Attributable risk represents one of the foundational metrics of modern epidemiological inquiry and population health analytics, bridging the statistical assessment of causation with pragmatic healthcare intervention. By estimating the excess rate of disease that can be directly ascribed to an environmental, behavioral, or genetic exposure, this quantitative measure enables researchers and health authorities to determine the expected societal benefit if that specific hazard were systematically eradicated. Consequently, attributable risk transforms abstract probabilistic correlations into concrete, actionable public health imperatives.
Attributable Risk
1. Concise Definition
Attributable risk (AR) is an epidemiological metric that quantifies the difference in the rate of an adverse health outcome between an exposed group and an unexposed group, representing the absolute incidence of disease directly ascribed to the specific exposure under a causal assumption. Formally expressed as the incidence in the exposed minus the incidence in the unexposed, it serves as an index of the excess disease burden generated by a given risk factor.
Beyond basic subtraction of incidence rates, attributable risk reflects the proportion of total pathology within an exposed cohort that would theoretically be prevented if the exposure were eliminated entirely. Unlike relative measures that gauge the statistical strength of an etiological association, attributable risk directly addresses public health impact by linking exposure prevalence and excess absolute risk to concrete clinical outcomes.
In broader population frameworks, the concept extends to population attributable risk (PAR) and attributable fraction (AF), which account for exposure distribution across an entire community. Through these mathematical extensions, attributable risk bridges individual clinical epidemiology and macroeconomic healthcare policy, establishing a baseline for resource allocation, preventative screenings, and targeted environmental interventions.
2. Etymology & Linguistic Origin
The term attributable risk unites two classical linguistic roots that entered modern scientific vocabulary through distinct historical pathways. The verb attribute derives from the Latin attribuere, composed of the prefix ad- (meaning “to” or “toward”) and tribuere (meaning “to assign, bestow, or allot,” originally tied to the administrative divisions of ancient Roman tribes). By the sixteenth century, the term had entered Middle English and Early Modern English to denote the intellectual act of assigning an effect to a distinct causative agent or source.
The noun risk entered the English lexicon in the seventeenth century via the French risque and the Italian risco or rischio, which traces its lineage to the Byzantine Greek rhizikon (referring to a maritime hazard or cliff, symbolizing danger encountered at sea). In actuarial and mercantile contexts, risk came to denote the quantifiable probability of incurring misfortune, physical peril, or financial loss under conditions of uncertainty.
The formal synthesis of these concepts into attributable risk occurred within mid-twentieth-century epidemiology. Biostatisticians sought unambiguous nomenclature to separate purely multiplicative metrics of association from additive measures of absolute disease liability. Epidemiologist Abraham Lilienfeld and biometrician Morton L. Levin were instrumental in introducing and standardizing variations of the term in English-language scientific literature during the early 1950s, defining it as the liability of disease occurrence that can be credited to or blamed upon a verified etiological exposure.
3. Pronunciation & Grammatical Form
The term is pronounced phonetically in Received Pronunciation as /əˈtrɪb.jʊ.tə.bəl rɪsk/ and in General American as /əˈtrɪb.jə.t̬ə.bəl rɪsk/. The primary stress falls on the second syllable of “at-trib-ut-a-ble” and directly on the single syllable of “risk.”
Grammatically, attributable risk functions as a compound noun phrase, categorized as an uncountable mass noun in general biostatistical contexts, though it admits countable pluralization (attributable risks) when comparing discrete estimates across multiple demographic strata or distinct exposures. The word attributable functions as a participial adjective modifying the head noun risk, demanding an explicit or implied prepositional complement introduced by “to” (for example, “the risk attributable to tobacco consumption”).
Common spelling and compositional variants appear across academic literature, including attributable risk percentage (ARP), attributable fraction, and excess risk. In causal inference frameworks, authors occasionally refer to it as excess incidence or risk difference, though purists maintain technical distinctions between simple differences in observed incidence and the counterfactual assertions inherent in true attribution.
4. Detailed Conceptual Explanation
To comprehend attributable risk in its full epidemiological scope, one must distinguish between the relative strength of a biological association and its practical, real-world consequence. A relative measure, such as the relative risk or hazard ratio, assesses how much more likely an exposed individual is to develop a condition compared to an unexposed peer. While fundamental for establishing biological plausibility and etiological causation, high relative risks do not automatically translate into widespread societal harm if the baseline incidence of the disease is vanishingly small or if the exposure is extraordinarily rare.
Attributable risk operates within the additive framework of risk assessment. By computing the mathematical difference between two absolute incidence rates—the incidence rate among the exposed ($I_e$) and the incidence rate among the unexposed ($I_u$)—attributable risk answers a vital question: How much excess disease occurs per unit of population as a direct consequence of this exposure? This difference ($AR = I_e – I_u$) measures the surplus caseload directly borne by the exposed population, grounding statistical findings in tangible units of human suffering and clinical demand.
Moreover, attributable risk is intrinsically tied to counterfactual reasoning within modern causal inference models. Under this paradigm, the unexposed cohort is conceptualized as an empirical surrogate for the exposed cohort had they not encountered the risk factor, assuming total exchangeability and the absence of residual confounding. The subtraction isolates the exposure-specific liability, asserting that if the causal relationship is genuine and reversible, complete cessation of the exposure would cause the disease rate among the exposed to regress precisely to the baseline rate seen in the unexposed.
The conceptual boundary of attributable risk becomes particularly salient when extending the concept from the exposed subset to the community as a whole. When absolute excess risk is scaled by the prevalence of the exposure across an entire society, it yields the population attributable risk (PAR). This broader measure enables public health officials to understand what share of community-wide morbidity or mortality stems from a prevalent, mild risk factor versus a rare, devastating environmental hazard, guiding comprehensive health intervention strategies.
5. Historical Development
The quantitative concept of attributable risk emerged from mid-twentieth-century epidemiological investigations into non-communicable, chronic diseases, particularly the post-World War II surge in cardiovascular illness and bronchogenic carcinoma. Prior to this period, classical infectious disease epidemiology focused primarily on necessary and sufficient causative agents, such as specific bacterial pathogens, where simple attack rates and transmission chains dominated analytic methodology.
A critical milestone occurred in 1953, when Morton L. Levin published his groundbreaking paper analyzing the occurrence of lung cancer in relation to tobacco smoking. Levin recognized that simply demonstrating high relative odds between smokers and non-smokers failed to articulate the proportion of all lung cancer cases in New York State driven by smoking. He formulated what is now recognized as Levin’s formula for population attributable fraction, providing a mathematical blueprint for evaluating how reducing population-level exposure diminishes aggregate disease incidence.
During the late 1950s and 1960s, epidemiologists Jerome Cornfield, Abraham Lilienfeld, and their contemporaries expanded this framework amid intense methodological debates with tobacco industry statisticians and Ronald A. Fisher. Fisher argued that observational correlations could stem from genetic confounders that simultaneously caused nicotine craving and lung malignancy. Cornfield published mathematical proofs establishing that for an unobserved confounder to fully explain a relative risk of high magnitude, that confounder would need an implausibly colossal prevalence and association with the outcome. This work solidified attributable risk as an indispensable tool for demonstrating causal culpability.
Throughout the 1970s and 1980s, biostatisticians including Kenneth Rothman and Sander Greenland further refined attributable risk theory within the potential outcomes framework. They exposed subtleties regarding multi-causality, demonstrating that because human diseases typically arise from complex constellations of synergistic components, the sum of attributable fractions for all individual causes of a single disease can substantially exceed 100%. In contemporary research, attributable risk metrics anchor international epidemiological endeavors, such as the Global Burden of Disease Study, mapping attributable morbidity worldwide.
6. Theoretical Foundations
Attributable risk is theoretically underpinned by the sufficient-component cause model, formalized extensively by Kenneth Rothman. In this conceptualization, a biological outcome does not result from a solitary, isolated trigger, but from the completion of a “sufficient cause”—a causal mechanism comprising a constellation of minimal conditions and biological events that inevitably produce the pathology. Within this construct, any single risk factor serves as a “component cause.”
Because a component cause can participate in multiple distinct sufficient causal complexes, eliminating that single factor prevents the completion of every causal mechanism that requires its presence. This theoretical foundation explains why attributable risk must not be treated as a zero-sum, mutually exclusive metric across multiple co-occurring hazards. If both exposure A and exposure B must interact to produce a pathology, that disease case is entirely attributable to exposure A (in the sense that removing A prevents it) and entirely attributable to exposure B, yielding individual attributable fractions that sum beyond unity.
A second foundational pillar is the Counterfactual or Potential Outcomes Framework, popularized by Donald Rubin, James Robins, and standard epidemiological theorists. This paradigm formally posits that each individual possesses hypothetical potential outcomes under alternative states of exposure: $Y(1)$ representing the health outcome if exposed, and $Y(0)$ representing the health outcome if unexposed. In this light, attributable risk directly estimates the average causal effect within the exposed group: the expected difference between the observed state $E[Y(1)|E=1]$ and the counterfactual, unobservable state $E[Y(0)|E=1]$.
Finally, attributable risk relies on established biostatistical theories of additive hazard models. Whereas traditional logistic regression and Cox proportional hazards models assume multiplicative effects on the probability scale, attributable risk operates on absolute difference equations. The theoretical justification for prioritizing additivity over multiplicativity rests on public health utilitarianism: human societies experience disease burdens in countable units of individual patients requiring clinical care, rendering absolute differences far more informative for population-level resource planning than pure relative ratios.
7. Key Components, Types & Dimensions
Attributable risk is operationalized through several distinct metrics, each answering specific clinical and epidemiological questions:
- Attributable Risk (Risk Difference, AR or RD): The basic arithmetic difference in cumulative incidence or incidence rates between the exposed and unexposed cohorts ($AR = I_e – I_u$). It reflects the absolute excess rate of pathology caused by the exposure among those directly subjected to it.
- Attributable Fraction for the Exposed ($AF_e$ / Attributable Risk Percent): The proportion of disease occurrences among exposed persons that can be directly attributed to the exposure, calculated as $(I_e – I_u) / I_e$, or equivalently $(RR – 1) / RR$, where $RR$ represents relative risk. This metric expresses individual liability as a clean percentage of the patient’s total hazard.
- Population Attributable Risk (PAR): The excess incidence of disease observed in the general population (both exposed and unexposed) that can be ascribed to the exposure. Calculated as the total population incidence minus the unexposed incidence ($PAR = I_t – I_u$), this metric incorporates exposure prevalence into absolute disease burden.
- Population Attributable Fraction (PAF / Population Attributable Risk Percent): The proportion of all disease cases occurring across an entire population that would be prevented if the exposure were completely eliminated. Calculated as $(I_t – I_u) / I_t$, or via Levin’s formula: $P(E)(RR – 1) / [1 + P(E)(RR – 1)]$, where $P(E)$ represents exposure prevalence.
- Generalized Impact Fraction (GIF): An advanced, modern dimension of attributable risk that estimates the fraction of disease prevented not under hypothetical total eradication, but under realistic policy scenarios where exposure prevalence is partially reduced or shifted along a continuous distribution.
- Etiologic Fraction: A distinct theoretical construct sometimes separated from attributable fraction, denoting the true biological proportion of disease events in which the exposure played an indispensable causal role, accounting for competitive risks and accelerated disease onset.
8. Examples & Illustrative Cases
A classic, intuitive application of attributable risk appears in longitudinal cohort studies tracking the adverse sequelae of cigarette smoking on cardiovascular and pulmonary pathology. Consider a hypothetical observational study tracking 100,000 adult participants over ten years. Suppose the incidence of lung cancer among heavy cigarette smokers is observed at 200 cases per 100,000 person-years ($I_e = 0.0020$), whereas the incidence among lifelong non-smokers is 10 cases per 100,000 person-years ($I_u = 0.0001$).
The calculation of basic attributable risk proceeds by direct subtraction: $AR = 200 – 10 = 190$ cases per 100,000 person-years. This figure informs clinicians that for every 100,000 smokers followed over one year, 190 excess cases of lung cancer emerge specifically because of their tobacco use. Converting this into the attributable fraction for the exposed yields $(200 – 10) / 200 = 0.95$ (or 95%). This informs an individual smoker that 95% of their personal risk of developing lung cancer is driven directly by smoking, and only 5% reflects baseline environmental and biological vulnerability.
To contrast absolute attributable risk with relative risk, consider the same cohort’s coronary artery disease outcomes. Suppose the incidence of myocardial infarction is 1,200 per 100,000 person-years in smokers ($I_e = 0.0120$) and 600 per 100,000 person-years in non-smokers ($I_u = 0.0060$). Here, the relative risk for coronary events is $1200 / 600 = 2.0$, which is far lower than the relative risk for lung cancer ($200 / 10 = 20.0$). However, the attributable risk for myocardial infarction is $1200 – 600 = 600$ cases per 100,000 person-years—more than triple the absolute excess caseload observed for lung cancer ($190$). Thus, despite a modest relative risk, coronary disease delivers a much greater absolute societal burden attributable to smoking.
At the population level, consider how exposure prevalence influences outcomes. If 25% of the adult population smokes ($P(E) = 0.25$), applying Levin’s formula for lung cancer yields a population attributable fraction (PAF) of approximately 82.6%. This reveals that more than four-fifths of all lung cancer hospitalizations across the community would disappear if tobacco usage were completely eradicated. Conversely, if only 1% of the population were exposed to a rare occupational solvent with a tenfold relative risk for leukemia, the corresponding PAF would remain below 9%, illustrating why low-prevalence hazards yield minor community-wide attributable impacts despite severe individual toxicity.
9. Measurement & Assessment
The practical calculation and statistical assessment of attributable risk depend heavily on study design and sampling methodology. In prospective cohort studies and randomized controlled trials, where investigators directly measure incidence rates or cumulative risks over well-defined time horizons, estimating attributable risk is mathematically straightforward through direct subtraction of empirical proportions.
In retrospective case-control studies, absolute incidence rates cannot be computed directly because the ratio of cases to controls is fixed artificially by study design rather than natural occurrence. In this setting, researchers must derive attributable risk metrics indirectly. Provided the disease outcome is sufficiently rare in the source population (the “rare disease assumption”), the odds ratio ($OR$) serves as an accurate proxy for relative risk ($RR$). Biostatisticians then estimate the population attributable fraction via Miettinen’s formula: $PAF = P(E|D) imes [(OR – 1) / OR]$, where $P(E|D)$ represents the prevalence of exposure among the documented disease cases.
Furthermore, calculating point estimates of attributable risk requires constructing robust confidence intervals to assess statistical precision. Because the attributable fraction is a non-linear combination of incidence parameters or odds ratios, standard variance estimation utilizes the delta method or log-transformation strategies, such as Walter’s variance estimator, to establish 95% confidence limits. In complex modern epidemiological designs featuring multiple continuous confounders, multivariable regression modeling is standard practice:
Researchers routinely employ multivariable additive hazard models, Poisson regression with identity link functions, or targeted maximum likelihood estimation (TMLE) to calculate adjusted attributable risks while controlling for demographic and lifestyle variables. Modern causal inference software leverages semi-parametric G-computation and bootstrapping techniques to draw accurate confidence bounds around population attributable fractions in the presence of complex, time-varying confounders.
10. Applications & Practical Significance
Attributable risk serves as a cornerstone of preventive medicine, economic modeling, and public health governance. While basic researchers rely on relative risk to unravel physiological pathways and determine cellular mechanisms, public health authorities rely on attributable risk to allocate budgetary resources, justify statutory regulations, and establish clinical prevention targets.
In health economics, attributable risk calculations are used to measure disease burden via Disability-Adjusted Life Years (DALYs) and Quality-Adjusted Life Years (QALYs). By multiplying disease-specific disability weights by the population attributable fraction for a specific risk factor, economists can estimate the financial and functional savings achievable through targeted prevention. This methodology underpins major legislative decisions, including excise taxes on alcohol and sugar-sweetened beverages, where projected revenue gains are weighed against the societal cost of the attributable disease burden.
In clinical practice, understanding attributable risk empowers physicians to communicate probabilistic medical advice effectively to patients. Presenting health risks using absolute differences—often translated into the “number needed to treat” ($NNT = 1 / AR$) or “number needed to harm”—consistently improves patient comprehension compared to relative risk reductions, which can inflate perceived therapeutic benefits. Clinicians explaining hyperlipidemia management, for example, can show patients that while statin therapy yields a constant relative risk reduction across demographics, the absolute attributable risk averted is significantly greater in patients with high baseline cardiovascular risk, justifying aggressive therapy.
Environmental regulators similarly depend on attributable risk calculations when establishing maximum allowable limits for air pollutants, such as particulate matter ($PM_{2.5}$). By mapping ambient air quality metrics against localized population attributable fractions for acute asthma, stroke, and chronic obstructive pulmonary disease, agencies calculate the net human lives preserved by tightening industrial emissions limits, directly guiding regional and national clean-air legislation.
11. Research & Empirical Evidence
The empirical deployment of attributable risk is best exemplified by the global surveillance of non-communicable diseases. The landmark Global Burden of Disease (GBD) studies, coordinated by the Institute for Health Metrics and Evaluation (IHME), use attributable risk frameworks to map exposure metrics worldwide. These large-scale analyses evaluate dozens of environmental, occupational, and behavioral hazards, demonstrating that high systolic blood pressure, tobacco use, and dietary risks represent the leading global contributors to attributable mortality.
In oncology, landmark investigations by Sir Richard Doll and Richard Peto established the empirical baseline for cancer attribution. Their seminal 1981 report on the causes of cancer in the United States concluded that approximately 30% of all cancer deaths were attributable to tobacco consumption, while dietary factors accounted for another substantial, preventable fraction. Subsequent genomic and epidemiological reassessments over the following four decades have validated these attributable fractions across international populations, refining them to reflect contemporary exposure patterns.
Cardiovascular research has heavily integrated attributable risk to validate primary prevention paradigms. The international INTERHEART study, directed by Salim Yusuf and colleagues, evaluated thousands of myocardial infarction cases across 52 nations. The investigators demonstrated that nine easily measurable, modifiable risk factors—including smoking, dyslipidemia, hypertension, abdominal obesity, and psychosocial stress—accounted for a collective population attributable fraction of over 90% for initial acute myocardial infarctions globally. This empirical finding proved that coronary artery disease is overwhelmingly driven by preventable, modifiable exposures rather than untreatable biological aging.
In psychiatric epidemiology, attributable risk metrics have clarified the population-level consequences of cannabis use, childhood trauma, and urban living on the subsequent emergence of psychotic disorders. By tracking cohort datasets, researchers have demonstrated that while the relative risk of psychosis associated with early, high-potency cannabis use is substantial, the population attributable fraction varies widely between geographic jurisdictions, correlating directly with recreational usage rates among adolescents.
12. Cultural & Cross-Cultural Considerations
The interpretation and practical application of attributable risk metrics vary substantially across diverse international contexts and socioeconomic settings. A central challenge is that the population attributable fraction is context-dependent: because PAF relies directly on exposure prevalence within a specific community, the exact same biological hazard will yield radically disparate attributable burdens in different countries.
For instance, in high-income regions where industrial vehicle electrification and clean energy mandates have reduced particulate emissions, the population attributable risk of ambient air pollution for cardiopulmonary death is relatively modest. Conversely, in rapidly industrializing or developing nations that rely heavily on low-grade fossil fuels and solid biomass for indoor cooking, the PAF for identical respiratory conditions climbs dramatically. Consequently, interventions that deliver major health dividends in one country may offer limited absolute utility in another with lower baseline exposure.
Cultural attitudes toward health behaviors also shape how attributable risk data are received by populations and policymakers. In nations that emphasize communitarian welfare and universal health insurance, high attributable risk statistics for preventable conditions often generate political consensus for structural public health measures, such as mandatory salt limits or smoking bans. Conversely, in cultures that emphasize individual autonomy and personal liberty, highlighting the disease burden attributable to lifestyle behaviors can face skepticism or spark concerns over state paternalism.
Furthermore, biostatistical assumptions of exchangeability and baseline risk comparability across racial and ethnic groups can lead to flawed policy if applied uncritically. Variations in genetic architecture, health literacy, access to primary care, and systemic socioeconomic stressors can alter baseline disease rates ($I_u$). As a result, calculating attributable risks using unadjusted baseline rates from one demographic group can misstate the actual disease burden present in historically disadvantaged populations.
13. Criticisms, Debates & Limitations
Despite its mathematical elegance and practical utility, attributable risk has drawn sharp methodological critique, particularly regarding its common misinterpretation among healthcare professionals and policymakers. The most widespread error is the assumption that attributable fractions across multiple distinct risk factors must sum to 100%. Because human illness involves synergistic multi-causality, where multiple component causes interact to initiate pathology, attributable risks regularly overlap. Eliminating hypertension or eliminating hyperlipidemia could both independently prevent the same catastrophic vascular event, causing their individual PAFs to sum beyond unity when considered together.
A second major limitation centers on the strict counterfactual assumptions required for valid causal attribution. Attributable risk calculations assume that eliminating an exposure removes its associated disease burden entirely, without triggering unintended behavioral substitutions or biological shifts. In practice, eliminating one harmful exposure can cause individuals to substitute alternative hazards—such as replacing cigarette smoking with increased alcohol intake or ultra-processed foods—thereby violating theoretical assumptions and failing to deliver the predicted decline in total mortality.
Statisticians have also challenged the traditional population attributable fraction for assuming the hypothetical complete eradication of a risk factor, an unrealistic scenario in complex human societies. Demanding the total elimination of dietary sodium, ambient air pollution, or sedentary behavior is practically impossible. Consequently, methodologists such as Sander Greenland have advocated replacing binary attributable fractions with Generalized Impact Fractions (GIFs), which model the health effects of realistic, incremental reductions in exposure distributions.
Finally, standard attributable risk formulations do not capture the time dimension of disease onset. Traditional equations treat disease occurrence as a binary, static endpoint, ignoring whether removing an exposure permanently prevents an illness or merely delays its onset by several months or years. In aging populations where non-communicable chronic diseases predominate, metrics that incorporate time-to-event dynamics—such as years of life lost (YLL) attributable to an exposure—provide a more accurate reflection of true biological impact than simple risk differences.
14. Related Terms & Distinctions
Understanding attributable risk requires maintaining clear distinctions among several closely related epidemiological concepts:
- Relative Risk (Risk Ratio, RR): The ratio of disease incidence among the exposed to that among the unexposed ($I_e / I_u$). Relative risk assesses the biological strength of an etiological relationship; in contrast, attributable risk ($I_e – I_u$) measures the absolute excess disease burden generated by that relationship.
- Odds Ratio (OR): The ratio of the odds of exposure among cases to the odds of exposure among non-cases. While the odds ratio mathematically approximates relative risk when outcomes are rare, it does not directly reflect absolute caseload differences unless combined with known baseline population incidences.
- Absolute Risk Reduction (ARR): The direct mathematical counterpart of attributable risk used primarily in clinical trials of therapeutic interventions. Whereas attributable risk measures the excess hazard generated by an exposure ($I_e – I_u$), absolute risk reduction measures the drop in disease incidence achieved by a beneficial treatment ($I_{ ext{control}} – I_{ ext{intervention}}$).
- Population Attributable Risk (PAR) vs. Attributable Risk in the Exposed (AR): AR evaluates the excess disease burden restricted solely to individuals who encounter the hazard, whereas PAR weights this excess burden by the hazard’s prevalence across the entire general population.
- Number Needed to Harm (NNH): The reciprocal of the attributable risk ($NNH = 1 / AR$). It expresses the absolute excess risk in intuitive clinical terms, representing the average number of persons who must be exposed to the hazard for one additional adverse event to occur.
- Etiologic Fraction vs. Excess Fraction: In advanced survival analysis, the excess fraction measures net disease incidence produced by an exposure at a specific point in time, whereas the etiologic fraction measures the proportion of cases where the exposure accelerated or biologically initiated the disease process, even if the condition would have developed later in life from other causes.
15. Summary / Key Takeaways
Attributable risk remains one of the most vital quantitative metrics in epidemiology and preventive medicine. By calculating the absolute difference in disease occurrence between exposed and unexposed populations, it transforms abstract statistical correlations into practical estimates of societal disease burden. Unlike relative risk, which measures etiological association strength, attributable risk links exposure prevalence with absolute baseline incidence, directly identifying how many cases of an illness can be prevented by specific interventions.
The metric extends naturally from individuals to whole communities through the population attributable fraction (PAF), which models the percentage of community-wide illness that would resolve if a specific hazard were eliminated. Although widely used, attributable risk calculations require careful methodological handling. Researchers must account for multicausal synergy, unmeasured confounding, the non-additivity of independent fractions beyond 100%, and the biological reality that exposures often accelerate disease onset rather than permanently preventing it.
Ultimately, attributable risk bridges theoretical biostatistics, clinical decision-making, and public policy. By grounding epidemiology in measurable units of human disease, it enables governments, health organizations, and clinicians to target resources effectively, justify public health regulations, and design interventions that maximize long-term population well-being.
References
- Cole, P., & MacMahon, B. (1971). Attributable risk percent in public health. Preventive Medicine, 1(4), 515–524.
- Greenland, S., & Robins, J. M. (1988). Conceptual problems in the definition and estimation of attributable fractions. American Journal of Epidemiology, 128(6), 1185–1197.
- Levin, M. L. (1953). The occurrence of lung cancer in man. Acta Unio Internationalis Contra Cancrum, 9(3), 531–541.
- Miettinen, O. S. (1974). Proportion of disease caused or prevented by a given exposure, trait or intervention. American Journal of Epidemiology, 99(5), 325–332.
- Rothman, K. J., Greenland, S., & Lash, T. L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins.
- Yusuf, S., Hawken, S., Ôunpuu, S., Dans, T., Avezum, A., Lanas, F., McQueen, M., Budaj, A., Pais, P., Varigos, J., & Lisheng, L. (2004). Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): Case-control study. The Lancet, 364(9438), 937–952.