Table 2 Comparison of three methods for estimating smoking-attributable fraction and mortality

Method comparison

Multivariable models relating exposure (and covariates) to outcome are created, then applied to current exposure data in the target population to predict total burden. The models may be created from earlier years of data from the same target population.

Rate of outcome in the total population is compared to the rate in the unexposed population to estimate the contribution of exposure to excess outcome.

Prevalence of exposure in the target population is combined with hazards relating exposure to outcome from an etiologic study. This is done to estimate proportion of burden attributable to the exposure in the population.

The models are applied to a counterfactual population where no one is exposed

(AF_p) = (I_t-I_u)/I_t, AF_p is multiplied by total outcome count (see text)

AF_p = [P_e(RR-1)]/[1 + P_e(RR-1)], AF_p multiplied by total outcome count (see text)

Population-based, routinely collected data on health outcomes that are linked at the individual level to exposure data, often from health surveys.

Not commonly available at the population level. Cohort studies, disease registries, or exposure data linked to outcome.

Ecological, summary measures of: prevalence from health surveys, hazards from the literature, and outcome counts from routinely collected data.

Study data sources

Canadian Community Health Survey (CCHS) 4.1

Not used

CCHS 4.1

CCHS 1.1 to 3.1 linked to death database

Not used

Cancer Prevention Study II, 2014 Surgeon General’s Report [8]

Predicted by algorithm

CCHS 1.1 to 3.1 linked to death database

Death database (RPDB)

Smoking-attributable fraction/deaths, 2009–2010

Males

26.1%

36.8%

24.1%

11 332

15 998

10 648

Females

21.4%

33.9%

15.8%

9 285

14 713

6 928

23.7%

35.4%

20.0%

20 573

30 711

17 576