Stull DE, Houghton KF, Gale R, Wiklund I, Capkun-Niggli G, Jones P. Application of innovative methods to identify and characterize differential responders in clinical trials of COPD: the use of mixture models. Poster presented at the 2011 ISPOR 14th Annual European Congress; November 2011. Madrid, Spain. [abstract] Value Health. 2011 Nov; 14(7):A501-2.

OBJECTIVES: Applying innovative methods to clinical trial data to identify and characterize unobserved subgroups of differential responders.

Data from three COPD clinical trials was retrospectively analysed using Growth Mixture Models (GMMs): INHANCE (indacaterol 150g and 300g vs tiotropium 18g and placebo); INLIGHT-2 (indacaterol 150g vs salmeterol 50g and placebo); and INVOLVE (indacaterol 300g and 600g vs formoterol 12g and placebo). GMMs were conducted on SGRQ Symptoms Domain data at baseline, 12 weeks, and six months to identify unobserved subgroups. Baseline characteristics were compared between emergent subgroups of differential responders in post hoc analyses.

RESULTS: Within INHANCE and INLIGHT-2, two subgroups of patients emerged per treatment arm: responders (improvement) and non-responders (little change/deterioration). Within INOLVE, three subgroups of patients emerged per treatment arm: responders, non-responders, and partial-responders. When responders were analysed separately, mean treatment effects in terms of SGRQ Symptom scores were generally larger than when all patients were included: INHANCE responder improvements ranged from 8 -12 units compared with 7-14 for all patients; INLIGHT-2 responder improvements were 3 -13 units versus 3 -8 for all patients; INVOLVE responder improvements were 5 -17 units vs 3 -11 for all patients. Within each trial, responders made up the largest proportion of the sample (55% - 82%) but non-/partial-responder groups were large enough and different enough to dampen treatment effects when group means were analyzed as a whole. Responders had significantly better baseline SGRQ Symptom scores than non-responders. Further significant differences were found between non-responders, partial-responders and responders in terms of smoking history, age, and breathlessness.

CONCLUSIONS: GMMs have the potential to increase understanding of treatment effects and identify patients more likely to benefit from treatment. The ability of baseline characteristics to predict responders/non-responders needs to be tested prospectively.

Share on: