Attributable Fractions

Measures of Effect and Counterfactuals

Much of epidemiologic research is aimed at making inferences about the net (causal) effect of one or more exposures on disease occurrence in a population. The dominant paradigm for defining such effects is the counterfactual or potential-outcomes model in which we contrast the frequency of disease (usually incidence) in the population under two or more exposure conditions such as everyone being exposed versus everyone being unexposed. Since individuals cannot be both exposed and unexposed simultaneously, at least one of those conditions is counter to fact or counterfactual. For example, suppose a group of Ne exposed persons at risk are followed without [Page 56]loss for a given period during which A new (incident) cases of the disease occur. Thus, therisk(Re)ofdisease in this exposed group is A=Ne. Thecentralcausal question is how many cases would have occurred during that period if no one in that population had been exposed. Suppose that counterfactual number is A0. Therefore, the counterfactual risk in the exposed group is A0/Ne, and the causal risk ratio (RRe), a measure of the exposure effect, is (A/Ne)= (A/0=Ne) = A= A0: If RRe > 1, the exposure is called acausalorpositive risk factor for the disease in that population; if RRe < 1, the exposure is called a protective or negative risk factor. It should be noted that without knowledge of biological mechanisms, this distinction between a causal and protective risk factor is generally arbitrary since exposure statuses can be reversed. For example, inferring that an active lifestyle is a protective risk factor for coronary heart disease is equivalent to inferring that a sedentary lifestyle is a causal risk factor.

...