---

More than half of the deaths in the United States have been attributed to nine well-known determinants (or causal risk factors): tobacco use, poor diet and physical inactivity, alcohol consumption, infectious agents, environmental toxins, firearms, unsafe sex, motor vehicles, and illicit drugs [Mokdad 2004]. Population members exposed to a causal risk factor have an increased risk of disease, disability, or death. Moreover, many of these causal risk factors contribute to more than one cause of death, and a cause of death can be attributed to multiple underlying determinants. To link the leading causes of premature death in San Francisco to these underlying determinants, a prevention attribution matrix was created based on a review of the biomedical literature and public health literature. For example, here is the prevention attribution matrix for SF females, 2003-4, and the matrix for SF males, 2003-4.
The more that deaths are attributable to preventable risk factors, the greater the potential benefit from reducing exposure to these factors. An intuitive approach is to estimate the population attributable fraction (PAF) to approximate how much of the burden of premature mortality due to a specific cause of death might have been prevented if a determinant (e.g., tobacco smoking) had been eliminated from the population. The PAF is estimated using the following, widely-employed equation:

     PAF = P_exposure(RR - 1) / [P_exposure(RR - 1) + 1]        Equation 1

where P_exposure is prevalence of exposure in the target population and RR is the relative risk of the outcome estimated from comparing an exposed to an unexposed study population. For example, to estimate the proportion of lung cancer deaths that might not have occurred if tobacco smoking (in the past) had been eliminated from the population, one needs to measure the prevalence of smoking (P_exposure) and the relative risk (RR) of lung cancer deaths comparing smokers to non-smokers. If lung cancer death rates are 10 times higher in tobacco smokers compared to non-smokers (relative risk equals 10) and 40% of the population smokes, then the proportion of lung cancer deaths attributable to smoking is about 80%, using this method.

Because the leading 15 causes of death differ by gender and ethnic group and ZIP code, the prevention attribution matrix is different for each demographic group. Depicted here are the attribution matrices for San Francisco males and San Francisco females. Each cause of death/determinate pair was systematically evaluated for potential relationships. First, textbooks, local expert opinion, and PubMed/MEDLINE searches were used to establish whether there is a known association between each leading cause of death and each of these determinants. Next, using Equation 1, the PAFs for the relevant cause of death/determinate pairs were roughly approximated using the PAF categories displayed in the legend to the attribution matrix ( = PAF greater than 40%,  = PAF  10% to 40%,  = PAF  2% to 10%, and ? = more than two studies but no consensus). Finally, in an ongoing process PubMed/MEDLINE searches continue to find data on relative risk and prevalence from comparable populations to approximate the proportion of deaths attributable to these determinants. Increasingly, PAFs from the Global Burden of Disease Study are being employed to calculate rough estimates of years of life lost (YLLs) for these determinants.

The Comparative Risk Assessment module of the Global Burden of Disease Study has established PAFs for various global regions, including high-income countries [links to Chapter 4 of the Global Burden of Disease and Risk Factors]. These PAFs, based on extensive literature reviews by global authorities, utilize a more nuanced version of Equation 1 that attempts to account for the ranges of exposure and relative risks that occur in various populations. In addition, these PAFs are based upon comparisons with populations experiencing ideal levels of exposure, levels that are well below the threshold for clinical concern [e.g., the ideal body mass index, BMI, is 21, rather than <25 (25-30 being overweight) or <30 (30 being the threshold for obesity)]. These PAFs are being compared with those which were previously calculated for San Francisco, and the prevention attribution matrix will ultimately be expanded to include a wider array of risk factors. Nevertheless, It remains uncertain whether PAFs for developed economies as a global region are appropriate to San Francisco’s populations.

The PAFs shown on these pages are the best qualitative estimates based on a review of the literature. Many of the estimates were made in the absence of the most approximate information on strength of risk and actual distribution of determinants in populations like those comprising San Francisco. The qualitative classification of attribution table cells will change with further information and analysis.