In developing our final model determining if statins were associated with a reduced risk for fractures, we included factors such as advanced age, female sex, and prednisone use because there was substantial evidence in the literature suggesting their association with fracture risk. For all other potential confounders, we first assessed whether the exposure of interest was associated with both statin use and the outcome in univariate models. Hydrochlorothiazide use was significantly associated with statin use but not with fracture after adjustment for duration of use. Therefore, we did not include hydrochlorothiazide use in our final model.
Scranton RE, Lawler E, Young M, Gaziano JM. Statins and Fracture: Are All Variables Accounted For?—Reply. Arch Intern Med. 2006;166(9):1041–1042. doi:10.1001/archinte.166.9.1041-b