Development of a Sensitive Outcome for Economical Drug Screening for Progressive Multiple Sclerosis Treatment

Abstract

Therapeutic advance in progressive multiple sclerosis (MS) has been very slow. Based on the transformative role magnetic resonance imaging (MRI) contrast-enhancing lesions had on drug development for relapsing-remitting MS, we consider the lack of sensitive outcomes to be the greatest barrier for developing new treatments for progressive MS. The purpose of this study was to compare 58 prospectively acquired candidate outcomes in the real-world situation of progressive MS trials to select and validate the best-performing outcome. The 1-year pre-treatment period of adaptively designed IPPoMS (ClinicalTrials.gov #NCT00950248) and RIVITaLISe (ClinicalTrials.gov #NCT01212094) Phase II trials served to determine the primary outcome for the subsequent blinded treatment phase by comparing 8 clinical, 1 electrophysiological, 1 optical coherence tomography, 7 MRI volumetric, 9 quantitative T1 MRI, and 32 diffusion tensor imaging MRI outcomes. Fifteen outcomes demonstrated significant progression over 1 year (Δ) in the predetermined analysis and seven out of these were validated in two independent cohorts. Validated MRI outcomes had limited correlations with clinical scales, relatively poor signal-to-noise ratios (SNR) and recorded overlapping values between healthy subjects and MS patients with moderate-severe disability. Clinical measures correlated better, even though each reflects a somewhat different disability domain. Therefore, using machine-learning techniques, we developed a combinatorial weight-adjusted disability score (CombiWISE) that integrates four clinical scales: expanded disability status scale (EDSS), Scripps neurological rating scale, 25 foot walk and 9 hole peg test. CombiWISE outperformed all clinical scales (Δ = 9.10%; p = 0.0003) and all MRI outcomes. CombiWISE recorded no overlapping values between healthy subjects and disabled MS patients, had high SNR, and predicted changes in EDSS in a longitudinal assessment of 98 progressive MS patients and in a cross-sectional cohort of 303 untreated subjects. One point change in EDSS corresponds on average to 7.50 point change in CombiWISE with a standard error of 0.10. The novel validated clinical outcome, CombiWISE, outperforms the current broadly utilized MRI brain atrophy outcome and more than doubles sensitivity in detecting clinical deterioration in progressive MS in comparison to the scale traditionally used for regulatory approval, EDSS.

Keywords: clinical trial; composite scale; disability scale; multiple sclerosis; outcome measure; progressive MS; quantitative MRI.

Figures

Figure 1

Correlation matrix for 14 clinical and MRI measures in the cohort of 98 progressive MS patients. Correlation matrix for Mo −12 cross-sectional data (above the diagonal) and relative percentage change (Δ) over 1 year (below the diagonal) in the progressive MS cohort. The heatmap shows positive (shades of blue) and negative (shades of red) Spearman correlations. Black border of a window indicates Spearman correlation with p < 0.0018 (the Bonferroni-adjusted significance level for multiple comparisons of 28 tested variables). For exact values of Spearman correlation coefficients, p-values, and number observations, see Table S2 in Supplementary Material.

Figure 2

Development of mathematically optimized CombiWISE scale. (A) A bean plot (25) of variability in relative weights constructed from 200 permutations of training data using the genetic algorithm. (B) A bean plot of variability in weights for optimization across 500 permutations of training data using reduced set of clinical scales. (A,B) Black lines represent the individual weight results, red lines show the average of weights for each contributing scale, yellow areas represent non-parametric density curves of distribution of individual weights. (C) Mean relative weights and computing weights from the contributing clinical scales for CombiWISE calculation. Relative weights allow for comparing the level of contribution of individual scales while the computing weights are rescaled versions that produce a metric ranging from 0 to 100. (D) A bean plot of 500 permutations of validation data test-statistics for CombiWISE in addition to individual clinical scales and the Multiple Sclerosis Functional Composite (MSFC) scale. Black lines represent individual test statistics, red lines show average of test statistics for each scale, yellow areas represent non-parametric density curves of individual test statistics, blue lines correspond to cut-offs for 5% significance level tests.

Figure 3

Combinatorial weight-adjusted disability score (CombiWISE) correlates highly with other clinical scales and shows statistically significant progression in three independent cohorts. (A) Spearman correlations between CombiWISE and standard clinical scores (EDSS, SNRS, 25FW, 9HPT, SDMT, and MSFC) in the cross-sectional cohort of 303 untreated subjects with different types of MS, other inflammatory and non-inflammatory CNS conditions, and healthy volunteers. The Y-axis scales for 25FW and 9HTP are log2-transformed; ****p < 0.0001. (B) Longitudinal data for CombiWISE calculated from clinical scores collected every 6 months (Mo −12, Mo −6, and Mo 0) during the pre-treatment baseline for PPMS subjects in the IPPoMS trial and for a cohort of 29 untreated SPMS subjects showing statistically significant worsening of the clinical status over periods of 6 to 12 months in both PPMS and SPMS subjects. Statistical significance was determined by one-way ANOVA test on repeated measures. The red bars show mean for each group, *p < 0.05, **p < 0.01, ***p < 0.001, displayed that p-values were adjusted for multiple comparisons by Holm–Sidak test. (C) Longitudinal data for CombiWISE collected during the pre-treatment baseline of the IPPoMS trial show statistically significant increase in CombiWISE over 1 year. CombiWISE data collected on healthy subjects with a yearly follow-up visit show no overlap with the data of moderately to severely disabled PPMS patients, as well as no appreciable change over 1 year. Red bars represent the mean of the group. (D) Comparison of measured change and the overlap of values between IPPoMS and HV cohorts for the best-performing DTI biomarker (axial diffusivity of the medulla). Red bars represent the mean of the group.

Figure 4

Power analysis for the selected seven statistically significant variables. Power analysis shows number of required subjects per arm (accumulated sample size as effect size drops across bars) considering 80% nominal power, statistical significance level of 5% and either (A) parallel or (B) baseline vs. treatment group design for the 2-year study considering 30% (sum of black, gray, hatched bars), 40% (sum of gray and hatched bars), and 50% (hatched bars) drug effect.