To the Editor Mansi and colleagues1 have published an analysis of musculoskeletal conditions, arthropathies, and injuries among users and nonusers of statins. They reported higher adjusted odds ratios (ORs) for statin users in all outcome groups. Are their findings clinically significant?
The OR is (Ps/1 − Ps)/(Pn/1 − Pn), where Ps is the probability of adverse outcomes in statin users and Pn is the probability in nonusers. The OR is not easy to interpret. More desirable for interpretative purposes is the relative risk (RR), Ps/Pn. It may not be widely known to nonepidemiologists that the OR is a good approximation to the RR when the outcome is “rare” (<10%), but that the OR greatly overestimates the RR when the outcome is common.2 Table 4 in the article by Mansi et al1 showed that 86.9% of statin users and 84.8% of nonusers had an adverse event (there is an error in the table, the number among nonusers should be 5905). This difference in incidence of 2% translates into an OR of 1.19, which “looks” much larger. Zhang and Yu3 have published a method to convert logistic regression ORs into RRs. I applied this method and calculated that the RR for “all musculoskeletal diseases” among statin users, compared with nonusers was 1.02 (95% CI, 1.01-1.04). The risk of adverse events among statin users was thus 2% higher than among nonusers. When presented in this manner, one wonders whether this small risk difference might be attributable to residual confounding not eliminated by the propensity score matching and whether this difference, if real, is clinically significant. It is true that, in the large population of statin users, small differences in risk translate into large numbers of affected individuals, but I think that stronger evidence is required before public health policy can be modified.