Problems With Odds Ratios

To the Editor: In the post hoc analysis section of his article, Dr. Bovasso identified the significant baseline covariates for the incidence of suicidal ideation. It is obvious that the risk set Dr. Bovasso used to calculate the incidence rate was the population that was free from a history of any of the nine depressive symptoms in the DSM-III depression module. However, the method for determining the risk set (denominator) of this analysis was not necessarily appropriate. Instead of excluding all of the Epidemiologic Catchment Area (ECA) participants with any baseline depressive symptoms from the post hoc analysis, only participants with a history of suicidal ideation should have been excluded. Under this definition, 1,708 ECA participants—instead of 849—would have been susceptible to the incidence of suicidal ideation. In an analysis by my colleagues and me (1), also of the Baltimore ECA sample, 89 participants reported new onset of suicidal ideation. Among all of the sociodemographic covariates, only age was significantly associated with the incidence of suicidal ideation (odds ratio=0.96, confidence interval [CI]=0.94–0.98; Wald χ²=16.2, df=1, p<0.001). Gender and race were not associated with suicidal ideation (1). However, our analysis did find that cannabis use at baseline was associated with the incidence of suicidal ideation; cannabis abusers were three times as likely to develop suicidal ideation as were nonabusers (odds ratio=3.00, CI=1.46–6.18; Wald χ²=8.9, df=1, p<0.01). Even after adjustment for the baseline diagnosis of any depressive episode, cannabis abuse remained a significant risk factor for new onset of suicidal ideation (odds ratio=3.14, CI=1.52–6.50; Wald χ²=9.6, df=1, p<0.01; data not published).