Once-Daily Tiotropium Respimat® Add-on to at Least Ics Maintenance Therapy in Patients with Symptomatic Asthma: Methodology of Modeling Analyses By Serum IgE and Blood Eosinophil Levels
Monday, March 7, 2016
South Exhibit Hall H (Convention Center)
Hendrik Schmidt, PhD, Petra Moroni-Zentgraf, MD, Michael Engel, MD, Ronald Dahl, MD, Huib A.M. Kerstjens, MD
Rationale: Conventional subgroup analyses in late-phase clinical trials are often conducted by calculating relative treatment effects for each trial endpoint within subgroups defined by baseline characteristics.  For continuous variables, selection of cut-off values by which to define subgroups is a matter of discussion and convention.  To overcome this issue, modeling of tiotropium Respimat® treatment effects was performed for biomarkers, i.e. total serum IgE and blood eosinophils, in patients with moderate or severe symptomatic asthma.

Methods: In 4 Phase III randomized, double-blind, placebo-controlled studies of tiotropium Respimat® add-on therapy in patients with moderate (2 replicate trials: NCT01172808/NCT01172821) or severe (2 replicate trials: NCT00772538/NCT00776984) asthma, total serum IgE and blood eosinophil data were collected at screening.  Modeling was applied to pre-defined trial endpoints: asthma exacerbations (Cox regression); patient-reported outcomes, e.g. ACQ/AQLQ (logistic regression); and lung function expressed as trough and peak FEV1 (mixed-model repeated measures).

Results: Rather than obtaining relative treatment effect estimates for each subgroup (e.g. hazard ratios, odds ratios, or mean differences, plus confidence intervals), modeling has the advantage that these estimates are calculated over the whole range of IgE and eosinophil values.  The beneficial effect of tiotropium Respimat® in moderate and severe asthma was observed for all endpoints across a broad range of IgE and eosinophil values.

Conclusions: Modeling of treatment effects, while yielding results comparable with conventional subgroup analyses, provides a continuous profile of relative treatment effects without explicit reference to cut-off values conventionally used to define subgroups.  All types of endpoint (time to event, binary, and continuous) can be modeled.