Background Blood eosinophil counts might predict response to inhaled corticosteroids (ICS) in patients with chronic obstructive pulmonary disease (COPD) and a history of exacerbations. We used data from the WISDOM trial to assess whether patients with COPD with higher blood eosinophil counts would be more likely to have exacerbations if ICS treatment was withdrawn. Methods WISDOM was a 12-month, randomised, parallel-group trial in which patients received 18 mu g tiotropium, 100 mu g salmeterol, and 1000 mu g fluticasone propionate daily for 6 weeks and were then randomly assigned (1:1) electronically to receive either continued or reduced ICS over 12 weeks. We did a post-hoc analysis after complete ICS withdrawal (months 3-12) to compare rate of exacerbations and time to exacerbation outcomes on the basis of blood eosinophil subgroups of increasing cutoff levels. The WISDOM trial is registered at ClinicalTrials.gov, number NCT00975195. Findings In the 2296 patients receiving treatment after ICS withdrawal, moderate or severe exacerbation rate was higher in the ICS-withdrawal group versus the ICS-continuation group in patients with eosinophil counts (out of total white blood cell count) of 2% or greater (rate ratio 1.22 [95% CI 1.02-1.48]), 4% or greater (1.63 [1.19-2.24]), and 5% or greater (1.82 [1.20-2.76]). The increase in exacerbation rate became more pronounced as the eosinophil cutoff level rose, with significant treatment-by-subgroup interaction reached for 4% and 5% only. Similar results were seen with eosinophil cutoffs of 300 cells per mu L and 400 cells per mu L, and mutually exclusive subgroups. Interpretation Blood eosinophil counts at screening were related to the exacerbation rate after complete ICS withdrawal in patients with severe to very severe COPD and a history of exacerbations. Our data suggest that counts of 4% or greater or 300 cells per mu L or more might identify a deleterious effect of ICS withdrawal, an effect not seen in most patients with eosinophil counts below these thresholds.