Design: self-controlled case series using territory-wide electronic health records, conditional Poisson regression, 2004–2023
Population: 2,086 people with bipolar disorder in Hong Kong who received both long-acting injectable and oral antipsychotics, from a cohort of 17,841
Primary outcome: health care utilisation, illness relapse and safety during injectable versus oral treatment periods
Effect: psychiatric admission aIRR 0.67 (95% CI 0.61–0.74), manic-episode admission 0.51 (0.44–0.59), mixed-episode 0.31 (0.14–0.66); depressive-episode 1.34 (0.94–1.93); extrapyramidal symptoms 2.95 (1.40–6.22), not significant beyond 90 days

Depot antipsychotics are standard in schizophrenia and treated as an exception in bipolar disorder, largely for want of evidence. A self-controlled case series using twenty years of Hong Kong electronic health records provides some. Of 17,841 people with bipolar disorder, 2,086 had received both a long-acting injectable and an oral antipsychotic, allowing each person to serve as their own control across the two treatment periods.

Injectable periods carried lower rates of almost everything measured: all-cause emergency department attendance (adjusted incidence rate ratio 0.85, 95% confidence interval 0.81 to 0.89), all-cause admission (0.79, 0.73 to 0.85), psychiatric admission (0.67, 0.61 to 0.74), admission for a manic episode (0.51, 0.44 to 0.59) and for a mixed episode (0.31, 0.14 to 0.66). Admissions for depressive episodes did not differ (1.34, 0.94 to 1.93), and neither did non-psychiatric or cardiovascular admissions — which addresses the standing worry that depot exposure carries a metabolic price.

The safety signal is in the timing. Extrapyramidal symptoms were nearly three times more common on injectable treatment (2.95, 1.40 to 6.22), but the difference was no longer significant beyond 90 days. That is a specific instruction rather than a general caution: the first three months after starting a depot in bipolar disorder need active movement-disorder review, and the patient should be told what to look for before the second injection, not after it.

Consider a long-acting injectable in bipolar disorder with a mania-predominant course and repeated relapse, not only in schizophrenia.
Set the expectation honestly: the benefit is on manic and mixed relapse, not on depressive episodes.
Review for extrapyramidal symptoms at every contact in the first 90 days, and rate them formally if you can.
Tell the patient what stiffness, restlessness and tremor feel like before the second injection.
Do not use a metabolic or cardiovascular concern as the reason to avoid depot here — non-psychiatric and cardiovascular admissions did not rise.

Why it matters

It removes the main reason depot treatment is reserved for schizophrenia, and it tells you which relapses it will and will not prevent.

The statistics, in plain English

A self-controlled case series compares each person against themselves at a different time, which removes all fixed differences between people — genetics, baseline severity, socioeconomic position — and is the reason these estimates are more credible than a between-person comparison of who gets a depot. What it cannot remove is anything that changes with time within a person, and treatment order is exactly that: depots are usually started after a bad period, so some of the apparent improvement may be regression to the mean. The depressive-episode estimate of 1.34 has an interval from 0.94 to 1.93 that crosses 1.0, so the honest reading is no demonstrated effect, not a hint of harm. The extrapyramidal interval (1.40 to 6.22) is wide because the events are few — the direction is clear, the magnitude is not.