Table 3– Intention-to-treat analysis of mean change (improvement) in transitional dyspnoea index (TDI) focal score according to concomitant treatment with a long-acting β₂-agonist (LABA) and frequency of previous exacerbations

	Mean TDI focal score		Difference score (95% CI)	Two-sided p-value
	Roflumilast	Placebo	Difference score (95% CI)	Two-sided p-value
All patients	0.662 (n=1470)	0.409 (n=1514)	0.253 (0.104–0.402)	0.0009
LABA use^#
With LABA	0.448 (n=718)	0.159 (n=775)	0.289 (0.081–0.497)	0.0066
Without LABA	0.904 (n=752)	0.680 (n=739)	0.224 (0.002–0.446)	0.0477
Previous exacerbations^#
Infrequent	0.604 (n=1070)	0.451 (n=1105)	0.153 (-0.026–0.332)	0.0948
Frequent	0.698 (n=400)	0.201 (n=409)	0.497 (0.207–0.786)	0.0008

TDI is graded from 0–4, where 0 is most severe; a change in TDI score of 1.0 unit is considered clinically relevant. Least square means, 95% CI and p-values are based on a repeated measurements ANCOVA model with the following factors and covariates: treatment, baseline value, age, sex, smoking status, time, treatment-by-time interaction, study and country pool (only for the overall population). For all groups apart from LABA usage, concomitant treatment with LABA was also included as a factor. Exacerbations in the previous year (based on patient recall) are defined as infrequent if fewer than two and frequent if two or more exacerbations. ^#: post hoc analyses.