Improving population attributable fraction methods: examining smoking-attributable mortality for 87 geographic regions in Canada

Publication type
Journal article
Authors
Tanuseputro P, Manuel DG, Schultz SE, Johansen H, Mustard C
Date published
2005 Apr 15
Journal
American Journal of Epidemiology
Volume
161
Issue
8
Pages
787-798
PMID
15800272
Open Access?
Yes
Abstract

Smoking-attributable mortality (SAM) is the number of deaths in a population caused by smoking. In this study, the authors examined and empirically quantified the effects of methodological problems in the estimation of SAM through population attributable fraction methods. In addition to exploring common concerns regarding generalizability and residual confounding in relative risks, the authors considered errors in measuring estimates of risk exposure prevalence and mortality in target populations and estimates of relative risks from etiologic studies. They also considered errors resulting from combining these three sources of data. By modifying SAM estimates calculated using smoking prevalence obtained from the 2000-2001 Canadian Community Health Survey, a population-based survey of 131,535 Canadian households, the authors observed the following effects of potential errors on estimated national SAM (and the range of effects on 87 regional SAMs): 1) using a slightly biased, mismatched definition of former smoking: 5.3% (range, 1.8% to 11.6%); 2) using age-collapsed prevalence and relative risks: 6.9% (range, 1.1% to 15.5%) and -15.4% (range, -7.9% to -21.0%), respectively; 3) using relative risks derived from the same cohort but with a shorter follow-up period: 8.7% (range, 4.5% to 11.8%); 4) using relative risks for all diseases with age-collapsed prevalence: 49.7% (range, 24.1% to 82.2%); and 5) using prevalence estimates unadjusted for exposure-outcome lag: -14.5% (range, -20.8% to 42.6%) to -1.4% (range, -0.8% to -2.7%), depending on the method of adjustment. Applications of the SAM estimation method should consider these sources of potential error