Meta-analytic evidence of systematic bias in estimates of neuroleptic malignant syndrome incidence.

OBJECTIVE: The aim of this study was to examine published reports for sources of excessive variance in neuroleptic malignant syndrome (NMS) incidence estimates. DATA SOURCES: An unrestricted computerized MEDLINE search was conducted with a comprehensive search logic and supplemented by secondary references and a manual search of an extensive personal library. STUDY SELECTION: Studies were analyzed if they presented original data and provided at least 2 of the following: number of NMS cases, number of patients at risk, or ratio of cases to patients at risk. Twenty-six of the 28 candidate studies met these minimal criteria. DATA EXTRACTION: Variables included incidence, year of study publication, mean year of NMS occurrence, patient population at risk, study design, diagnostic criteria, and country of origin. DATA SYNTHESIS: Standard error, which reflects study size, accounted for 90.8% of the variance (beta = .953, P < .001) in this international series of 26 NMS incidence estimates. Incidence was significantly lower in 7 studies the time end points of which were set in advance of case identification (chi(2) = 71.08, P < .001). No other variable was significantly related to incidence.
CONCLUSIONS: Neuroleptic malignant syndrome incidence estimates to date are non-trivially biased such that larger study size (patients at risk) is strongly related to lower observed incidence. Future studies can minimize the contribution of this and other sources of experimental error by incorporating several very feasible recommendations.