First, the frequency of achieving primary and secondary asthma outcomes (ie, asthma exacerbations, at least two rescue inhalers in 4 months, returning to or exceeding baseline asthma controller medication use) over the 24 months following the step-down event was calculated for the entire cohort, as well as according to length of stability prior to step-down. Then, a time-to-event analysis was conducted with time 0 defined as the end of the 4-month interval during which the step-down occurred and failure defined by first asthma exacerbation. To determine the most appropriate time-to-event model, we first estimated a Cox proportional hazards model including only the stability group and rejected the proportional hazards assumption based on Schoenfeld residuals.12 We then compared parametric time-to-event models with exponential, Gompertz, log-logistic, Weibull, log-normal, and γ distributions, again including only the stability group, and compared the models using Akaike information criterion, Bayesian information criterion, and log likelihood; based on all three statistics, the γ distribution was selected as the most appropriate for time-to-asthma exacerbation. We then estimated a separate time-to-event parametric model with a γ distribution for three outcome categories: asthma exacerbation (primary outcome), inpatient or ED visit for asthma, or meeting any possible of the primary or secondary outcomes. In addition to prior stability, each model included age category, race, sex, season of step-down, Charlson comorbidity index, and number of days of asthma controller mediation dispensed in the pre-step-down period. All analyses were performed using Stata 12.1 (StataCorp LP) and SAS 9.3 (SAS Institute Inc).