Dichotomous data were collected for both groups regarding use of PPIs, use of antibiotics within the 30 days preceding diagnosis of the CDAD infection, diagnosis of diabetes mellitus, use of enteral feeding within the 30 days preceding the infection, diagnosis of cancer, previous gastrointestinal surgery, diagnosis of inflammatory bowel disease, and previous residence or care in an institution. Oral lansoprazole and IV pantoprazole were the formulary PPIs at the study institution. However, patients might also have been taking omeprazole, esomeprazole, or rabeprazole during or before admission. The term “antibiotic” was chosen to represent all antimicrobials, classified as follows: cephalosporins, clindamycin, fluoroquinolones, penicillins, and other. Previous residence or care in an institution included admission from another hospital or health care centre, long-term care centre, or nursing home.
The sample size was calculated a priori by estimating the statistical power for the primary objective. Assuming 80% power with 5% error, 139 patients were required to fulfill the estimated rate of exposure of 20%. One hundred and fifty patients with CDAD were included to account for potential mismatching of cases and controls. Each case patient had to be matched to a control for all 4 matching criteria, to minimize potential confounding and selection bias. Approval was received from the Research and Ethics Board at Trillium Health Centre before study commencement. cialis canadian pharmacy
Selection bias was minimized by choosing control patients who had been admitted to the study institution, regardless of exposure to PPIs. Statisticians who were otherwise not involved with the study performed the statistical analysis on the matched pairs. A main effects forward stepwise logistic regression was performed, using SAS version 9.1 (SAS Institute, Cary, North Carolina) to determine odds ratios (ORs) and Wald confidence intervals (CIs) for each of the risk factors. Only the first-order terms for current and past use of PPIs, use of antibiotics, diabetes, enteral feeding, cancer, gastrointestinal surgery, inflammatory bowel disease, and previous residence or care in an institution were included in the analyses to ensure at least 10 observations per model variable, as suggested by Stokes and others (there was a maximum of 14 [142/10] variables that could be fitted in the model). For the forward stepwise regression, ap value < 0.10 for Wald’s test was used for entry of a variable, whereas a p value < 0.15 for Wald’s test was used for retention of variables in the model. The p values for inclusion or exclusion of variables were derived from Hosmer and Lemeshow, who stated that ap value of 0.15 to 0.20 is appropriate to ensure entry of variables with coefficients that are different from zero. In addition, a post hoc forward stepwise regression was conducted on the interactions between variables in the model that had the smallest p values in the model described above. We were also interested in determining if there were any interactions among the variables contributing to the model, to determine if one risk factor might have had an effect on another.