Fig. 2

Time trend in annualized exacerbation rates based on the Bayesian regression analysis (without adjustment for study or baseline characteristics). The dashed line shows the estimated rate over time along with a 95% credible region. The points and vertical whiskers indicate each individual study’s estimated rate along with a 95% credible interval. The shown (“shrinkage”) estimates are based not only on the corresponding study’s provided data (which in some cases may not be sufficient to derive a rate estimate), but these are supported by the complete data set. Different types of data sources are indicated by different symbols (some references directly provide rate estimates along with standard errors, some report the total number of exacerbations and the number of exacerbation-free patients, and some only one of the two). Within each publication year, estimates are evenly distributed so that they are identifiable by their respective index (see axis at top and Additional file 1: Table S1). Inserted below are two smaller figures illustrating the time trend for the two subsets of “true” placebos (21 studies) and ICS-placebos (34 studies)