Attributable Risk Estimation in Epidemiological Studies

Attributable risk estimation quantifies the proportion of disease incidence or prevalence in a population that can be linked to one or more exposures. Central to public health planning and policy, these measures—often expressed as population attributable fractions (PAFs) or impact fractions—combine information on exposure prevalence and the strength of association between exposure and disease. Classical approaches, such as Levin’s formula, require only aggregate data on risk factor prevalence and relative risk but may be biased in the presence of confounding. Modern methods address these biases through re-expressed estimands, causal inference frameworks and Bayesian networks, allowing for sequential, joint and average attributable fractions under complex exposure interactions. Recent advances also extend estimators to continuous exposures, pathway-specific attributions and visualisation tools that integrate prevalence and effect size. Together, these developments support more accurate burden estimates, inform targeted interventions and enhance global comparability of risk assessments.

Research from Nature Portfolio

Seminal methodological frameworks have formalised the partitioning of exposure–disease associations into categories reflecting both the direction and magnitude of risk. By dividing the odds ratio parameter space into five exclusive regions—ranging from strong protective to strong risk factors—these methods provide tailored hypothesis tests, confidence intervals and power calculations for each category. This multi-region demarcation enhances interpretability of relative effect measures and underpins clearer communication of attributable risk estimates in diverse epidemiological settings.

Research from all publishers

Contemporary analyses have quantified the asymptotic bias inherent in Levin’s population attributable fraction formula when applied to confounded associations. By re-expressing the estimand in terms of both unadjusted and causal relative risks, researchers derived analytic formulae for relative and absolute bias and extended these to multilevel and continuous exposures, thereby offering corrective adjustments for more reliable PAF estimates.

To facilitate practical application, a comprehensive R package now supports estimation, inference and graphical display of PAFs and impact fractions. It accommodates standard and pathway-specific attributable fractions, continuous exposures, nomograms and fan-plot visualisations, and implements Bayesian network methods for joint, sequential and average attributable fractions across multiple risk factors.

Advances in mediation-aware metrics introduced the pathway-specific population attributable fraction (PS-PAF), which quantifies disease burden attributable to individual causal pathways. Under a potential outcomes framework, PS-PAFs measure the relative reduction in disease prevalence when a mediating pathway is disabled, offering novel insights into the mechanistic contributions of intermediate risk factors and guiding targeted prevention strategies.

Attributable Risk Estimation in Epidemiological Studies publication trend

The graph below shows the total number of articles in attributable risk estimation in epidemiological studies across all publications each year (not limited to Nature Index journals).

Technical terms

Population attributable fraction (PAF): The proportion of cases in a population that would be prevented if a specific exposure were eliminated.

Relative risk (RR): The ratio of disease incidence among the exposed group to that among the unexposed group.

Confounding: A situation in which the observed association between exposure and disease is distorted by a third variable related to both.

Impact fraction: A generalisation of PAF estimating the reduction in disease prevalence after partial risk factor modification.

Pathway-specific population attributable fraction (PS-PAF): The fraction of disease burden attributable to a particular causal pathway linking exposure, mediator and outcome.

References

1. Bias assessment and correction for Levin’s population attributable fraction in the presence of confounding. European Journal of Epidemiology (2024).
2. Pathway-specific population attributable fractions. International Journal of Epidemiology (2022).
3. Estimating and displaying population attributable fractions using the R package: graphPAF. European Journal of Epidemiology (2024).
4. A Five-Region Hypothesis Test for Exposure-Disease Associations. Scientific Reports (2017).