Descriptive statistics were examined for each study variable. A natural log transformation was applied when the outcome measures were skewed, (eg, for the CCR). When the skew was more extreme, nonparametric methods were used. Because of the highly skewed distributions of the number of asthma visits, children were classified in each of the 2 years (pre-enrollment and postenrollment) as having more than one visit per year or fewer than or equal to one visit per year (one being the minimum number, pre-enrollment, for eligibility). The proportions of children with higher utilization then were compared between the intervention and control groups using logistic regression.
Because of the seasonal variability in asthma and the correspondence of the 12-month active follow-up with the end point of the year-long health-care utilization follow-up, the analyses focused on the 12-month follow-up. Study hypotheses were tested using logistic regression (dichotomous variables) or t tests (continuous variables), with and without adjustment/control for baseline differences on the outcome variable. All analyses were performed by intent to treat and, except as noted below, were performed using computer software (SAS, version 6.12; SAS Institute; Cary, NC). The analysis used all cases for which a particular outcome variable was available. In addition, the analyses of all the outcome measures were repeated for the subset of children for whom cotinine data were available at 12 months. my canadian pharmacy.com
Attrition rates on the cotinine data were equivalent in the intervention and control groups (see “Results” section).
Asymptotic tests of the significance of treatment effects can be unreliable in logistic regression models that adjust for an influential baseline covariate. We used the nonparametric bootstrap estimates of confidence intervals by Efron and Tibshirani to check the results. The bootstrap analyses were performed with a general nonparametric bootstrapping routine (S-Plus, version 4.0; MathSoft; Cambridge, MA).