Calculating Population Attributable Fraction

Estimate the proportion of disease burden in a population attributable to a specific exposure using standard epidemiologic formulas.

Results

Interpretation: PAF approximates the proportion of all cases that could be prevented if the exposure were eliminated, assuming a causal relationship and no major bias or confounding.

How to Calculate Population Attributable Fraction (PAF): Practical Guide for Public Health and Clinical Epidemiology

Population Attributable Fraction (PAF) is one of the most useful metrics in epidemiology, prevention science, and health policy planning. While relative risk tells you how strongly an exposure is associated with disease at the individual level, PAF answers a different and often more policy-relevant question: how much disease burden in the entire population is attributable to a given exposure? In plain terms, it estimates the proportion of all cases that could potentially be prevented if the exposure were removed, assuming the association is causal and estimated correctly.

This distinction matters. A very high risk factor with low population prevalence may produce a modest total burden, while a moderate risk factor with high prevalence can drive a large share of disease. That is exactly why PAF is so central in prevention strategy. It links individual-level risk estimates with population-level impact.

Core formula used in this calculator

For a dichotomous exposure with prevalence in the population and an effect estimate treated as a risk ratio, the standard Levin equation is:

PAF = [Pe x (RR – 1)] / [Pe x (RR – 1) + 1]

• Pe = prevalence of exposure in the population (as a proportion, not percent)
• RR = risk ratio (or hazard ratio approximation under common assumptions)

When you only have an odds ratio from case-control data, this tool can convert OR to an approximate RR if you provide baseline risk among unexposed (P0):

RR = OR / ((1 – P0) + P0 x OR)

This adjustment reduces inflation that can occur when OR is interpreted directly as RR in non-rare outcomes.

What PAF is best used for

• Prioritizing prevention targets in national and regional health plans
• Estimating potential impact of reducing harmful exposures
• Comparing contribution of multiple risk factors to one disease outcome
• Translating epidemiologic evidence into health system planning language
• Communicating disease burden in absolute terms by multiplying PAF by case counts

Step-by-step calculation workflow

1. Define the exposure clearly and ensure consistent case definition.
2. Use prevalence estimates representative of your target population and time period.
3. Choose the most defensible effect size (RR preferred; adjusted estimates from high-quality studies).
4. If using OR, convert to RR when baseline risk is not negligible.
5. Compute PAF using the formula above.
6. Multiply PAF by total disease cases to estimate attributable case count.
7. Report assumptions, data sources, and uncertainty.

Comparison table: real-world statistics and illustrative PAF values

Exposure and outcome	Population prevalence estimate	Representative effect estimate	Approximate PAF	Why this matters
Cigarette smoking and lung cancer mortality (U.S.)	11.5% current adult smoking prevalence (CDC, 2022)	RR around 20 for heavy long-term smoking versus never smoking (major cohort evidence summarized in federal reports)	About 68.6%	Even with declining prevalence, very high RR yields a large attributable burden.
Residential radon and lung cancer risk	About 1 in 15 U.S. homes estimated above EPA action level (approximately 6.7%)	Relative risk increase is modest per exposure increment, often near 1.1 to 1.2 depending on concentration models	Roughly 1.0% to 1.3% in simplified single-threshold framing	Small individual effects can still justify policy because exposure is widespread and modifiable via building remediation.
Adult obesity and type 2 diabetes risk	U.S. adult obesity prevalence around 41.9% (CDC NHANES period estimate)	RR commonly in the 3 to 4 range in epidemiologic literature	Around 45% to 56% depending on exact RR	High prevalence plus strong association produces high preventive potential.

Values above are educational illustrations based on widely reported U.S. prevalence statistics and representative risk estimates. Exact attributable fractions vary by age, sex, intensity, lag effects, and adjustment strategy.

Interpreting PAF correctly

A PAF of 30% does not mean every third case is individually caused by the exposure in a deterministic sense. It means that, under the model assumptions, around 30% of cases in the population are attributable to exposure and could theoretically be avoided if exposure were fully eliminated and everything else remained equal. Real implementation effects are usually smaller because interventions are incomplete, delayed, or affected by behavior and structural factors.

PAF is also sensitive to both prevalence and effect size. If prevalence drops, PAF can fall even when RR remains constant. If RR rises (for example, due to stronger exposure intensity), PAF increases even if prevalence is stable. This is why surveillance systems need regular updates rather than one-time estimates.

Second comparison table: sensitivity of PAF to prevalence and RR

Scenario	Exposure prevalence (Pe)	Risk ratio (RR)	Calculated PAF	Attributable cases if total cases = 100,000
Low prevalence, high RR	5%	10.0	31.0%	31,034
Moderate prevalence, moderate RR	20%	2.0	16.7%	16,667
High prevalence, moderate RR	50%	2.0	33.3%	33,333
High prevalence, strong RR	40%	4.0	54.5%	54,545

This table demonstrates why high-prevalence exposures are often high-yield prevention targets, even when individual risk increase is not extreme.

Common methodological pitfalls

• Confounding and bias: If the RR estimate is biased, PAF will be biased.
• Mismatched populations: Using RR from one population and prevalence from another can distort estimates.
• Time lag neglect: Some exposures (for example tobacco or asbestos) have long latency; immediate elimination does not instantly erase all risk.
• Using crude instead of adjusted effects: Prefer adjusted estimates that account for key confounders.
• Ignoring joint effects: Separate PAFs for multiple exposures usually cannot be summed directly without methods for overlap and interaction.

Advanced considerations for experts

In advanced burden-of-disease work, analysts frequently use category-specific exposures, continuous risk distributions, and counterfactual minimum-risk exposure distributions rather than a single exposed/unexposed cut point. For those settings, generalized impact fraction methods are often preferable to basic Levin PAF. Still, the classic PAF framework remains an excellent foundation for communication and for first-pass policy modeling.

Uncertainty should be reported whenever possible. A good practice is to compute confidence intervals by propagating uncertainty from prevalence and effect estimates, often with bootstrap or Monte Carlo approaches. Decision-makers respond better to ranges than point estimates alone, especially in resource allocation decisions.

Recommended authoritative references

• CDC: Smoking-related mortality and health effects
• U.S. EPA: Health risk of radon
• Boston University School of Public Health: Attributable risk and attributable fraction concepts

Practical reporting template

When publishing or presenting your estimate, include: exposure definition, prevalence source and year, effect estimate source, formula used, assumptions, total case count used for absolute burden conversion, and sensitivity analyses. A transparent PAF report is far more credible than a single headline figure without context.

Used responsibly, PAF is a high-value bridge from epidemiologic evidence to prevention policy. It helps answer the question health leaders care about most: where can intervention produce the greatest reduction in real-world disease burden?