Eduovisual
Biostatistics & Population Health
Incidence vs prevalence: calculation and clinical use
— Incidence = NEW cases of disease arising in a previously disease-free population over a defined time period. Measures risk of developing disease.
— Prevalence = ALL existing cases (new + old) in a population at a point in time (point prevalence) or over an interval (period prevalence). Measures burden of disease.
— Stem mentions "newly diagnosed cases this year" → incidence.
— Stem mentions "currently living with," "as of January 1," or "total cases in the community" → prevalence.
— Stem changes a treatment that prolongs survival without curing (e.g., antiretrovirals for HIV, insulin for T1DM) → prevalence rises, incidence unchanged.
— Stem introduces a new vaccine or primary prevention (HPV vaccine, statin campaign) → incidence falls first; prevalence falls only after a lag.
— Public health director allocating resources → uses prevalence (how many beds, clinics, drugs needed).
— Researcher studying causation/risk factors → uses incidence (cohort studies, RCTs).
— Screening program impact → tracked by incidence of advanced disease and disease-specific mortality, not crude prevalence.
— Prevalence ≈ Incidence × Average Duration of disease (P ≈ I × D).
— Useful when the question gives you two of three variables and asks for the third.
Board pearl: A disease that is highly fatal and short-lived (e.g., pancreatic cancer, rabies, Ebola) has high incidence but low prevalence because patients die before accumulating. A chronic, manageable disease (HTN, HIV on ART, T2DM) has low incidence relative to prevalence because patients accumulate over decades.
— "In 2023, 400 residents of a town of 10,000 were diagnosed with type 2 diabetes" → if these are new diagnoses, it's incidence (400/10,000/year). If it's everyone currently carrying the diagnosis, it's prevalence.
— "On January 1, a survey identified..." → point prevalence.
— "Between January and December, the number of people who developed..." → cumulative incidence (a proportion, no units of time in denominator beyond the period).
— "Per 1,000 person-years" → incidence rate (density). Person-time denominator is the giveaway.
— Cumulative incidence (incidence proportion) = new cases / population at risk at start. Range 0–1. Assumes closed cohort, fixed follow-up.
— Incidence rate (density) = new cases / total person-time at risk. Units = cases per person-time. Used when follow-up varies or population is dynamic.
— Including prevalent cases in the denominator of an incidence calculation. Denominator must be disease-free at baseline.
— Counting recurrences as new cases when the question defines "new" as first-ever diagnosis.
— Mixing up survey-derived prevalence (cross-sectional) with cohort-derived incidence.
— Cross-sectional study design, NHANES data, "snapshot," "currently."
— Cohort design, "followed for X years," "developed," person-years.
Key distinction: Cross-sectional studies measure prevalence and associations only, never incidence. Cohort and RCT designs measure incidence. Case-control studies measure neither directly — they yield odds ratios that approximate relative risk only when disease is rare (<10%).
— Numerator: cases (new vs total).
— Denominator: at-risk population (must exclude existing cases for incidence; includes everyone for prevalence).
— Time frame: instant, interval, or person-time.
— Town of 50,000 adults. On Jan 1, 2,000 already have HTN. During the year, 500 new cases diagnosed.
— Population at risk = 50,000 − 2,000 = 48,000.
— Cumulative incidence = 500 / 48,000 = 1.04% per year.
— A common wrong answer divides by 50,000 (forgetting to subtract prevalent cases).
— Same town on Dec 31. Existing 2,000 + new 500 − 100 who died or moved = 2,400.
— Point prevalence = 2,400 / 50,000 = 4.8%.
— Cohort of 1,000 smokers followed for lung cancer. 200 develop it after avg 5 yr, 800 disease-free after 10 yr.
— Person-years ≈ (200 × 2.5) + (800 × 10) = 500 + 8,000 = 8,500 person-years.
— Incidence rate = 200 / 8,500 = 23.5 per 1,000 person-years.
— Does the answer have correct units? Incidence rate has time in denominator; prevalence is unitless proportion.
— Is the denominator the correct at-risk group?
— Did you account for censoring (loss to follow-up, competing death)?
Step 3 management: When a vignette gives you a screening cohort and asks for "risk of developing disease," compute cumulative incidence, not prevalence. When it asks "what proportion of the community currently has...," compute prevalence. Mismatching the metric to the clinical question is the most tested error.
— Cumulative incidence = (new cases during period) / (population at risk at start of period).
— Incidence rate = (new cases) / (total person-time at risk).
— Point prevalence = (existing cases at time t) / (total population at time t).
— Period prevalence = (existing + new cases during interval) / (average or mid-interval population).
— Prevalence ≈ Incidence × Duration, or more precisely P/(1−P) = I × D.
— When prevalence is small (<10%), P ≈ I × D is acceptable.
— Implications:
▪ Better treatment that prolongs life (↑D) without curing → ↑P, I unchanged.
▪ A cure (↓D toward 0) → ↓P, I unchanged.
▪ Primary prevention (↓I) → ↓P after lag of ~D.
▪ Faster mortality (↓D) → ↓P, I unchanged.
— Given P = 0.05 and avg disease duration 10 yr → I ≈ 0.005/yr = 5 per 1,000 per year.
— Given I = 10/100,000/yr and D = 20 yr → P ≈ 200/100,000 = 0.2%.
Board pearl: If a vignette says "5-year survival improved from 20% to 80% after a new drug," and asks what happens to the disease's prevalence vs incidence, the answer is prevalence rises, incidence unchanged. The drug doesn't prevent new cases — it just keeps existing patients alive longer, inflating D in P = I × D.
— Attack rate = cumulative incidence in an outbreak setting, usually short time frame (foodborne illness, norovirus on a cruise). Numerator = ill; denominator = exposed.
— Secondary attack rate = new cases among household/close contacts of an index case ÷ susceptible contacts. Measures transmissibility.
— Case-fatality rate (CFR) = deaths from disease / total cases of disease. Not the same as mortality rate.
— Mortality rate = deaths / total population (a type of incidence of death).
— Lifetime prevalence = proportion who have ever had disease, regardless of current status (used for psych: MDD, schizophrenia).
— Crude rate = raw cases / raw population. Misleading when populations differ in age/sex distribution.
— Age-adjusted (standardized) rate = weighted to a standard population. Allows comparison between populations with different demographics (e.g., Florida retirees vs Utah young families).
— Direct standardization: apply observed age-specific rates to a standard population.
— Indirect standardization: apply standard rates to observed population → Standardized Mortality Ratio (SMR) = observed / expected deaths.
— Comparing cancer rates between counties → age-adjusted.
— Tracking a single population's trend over time → either, but age-adjusted controls for an aging cohort.
Key distinction: CFR ≠ mortality rate. CFR's denominator is sick people; mortality's denominator is the whole population. A disease can have high CFR (Ebola ~50%) but low population mortality if rare, or low CFR (influenza ~0.1%) but high mortality due to massive incidence. Step 3 frequently swaps these in answer choices.
— Etiology / causation / risk factor identification → incidence (cohort, RCT). You need to watch disease arise after exposure.
— Resource planning, health-services demand, staffing, drug budgets → prevalence. You need to know how many people need care right now.
— Screening program evaluation → primary endpoints are incidence of advanced/metastatic disease and disease-specific mortality, not prevalence (which is inflated by lead-time bias and overdiagnosis).
— Public health emergency / outbreak control → incidence (attack rate) to track transmission dynamics.
— Quality of care, chronic disease management programs → prevalence of controlled vs uncontrolled disease (e.g., % of diabetics with A1c <8).
— A new screening test detects indolent, slow-growing cases that would never have surfaced clinically → prevalence rises (overdiagnosis) without true benefit.
— Lead-time bias makes survival appear longer but doesn't change incidence of death.
— Therefore guideline bodies (USPSTF) require RCT evidence of reduced disease-specific mortality or incidence of advanced disease, not improved prevalence-based survival.
— Knowing 50 new HIV cases/year doesn't tell a clinic how many patients need ART monthly — for that, you need prevalence (~5,000 living with HIV).
Step 3 management: When asked "which study design would best determine whether [exposure] causes [disease]," pick the design that measures incidence — prospective cohort or RCT. When asked "what is the burden of [disease] in this community," pick the cross-sectional prevalence survey.
— Primary prevention (vaccines, statins, antihypertensives, smoking cessation, HPV vaccination, PrEP for HIV) → ↓Incidence. Prevalence falls only after average disease duration elapses.
— Curative therapy (antibiotics for strep, H. pylori eradication, hepatitis C DAAs) → ↓Duration dramatically → ↓Prevalence; incidence unchanged unless transmission also drops.
— Disease-modifying / life-prolonging therapy without cure (ART for HIV, insulin for T1DM, dialysis for ESRD, ICDs in HF, modern oncology) → ↑Duration → ↑Prevalence; incidence unchanged.
— Screening with early treatment (colonoscopy with polypectomy, Pap with LEEP) → ↓Incidence of invasive cancer (because precursors removed); prevalence of in-situ disease may transiently rise.
— A new HF drug reduces 1-year mortality from 30% to 15%. Effect on HF prevalence? Increases (longer D). Effect on HF incidence? Unchanged.
— Universal hepatitis B vaccination at birth introduced 30 years ago. Effect today? Both incidence and prevalence are lower, incidence having dropped first.
— Sofosbuvir cures HCV in 95% of treated. Effect on HCV prevalence? Decreases. On incidence? Decreases secondarily via reduced transmission (treatment-as-prevention).
Board pearl: If a question stem highlights a new chronic disease therapy that improves survival, expect the correct answer to read "prevalence increases, incidence unchanged." This is one of the most repeated Step 3 epi distractors.
— Cross-sectional study: Snapshot in time. Measures prevalence and associations (prevalence odds ratio). Cannot establish temporality or incidence. Cheap, fast, good for hypothesis generation.
— Cohort study (prospective): Disease-free at baseline, followed forward. Directly measures incidence, relative risk, incidence rate ratio. Best for common outcomes and multiple outcomes from one exposure. Limited for rare diseases (need huge N).
— Cohort study (retrospective): Uses historical records; still measures incidence within the assembled cohort.
— Case-control study: Starts with diseased and non-diseased, looks backward at exposure. Cannot calculate incidence or prevalence directly. Yields odds ratio, which approximates relative risk only when disease is rare. Best for rare diseases and multiple exposures.
— Randomized controlled trial (RCT): Gold standard for causal incidence comparisons. Controls confounding by randomization.
— Ecological study: Population-level. Compares incidence/prevalence between groups. Vulnerable to ecological fallacy.
— Rare disease, multiple risk factors → case-control.
— Rare exposure, multiple outcomes → cohort.
— Burden estimate → cross-sectional.
— Causal inference / new drug → RCT.
CCS pearl: When an outpatient vignette asks you to design a study to determine whether a workplace chemical causes bladder cancer (rare disease), order a case-control study. If it asks whether a common dietary pattern causes hypertension (common outcome), order a prospective cohort.
— Prevalence of most chronic diseases (HTN, T2DM, CAD, osteoarthritis, dementia, CKD) rises steeply with age because duration is long and survival is being prolonged by modern care.
— Incidence may also rise with age (biology of senescence) but the prevalence-to-incidence ratio widens in old age.
— A nursing home will have a much higher crude dementia prevalence than a college town, but age-specific prevalence at age 75 is comparable. Always check whether the question uses crude or age-adjusted data.
— Standardized Mortality Ratio (SMR): When comparing mortality in an elderly cohort (e.g., dialysis patients) to the general population, the comparator must be age-matched. SMR >1 = excess mortality after adjusting for age.
— Chronic kidney disease prevalence ≈ 15% of US adults; incidence of ESRD ≈ 350 per million per year. The huge prevalence-incidence gap reflects long disease duration on dialysis.
— Cirrhosis incidence has fallen (HCV cures) but prevalence lags — the cured population still exists in registries until reclassified.
— In adults >65, average of 3 chronic conditions. Prevalence of "any chronic disease" approaches 80%, but incidence of any single new condition per year is far lower.
Key distinction: A high prevalence in elderly populations does not imply a high current incidence — it often reflects accumulated survival. Don't conclude that "dementia is on the rise" from prevalence data alone; you need age-specific incidence trends, which have actually been flat or falling in some cohorts despite rising prevalence.
— Maternal mortality ratio = maternal deaths / 100,000 live births (denominator is births, not population).
— Infant mortality rate = deaths <1 year / 1,000 live births.
— Neonatal mortality = deaths <28 days / 1,000 live births.
— Perinatal mortality = stillbirths ≥28 wk + early neonatal deaths / 1,000 total births.
— Fetal mortality (stillbirth) rate = fetal deaths ≥20 wk / 1,000 live births + fetal deaths.
— All function as incidence rates with non-standard denominators.
— Asthma prevalence in US children ≈ 8%; incidence varies by age (peaks early childhood).
— Autism spectrum prevalence has risen markedly (now ~1 in 36); much of this reflects broader diagnostic criteria and ascertainment rather than true incidence rise. This is a classic confounder Step 3 may test.
— Homeless and uninsured populations are under-counted in prevalence surveys → underestimation.
— Increased screening in a group artificially raises measured prevalence (e.g., chlamydia screening expansion in young women).
— Attack rate in schools/daycares is the operational incidence for response planning.
— Vaccine coverage is reported as prevalence of vaccinated but its effect is measured in incidence reduction of disease.
Board pearl: If a stem asks "why has autism prevalence quadrupled since 1990?" the correct Step 3 answer emphasizes broadened diagnostic criteria, improved ascertainment, and earlier screening — not necessarily a true rise in incidence. This generalizes: rising prevalence ≠ rising incidence whenever case-finding changes.
— Length-time bias (screening): Slow-growing, indolent cases are over-represented in screen-detected cohorts → inflates apparent prevalence of "screen-detected" disease and overestimates survival.
— Lead-time bias: Earlier detection moves the diagnosis date earlier without changing time of death → survival looks longer; incidence and mortality unchanged.
— Overdiagnosis: Detection of disease that would never have caused harm (e.g., low-grade prostate cancer, thyroid microcarcinoma) → inflates incidence and prevalence without affecting mortality.
— Survivor bias / Neyman bias: Prevalence studies miss patients who died quickly → underestimates severe disease prevalence. Cross-sectional studies of MI underrepresent fatal MIs.
— Ascertainment / detection bias: Increased testing finds more cases (PSA era for prostate cancer, COVID testing scale-up).
— Ecological fallacy: Inferring individual risk from population-level prevalence (e.g., countries with high fat intake have high CAD doesn't mean high-fat individuals have CAD).
— Berkson bias: Hospital-based prevalence overestimates disease association because sicker patients are over-sampled.
— Thyroid cancer "epidemic" in South Korea after universal ultrasound screening — incidence quadrupled, mortality flat = overdiagnosis.
— Apparent improved 5-year survival of screen-detected lung cancer — partially lead-time bias.
Step 3 management: When a question describes "5-year survival increased from 40% to 80% after a new screening test was adopted" but mortality is unchanged, the correct answer is lead-time bias and/or overdiagnosis — not a genuine treatment benefit. Demand mortality and advanced-disease incidence endpoints before endorsing a screening program.
— Nationally notifiable list includes HIV, TB, syphilis, gonorrhea, chlamydia, hepatitis A/B/C, measles, mumps, pertussis, meningococcus, foodborne illness clusters, COVID-19 (currently), and many others. Reporting is to the local/state health department, who forwards to CDC.
— Reporting generates incidence data in near-real-time.
— Endemic = expected baseline incidence in a community.
— Outbreak / epidemic = incidence above expected for that time/place.
— Pandemic = epidemic across multiple countries/continents.
— Step 3 may ask: 3 cases of measles in a county where 0 expected = outbreak, mandates immediate public health response.
— Suspect a reportable disease → notify health department even before lab confirmation (e.g., suspected measles, meningococcus).
— Cluster of unusual cases → call public health, even if not on the formal notifiable list (cluster of pediatric leukemia, environmental exposures).
— Rising incidence rate of pertussis → trigger vaccination catch-up campaign.
— High prevalence of uncontrolled HTN in a clinic → quality improvement initiative, not outbreak response.
CCS pearl: When you see a vignette with a single case of measles, meningococcal meningitis, pertussis, TB, or suspected foodborne outbreak, the very next CCS action after isolation/treatment is "Notify public health department." Failing to report is both a public-health failure and a tested patient-safety item.
— Cumulative incidence vs incidence rate: Proportion (unitless, bounded 0–1) vs rate (per person-time). Cumulative incidence assumes everyone followed for the same period; incidence rate accommodates variable follow-up.
— Prevalence vs incidence: Existing cases vs new cases. Stock vs flow.
— Risk vs rate: Risk = cumulative incidence. Rate = incidence density per person-time.
— Relative risk (RR) vs odds ratio (OR): RR from cohort/RCT (incidence-based); OR from case-control. OR ≈ RR when disease rare.
— Attributable risk vs relative risk: AR = incidence_exposed − incidence_unexposed (absolute difference). RR = ratio. AR drives NNT and public health impact; RR drives strength-of-association arguments.
— Population attributable risk (PAR): Incidence in population − incidence in unexposed. Reflects what would happen if exposure eliminated population-wide.
— Case-fatality rate vs mortality rate: Denominator is sick people vs whole population (see chunk 5).
— Crude vs age-adjusted rate: Raw vs demographic-standardized.
— Sex ratio, dependency ratio, fertility rate — demographic, not disease incidence, but tested in population health.
— Standardized Mortality Ratio (SMR): observed / expected deaths after age-adjustment.
Key distinction: Relative risk tells you strength of association; attributable risk tells you public health impact. A doubling of a rare disease incidence (RR=2) may have trivial absolute impact, while a 20% increase in a common condition (RR=1.2) may translate to thousands of cases. Step 3 expects you to use AR/NNT when discussing real-world clinical decisions.
— Sensitivity and specificity are properties of a test, not of disease frequency, but positive and negative predictive values depend on prevalence:
▪ PPV = TP / (TP + FP). Rises with prevalence.
▪ NPV = TN / (TN + FN). Falls with prevalence.
▪ This is why screening tests perform poorly when prevalence is low (more false positives than true positives).
— Likelihood ratios are prevalence-independent; combine with pre-test probability (essentially prevalence in the relevant population) to get post-test probability via Bayes/Fagan nomogram.
— Life expectancy is derived from age-specific mortality (incidence of death) rates in a life table.
— DALYs (disability-adjusted life years) combine incidence of disease, duration, and severity — a hybrid metric for global burden.
— Years of potential life lost (YPLL) weights early deaths more heavily; an incidence-of-mortality variant.
— Confusing screening test performance (sensitivity/specificity) with disease burden (prevalence). Both matter for PPV.
— Confusing incidence reduction (true prevention) with mortality reduction (could be treatment improvement).
— Confusing prevalence with risk — prevalence is a snapshot, not a probability of developing disease.
— A test with 99% sensitivity and specificity, applied to a disease with 0.1% prevalence, yields PPV ≈ 9% (huge false-positive problem). Step 3 loves this trap.
Board pearl: PPV depends on prevalence; sensitivity does not. A common Step 3 stem describes a "highly accurate" screening test applied to a low-prevalence population — the correct answer addresses low PPV / high false-positive rate, the principal reason USPSTF restricts screening to higher-risk groups.
— Cancer screening programs: Track incidence of advanced-stage/metastatic disease and disease-specific mortality. Falling rates = real success. Rising overall prevalence may simply reflect detection of indolent disease.
— Vaccination programs: Track incidence of vaccine-preventable disease. Measles incidence in US fell from ~500,000/yr pre-vaccine to ~50/yr post-vaccine.
— HIV prevention (PrEP, condoms, U=U): Track HIV incidence, not prevalence (which will rise as treated patients survive longer).
— Smoking cessation campaigns: Track incidence of lung cancer, MI, COPD exacerbation in cohorts.
— Diabetes prevention programs: Track incidence of T2DM in pre-diabetic cohorts.
— Communicate absolute risk reduction (ARR) and NNT to patients, not just relative risk. Patients understand "1 in 50 will avoid a heart attack" better than "30% lower risk."
— For chronic disease management programs, prevalence of controlled disease (e.g., % HTN patients at goal BP) is a valid quality metric (HEDIS, MIPS).
— Define a denominator (panel of patients).
— Choose incidence (new events) for prevention metrics; prevalence (controlled vs uncontrolled) for management metrics.
Step 3 management: When evaluating whether a mammography program is working in a community, the correct outcome is reduction in breast cancer mortality and incidence of advanced (stage III–IV) disease, not improvement in 5-year survival (corrupted by lead-time bias) and not raw prevalence of breast cancer (inflated by overdiagnosis of DCIS).
— Time-series plots of incidence reveal outbreaks, seasonality (flu winter peak), and intervention effects.
— Joinpoint regression identifies inflection points in cancer incidence trends (e.g., the post-PSA prostate cancer incidence drop).
— Age-period-cohort analysis separates effects of aging, calendar time, and birth cohort on incidence.
— Panel-level prevalence of chronic conditions: HTN, T2DM, CKD, HF — drives staffing, formulary, care management.
— Incidence of complications: new strokes in your HTN panel, new ESRD in your CKD panel, new amputations in diabetics — drives quality improvement.
— Readmission rate = incidence of readmission within 30 days; CMS-penalized metric.
— Healthcare-associated infection (HAI) rates: incidence per 1,000 device-days (CLABSI, CAUTI, VAP) — patient safety indicators.
— Patients in cardiac rehab: monitor incidence of recurrent MI, readmission, not just prevalence of risk factors.
— Smoking cessation: track 6- and 12-month abstinence rates (an incidence-style metric of relapse).
— Notifiable diseases — immediate to weekly.
— Quality metrics — quarterly, annually.
— National surveillance (NHANES, BRFSS) — annual or biennial cycles.
CCS pearl: When a clinic's diabetes care metric is "% of patients with A1c <8," that is a prevalence-of-control measure. When the metric is "new amputations per 1,000 patient-years," that is an incidence-of-complication measure. Both are valid, but they answer different quality questions, and a high-performing clinic monitors both.
— Mandatory reporting of notifiable diseases (TB, HIV in most states, syphilis, gonorrhea, measles, COVID-19) overrides patient confidentiality under public health law. Clinicians must report even without patient consent.
— Partner notification for STIs: most states permit or require provider-initiated or health-department-led contact tracing. Patient anonymity preserved but partners are warned.
— HIV-specific laws vary: some states criminalize nondisclosure of HIV status to sexual partners; clinicians should know local statutes.
— Cohort studies and RCTs require IRB approval and informed consent.
— De-identified retrospective database studies may qualify for waiver of consent under HIPAA and Common Rule, but identifiable data needs consent.
— Cluster-randomized trials raise consent challenges — communities, not individuals, are randomized.
— Offering screening with high overdiagnosis rates (e.g., PSA in elderly men with limited life expectancy) can cause net harm. Shared decision-making is the Step 3 standard for PSA, mammography in 40–49, lung CT.
— Disclose absolute benefits (ARR, NNT) and harms (false positives, biopsies, overdiagnosis) — not just relative risk reduction.
— Patients with newly diagnosed reportable disease may fall through cracks between inpatient and outpatient care. Confirm reporting was completed, and ensure partner notification and contact tracing were initiated before discharge.
— Crude prevalence and incidence data can mask disparities; stratified reporting by race/ethnicity, SES, geography is ethically required to avoid masking inequity.
Board pearl: A patient with newly diagnosed pulmonary TB refuses public health follow-up. The correct answer is report to the health department regardless of consent; mandatory reporting and even directly observed therapy with detention authority override individual confidentiality when public health is at stake.
— P ≈ I × D in steady state for rare disease.
— Rising prevalence + stable incidence = treatment improving (longer survival) or improved ascertainment.
— Rising incidence + stable prevalence = disease becoming more fatal or shorter duration.
— Falling prevalence + stable incidence = cure or increased mortality.
— HIV in US: incidence ~30,000/yr (falling), prevalence ~1.2 million (rising due to ART survival).
— T2DM in US: prevalence ~11% adults; incidence ~7 per 1,000 adult-years.
— Hypertension: prevalence ~47% adults under newer thresholds.
— Alzheimer disease: prevalence ~10% of those ≥65; doubles every 5 years of age after 65.
— Breast cancer: lifetime risk ~12% (1 in 8); annual incidence ~130/100,000.
— Lung cancer: incidence ~50/100,000/yr; CFR ~80% within 5 years.
— Influenza attack rate: 5–20% per season in US.
— Measles R₀: 12–18 (highest of common pathogens); requires ~95% vaccination coverage for herd immunity.
— Per 100,000 → per 1,000 by dividing by 100.
— Per person-year × population × years = expected cases.
— Doubling of incidence rate doubles expected cases; doubling of population also doubles expected cases.
— Confusing "5-year survival" (often lead-time biased) with mortality.
— Calling case-control studies a measure of incidence.
— Reporting OR as RR when disease is common.
Key distinction: R₀ (basic reproduction number) is a transmission parameter, not an incidence/prevalence directly, but drives outbreak incidence dynamics. Herd immunity threshold ≈ 1 − 1/R₀. Measles (R₀ ~15) needs ~93% immunity; polio (R₀ ~6) needs ~83%.
— "A new drug reduces mortality from disease X by 50%. Other factors unchanged. What happens to prevalence and incidence?"
— Answer: Prevalence increases (longer duration), incidence unchanged.
— "A test with 99% sensitivity and 99% specificity is applied to a population with 0.1% disease prevalence. What is the PPV?"
— Compute: TP=99, FP=999, PPV≈9%. Lesson: low prevalence destroys PPV.
— Gives population, prevalent cases, and new cases. Asks for incidence.
— Trick: subtract prevalent cases from denominator.
— Rare disease + multiple exposures → case-control.
— Causal claim → RCT or cohort (incidence-based).
— Community burden → cross-sectional.
— "5-year survival improved but mortality unchanged after new screening test."
— Answer: lead-time bias and/or overdiagnosis.
— Improved ascertainment or broader diagnostic criteria.
— Case-control study reports OR. If disease prevalence is 30%, OR overestimates RR.
— Newly diagnosed TB/measles/syphilis/HIV → report to health department is the next step.
— Foodborne illness at a wedding: attack rate per food item, identify highest RR → implicated food.
— Given control and treatment incidences, compute ARR = difference, NNT = 1/ARR.
Step 3 management: When a stem gives you raw counts, always orient: numerator, denominator, time frame. Then identify whether the question wants risk (incidence), burden (prevalence), or association (RR/OR). This 3-step framework solves the vast majority of epidemiology vignettes on Step 3.
Incidence measures new cases (risk/flow) and is the right metric for causation, prevention, and outbreak control; prevalence measures all existing cases (burden/stock) and is the right metric for resource planning — and they're linked by P ≈ I × D, which explains why life-prolonging therapies raise prevalence without changing incidence.
— Cumulative incidence = new cases / at-risk population (start). Subtract prevalent cases from denominator.
— Incidence rate = new cases / person-time.
— Point prevalence = existing cases / total population at time t.
— P ≈ I × D in steady state for rare disease.
— Causation/prevention → incidence (RCT, cohort).
— Resource planning, chronic-care burden → prevalence (cross-sectional).
— Screening program success → incidence of advanced disease + mortality, not survival or crude prevalence.
— Outbreak response → attack rate (incidence in exposed).
— Lead-time bias and overdiagnosis inflate apparent screening benefit.
— Length-time bias overrepresents indolent cases.
— Berkson, Neyman/survivor, ecological fallacy distort observational estimates.
— PPV depends on prevalence; sensitivity/specificity/LRs do not.
— Notifiable disease → report to public health regardless of consent.
— Rising prevalence with stable incidence → better survival or better detection, not necessarily more disease.
— Low prevalence + good test → expect poor PPV; counsel patients on false-positive risk before screening.
Board pearl: If you remember only one equation from biostatistics for Step 3, make it Prevalence ≈ Incidence × Duration — it explains nearly every "what happens to incidence/prevalence when..." question the exam will throw at you, from ART in HIV to thyroid cancer overdiagnosis to hepatitis C cures.