- Design
- territory-wide retrospective cohort with inverse probability of treatment weighting, plus pairwise and network meta-analysis
- Population
- 1,379 lithium and 5,556 valproate monotherapy initiators with bipolar disorder, Hong Kong 2003-2023, plus 12 observational studies
- Primary outcome
- suicide
- Effect
- cohort weighted HR 0.76 (95% CI 0.54-1.08); network lithium 0.46 (0.33-0.64) and valproate 0.63 (0.45-0.88) vs no treatment
Electronic health records from the Hong Kong Hospital Authority identified patients with bipolar disorder starting lithium or valproate monotherapy between 2003 and 2023 — 1,379 on lithium and 5,556 on valproate. Inverse probability of treatment weighting balanced baseline characteristics, and weighted Cox regression estimated suicide risk. The same team then searched for trials and observational studies comparing the two agents and no treatment, and combined them pairwise and in a network.
The cohort showed no significant difference (weighted HR 0.76, 95% CI 0.54-1.08). Pairwise pooling of eight observational studies agreed (HR 0.87, 0.69-1.09). In the network analysis both drugs were associated with lower suicide risk than no treatment — lithium HR 0.46 (0.33-0.64) and valproate 0.63 (0.45-0.88) — and the head-to-head estimate again showed no separation (0.73, 0.52-1.02). The three randomised trials found could not be pooled because their methods were incompatible.
The practical consequence is a narrowing of what the choice turns on. Lithium's reputation as the anti-suicide agent has, in some clinics, been the argument that overrode tolerability concerns, monitoring burden and renal or thyroid risk. On this evidence that argument is weaker than it was: the difference between the two, if there is one, was not detectable in a 20-year territory-wide cohort or in pooled observational data.
What survives intact is the comparison that matters most. Both agents were associated with substantially lower suicide risk than being on neither — so the decision to treat is far more consequential than the decision of which. Valproate's teratogenicity is a hard constraint in anyone who could become pregnant, and regulatory restrictions on its use in that group have not changed on the basis of this study.
- Base the choice between lithium and valproate on tolerability, comorbidity and monitoring feasibility rather than on a presumed suicide-prevention edge
- Treat being on neither agent as the position to avoid — both were associated with substantially lower suicide risk than no treatment
- Do not use this to reopen valproate in anyone who could become pregnant; the teratogenic restrictions are unchanged
- Where lithium is poorly tolerated or monitoring is impractical, discuss valproate as a reasonable alternative rather than a compromise
- Continue structured suicide risk assessment regardless of which mood stabiliser is chosen — neither drug replaces it
Why it matters
If lithium carries no suicide-prevention edge over valproate, its monitoring burden has to be justified on other grounds.
Don't overread it
These are observational comparisons with wide intervals; failing to detect a difference is not the same as showing there is none.
The statistics, in plain English
Every head-to-head estimate here crosses 1.0 — the cohort's 0.54 to 1.08, the pairwise 0.69 to 1.09, the network's 0.52 to 1.02. That means each is consistent with either drug being modestly better, so no difference was established; it does not prove the two are identical, and a real advantage of the size these intervals still permit could be clinically meaningful. The comparisons with no treatment, by contrast, sit entirely below 1.0. All of this is observational, and weighting can balance recorded characteristics but not the reasons a clinician chose one drug over the other.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for psychiatry, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free