Change in FEV1 (Percent of predicted) from baseline to week 12. FEV1 was measured only in children between 6 to \<12 years only. Least Squares Mean were adjusted for Baseline FEV1, age \[yrs\], pooled center, previous corticosteroid therapy and holding chamber.

PEF is the maximum speed of expiration. A portable electronic PEF meter was used for the home PEF readings. The patients recorded PEF daily, in the morning immediately after getting up. Readings were done preferably at least 4 hours after use of rescue medication and before inhalation of the study medication. At each measurement, three readings were obtained in the standing position. All three values were recorded in the diary; the highest value was used for evaluation. The higher change from Baseline values are the best. Analysis of covariance (ANCOVA) model with the baseline value and age as covariates was used for analysis. Last observation carried forward.

Mean change of Provocative concentration of Adenosine-5'-monophosphate (PC20 AMP) leading to a 20 percent decrease in Forced expiratory volume in one second (FEV1) between post-treatment and baseline using two different particle sizes. - Small particles = Mass mean aerodynamic diameter (MMAD) of approximately 1.04-1.08 micron - Large particles = MMAD of approximately 9.9-10.6 micron

Time to first asthma exacerbation is defined as the time in days until the first asthma exacerbation, or to the end of treatment visit. In the absence of an exacerbation, an early treatment discontinuation is treated as a censored observation on the day following the last use of study drug.

A model-based analysis of asthma exacerbation was performed to adjust to important covariables. The distribution of the data suggested a Poisson regression modeling (zero inflated) strategy. After a variable selection process considering also variable-by-treatment interactions, the variables centre, age \[years\] and race were identified to be important beside treatment. The parameters centre and age \[years\] were allocated to zero-model part and the variables treatment and race to the Poisson model part. The estimates of the per-treatment rates are based on a negative-binomial distribution.