Comorbidity and Clinical Trials of Disease-Modifying Therapies in Multiple Sclerosis

Throughout, comorbidity will be classified in the following ways: (i) any; (ii) a count (0, 1, 2, ≥3); and as (iii) individual comorbidities with the priority conditions being depression, anxiety disorders, hypertension, dyslipidemia, diabetes, migraine, chronic lung disease. Comorbidity information will be derived from medical history, concomitant medications and adverse event forms as reported previously. Comorbidity counts will be based on the same group of comorbidities across trials.C1. Aim 1. To determine the prevalence of comorbidity at enrollment in Phase 3 clinical trials of DMTs for MS.C1.1 Outcome: Prevalence of comorbidity, as defined above.Approach: 1. Univariate: For each clinical trial, we will report the prevalence of comorbidity. The use of any comorbidity and a count of comorbidities will facilitate summary comparisons across clinical trials. We will report the prevalence of comorbidity at enrollment overall, at the end of years 1 and 2 (and 3 where applicable), and stratified by year the study was initiated, gender, age group, clinical course (CIS, RRMS, SPMS, PPMS) and EDSS at enrollment; 95% confidence intervals (95%CI) will be reported based on the binomial distribution. The categories for study year, EDSS and age groups will be determined based on the distributions observed in the clinical trials. We will report prevalence ratios (PR) and 95%CI for the comparison of comorbidity status across groups.2. Meta-analysis: We will combine findings across clinical trials using random-effects meta-analysis using the Der-Simonian and Laird estimator. The heterogeneity of findings across studies will be assessed using the I2 statistic. C2. Aim 2. To determine the association between comorbidity and relapses, disability progression and MRI outcomes in Phase 3 clinical trials of DMTs for MS.C2.1 Comorbidity status (as defined above).C2.2 Covariates (if available in most datasets): DMT (yes vs. no), age at enrollment (continuous), gender (females as reference group), race (White as reference group), body mass index (categorized as <18.5, underweight; 18.5-24.9, normal [reference]; 25.0-29.9, overweight; ≥30, obese), smoking status. In sensitivity analyses, we will include the number of relapses in the 12 months prior to enrollment. We will also consider finer categories for classification of body mass index.C2.3 Approach: 1. Univariate: We will summarize participant enrollment characteristics in each trial using descriptive statistics. We will compare enrollment characteristics according to comorbidity status using student's t-tests, chi-square tests, and Kruskal-Wallis tests as appropriate.2. Multivariable: We will use multivariable models to test the association between comorbidity and the outcomes of interest. The specific model chosen and distribution used will be based on theoretical and empirical considerations as we evaluate the clinical trial datasets. The outcomes below are listed in order of priority. Covariates would be those described in C2.2; they will be included as time-varying where possible. We will specifically test whether there is effect modification by gender and race/ethnicity. Model assumptions will be tested using standard methods. We will be conducting multiple analyses, thus we will employ a false discovery rate correction (FDR = 0.10) for secondary analyses.C3: Aim 2 outcomes and likely statistical modelsOutcome   ModelPrimary: No evidence of disease activity (based on relapses, sustained disability worsening, MRI changes) -Cox proportional hazards modelSecondary: Annualized relapse rate-Negative binomial regression model with generalized estimating equations, with exchangeable correlation coefficientSecondary: Sustained disability worsening (1-point increase in the EDSS if the baseline EDSS was ≤5.0 and half point increase in the EDSS if the baseline EDSS was 5.5, sustained for 3 or 6 months [based on visit intervals])-Cox proportional hazards modelSecondary: • Change in T2 lesion volume• Annualized combined unique active lesions (sum of the number of new enhancing lesions, new non-enhancing T2 lesions, and enlarged non-enhancing T2 lesions observed after enrollment)- Negative binomial regression model with generalized estimating equations, with exchangeable correlation coefficientTertiary: Time to relapse - Cox proportional hazards model. We will also consider Cox model with Anderson Gill adjustment to allow repeated relapse eventsTertiary: Annualized number of new enhancing lesions - Negative binomial regression model with generalized estimating equations, with exchangeable correlation coefficient3. Meta-analysis: We will combine findings across clinical trials using random-effects meta-analysis using the Der-Simonian and Laird estimator. The heterogeneity of findings across studies will be assessed using the I2 statistic. C3. Aim 3. To determine the association between comorbidity and effectiveness of DMTs in Phase 3 clinical trials.C3.1 Primary Outcome: No evidence of disease activity (based on relapses, disability worsening, MRI changes)C3.2 Covariates (if available in all datasets): age at enrollment (continuous), gender (females as reference group), race (White as reference group), body mass index (categorized as <18.5, underweight; 18.5-24.9, normal [reference]; 25.0-29.9, overweight; ≥30, obese), smoking status. In sensitivity analyses, we included the number of relapses in the 12 months prior to enrollment.C3.3 Approach: 1. This analysis will be restricted to clinical trials which include a placebo group. We will repeat the analyses testing the association between comorbidity and no evidence of disease activity as described in Aim 2, after adding interaction terms. To evaluate the individual and joint effects of comorbidity and DMT on outcomes, we will create four groups where R = relative risk: (i) DMT and no comorbidity (reference group [R00]); (ii) no DMT and no comorbidity [R10]; (iii) with any comorbidity but no DMT [R01]; and (iv) with DMT and any comorbidity [R11]. To assess a possible biological interaction of comorbidity and DMT on outcomes we will calculate the relative excess risk of interaction (RERI) as R11-R01-R10 + 1. When one of the exposures of interest is protective, this requires recoding of the protective factor. We will also report the attributable proportion due to interaction (the excess risk of the outcome due to both exposures). These analyses will also be repeated for comorbidities identified as being of particular relevance in Aim 2.2. Meta-analysis: First, we will combine findings across clinical trials using random-effects meta-analysis using the Der-Simonian and Laird estimator. The heterogeneity of findings across studies will be assessed using the I2 statistic. Second, we will stratify the analysis by year of study initiation (using the groups from Aim 1) to determine whether this contributes to heterogeneity across studies.C4. Aim 4. To determine the association between comorbidity and premature discontinuation from clinical trials of DMTs for MS.C4.1 Primary Outcome: Rate of discontinuation of trial participation. C4.2 Secondary Outcome: Rate of missed visitsC4.2 Approach: 1. Univariate: We will report the cumulative incidence rate of the primary outcome overall, and stratified by reason by discontinuation (withdrew, loss to follow-up, adverse event, other). To account for differences in clinical trial duration we will report the incidence per person-year, where person-year represents the participant's maximum follow-up time if clinical trial participation were complete, based on their date of enrollment. We will age-standardize these rates to the 2019 US or world population to enhance comparability across studies. We will compare crude age-standardized incidence rates by comorbidity status using incidence rate ratios (IRR) and 95%CI. We will use a similar approach for the rate of missed study visits.2. Multivariable: We will compare the cumulative incidence rates of trial discontinuation (or annual incidence of missed visits) by comorbidity status using Poisson models (or negative binomial regression distribution if overdispersion is observed). We will include the log of person-years as an offset to account for variable duration of follow-up. We will also consider sensitivity analyses using a Cox proportional hazards model.3. Meta-analysis: We will combine cumulative incidence rate ratios across clinical trials using random-effects meta-analysis using the Der-Simonian and Laird estimator. The heterogeneity of findings across studies will be assessed using the I2 statistic. C5. Aim 5. To determine the association between comorbidity and safety of DMTs in Phase 3 clinical trials.C5.1 Outcomes (Adverse Events): Incidence of (i) infection; (ii) treatment-emergent autoimmune disease; (iii) cancer; (iv) elevated transaminases; and (v) lymphopenia (classified as any, then as grade I, II, ≥III sample size permitting). Laboratory abnormalities will be classified using the Common Terminology Criteria for Adverse Events (CTCAE) V5.0 for consistency across trials.We will focus on these outcomes because these are of particular interest as recommended by the International Advisory Committee on Clinical Trials, do not overlap with the comorbidities of interest, and/or have implications for monitoring therapy, and to reduce the number of comparisons and thereby the risk of false positive findings.C5.2 Covariates (if available in all datasets): age at enrollment (continuous), gender (females as reference group), race (White as reference group), body mass index (categorized as <18.5, underweight; 18.5-24.9, normal [reference]; 25.0-29.9, overweight; ≥30, obese), smoking status.C5.3 Approach: 1. Univariate: We will report the incidence rate for each adverse event of interest in the following groups: risk: (i) DMT and no comorbidity (reference group [R00]); (ii) no DMT and no comorbidity [R10]; (iii) with any comorbidity but no DMT [R01]; and (iv) with DMT and any comorbidity [R11]. We will compare incidence rates across groups using IRR (95%CI).2. Multivariable: We will test the association between incidence of adverse events and comorbidity status using Poisson models (or negative binomial regression distribution if overdispersion is observed) and the 4 comorbidity/DMT groups defined in #1. We will include the log of person-years (from enrollment to the date the adverse event is first reported) as an offset to account for variable duration of follow-up. Covariates will include those listed above (C5.2). This will allow us to determine how comorbidity affects the incidence of the adverse events, and whether the combination of comorbidity and DMT exposure has a synergistic effect on the risk of these adverse events. This analysis can readily be extended to an ordinal comorbidity count variable. 3. Meta-analysis: We will combine IRRs across clinical trials using random-effects meta-analysis using the Der-Simonian and Laird estimator. The heterogeneity of findings across studies will be assessed using the I2 statistic. Sample Size/Power CalculationsIndividual studiesSince we are conducting secondary analyses of existing data, the sample size is fixed and we estimate study power for our primary outcome, no evidence of disease activity (NEDA). We conservatively assume a sample size of 500 as the smallest Phase 3 clinical trial in MS had a sample size of 563. For time to event analyses, if we assume ≥30% of the population is exposed (affected by) to comorbidity, a hazard ratio of 1.20 as the smallest clinically meaningful effect, that other variables and minimum trial sample size of 500, then power to detect the effect is >0.85.For Poisson regression analyses, if we assume ≥30% of the population is exposed (affected by) to comorbidity, and a rate ratio of 1.20 as the smallest clinically meaningful effect, that other variables account for up to 30% of the variance in the outcome, and a minimum trial sample size of 500, then power to detect the effect is 0.79. Increasing the sample size or the proportion of the population exposed increases the power.Two step meta-analysisIf we assume meta-analysis of findings across studies, and that we can combine at least 3 studies (based on study population, phase, and other similarities), that the average effect size is small (Cohen's D = 0.20), and that each trial includes at least 500 participants, then the power to detect a small effect is:• 0.99 in the presence of low heterogeneity across studies• 0.97 in the presence of moderate heterogeneity• 0.78 in the presence of high heterogeneityIncreasing the study sample size or increasing the number of studies increases the power to >0.85 even if heterogeneity is high.