Contributions
Abstract: P768
Type: Poster
Abstract Category: Therapy - disease modifying - 30 Tools for detecting therapeutic response
Background: Identifying patient subgroups that explain individual response to disease modifying therapy (DMT) is an important step in transitioning from population-level randomized clinical trials (RCTs) to personalized medicine in MS. Classical approaches for subgroup analysis use predefined or algorithm-selected cut-offs, but are often insufficiently powered. A novel modelling approach by Li et al. (Biometrics, 2016) yields a continuous patient-specific score that predicts treatment response according to baseline characteristics.
Objectives: To assess the utility of a continuous patient-specific treatment response score to predict annualised relapse rate (ARR) in MS patients.
Methods: The DEFINE (Dimethyl Fumarate [DMF] vs placebo) RCT (n=1234) was used to build the prediction score and the CONFIRM (DMF vs placebo) RCT (n=1066) was used to validate the score. The prediction score was developed by regressing ARR on baseline age, sex, ethnicity, number of prior relapses, disease duration, time since relapse, prior treatment, EDSS, MSFC-4, Gd+ lesions, T2 and T1 lesion volume, brain volume, SF-36 PCS and MCS using negative binomial regression. Regression models were based on a fully specified additive model; stepwise, ridge and LASSO regression; and elastic nets. Treatment-by-score interaction models were developed and cut-offs for high responders based on the top quartile of scores.
Results: The ARR risk ratios (treatment vs. placebo) ranged from 0.18 to 0.52 in DEFINE and the validation ARR risk ratios ranged from 0.22 to 0.55 in CONFIRM. Treatment-by-score interaction p-values were < 0.05 in both RCTs. The group of patients in the top quartile of score identified a high-responders group, with an ARR risk ratio of 0.20 (95%CI 0.12-0.32) vs. 0.69 (95%CI 0.54-0.88) for all others (interaction p< 0.0001). These findings were validated in CONFIRM with corresponding ARR ratios of 0.36 (95%CI 0.24-0.55) and 0.64 (95%CI 0.48-0.86) (interaction p=0.026). Significant treatment effect modifiers were SF-36 PCS, age, EDSS, prior treatment, brain volume z-score, LCLA, Gd+ lesions, and log T2 lesion (all p< 0.0001 except Gd+ lesions p=0.002).
Conclusions: This proof-of-concept application of a powerful modelling strategy successfully detected high responders, which was validated in an independent RCT. The individual response score is useful for personalized medicine, treatment response calculators, and identifying patient sub-groups for RCTs.
Disclosure: Supported by: Biogen, Inc.
Author disclosures:
Francesca Bovis has nothing to disclose.
Massimiliano Copetti received consulting fees from Biogen, Teva and Eisai.
Fabio Pellegrini, Ulrich Freudensprung and Carl de Moor are employees of, and stockholders in, Biogen.
Maria Pia Sormani received consulting fees from Biogen, Novartis, Genzyme, Roche, Teva, GeNeuro, Merck Serono and Medday.
Abstract: P768
Type: Poster
Abstract Category: Therapy - disease modifying - 30 Tools for detecting therapeutic response
Background: Identifying patient subgroups that explain individual response to disease modifying therapy (DMT) is an important step in transitioning from population-level randomized clinical trials (RCTs) to personalized medicine in MS. Classical approaches for subgroup analysis use predefined or algorithm-selected cut-offs, but are often insufficiently powered. A novel modelling approach by Li et al. (Biometrics, 2016) yields a continuous patient-specific score that predicts treatment response according to baseline characteristics.
Objectives: To assess the utility of a continuous patient-specific treatment response score to predict annualised relapse rate (ARR) in MS patients.
Methods: The DEFINE (Dimethyl Fumarate [DMF] vs placebo) RCT (n=1234) was used to build the prediction score and the CONFIRM (DMF vs placebo) RCT (n=1066) was used to validate the score. The prediction score was developed by regressing ARR on baseline age, sex, ethnicity, number of prior relapses, disease duration, time since relapse, prior treatment, EDSS, MSFC-4, Gd+ lesions, T2 and T1 lesion volume, brain volume, SF-36 PCS and MCS using negative binomial regression. Regression models were based on a fully specified additive model; stepwise, ridge and LASSO regression; and elastic nets. Treatment-by-score interaction models were developed and cut-offs for high responders based on the top quartile of scores.
Results: The ARR risk ratios (treatment vs. placebo) ranged from 0.18 to 0.52 in DEFINE and the validation ARR risk ratios ranged from 0.22 to 0.55 in CONFIRM. Treatment-by-score interaction p-values were < 0.05 in both RCTs. The group of patients in the top quartile of score identified a high-responders group, with an ARR risk ratio of 0.20 (95%CI 0.12-0.32) vs. 0.69 (95%CI 0.54-0.88) for all others (interaction p< 0.0001). These findings were validated in CONFIRM with corresponding ARR ratios of 0.36 (95%CI 0.24-0.55) and 0.64 (95%CI 0.48-0.86) (interaction p=0.026). Significant treatment effect modifiers were SF-36 PCS, age, EDSS, prior treatment, brain volume z-score, LCLA, Gd+ lesions, and log T2 lesion (all p< 0.0001 except Gd+ lesions p=0.002).
Conclusions: This proof-of-concept application of a powerful modelling strategy successfully detected high responders, which was validated in an independent RCT. The individual response score is useful for personalized medicine, treatment response calculators, and identifying patient sub-groups for RCTs.
Disclosure: Supported by: Biogen, Inc.
Author disclosures:
Francesca Bovis has nothing to disclose.
Massimiliano Copetti received consulting fees from Biogen, Teva and Eisai.
Fabio Pellegrini, Ulrich Freudensprung and Carl de Moor are employees of, and stockholders in, Biogen.
Maria Pia Sormani received consulting fees from Biogen, Novartis, Genzyme, Roche, Teva, GeNeuro, Merck Serono and Medday.