A Methodological Approach to Adjust for Channeling Bias in Nonrandomized Observational Studies: Application to Cyclooxygenase-2 (COX-2) Specific Inhibitors

Abstract

Background: Channeling bias may confound the interpretation of results of nonrandomized observational studies. We undertook this analysis to develop a methodology for examining potential channeling bias in the prescription of cyclooxygenase-2 specific inhibitors (COX-2 SIs) and to determine the effect of such bias on measured estimates of drug safety. Methods: The Protocare Sciences Propriety Research database, a managed care claims database that has integrated information on patient eligibility and claims, was used as the data source. Patients were identified based on dispensation of either COX-2 SIs (n 43,653) or nonspecific (ns) NSAIDs (n 492,831) between January 1, 1999 and December 31, 2001. Odds ratios were calculated to determine the association of health status and gastrointestinal (GI) history with COX-2 SI and ns NSAID dispensation. Propensity scores were derived from multivariate logistic regression analyses to adjust for risk factors associated with COX-2 SI and ns NSAID dispensation. Poisson regression accounting for duration of exposure was used in conjunction with propensity scores to examine the incidence of NSAID-related and ulcer specific hospitalizations during therapy. Results: Among patients with a history of GI hospitalization, the crude incidence rates were approximately 4-fold higher in the ns NSAID compared with the COX-2 SI cohort (incidence rate/100 pt-years: 4.06, 95% CI 3.10–5.32 and 0.88, 95% CI 0.33–2.34, respectively). Disease state severity based on advancing age and history of GI healthcare resource utilization or NSAID intolerance were strongly associated with COX-2 SI dispensation. After adjusting for these predictors of COX-2 prescribing or factors associated with selective COX-2 prescribing (channeling bias), the relative risk (RR) of both NSAID-related and ulcer-specific GI hospitalizations was reduced among COX-2 SI users compared with ns NSAID users (RR 0.61, 95% CI 0.48–0.78 and 0.68, 95% CI 0.48–0.96, respectively). Similar results were obtained regardless of exposure to gastroprotective agents (H2RAs, PPIs, or misoprostol) (RR 0.55, 95% CI 0.40–0.77 and 0.52, 95% CI 0.31–0.87 for NSAID-related and ulcer-specific hospitalizations, respectively). Conclusions: Managed care patients with a history of GI problems or ns NSAID intolerance are being channeled to COX-2 SIs. After adjusting for channeling bias, COX-2 SI therapy was associated with fewer GI events, even though these patients had a higher background risk of GI complications. 4 Flagellin Stimulates IL-8 Production in Native Human Colon and Enhances Ongoing Colonic Inflammation in Mice Sang Hoon Rhee, Martin Riegler, and Charalabos Pothoulakis