Low-dose lithium vs valproate and incident dementia diagnoses in older adults: A propensity-matched cohort study using electronic health record data

Background: Whether low-dose lithium lowers dementia incidence in older adults, relative to an active comparator, is unestablished in real-world data. We compared incident dementia in older adults receiving low-dose lithium versus valproate using a target trial emulation of multicenter electronic health record data. Methods: Adults aged ≥60 years receiving lithium carbonate (all serum measurements ≤0.40 mmol/L) or valproate between 2015 and 2025 in the TriNetX network, free of prior dementia, mild cognitive impairment, bipolar disorder, Parkinson's disease, or related neurological conditions, were eligible. Cohorts were balanced by 1:1 propensity score matching on 23 covariates. The primary outcome was incident all-cause dementia at 730 days; secondary outcomes were mortality and hospitalization. Cause-specific hazard ratios (HRs), Aalen-Johansen cumulative incidence, and an E-value were computed, with prespecified sensitivity analyses. Results: After matching, 853 patients per cohort remained (all standardized mean differences <0.10). All-cause dementia occurred in 25/795 (3.1%) lithium versus 69/832 (8.3%) valproate users (HR, 0.38; 95% CI, 0.24–0.60; P < .001). Mortality was lower with lithium (HR, 0.56; 95% CI, 0.39–0.80), whereas hospitalization did not differ (HR, 0.81; 95% CI, 0.58–1.13). The E-value was 4.69 (2.71 for the upper bound). Sensitivity analyses (90-day lag, 0.36; 180-day lag, 0.40; non-AD subtype, 0.23; 3-year follow-up, 0.47) were directionally consistent. Conclusion: Low-dose lithium was associated with a lower observed incidence of dementia diagnoses than valproate. These observational findings cannot distinguish a direct effect from prescribing or surveillance bias, depletion of susceptibles, or differential healthcare engagement; they are hypothesis-generating, and randomized trials are required before causal interpretation.