Research and practice methodsWhy population attributable fractions can sum to more than one
Introduction
P opulation attributable fractions (PAFs), also known as population attributable risks, are useful for estimating the proportion of disease cases that could be prevented if one or more risk factors for that disease were reduced or eliminated. For diseases with multiple risk factors, PAFs of individual risk factors can sum to more than 1. For example, one published table1 of modifiable risk factors for coronary heart disease shows that the PAFs of hypertension, cigarette smoking, elevated cholesterol, diabetes, obesity, physical inactivity, and environmental tobacco smoke exposure are 0.25, 0.22, 0.43, 0.08, 0.17, 0.35, and 0.18. Other examples exist for cancer,1, 2 hepatitis,3 and salmonellosis,4 and more examples could emerge as new risk factors for diseases are identified. Several investigators have commented on the troubling nature of results such as these that suggest the impossible situation in which more than 100% of cases are preventable.5, 6, 7, 8 Explanations have included inappropriate methods,5 risk factor overlap,7 and “multicausal” models.6, 8
This report examines a hypothetical example in which risk factors for a disease are eliminated one at a time in different sequences such that by the end of each sequence (or path) all preventable cases of the disease are prevented and only cases attributable to unknown factors remain. The example was analyzed by using sequential attributable fractions (SAFs) (i.e., the proportion of the original number of disease cases attributable to a risk factor9) and show that the PAF of a risk factor equals the SAF when the risk factor is the first to be eliminated. Because only one risk factor can be eliminated first, PAFs imply mutually exclusive scenarios; thus, PAFs cannot be summed meaningfully (except under special circumstances, such as when risk factors do not overlap10). It is also shown that SAFs depend on the order in which risk factors are eliminated and that the SAF is highest for a risk factor when it is eliminated first (i.e., when the SAF equals the PAF) and lowest when it is eliminated last. On the basis of this finding, expressions for upper and lower SAF limits are presented. Use of these limits could help prevent confusion and unrealistic expectations arising from the mistaken belief that PAFs sum to the total proportion of cases attributable to known risk factors.
Section snippets
A hypothetical example
With the use of Rothman's general model of causation,8 we will consider a hypothetical disease with only four risk factors (or component causes), A to D, distributed among a stable dynamic disease-free population of 9000 people (Table 1). Assume a disease case occurs when and only when an individual is exposed for exactly 1 year to at least one of the following four sufficient causes: [A and B], [A and C], [B and D], and [C and D]. For example, disease always occurs among individuals exposed to
Estimating upper and lower SAF limits (i.e., the proportion of original cases that can be prevented by eliminating a risk factor)
The paths shown in Figure 1, which assume complete elimination of one risk factor while the prevalence of the other risk factor remains static, do not represent realistic scenarios. More realistic situations, such as reducing risk factors A and B in small increments either singly or in combination, and scenarios involving more than two risk factors can also be analyzed by estimating path-specific proportions of original cases that could be prevented by a eliminating a risk factor, or SAFs.
Discussion
Despite potential pitfalls in their estimation and interpretation, PAFs are useful for translating surveillance data on risk factor prevalence and disease occurrence into numbers that can help policymakers and the public appreciate the potential benefits to be gained by risk factor reduction and health promotion. To highlight and explain one of these pitfalls, the problem of summing PAFs inappropriately, risk factor reduction was conceptualized as a series of steps along a path ending in the
References (15)
- et al.
Sequential and average attributable fractions as aids in the selection of preventive strategies
J Clin Epidemiol
(1995) - et al.
Attributable risk for symptomatic liver cirrhosis in Italy
J Hepatol
(1998) - et al.
Cardiovascular disease
- et al.
The causes of cancerquantitative estimates of avoidable risks of cancer in the United States today
J Natl Cancer Inst
(1981) - et al.
Transmission of hepatitis C virus in Taiwanprevalence and risk factors based on a nationwide survey
J Med Virol
(1999) - Rowe SY, Rocourt JR, Shiferaw B, et al and the FoodNet Working Group. Breast-feeding decreases risk of sporadic...
Modeling of risk factors for ischemic stroke
Stroke
(1997)
Cited by (102)
Towards a framework of driver fitness: Operationalization and comparative risk assessment
2024, Transportation Research Interdisciplinary PerspectivesFour-month incidence of suicidal thoughts and behaviors among healthcare workers after the first wave of the Spain COVID-19 pandemic
2022, Journal of Psychiatric ResearchThe contribution of potentially modifiable risk factors to acute ischemic stroke burden - Comparing young and older adults
2022, Preventive MedicineCitation Excerpt :In line with other studies (Aigner et al., 2017; O'Donnell et al., 2016; Owolabi et al., 2018), we found that the combined PAR for the five studied risk factors was lower than the sum of their individual PARs that exceeded 100%. This suggests that the fraction of cases that is attributed to each risk factor overlaps with the fractions attributable to other risk factors, indicating that stroke can simultaneously be attributed to or prevented by more than one risk factor (Dehghan et al., 2007; Rowe et al., 2004). We further showed that the extra attributable risk for every single risk factor was much lower than its individual PAR.
Fraction of COVID-19 hospitalizations and deaths attributable to chronic diseases
2022, Preventive MedicineCitation Excerpt :In Mexico, prior population attributable fraction (PAF) estimates have shown that 1.1%, 14.3% and 16.8% of deaths from COVID-19 were attributable to diabetes, hypertension, and obesity, respectively among ambulatory patients (Peña et al., 2020). Estimating PAFs for different risk factors is informative, yet insufficient to capture the total burden of multiple chronic diseases (Rowe et al., 2004). Few studies have estimated the impact of multiple chronic diseases or multimorbidity on COVID-19 (Lai et al., 2020; Hamer et al., 2020; Adams et al., 2020).
Contribution of socioeconomic, lifestyle, and medical risk factors to disparities in dementia and mortality
2021, SSM - Population Health