Background: Haloperidol remains the default antipsychotic for hospital delirium in much of the world, yet prescribing practice and measured outcomes have rarely been described together in Latin American hospitals.

Methods: Retrospective cohort of consecutive adults with delirium seen by consultation-liaison psychiatry at a Mexico City tertiary referral centre (March 2021 to August 2022). Severity was the summed CAM-S short-form score (0–7) at each psychiatric visit; each patient contributed one index antipsychotic regimen. We modelled CAM-S with linear mixed models and estimated cumulative incidence of first resolution (CAM-S = 0) with the Aalen–Johansen estimator, treating in-hospital death as a competing event. No comparative-effectiveness hypothesis was prespecified.

Results: Of 195 screened records, 136 patients contributing 724 assessments were analysed; haloperidol accounted for 115 regimens (84.6%). CAM-S fell by 0.38 points per day (95% CI 0.30–0.45; p < 0.001), from 3.79 to 2.21. Adjusted for baseline severity, no agent differed from haloperidol (olanzapine −0.50, 95% CI −1.04 to 0.05; risperidone −0.23, −0.69 to 0.24). Cumulative incidence of first resolution reached 38.3% (95% CI 27.1–49.4) by day 7 and 46.1% (31.8–59.2) by day 14. Baseline CAM-S was the only variable associated with resolution (hazard ratio 0.49 per point, 0.36–0.67). Physician-documented improvement at discharge (77.9%) far exceeded instrument-based severity control (25.0%).

Conclusions: Prescribing was concentrated on haloperidol while measured severity improved independently of which agent was chosen. The gap between charted improvement and measured outcomes reflects two different constructs, not a treatment effect. These data describe local practice and cannot support causal comparison between agents.