explanations. Pair this guide with active recall and practice questions and medical ethics and law for research ethics.
Summary
- Draw the 2ร2 table and apply the formulas.
- Prevalence changes predictive values, not sensitivity or specificity.
- NNT = 1/ARR; odds ratios belong to case-control studies.
- Know the main study designs, biases and what a p-value and confidence interval mean.
Practice questions in MediFlow Exam Prep
Daily questions with explanations, a study plan and mock exams. Start free.
App StoreFrequently asked questions
What is the difference between sensitivity and specificity?
Sensitivity is the proportion of people with the disease who test positive. Specificity is the proportion of people without the disease who test negative. Both are properties of the test and do not change with prevalence.
How does prevalence affect predictive values?
As prevalence rises, the positive predictive value rises and the negative predictive value falls. The same test performs differently in a high-risk clinic than in population screening.
How do you calculate the number needed to treat?
NNT is 1 divided by the absolute risk reduction. If an outcome occurs in 10% of controls and 6% of treated patients, the absolute risk reduction is 4% (0.04) and the NNT is 25.
When is an odds ratio used instead of relative risk?
In case-control studies, where participants are selected by outcome, so incidence and relative risk cannot be calculated directly. When the outcome is rare, the odds ratio approximates the relative risk.
Do I need to know complex statistical tests?
Mostly no. Exams test choosing the right test for the type of data, interpreting p-values and confidence intervals, and recognising bias and study design.
Educational content only, not individual medical, legal or immigration advice. Fees, formats and rules change: always confirm with the responsible authority before you apply.
MediFlow Editorial Team
Medical education editors
The MediFlow editorial team researches licensing exams, medical education and healthcare careers worldwide. Every figure is checked against the official exam body, regulator or statistics office and dated; articles are re-reviewed when rules change. MediFlow is operated by ICSL FZE, SPC Free Zone, Sharjah, UAE, with a branch office: Exchange Tower, Business Bay, Dubai.
Related guides
Learn
High-Yield Medical Ethics and Law for Licensing Exams
High-yield medical ethics revision: the four principles, consent and capacity, confidentiality, minors, end-of-life decisions, duty of candour and exam answer patterns.
7 min read ยท Last reviewed 30 October 2026
Learn
Active Recall and Practice Questions for Medical Exams
Why testing yourself beats rereading, what the research shows, and how to use practice questions, explanations and error logs for medical licensing exams.
5 min read ยท Last reviewed 30 October 2026
Learn ยท US, CA
How to Study for USMLE Step 1 in 2026: A Practical Plan
A phase-by-phase USMLE Step 1 study plan: resources by type, spaced repetition, question banks, NBME self-assessments and weekly schedule tables.
8 min read ยท Last reviewed 30 October 2026
New guides, exam rule changes, study plans
We are setting up a double opt-in newsletter. Until then, get daily practice questions and exam updates in the free MediFlow app.