Biostatistics & Population Health

Population attributable risk and fraction

Clinical Overview and When to Suspect High Population Attributable Risk

— Formula: PAR = Incidence(total population) − Incidence(unexposed)

— Units: cases per person-time (e.g., cases per 100,000 person-years)

— Answers: "How many cases per year in the whole population are due to this exposure?"

— PAF = PAR / Incidence(total population)

— Alternative formula using prevalence of exposure (Pe) and relative risk (RR): PAF = Pe(RR−1) / [1 + Pe(RR−1)]

— Answers: "What fraction of disease in the population would disappear if we eliminated the exposure?"

— Public health prioritization questions ("Which intervention would prevent the most cases?")

— Tobacco, hypertension, obesity, alcohol, vaccine-preventable disease vignettes

— Quality improvement / value-based care stems asking which risk factor to target first in a panel

— Comparing individual-level risk (relative risk, attributable risk) with population-level impact

Population attributable risk (PAR) quantifies the absolute excess disease burden in a population that can be attributed to a specific exposure, assuming causality

Population attributable fraction (PAF), also called population attributable risk percent (PAR%), expresses the same idea as a proportion of total disease in the population attributable to the exposure

When Step 3 invokes PAR/PAF concepts:

Key distinction: A risk factor with a modest RR but very high prevalence (e.g., hypertension causing stroke) often has a larger PAF than a rare exposure with a huge RR (e.g., a rare occupational toxin). This is the single most-tested concept on this topic.

Board pearl: PAR/PAF assume the exposure–disease association is causal, that confounding has been controlled, and that removing the exposure would actually remove the attributable cases. If the question hints at residual confounding or reverse causation, the PAF estimate is biased upward.

Clinical translation: Use PAF to counsel policymakers and ACOs; use number needed to treat (NNT) or absolute risk reduction to counsel individual patients in clinic.

Presentation Patterns and Key History — How PAR/PAF Show Up in Stems

— A public health officer or medical director asks which modifiable risk factor to target in a defined population to maximally reduce disease incidence

— A table provides prevalence of exposure and relative risk for several risk factors; you must compute or rank PAFs

— A cohort study reports incidence in exposed and unexposed groups; you are asked the proportion of cases in the population attributable to the exposure

— A vignette contrasts an individual patient's smoking cessation benefit (use ARR/NNT) versus the community's benefit (use PAF)

— Incidence in exposed (Ie) and unexposed (Iu) → enables attributable risk (AR = Ie − Iu) and attributable risk percent in the exposed (ARP = (Ie−Iu)/Ie)

— Incidence in the total population (It) and unexposed (Iu) → enables PAR = It − Iu

— Prevalence of exposure (Pe) plus RR → enables PAF via Levin's formula

— Attributable risk (AR) = Ie − Iu → excess risk in exposed individuals

— Attributable risk percent (ARP) = (Ie − Iu)/Ie → fraction of disease in the exposed due to exposure

— Population attributable risk (PAR) = It − Iu → excess risk in the whole population

— Population attributable fraction (PAF) = PAR/It → fraction of disease in the whole population due to exposure

Classic stem structures you will see on Step 3:

Information the stem will give you:

Key distinction — four "attributable" terms students confuse:

Board pearl: If the stem says "among smokers, what percent of lung cancer is due to smoking?" → that is ARP, not PAF. If it says "what percent of lung cancer in the United States is due to smoking?" → that is PAF (~80–90%).

Step 3 management: When a population health question lists multiple risk factors with their prevalences and RRs, the highest PAF wins the intervention priority — not the highest RR alone. This mirrors real ACO and CDC prioritization logic.

Physical Exam Findings — Conceptual "Exam" of a PAR/PAF Calculation

• Since PAR/PAF is an epidemiologic construct, the "physical exam" equivalent is inspecting the 2×2 table and the population structure before computing anything.
• Step-by-step inspection of a cohort 2×2 table:
Disease +	Disease −	Total
Exposed	a	b	a+b
Unexposed	c	d	c+d
— Incidence in exposed: Ie = a/(a+b)
— Incidence in unexposed: Iu = c/(c+d)
— Total incidence: It = (a+c)/(a+b+c+d)
— Relative risk: RR = Ie/Iu
— Prevalence of exposure: Pe = (a+b)/(a+b+c+d)
• Hemodynamic analogy: Just as you check blood pressure, HR, and perfusion before treating shock, in epidemiology you check:
— Is the design a cohort (RR available) or case-control (only OR available, use OR as RR approximation only if disease is rare, <10%)?
— Is the exposure prevalent enough to drive a meaningful PAF?
— Is the RR statistically significant (CI excludes 1)?
• Key distinction: In case-control studies, you cannot calculate true incidence, so PAR in absolute terms is not directly obtainable; PAF can be estimated using OR ≈ RR (rare disease assumption) plus prevalence of exposure among controls as a proxy for population exposure prevalence.
• Board pearl: A high RR with Pe near zero yields a tiny PAF. Example: a workplace carcinogen with RR = 20 but Pe = 0.001 produces PAF ≈ 1.9%, whereas hypertension with RR = 2 and Pe = 0.45 produces PAF ≈ 31%. Always multiply prevalence by (RR−1) mentally before ranking.
• Step 3 management: Before answering, restate the question — "Are they asking about risk in the exposed or risk in the whole population?" This single check prevents the most common wrong-answer trap (choosing ARP when PAF is asked, or vice versa).

Diagnostic Workup — Core Formulas and First-Pass Calculations

— Attributable risk (AR/risk difference): AR = Ie − Iu

— Attributable risk percent (ARP): ARP = (Ie − Iu)/Ie = (RR − 1)/RR

— Population attributable risk (PAR): PAR = It − Iu

— Population attributable fraction (PAF), direct: PAF = (It − Iu)/It

— PAF, Levin's formula: PAF = Pe(RR − 1) / [1 + Pe(RR − 1)]

— PAF, alternative: PAF = Pd × (RR − 1)/RR, where Pd = proportion of cases exposed

— Smoking and lung cancer: Ie = 180/100,000, Iu = 10/100,000, It (population with 25% smokers) = 52.5/100,000

— PAR = 52.5 − 10 = 42.5 per 100,000/year

— PAF = 42.5/52.5 = 81% of lung cancer cases attributable to smoking

— Hypertension and stroke: Pe = 0.40, RR = 2.5

— PAF = 0.40(1.5) / [1 + 0.40(1.5)] = 0.60/1.60 = 37.5%

— Of 1000 MI cases, 700 are smokers; RR for MI from smoking = 3

— PAF = (700/1000) × (3−1)/3 = 0.70 × 0.667 = 46.7%

Formula toolkit (memorize cold):

Worked example 1 — direct method:

Worked example 2 — Levin's formula:

Worked example 3 — proportion of cases exposed:

Key distinction: The three PAF formulas give the same answer when assumptions hold (no confounding, exposure measured correctly). Pick whichever matches the data the stem provides.

Board pearl: ARP applies only to the exposed subgroup; PAF applies to the entire population. ARP is always ≥ PAF because the exposed subgroup is enriched for the exposure. The two converge as Pe → 1.

Step 3 management: If the stem gives you RR and exposure prevalence, default to Levin's formula. If it gives you incidences, default to PAR = It − Iu and PAF = PAR/It. Don't waste time hunting for a formula that fits unavailable data.

Diagnostic Workup — Advanced Calculations and Adjustment for Confounding

— Because a single case can be attributable to several causes (multicausality), the sum of PAFs across risk factors often exceeds 1.0

— Example: smoking PAF for cardiovascular death = 20%, hypertension = 35%, dyslipidemia = 30%, diabetes = 15% → sum = 100%, but joint elimination would not eliminate 100% of deaths

— Use combined (joint) PAF formula: PAF_joint = 1 − Π(1 − PAFi) when exposures are independent

— Use adjusted RR (from multivariable regression — Cox, logistic, Poisson) in Levin's formula instead of crude RR

— Failing to adjust → overestimates PAF if confounders inflate the crude RR

— Miettinen's formula: PAF = Pd(RR_adj − 1)/RR_adj uses proportion of cases exposed and the adjusted RR, and is preferred when confounding is present

— Reported as point estimate with 95% CI (e.g., PAF = 35%, 95% CI 28–42%)

— Wide CI → imprecise estimate, often due to small exposed group or borderline RR

— If the CI for the underlying RR crosses 1, the PAF CI will cross 0 → exposure may not contribute meaningfully

— PAF assumes complete elimination of a harmful exposure

— Preventable fraction (PF) = Pe(1 − RR) / [1 − Pe(1 − RR)] is used for protective exposures (RR < 1), e.g., vaccination, statins — estimates fraction of disease that would be prevented if everyone received the protective factor

Multiple exposures — summed PAFs can exceed 100%:

Adjusted PAF (for confounders):

Confidence intervals for PAF:

Key distinction — PAF vs. preventable fraction (PF):

Board pearl: A protective intervention's impact is summarized by the preventable fraction, not PAF. Watch for vaccine, screening, or statin stems asking "what fraction of cases could be prevented if uptake reached 100%?" — that's PF.

CCS pearl: In CCS-style population health vignettes, ordering a registry-based risk factor audit before launching a screening program lets you estimate local PAFs and prioritize high-impact targets.

Risk Stratification — Choosing Which Risk Factor to Target

— Step 1: List candidate modifiable exposures with their local prevalence and RR for the outcome

— Step 2: Compute PAF for each (Levin's formula)

— Step 3: Multiply PAF × baseline disease incidence × population size → expected absolute cases preventable

— Step 4: Weight by intervention efficacy (rarely 100%), feasibility, and cost

— Step 5: Choose the intervention with the largest preventable case burden per dollar

— Hypertension: Pe ~45%, RR for stroke ~2.5 → PAF ~40%

— Atrial fibrillation: Pe ~2%, RR for stroke ~5 → PAF ~7%

— Even though AF confers a stronger individual risk, hypertension control prevents more strokes population-wide

— A preventive measure that benefits the population substantially often offers little benefit to each participating individual (e.g., universal sodium reduction)

— Conversely, high-risk strategies (treat only people with BP >160) miss the large reservoir of moderate-risk individuals who collectively generate most events

— Population strategy maximizes PAF reduction; high-risk strategy maximizes individual ARR

Population health prioritization algorithm:

Why prevalence dominates rare-but-strong risks at the population level:

Rose's prevention paradox:

Key distinction: Clinical decisions are guided by individual absolute risk reduction and NNT. Policy and ACO decisions are guided by PAF and total preventable cases. Step 3 stems frequently force this distinction.

Board pearl: When asked "which intervention has the greatest public health impact?" — choose the option with highest PAF, even if its RR is modest. When asked "which is most beneficial for this patient?" — choose by ARR/NNT.

Step 3 management: As an ACO medical director, allocate the next quality dollar to the risk factor with the largest PAF × population × intervention efficacy, not the rarest exotic exposure with a high RR.

Pharmacotherapy — Translating PAF into Intervention Choice

— Dyslipidemia: PAF ~49% → statin therapy is the single highest-yield pharmacologic lever

— Smoking: PAF ~36% → varenicline, bupropion, nicotine replacement

— Hypertension: PAF ~18% → ACEi/ARB, thiazide, CCB per JNC 8/ACC/AHA

— Diabetes: PAF ~10% → metformin, GLP-1 RA, SGLT2 inhibitor

— Abdominal obesity: PAF ~20% → GLP-1 RA, lifestyle, bariatric referral

— Combined modifiable PAF: ~90% of first MI

— Hypertension: ~48% → antihypertensives are the #1 population-level stroke preventive

— Physical inactivity: ~36%

— Dyslipidemia: ~27%

— Diet, smoking, cardiac causes, alcohol, stress, diabetes round out >90%

PAR/PAF do not prescribe drugs directly, but they determine which drug-treatable risk factor a system should prioritize in formulary, panel management, and quality metrics.

Cardiovascular example — global PAFs (INTERHEART-type estimates) for MI:

Stroke (INTERSTROKE) PAFs:

Lung cancer: Smoking PAF ~85% → tobacco cessation pharmacotherapy is the single most impactful intervention; low-dose CT screening (USPSTF: age 50–80, ≥20 pack-years, quit ≤15 years) further reduces mortality

Key distinction: A drug with a large RRR in trials (e.g., PCSK9 inhibitors, RRR ~15%) may have low PAF impact if eligibility is narrow (Pe small); generic statins have lower per-patient RRR but enormous PAF reduction because eligible Pe is huge.

Board pearl: When a Step 3 stem asks "which medication class would most reduce community cardiovascular mortality?" — the answer is almost always statins (highest PAF × treatable prevalence), not the most potent niche agent.

Step 3 management: For panel-level quality improvement, prioritize statin prescription rates, BP control to <130/80, and tobacco cessation referrals — these three target the three highest-PAF cardiovascular exposures.

Procedures and System-Level Interventions — Maximizing PAF Impact

— Tobacco taxation and smoke-free laws: cut smoking prevalence by 10–20% → large drop in lung cancer, COPD, CVD PAFs

— Universal childhood vaccination: drives PAF for measles, polio, HPV-related cancers, Hib meningitis toward zero

— Folic acid fortification: reduced neural tube defect PAF ~30–50%

— Trans-fat bans: measurable decline in CHD events

— Lead abatement: dramatic reduction in pediatric neurodevelopmental disease PAF

— Seatbelt and airbag mandates: large PAF reduction in motor-vehicle mortality

— Screening reduces disease burden when the screenable risk factor or precursor has high prevalence and effective intervention exists

— Examples: colonoscopy (CRC PAF reducible by ~50–60% with population screening), mammography, cervical cytology/HPV co-testing, AAA ultrasound in male smokers 65–75

— HPV vaccine: PF for cervical cancer approaches 90% with high uptake

— Pneumococcal vaccine: substantial reduction in invasive pneumococcal disease PAF in elderly

— Influenza vaccine: modest PF (~40–60%) but huge absolute case prevention given enormous Pe of exposure

— A CABG benefits the individual via large ARR for the multivessel-disease patient

— A statewide hypertension control program benefits the population via large PAF reduction across millions of moderate-risk individuals

Population-level interventions ranked by historical PAF reduction:

Screening programs and PAF logic:

Vaccination PAF (preventable fraction framing):

Key distinction — individual procedure vs. population program:

Board pearl: The "bang for the buck" question on Step 3 — which intervention prevents the most deaths nationally? — is almost always answered by the option that changes a high-prevalence behavior or exposure, not the option that offers the largest per-person RRR.

CCS pearl: When designing a public health order set, prioritize tobacco quitline referral, BP medication intensification, statin initiation, vaccination, and cancer screening — each maps to a top-ten PAF target.

Special Populations — Elderly and Renal/Hepatic Impairment

— As populations age, prevalence of exposures (HTN, DM, AF, polypharmacy) rises → PAF for downstream outcomes like stroke, dementia, falls increases

— Competing mortality from other causes can paradoxically reduce PAF for any single exposure (people die of something else before the studied outcome occurs) — this is competing risk bias

— Hypertension for stroke and heart failure (Pe >60% over age 65)

— Atrial fibrillation for stroke (Pe ~10% over age 80, RR ~5 → PAF ~30% of strokes in the very old)

— Polypharmacy for falls, delirium, hospitalization

— Frailty and sarcopenia for postoperative complications

— Untreated hearing loss for incident dementia (Lancet Commission PAF ~8%)

— CKD inflates RR for cardiovascular events and bleeding; Pe of CKD (~15% US adults) gives meaningful PAF for cardiovascular mortality

— Hepatic impairment increases medication-related adverse event PAF — relevant for QI programs that audit hepatotoxic drug prescriptions in cirrhosis panels

— Less education (early life), hearing loss, hypertension, obesity, smoking, depression, physical inactivity, social isolation, diabetes, excessive alcohol, traumatic brain injury, air pollution, untreated visual impairment

— Step 3 may ask: "Which modifiable factor has the largest PAF for dementia?" → hearing loss (most recent estimates) or hypertension depending on midlife window

Why PAF estimates shift with age and comorbidity:

Elderly-specific high-PAF exposures:

Renal/hepatic impairment context:

Lancet Commission modifiable dementia PAFs (combined ~40%):

Key distinction: In the elderly, absolute risk and PAR rise (incidence is higher), even when PAF stays similar — because PAF is a ratio. Don't conflate "more cases prevented" (PAR) with "larger fraction prevented" (PAF).

Board pearl: Treating hypertension to <130/80 in adults ≥65 (SPRINT extrapolation, with care for orthostasis) yields one of the largest PARs of any single clinical intervention because of high Pe and high baseline event rate.

Step 3 management: In a geriatric panel, target hearing-aid access, BP control, fall prevention, and deprescribing — each a high-PAF lever.

Special Populations — Pregnancy, Pediatrics, and Demographic Subgroups

— Hypertensive disorders of pregnancy — large PAF for maternal mortality and preterm birth; Pe ~10% of pregnancies

— Smoking during pregnancy — high PAF for low birth weight and SIDS

— Inadequate prenatal care — high PAF for preventable maternal-fetal complications, particularly in under-resourced populations

— Untreated maternal syphilis — near-100% PAF for congenital syphilis cases

— Folate deficiency — PAF for NTDs cut dramatically by fortification

— Unsafe sleep environments — large PAF for SIDS; "ABC" counseling (Alone, Back, Crib) is the highest-yield intervention

— Unvaccinated status — drives PAF for measles, pertussis outbreaks in communities with declining coverage

— Secondhand smoke exposure — substantial PAF for otitis media, asthma exacerbations

— Lead exposure — PAF for cognitive deficits, behavioral disorders

— Childhood obesity — rising PAF for adult cardiometabolic disease

— PAFs for many outcomes (preterm birth, maternal mortality, hypertensive complications) are disproportionately driven by structural racism, food insecurity, housing instability — these social exposures often have higher PAF than biologic risk factors

— Step 3 increasingly tests recognition of social determinants of health as modifiable, measurable exposures

PAF stratifies by subgroup because both prevalence and RR vary by demographics. A single national PAF may mislead local decision-making.

Pregnancy / maternal health:

Pediatrics:

Racial/ethnic and social determinants:

Key distinction: A national PAF mixes subgroups; subgroup-specific PAFs (e.g., by race, income, geography) reveal where targeted intervention has the greatest yield. One-size-fits-all programs may miss high-PAF subpopulations.

Board pearl: For maternal mortality in the US, the leading modifiable PAFs include hypertensive disorders, hemorrhage, cardiomyopathy, and mental health/overdose — and racial disparities amplify each.

Step 3 management: When designing a community intervention, compute PAFs within the target demographic, not from national averages.

Complications and Adverse Outcomes — Misuse of PAR/PAF

— Mistaking association for causation: PAF assumes the exposure–outcome relationship is causal. If confounding, reverse causation, or selection bias inflates the RR, PAF is overestimated

— Ignoring the rare-disease assumption when using OR from case-control studies as RR — overestimates PAF for common outcomes

— Summing PAFs across multiple exposures and concluding total preventable burden — sums often exceed 100% because of multicausality; use joint PAF formula

— Assuming 100% intervention efficacy: real interventions reduce exposure imperfectly; the achievable preventable fraction is PAF × intervention coverage × intervention efficacy

— Generalizing PAF across populations: PAF depends on Pe, which varies geographically and temporally — a US PAF may not apply to a low-income country, or to 2025 versus 1990

— Wide confidence intervals when exposed group is small

— Residual confounding from unmeasured variables (socioeconomic status, genetics)

— Misclassification of exposure (recall bias in case-control studies) → biases RR toward null → underestimates PAF

— Effect modification: PAF may differ across strata (e.g., smoking PAF for MI is higher in young adults than elderly)

— Over-aggressive targeting of a high-PAF risk factor can divert resources from outcomes with lower PAF but high severity (rare cancers, suicide)

— Risk-factor-focused metrics may stigmatize patients (e.g., obesity, addiction) — ethical concern in QI programs

Common interpretive errors that generate exam distractors:

Statistical pitfalls:

Adverse population-level consequences:

Key distinction: A high PAF in an observational study does not guarantee a clinical trial will show the corresponding intervention prevents disease — only randomization confirms causality and intervention efficacy.

Board pearl: If a stem says "removing exposure X would prevent Y% of cases" but the study is observational with potential confounding, the correct critique is "PAF may be overestimated due to uncontrolled confounding."

Step 3 management: Always pair PAF estimates with causal evidence (RCT, Bradford Hill criteria) before launching costly interventions.

When to Escalate — From PAF Calculation to Policy Action

— High PAF (>20%) for a serious outcome with effective, scalable intervention → strong case for public health program

— Rising PAF over time (e.g., obesity-related cancers) → signal for new policy

— High PAF concentrated in vulnerable subgroup → equity-focused intervention

— Low PAF but high severity (e.g., rare lethal exposures) → targeted regulation rather than population campaign

— Quantify PAF with adjusted RR and best-available Pe

— Estimate intervention impact fraction (IIF) = PAF × coverage × efficacy

— Compare cost-effectiveness across candidate interventions (cost per QALY, cost per case prevented)

— Engage stakeholders, equity review, and implementation science

— Just as a hypotensive patient escalates from PO → IV → vasopressors → ICU consult, a public health priority escalates from clinical counseling → office-based protocol → community program → policy/legislation

— Tobacco trajectory: physician counseling (low reach) → pharmacotherapy (moderate) → quitlines (broad) → taxation and smoke-free laws (population-level, largest PAF impact)

— Health economists for cost-effectiveness analysis

— Implementation scientists for uptake and fidelity

— Community partners for cultural tailoring

— Legislators for policy levers (taxation, mandates, fortification)

Trigger points to move from epidemiologic estimate to system-level intervention:

Decision hierarchy for medical directors and public health officers:

Escalation analogues in clinical CCS-style management:

When to involve other disciplines:

Key distinction: PAF tells you what to target; implementation science tells you how to actually reduce exposure prevalence. A theoretically high PAF is useless if no feasible intervention can lower Pe.

Board pearl: The intervention with the largest realized impact in modern public health (e.g., tobacco control, vaccination, sanitation) combined high PAF with feasible, scalable population-level delivery — not just individual clinical encounters.

CCS pearl: When CCS-style population health questions appear, "order" a PAF analysis and cost-effectiveness review before committing system resources.

Key Differentials — Other Epidemiologic Measures (Same Category)

— Compares risk in exposed vs unexposed

— Measures strength of association, not population burden

— Used in cohort studies; in RCTs and prospective designs

— Used in case-control studies

— Approximates RR when disease is rare (<10%)

— Cannot directly calculate PAR; PAF can be estimated using OR with caveats

— Time-to-event analog of RR, from Cox proportional hazards

— Used for survival analyses (cancer, transplant)

— Same population impact translation applies

— Ie − Iu (when exposure is harmful) or Iu − Ie (when intervention is protective)

— Drives the NNT = 1/ARR

— Individual-level counseling metric

— Limited to exposed subgroup

— Useful for occupational or niche-exposure questions

— Whole-population perspective

— Used for policy, allocation, and quality-improvement decisions

— For protective exposures (RR < 1) — vaccines, statins, screening

— Estimates population-level benefit if uptake were universal

— Strength of association → RR/OR/HR

— Individual benefit → ARR/NNT

— Burden in exposed → AR/ARP

— Burden in population → PAR/PAF

— Benefit of protective exposure → PF

Differentiating measures of association and impact is the highest-yield content area:

Relative risk (RR) / risk ratio:

Odds ratio (OR):

Hazard ratio (HR):

Risk difference (absolute risk reduction, ARR):

Attributable risk in the exposed (AR) and ARP:

Population attributable risk (PAR) and fraction (PAF):

Preventable fraction (PF):

Key distinction quick-table:

Board pearl: Step 3 stems test the same dataset with different question framings. Identify which subgroup (individual exposed person vs. whole population) the question targets — that single keyword (e.g., "in the community," "among smokers," "for this patient") determines the correct measure.

Step 3 management: Build a mental decision tree: scope (individual vs. population) → harmful or protective exposure → choose the measure.

Key Differentials — Concepts From Adjacent Categories

— Incidence = new cases per person-time → drives PAR calculations

— Prevalence = existing cases at a point in time → not directly used in PAR; prevalence of exposure (Pe) is, however, central to PAF

— These are test performance measures, not exposure-impact measures

— PPV depends on disease prevalence; commonly confused with PAR because both involve prevalence, but they answer different questions (test accuracy vs. population disease burden)

— Diagnostic test domain; unrelated to PAR/PAF

— Individual-level intervention effect metrics

— NNT = 1/ARR; useful for patient counseling, not for population-level prioritization

— Effect modification (interaction) — RR differs across strata; report stratum-specific PAFs

— Confounding — distorts crude RR; use adjusted RR in PAF calculations

— Confusing the two leads to wrong PAF interpretation

— Bias in screening/observational studies that can inflate apparent RR → overestimate PAF

— Always check for these before accepting a published PAF

— Rose's prevention paradox: large population gains from small individual shifts (PAF lens)

— High-risk strategy: clinical management of individuals at greatest absolute risk (NNT lens)

Concepts confused with PAR/PAF that come from different epidemiologic domains:

Incidence vs. prevalence:

Sensitivity, specificity, PPV, NPV:

Likelihood ratios (LR+, LR−):

Number needed to treat (NNT) and number needed to harm (NNH):

Effect modification vs. confounding:

Lead-time bias, length bias, selection bias:

Public health vs. clinical paradigms:

Key distinction: PAR/PAF is an impact measure, not a diagnostic measure, not a screening performance metric, and not an individual treatment effect. If the stem is asking about "test accuracy" or "screening yield," you are in the wrong domain.

Board pearl: When a question mixes screening test performance with population disease burden, separate the two questions: "Is the test accurate?" (Sn/Sp/PPV) versus "How much disease in the population is attributable to the risk factor?" (PAF).

Step 3 management: Recognize domain before reaching for a formula.

Secondary Prevention — Sustaining PAF Reductions

— Continued surveillance of exposure prevalence (BRFSS, NHANES, registries)

— Monitoring of disease incidence for trend reversal

— Maintenance funding — many successful programs (e.g., tobacco quitlines, vaccine outreach) lose efficacy when budget cuts reduce coverage

— Equity audits to ensure benefits reach high-PAF subgroups, not only low-risk ones

— Tobacco control (US, 1965–present): adult smoking prevalence fell from 42% to <12%; lung cancer mortality declining; PAF for many CVD/cancer outcomes shrinking

— HPV vaccination: declining HPV prevalence and pre-cancer rates in vaccinated cohorts

— Childhood lead exposure: mean blood lead in children fell >90% after leaded gasoline phaseout

— Folate fortification: sustained reduction in neural tube defects since 1998

— Obesity prevalence continues rising → PAFs for diabetes, NAFLD, colon cancer, endometrial cancer increasing

— Opioid use disorder → rising PAF for overdose mortality

— Vaccine hesitancy → rising PAF for measles, pertussis

— Climate-related exposures (heat, air pollution, vector-borne disease) → expanding PAFs

— Just as a post-MI patient needs aspirin, statin, beta-blocker, ACEi, and cardiac rehab indefinitely, populations need ongoing investment in high-PAF interventions — these are not "one and done"

Once a population intervention is implemented, sustained PAF reduction requires:

Examples of successful long-term PAF reductions:

Examples of stalled or reversing trends (rising PAFs):

Long-term clinical plan analog:

Key distinction: Achieving PAF reduction is different from maintaining it. New exposures (e-cigarettes, ultra-processed food) can replace eliminated ones, shifting the PAF landscape.

Board pearl: A "successful" public health intervention often reduces incidence dramatically and then plateaus — additional gains require addressing the next-highest PAF, not redoubling effort on the already-controlled exposure.

Step 3 management: Build sustained surveillance, equity-stratified reporting, and funding into any program designed around PAF estimates.

Follow-Up, Monitoring Parameters, and Counseling

— Exposure prevalence (Pe) over time — primary process measure

— Disease incidence — primary outcome measure

— Disease mortality — long-term outcome

— Subgroup disparities — equity measure

— Cost per case prevented / cost per QALY — efficiency measure

— Intervention coverage and adherence — implementation fidelity

— Translate PAF into personally relevant terms: "Smoking causes about 85% of lung cancer cases in the US. For you specifically, quitting now reduces your lung cancer risk by about half within 10–15 years."

— Use absolute risk and NNT/NNH for individual decisions; use PAF for explaining societal stakes

— Acknowledge uncertainty: PAFs are estimates with confidence intervals

— Smoking: pack-years documented, cessation status at every visit (5 A's: Ask, Advise, Assess, Assist, Arrange)

— Hypertension: home BP monitoring, follow-up every 1 month until controlled, then every 3–6 months

— Diabetes: A1c every 3–6 months

— Obesity: BMI, waist circumference, lifestyle counseling

— Alcohol: AUDIT-C annually

— Vaccination status: reviewed at every preventive visit

— Quality measures (HEDIS, MIPS, CMS Star Ratings) often track high-PAF exposures: BP control, statin use, tobacco screening, A1c control, vaccination rates

— These metrics are deliberately chosen because moving them improves population health

Metrics to track after a PAF-guided intervention:

Counseling individual patients about population data:

Patient-level monitoring tied to high-PAF exposures:

System-level monitoring:

Key distinction: Process measures (Pe, screening rates) shift quickly; outcome measures (incidence, mortality) lag by years. Both must be monitored.

Board pearl: When a quality measure asks "tobacco use screening and cessation intervention," recognize that this metric directly addresses one of the highest-PAF exposures in clinical medicine.

Step 3 management: At every visit, screen and intervene on the top-PAF modifiable exposures — tobacco, BP, lipids, weight, alcohol, physical activity, vaccination — as the highest-yield use of clinic time.

Ethical, Legal, and Patient Safety Considerations

— Stigmatization: Targeting high-PAF behaviors (obesity, substance use, sexual practices) risks blaming individuals for socially patterned exposures. Frame interventions around structural drivers, not personal failure

— Equity vs. efficiency: Maximizing total PAF reduction may neglect smaller, marginalized groups whose disease burden is concentrated. Public health ethics requires balancing utility and justice

— Autonomy: Mandatory interventions (vaccine mandates, sugar taxes, motorcycle helmet laws) reduce PAF but may conflict with individual liberty. Step 3 frequently tests recognition of this tension

— Informed consent in screening: Programs justified by population PAF data must still disclose individual-level benefits, harms (false positives, overdiagnosis), and uncertainty

— Mandatory reporting of communicable diseases (TB, syphilis, measles, HIV in most states) enables PAF surveillance

— De-identification of public health data must comply with HIPAA; aggregate PAF reporting is permitted

— Workplace and product safety regulation (OSHA, FDA, EPA) is grounded in PAR estimates — e.g., asbestos bans, lead removal, food safety standards

— Transition-of-care risk (hospital discharge): medication reconciliation, follow-up scheduling, and identification of high-PAF readmission drivers (heart failure, COPD, sepsis) are central CMS quality measures

— A discharged HF patient should have a follow-up appointment within 7–14 days, weight-monitoring plan, and reinforced medication adherence — addressing the high-PAF readmission exposures

— Concrete Step 3 example: A patient is discharged with newly diagnosed AF and a CHA₂DS₂-VASc of 4 but is not started on anticoagulation. AF carries a high stroke PAF in the elderly; failure to anticoagulate at discharge is both a patient safety event and a missed population-level opportunity — the correct action is to start anticoagulation before discharge and ensure outpatient follow-up

Ethical issues when applying PAR/PAF to populations:

Legal and regulatory dimensions:

Patient safety and transitions of care:

Key distinction: Ethical public health practice combines PAF-guided prioritization with respect for autonomy, equity, and informed consent.

Board pearl: The most defensible interventions have high PAF, strong evidence, voluntary uptake, equitable delivery, and transparent risk communication — e.g., adult vaccination programs.

High-Yield Associations and Rapid-Fire Clinical Facts

— Lung cancer → smoking (~85%)

— COPD → smoking (~80%)

— Cervical cancer → HPV (~99%)

— Hepatocellular carcinoma → hepatitis B/C, alcohol, NAFLD (combined >80%)

— Stroke → hypertension (~48%)

— MI → dyslipidemia + smoking + HTN + DM (combined ~90%)

— Type 2 diabetes → obesity + physical inactivity (>70%)

— HIV → unprotected sex + IV drug use (population-dependent)

— Mesothelioma → asbestos (~80%)

— Bladder cancer → smoking (~50%)

— Dementia (modifiable share) → 12 Lancet factors (~40%)

— AR = Ie − Iu

— ARP = (RR − 1)/RR

— PAR = It − Iu

— PAF = (It − Iu)/It = Pe(RR−1)/[1 + Pe(RR−1)]

— PF = Pe(1−RR)/[1 − Pe(1−RR)] (protective exposure)

— RR = 2, Pe = 0.5 → PAF = 33%

— RR = 3, Pe = 0.3 → PAF = 38%

— RR = 10, Pe = 0.01 → PAF = 8%

— RR = 1.5, Pe = 0.6 → PAF = 23%

— RR = 5, Pe = 0.2 → PAF = 44%

— A common risk factor with modest RR may have larger PAF than a rare risk factor with huge RR

— PAF assumes causality

— ARP applies to exposed only, PAF applies to population

— Summed PAFs across exposures can exceed 100% (multicausality)

— Levin's formula uses prevalence of exposure and RR

— In case-control studies, use OR as RR approximation only if disease is rare

Highest-PAF exposures in US adults (cause → exposure):

Quick formula recall:

Numeric intuition:

Concepts most often tested:

Key distinction: Memorize the difference between "in the exposed" (ARP, AR, NNT) and "in the population" (PAR, PAF) — this single axis answers most stems.

Board pearl: When in doubt, calculate Pe(RR−1)/[1 + Pe(RR−1)] — the most likely correct answer on a PAF question.

Board Question Stem Patterns

— Table with several exposures, each with Pe and RR

— Compute PAF for each; pick the largest PAF (assuming intervention efficacy similar)

— Trap: choosing the exposure with the highest RR rather than the highest PAF

— Direct PAF computation

— Trap: confusing with ARP (only the exposed subgroup)

— Wants ARP = (RR−1)/RR

— Trap: applying PAF, which is lower because it includes nonsmokers

— Wants PAR (absolute units) × population size

— Trap: reporting PAF (a fraction) instead of an absolute count

— Combine PAF with intervention efficacy and coverage to estimate realized impact

— Trap: assuming intervention efficacy is 100%

— Compare preventable fraction × Pe × intervention efficacy × cost

— Trap: picking the more sensitive test without considering disease prevalence and PAF

— Possible flaws: residual confounding, reverse causation, recall bias, generalizability, rare-disease assumption violation

— Trap: accepting the headline PAF without methodological scrutiny

— Use preventable fraction formula

— Trap: applying PAF formula meant for harmful exposure (RR > 1)

— Wants individual-level measure (ARR, NNT)

— Trap: quoting PAF to an individual (relevant only at population scale)

Stem pattern 1 — "Which intervention prevents the most cases?"

Stem pattern 2 — "What proportion of disease in the population is attributable to X?"

Stem pattern 3 — "Among smokers, what percent of lung cancer is caused by smoking?"

Stem pattern 4 — "How many cases per year would be prevented?"

Stem pattern 5 — "An RCT shows intervention reduces disease incidence by Z%. What's the population impact?"

Stem pattern 6 — "The medical director must choose between two screening programs..."

Stem pattern 7 — Critique of a published PAF

Stem pattern 8 — Protective exposure (vaccine, statin)

Stem pattern 9 — Patient counseling

Key distinction: Always read the denominator population in the question — exposed, total, or eligible — to pick the correct measure.

Board pearl: When stuck, ask: "Is this question about an individual patient, or about a community/population?" That choice eliminates half the answer options instantly.

One-Line Recap

Population attributable risk (PAR) and population attributable fraction (PAF) quantify how much disease in an entire population is attributable to a specific causal exposure, and they are driven jointly by the strength of association (RR) and the prevalence of the exposure (Pe) — making common, modestly harmful exposures often the highest-yield public health targets.

— PAR = It − Iu (absolute population excess)

— PAF = (It − Iu)/It = Pe(RR−1)/[1 + Pe(RR−1)] (Levin's formula)

— ARP = (RR−1)/RR applies only to the exposed subgroup

— Preventable fraction (PF) is the protective-exposure analog

— A high-prevalence, modest-RR exposure (hypertension, dyslipidemia) typically has a larger PAF than a rare, high-RR exposure

— PAF assumes causality and complete elimination of the exposure; real-world impact is PAF × coverage × efficacy

— Sums of PAFs across multiple exposures can exceed 100% due to multicausality — use joint PAF when needed

— Adjust for confounding using multivariable-adjusted RR before computing PAF

— In case-control studies, OR approximates RR only when disease is rare

— Population/policy decisions → PAF (which exposure to attack first)

— Individual patient decisions → ARR and NNT (whether to treat this person)

— Quality measures (HEDIS, MIPS) deliberately target the highest-PAF clinical exposures: tobacco, BP, lipids, diabetes, vaccination, cancer screening

Formula essentials:

Conceptual essentials:

Decision essentials:

Board pearl: When a Step 3 stem asks which intervention to prioritize at the population level, the highest-PAF option wins — and that's usually the one targeting a common modifiable exposure with a moderate RR, not the rarest dramatic one.