Methods: We used MedStat Marketscan data for patients identified with a diagnosis of Crohn's disease (ICD9=555.*). Patient's who received any infliximab in 2001 were identified as the treatment group. This resulted in 205 patients in the treatment group and 7,577 patients in the control group. Resource utilization was measured for the years 2002 and 2003 as the total number of hospitalizations, number of emergency room visits, and the number of outpatient visits. In this study, we compared the results for the following propensity score methods: propensity-score matching, weighting using propensity scores, covariate adjustment using propensity score and standard regression analyses. We further compared the results from two alternative propensity-score matching methods: the nearest neighbor matching and kernel matching method. A non-parametric bootstrap was performed to calculate the confidence interval around the treatment effect for the matching methods. We used zero-inflated Poisson (ZIP) regression models for the total number of hospitalizations and number of emergency room visits and negative binomial regression models for the number of outpatient visits.
Results: The average treatment effect from the standard regression models estimated a difference of 0.17 hospitalizations, 0.59 emergency room visits, and 5.75 outpatient visits between the infliximab and non-infliximab group. However, the alternative propensity score methods estimated varying results. Average treatment effect for propensity score adjusted number of hospitalizations ranged from 0.14-1.43. Similarly, the average treatment effect for number of emergency room visits ranged from 0.10-1.85 and for number of outpatient visits ranged from 5.75-12.96.
Conclusion: We find that patients treated with infliximab had significantly higher number of hospitalizations, emergency room visits, and outpatient visits compared to patients with Crohn's disease that were not treated with infliximab. We also find that the results after propensity score adjustment varied greatly. Our findings suggest that caution needs to be exercised when interpreting results of these types of studies as causal.