Published in:
Open Access 01-12-2019 | Multiple Sclerosis | Research article
Defining responders to therapies by a statistical modeling approach applied to randomized clinical trial data
Published in: BMC Medicine | Issue 1/2019
Login to get accessAbstract
Background
Personalized medicine is the tailoring of treatment to the individual characteristics of patients. Once a treatment has been tested in a clinical trial and its effect overall quantified, it would be of great value to be able to use the baseline patients’ characteristics to identify patients with larger/lower benefits from treatment, for a more personalized approach to therapy.
Methods
We show here a previously published statistical method, aimed at identifying patients’ profiles associated to larger treatment benefits applied to three identical randomized clinical trials in multiple sclerosis, testing laquinimod vs placebo (ALLEGRO, BRAVO, and CONCERTO). We identified on the ALLEGRO patients’ specific linear combinations of baseline variables, predicting heterogeneous response to treatment on disability progression. We choose the best score on the BRAVO, based on its ability to identify responders to treatment in this dataset. We finally got an external validation on the CONCERTO, testing on this new dataset the performance of the score in defining responders and non-responders.
Results
The best response score defined on the ALLEGRO and the BRAVO was a linear combination of age, sex, previous relapses, brain volume, and MRI lesion activity. Splitting patients into responders and non-responders according to the score distribution, in the ALLEGRO, the hazard ratio (HR) for disability progression of laquinimod vs placebo was 0.38 for responders, HR = 1.31 for non-responders (interaction p = 0.0007). In the BRAVO, we had similar results: HR = 0.40 for responders and HR = 1.24 for non-responders (interaction p = 0.006). These findings were successfully replicated in the CONCERTO study, with HR = 0.44 for responders and HR=1.08 for non-responders (interaction p = 0.033).
Conclusions
This study demonstrates the possibility to refine and personalize the treatment effect estimated in randomized studies by using the baseline demographic and clinical characteristics of the included patients. The method can be applied to any randomized trial in any medical condition to create a treatment-specific score associated to different levels of response to the treatment tested in the trial. This is an easy and affordable method toward therapy personalization, indicating patient profiles related to a larger benefit from a specific drug, which may have implications for taking clinical decisions in everyday clinical practice.