Patient selection factors in angiographic studies: a conceptual formulation and empirical test.

Studies using coronary angiography patients as the study population to assess etiologic relationships are vulnerable to bias produced by patient selection factors, but the direction and magnitude of the bias in any particular study are uncertain. Partly for this reason, some authors have questioned the use of angiography data for studies of etiology. Nevertheless, this type of data continues to be used in etiologic studies. Therefore, in this paper a conceptual framework is described to quantify the degree of bias under different assumptions about the process whereby patients are selected for angiography. The direction and degree of bias depend on the pattern of patient selection, as well as the measure of association used (e.g., the risk ratio or odds ratio). One method is then described, based on assessing how physicians refer patients to angiography, which estimates the degree of bias produced by selection factors. This study is limited to the situation of a two-level risk factor (e.g., type A/B behavior) in relation to a two-level outcome measure (e.g., significant coronary artery occlusion versus no significant coronary artery occlusion). While type A/B behavior is used as an example in this paper, the concepts apply to any behavioral or nonbehavioral predictor variable.