The data synthesis section of our protocol initially read as follows "we will report fixed‐effect rate ratios, such as the rate of the need for rescue systemic corticosteroids per person‐years of follow up in treatment and control groups". There were two changes made to this statement. The fixed‐effect model was used if there was no significant heterogeneity, otherwise the random‐effects model was used. As we could not obtain individual patient level data, event rates could not be calculated and therefore we defaulted to comparing frequency of events (number of patients requiring rescue systemic corticosteroids in treatment and control groups).

We did not discuss unit of analysis issues for parallel and cross‐over studies in the protocol. We did not pool parallel and cross‐over study results. For dichotomous outcomes, studies were pooled using M‐H odds ratios unless there was a low number of events reported requiring Peto odds ratios. Marginal odds ratio was obtained from the cross‐over trial by comparing the number of patients who needed OCS on increased dose (but not on placebo) with those who needed OCS on placebo (but not on double dose ICS).

For the primary outcome (Analysis 1.1), the denominator was changed from number of patients requiring the study inhaler to the number of patients randomised, consistent with ITT analysis. As a secondary outcome (Analysis 1.9), treatment failure was re‐analysed based on the number of patients requiring the study inhaler.

Magnitude of ICS dose increase (two‐fold versus four‐fold) was added as a subgroup analysis (Analysis 1.7). Post‐hoc subgroup analyses were also performed on the modified ITT analysis of the primary outcome (Analysis 1.10 to Analysis 1.15).

An additional review author, NS, was added to assist with selection of studies, data extraction and management and assessment of risk of bias in included studies.