PEFR is expressed as percentage of a patient’s predicted value, based on age, gender, and height. Changes in PEFR are expressed as the absolute change in percent predicted (ie, final PEFR as percent predicted minus initial PEFR as percent predicted). Asthma-related ED visits and urgent clinic visits over the past year were treated as continuous variables. A high percentage of subjects had no hospital admissions for asthma over the past year (74%) and this was therefore treated as a dichotomous variable.

Triggers of the patient’s asthma, in general, were assessed using a standardized list of potential triggers: respiratory tract infections, environmental allergens, other environmental factors, tobacco, exercise, ingested substances, reproductive, psychosocial, and other factors Online pharmacy viagra. The individual triggers as well as total number reported (range, 0 to 9) were recorded. Relapse was defined as any urgent or unscheduled visit to any physician for worsening asthma symptoms during the 2-week follow-up period.

All analyses were performed using software. Data are summarized using proportions, mean ± SD, and median with interquartile range (IQR). Univariate analyses of the relation of various factors to risk for relapse employed test, Student’s t test, and Wilcoxon rank sum test where appropriate. Variables that were associated with relapse at a two-tailed p < 0.1 in univariate analysis were evaluated for inclusion in a multivariate logistic regression model. This model was built with both forward and backward steps but was not done using the stepwise software function. Initially, variables were grouped into categories and assessed for colinearity by Spearman correlation and simultaneous inclusion in logistic regression models.

Variables that were independently associated with relapse in these initial models were included in the model building process. When groups of similar variables showed significant colinearity, with no single variable attaining statistical significance in the initial logistic regression model, a representative variable from each group that showed the strongest association with relapse was chosen for inclusion in the model building process. The final model included all independently associated variables as well as age, gender, and race, which were chosen for their clinical significance. The possibility of a period effect was examined by adjusting for period of enrollment, but this did not materially.