A Comparison of Two Analytic Approaches for Identifying Risk for Suicide Among High Risk Veterans

The FDA has suggested closer clinical monitoring following antidepressant initiation to reduce suicide risks; however, few data support this recommendation. Only large observational studies have sufficient sample sizes to examine the relationship between more intensive monitoring and completed suicide, but such studies are often affected by confounding. We compared two analytic approaches for examining the relationship between suicide and monitoring during high-risk periods. Data were obtained from Veterans Health Administration patients receiving depression treatment between 1999 and 2004 (N=887,859). We used a traditional case-control (CC) design to assess the relationship between monitoring and suicide, matching cases with controls by age, gender, time since cohort entry, and number of high-risk days in the prior year (1,032 cases and 2,058 controls). We also used an instrumental variable (IV) approach where each patient in the cohort (N=714,106) was assigned an IV based on their geographic access to or use of VA facilities with different monitoring practices. The CC design indicated a modest increase in risk with each additional monitoring visit during the high-risk periods (OR=1.02; 95% CI=1.002, 1.04). The IV estimate using facility level monitoring indicated a 0.0002% increase (p=0.97), and the IV estimate using distance inverse-weighted facility practice pattern indicated a 0.012% decrease (p=0.14) for each additional visit. Case-control and IV approaches provided different point estimates of the relationship between monitoring and suicide, suggesting the importance of using alternative approaches to address confounding. None of the study analyses indicated that closer monitoring during high-risk periods was significantly associated with reduced suicide risks.