Comparative effectiveness analyses using retrospective databases may be highly sensitive to common design decisions employed by researchers.
To test the sensitivity of statistical results to common research methods in retrospective database analyses. Comparisons of time to all-cause discontinuation (TTAD) across antipsychotic drug therapies are used to illustrate these effects.
Data from the California Medicaid Program were used to identify 231,635 episodes of antipsychotic drug therapy. Four sequential analyses of TTAD were performed on all patients, patients with 1 year of post-treatment data, and patients with schizophrenia and using models that included variables documenting drug treatment history.
Patients using atypical antipsychotics consistently achieve longer TTAD than patients treated with conventional antipsychotics. Nevertheless, estimated differences narrowed when analyses included only patients with schizophrenia. Risperidone performed better than olanzapine when diagnosis was not limited to schizophrenia, and quetiapine outperformed olanzapine and risperidone when the analysis did not control for treatment history. This latter result reflects the disproportionate use of quetiapine in long-duration augmentation episodes. There were no statistical differences across alternative atypical antipsychotics once the analysis excluded patients without a diagnosis of schizophrenia and included patient treatment history in the analysis.
Comparative effectiveness analyses of alternative drug therapies are sensitive to diagnosis and patient drug treatment history. Data on these factors can be derived from paid claims data and should be used to provide more accurate comparisons of effectiveness across drugs and to provide results that cover the full range of clinical scenarios that clinicians face.