Antipsychotic drug utilization patterns and treatment-emergent diabetes: a methodological comparison of incidence using a claims database

The study purposes were to: 1) examine antipsychotic utilization trends; and 2) evaluate relationships between antipsychotic utilization and treatment-emergent diabetes when methodologies varied while controlling for covariates. Claims databases from North and South Texas Veterans Administration (1993-2004) were used to identify adults newly initiated on antipsychotics. For Purpose 2, only patients with no previous diabetes were eligible. Seven models were created based on the following methodological variations: 1) study designs (retrospective cohort and case-control); 2) treatment exposure assignment (intent-totreat (ITT) and as-treated (AT)); and 3) statistical approaches (propensity scoring, standard and conditional logistic regression, and Cox proportional hazards function). Covariates included: demographics, general health comorbidities, mental health comorbidities, drug utilization patterns, treatment duration, medication re-exposure, and treatment initiation year. Regarding Purpose 1, from 1997-2003, antipsychotic utilization shifted from first to second generation, with olanzapine and risperidone most frequently prescribed. Monotherapy was the predominant utilization pattern with switching and combination therapy used infrequently. Of the eligible patients (N=8,949) for Purpose 2, regardless of variations in methodologies of the seven tested models (eight models were proposed), there were no significant differences in diabetes risk among patients who were: 1) initiated on SGAs compared to those on FGAs; 2) initiated on olanzapine compared to those on risperidone; and 3) exposed to olanzapine or quetiapine compared to those exposed to FGAs. Inconsistent results among the seven models were observed when comparisons were made between: 1) patients initiated on quetiapine compared to those on risperidone; and 2) patients exposed to risperidone compared to those exposed to FGAs. Differences occurred among the following methods: ITT retrospective cohort and ITT case-control; and AT retrospective cohort and AT case-control. With respect to antipsychotic utilization, results of the various models using different methodologies were largely consistent.