Evaluation of three statistical estimation procedures in the assessment of the association between the fda suicidality warning and antiepileptic drug prescription claims in a state medicaid program

Introduction: In January 2008, the Food and Drug Administration (FDA) communicated concerns and later in May 2009, issued a warning about an increased risk of suicidality (suicidal ideation and behavior) related to all antiepileptic drugs (AEDs) used to treat epilepsy or psychiatric disorders. It is common to evaluate the impact of FDA risk communications on medication prescribing; often different analytical approaches have been used. However, evaluating the outcomes of a FDA risk communication with different estimation approaches may yield very different results, even when identical datasets are used. As a result, health service researchers may potentially reach very conflicting conclusions.;Objectives: This study evaluated three statistical estimation procedures for assessing the association between AED prescription claims and the aforementioned FDA suicidality warning; communicated in January 2008 and implemented in May 2009. The purpose of this study was to compare and contrast estimates derived from three different techniques to determine whether the different procedures provide consistent and/or inconsistent estimates and inferences.;Methods: A longitudinal interrupted design was utilized to evaluate Oklahoma Medicaid claims data from January 2006 through December 2009. The study included 9,289 continuously eligible individuals with a diagnosis of epilepsy and/or psychiatric disorder and at least one AED prescription claim. Trends, expressed as monthly changes in the log odds of AED prescription claims, were compared among three time periods: before the FDA warning (January 2006 to January 2008); during the FDA warning (February 2008 to May 2009); and after the FDA warning (June 2009 to December 2009). Three estimation procedures [generalized linear model (GLM), generalized estimation equations (GEE), and generalized linear mixed model (GLMM)] were used to estimate trends in log odds of AED prescription claims while adjusting for covariates. Data management and analysis employed PC-SAS (v9.2), with the a-priori alpha set at 0.01.;Results: All three statistical procedures estimated an increasing trend (p<0.0001) in log odds of AED prescription claims before the FDA warning period. None of the procedures detected a significant change in estimated trend of log odds of AED prescription claims during the FDA warning period (GLM: -30.0%, 99% CI: -60.0% to 10.0%, p=0.0446; GEE: -20.0%, 99% CI:-70.0% to 30.0%, p=0.3472; GLMM: -23.5%, 99% CI: -58.8% to 1.2%, p=0.0136) and after the FDA warning period (GLM: 50.0%, 99% CI: -70.0% to 160.0%, p=0.2869; GEE: 80.0%, 99% CI: -20.0% to 200.0%, p=0.0323; GLMM: 47.1%, 99% CI: -41.2% to 135.3%, p=0.1725) when compared to pre-warning period.;Conclusions: Although three estimation procedures provide consistent inferences, the magnitude of point estimates and precision of confidence interval varies. Among three different statistical estimation procedures, GEE appears to produce more robust and valid estimates for the research question studied because it accounts for autocorrelation between repeated observations and addresses scientific questions that are concerned to make inferences at the population level.