Smith EG, Center for Healthcare Organization and Implementation Research, ENRM VAMC, Bedford, MA, and UMass Medical School Dept. of Psychiatry, Worcester, MA; Austin KL, Center for Clinical Management Research and VA Serious Mental Illness Treatment Resource and Evaluation Center, Ann Arbor, MI; Kim HM, Center for Clinical Management Research, Ann Arbor, MI, and University of Michigan Center for Statistical Consultation and Research, Ann Arbor, MI; Eisen SV, Center for Healthcare Organization and Implementation Research, ENRM VAMC, Bedford, MA, & Dept. of Health Policy & Mngt, Boston Univ Sch Pub Health; Kilbourne AM, Quality Enhancement Research Initiative (QUERI), Washington, DC, & University of Michigan Dept of Psychiatry; Miller DR, Center for Healthcare Organization and Implementation Research, ENRM VAMC, Bedford, MA, & Dept. of Health Policy & Mngt, Boston Univ Sch Pub Health; Christiansen CL, Center for Healthcare Organization and Implementation Research, ENRM VAMC, Bedford, MA, & Dept. of Health Policy & Mngt, Boston Univ Sch Pub Health; McCarthy JF, Center for Clinical Management Research, Ann Arbor, MI, VA Serious Mental Illness Treatment Resource & Evaluation Ctr, & UMichigan Dept of Psychiatry; Sauer BC, VA IDEAS2.0 Center, Salt Lake City, UT, & University of Utah Department of Internal Medicine; Valenstein M, Center for Clinical Management Research, Ann Arbor, MI, VA Serious Mental Illness Treatment Resource & Evaluation Ctr, & UMichigan Dept of Psychiatry
Objectives:
To determine if lithium is associated with lower risk of suicide death among Veterans Health Administration (VHA) patients with mental health diagnoses.
Methods:
Historical prospective cohort study (1999-2008) of 93,000 VHA patients with mental diagnoses newly initiating lithium or valproate. High-dimensional propensity score matching was used to balance the treatment groups on > 900 VHA database covariates. Hazards of suicide death (based on National Death Index files) over 365 days of treatment were estimated by stratified Cox regression, except for analyses of patients discontinuing treatment (non-stratified Cox regression).
Results:
No significant differences in risk of suicide death were observed over 365 days (lithium versus valproate intent-to-treat Hazard Ratio (HR) = 1.22 [95% Confidence Interval = 0.81-1.82]). Stratifying by diagnosis indicated that lithium was associated with greater intent-to-treat risks of suicide death among patients with bipolar disorder (HR = 1.50 [1.05-2.15]) but not among patients without bipolar disorder (HR = 0.77 [0.49-1.21]). The greater risks associated with lithium initiation among patients with bipolar disorder were restricted to the period after discontinuation of lithium, compared to valproate, treatment (0-365 day HR = 2.05 [0.88-4.79]); these post-discontinuation differences in risk were statistically significant over 0-180 days (HR = 6.01 [1.37-27.3]).
Implications:
In intent-to-treat analyses, lithium was associated with higher risk of suicide among patients with bipolar disorder, attributable to the risks observed after discontinuation. Such risks may indicate new-onset risks from discontinuation or confounding. Several lines of evidence suggest that any residual confounding biases against lithium. Therefore, benefits from lithium may be underestimated and risks upon discontinuation overestimated. Nevertheless, the magnitude of discontinuation-associated risks suggests that patients (especially those with bipolar disorder) should be educated about potential risks of lithium discontinuation, and closely monitored after discontinuation. This study also suggests that if significant benefits of lithium against suicide exist (concealed by confounding), from an intent-to-treat perspective those benefits will be greater among patients without bipolar disorder than with bipolar disorder.
Impacts:
These results have informed the VHA's suicide prevention efforts and contributed to the planning of a collaborative studies trial (CSP 590) examining lithium for suicide prevention.