| The principal aims of this study were as follows: (i) To determine factors responsible for channeling of patients to the COX-2 specific inhibitors versus NSAIDs, (ii) To assess the impact of channeling bias on the estimates of treatment with respect to gastrointestinal (GI) outcomes, and (iii) To explore the effectiveness of propensity scores analysis in eliminating the effects of channeling.; This study employed a longitudinal, retrospective, cohort type design, using claims data from a large US Managed Care plan. Data were collected separately across three different periods to operationalize the effect of various periods of information on treatment choice. Cox proportional hazards models were used to estimate the hazard of two types of GI events as a function of treatment. Two-part models were used to estimate the impact of treatment on GI costs. Propensity scores analysis was used to correct for the channeling bias. In each time period, the treatment estimates from the bias uncorrected and bias corrected models were compared qualitatively to assess the extent to which channeling impacted the estimates.; Across the three timeperiods, preferential prescription of the COX-2 specific inhibitors seemed to be influenced by the patient characteristics like age, prior history of GI, renal or cardiovascular events, prior use of gastroprotective agents, anti-coagulants, or corticosteroids, and arthritic condition. Regardless of controlling for channeling bias, the Hazard of both types of GI events remained higher for the Coxib cohort. Differences between the predicted average GI costs for the two cohorts were lower in models employing the matched samples (ranging from {dollar}5–{dollar}18) as compared to models using the unmatched samples (ranging from {dollar}44–{dollar}62).; Overall, as a result of channeling, the two cohorts were severely imbalanced on several confounding covariates. The utility of propensity scoring techniques in correcting for the channeling bias was mixed. Multiple regression techniques without any formal selection correction were sufficient for mitigating channeling bias in models assessing the hazard of GI events. However, the propensity score matching procedure was more efficient in correcting channeling bias in the two part models estimating GI costs. |
