Brain volume loss in relapsing multiple sclerosis: indirect treatment comparisons of available disease-modifying therapies

Abstract

Background

Brain volume loss (BVL) has been identified as a predictor of disability progression in relapsing multiple sclerosis (RMS). As many available disease-modifying treatments (DMTs) have shown an effect on slowing BVL, this is becoming an emerging clinical endpoint in RMS clinical trials.

Methods

In this study, a systematic literature review was conducted to identify BVL results from randomized controlled trials of DMTs in RMS. Indirect treatment comparisons (ITCs) were conducted to estimate the relative efficacy of DMTs on BVL using two approaches: a model-based meta-analysis (MBMA) with adjustment for measurement timepoint and DMT dosage, and a network meta-analysis (NMA).

Results

In the MBMA, DMTs associated with significantly reduced BVL versus placebo at two years included fingolimod (mean difference [MD] = 0.25; 95% confidence interval [CI] = 0.15 – 0.36), ozanimod (MD = 0.26; 95% CI = 0.12 – 0.41), teriflunomide (MD = 0.38; 95% CI = 0.20 – 0.55), alemtuzumab (MD = 0.38; 95% CI = 0.10 – 0.67) and ponesimod (MD = 0.71; 95% CI = 0.48 – 0.95), whereas interferons and natalizumab performed the most poorly. The results of NMA analysis were generally comparable with those of the MBMA.

Conclusions

Limitations of these analyses included the potential for confounding due to pseudoatrophy, and a lack of long-term clinical data for BVL. Our findings suggest that important differences in BVL may exist between DMTs. Continued investigation of BVL in studies of RMS is important to complement traditional disability endpoints, and to foster a better understanding of the mechanisms by which DMTs can slow BVL.

Background

Brain volume loss (BVL) detected through magnetic resonance imaging (MRI) can be identified in the earliest stages of multiple sclerosis (MS), and occurs at a faster rate than in healthy adults [1, 2]. De Stefano et al. reported an annual BVL rate of 0.5 to 1.35% for MS patients, compared to a rate of 0.01–0.3% for individuals without the disease [1]. This marker of degeneration has been reliably correlated with future physical and cognitive disability [3, 4], and has been suggested as a key mechanism by which disease-modifying therapies (DMTs) can prevent or delay disability progression [5]. Predictors of disability progression in relapsing multiple sclerosis (RMS) are important to characterize, given that the accumulation of disability increases healthcare costs and greatly diminishes health-related quality of life [6]. Many DMTs have been noted to slow BVL in MS patients [7], potentially through mechanisms that are distinct from the effects of DMTs on central nervous system (CNS) lesions and require further research [8]. Recent evidence indicates that such protective impacts can be sustained over the long-term [9]. As such, BVL is an emerging clinical endpoint that should be considered when evaluating the efficacy of available and emerging DMTs.

While BVL is a commonly evaluated outcome measure in Phase 3 clinical trials in RMS, there is a paucity of head-to-head trials of DMTs. In such cases, indirect treatment comparison (ITC) methods are suitable for comparing treatments, and are often used to inform clinical, regulatory and reimbursement decisions. These methods leverage the existing direct evidence from clinical trials in order to estimate relative treatment effects for interventions that have not been directly compared [10, 11]. Currently, there are several approaches for conducting ITCs. Given the large number of randomized controlled trials (RCTs) reporting BVL for various DMTs, ITC approaches leveraging a network of multiple interventions and underlying RCTs, connected via common comparators, allow for the most comprehensive assessment of relative treatment efficacy. In this study, both model-based meta-analysis (MBMA) and network meta-analysis (NMA) were employed.

NMA is a well-accepted methodology that combines direct and indirect evidence across the network of studies to compare interventions [12,13,14], and has been extensively presented in the MS literature [15,16,17,18]. MBMA similarly utilizes the network of studies to compare interventions. However, MBMA provides a more flexible analytical framework in which additional variables can be readily incorporated in the statistical model [19, 20]. Specifically, MBMA can incorporate data pertaining to various dosages of the same DMT, as well as longitudinal data. This makes MBMA particularly advantageous for an investigation of BVL in RMS, since a majority of large RMS trials have assessed investigational agents at multiple dosages over a range of study durations, with varied measurement timepoints.

The objective of this study was to systematically collect published BVL data from RCTs in RMS, and to compare the effects of various DMTs on BVL using MBMA and NMA.

Methods

Systematic literature review methods

The systematic literature review (SLR) was reported in accordance with PRISMA reporting guidelines [21]. Searches of EMBASE, and MEDLINE®, and the Cochrane Central Register of Controlled Trials were designed by an information specialist and peer reviewed by a second information specialist [22]. Search strategies used a combination of controlled vocabulary (e.g., "multiple sclerosis") and keywords (e.g., "relapsing remitting"), with no date restrictions. The database searches were conducted in August, 2021 (search strategies are provided in Online Resource 1.1). Searches of several additional sources were also conducted, including the reference lists of included primary publications and relevant systematic reviews, as well as proceedings of large multiple sclerosis congresses, clinical trial registries (e.g., ClinicalTrials.gov), and regulatory and health technology assessment (HTA) agency websites. These additional sources were searched in late 2021; further details including date restrictions are provided in Online Resource 1.2.

The eligibility criteria for the SLR are detailed in Online Resource 1.3. English language citations reporting data from RCTs comparing two or more regimens of interest (e.g., placebo, interferon DMTs, orally administered DMTs, or monoclonal antibody DMTs) on MRI outcomes in adult RMS populations were eligible for inclusion. RMS studies were eligible for inclusion if the proportion of patients with relapsing–remitting multiple sclerosis (RRMS) was ≥ 80%. Two reviewers independently screened citations using Distiller SR (Evidence Partners; Ottawa, Canada), first based on titles and abstracts, then based on full text review. A single reviewer collected data regarding study populations, treatment regimens, and MRI outcomes using a structured data extraction form in Microsoft Excel (Microsoft, Seattle, Washington, USA), which was validated by a second reviewer. Results for BVL collected using any technique (e.g., Structural Image Evaluation using Normalization of Atrophy [SIENA], or brain parenchymal fraction calculation [BPF]), where reported as a mean change from baseline per treatment group, were abstracted from included studies. Otherwise, median change from baseline in BVL was collected, or data which lacked such descriptors. Variance measures were also collected for each estimate, where available (e.g., standard deviation [SD], standard error [SE], 95% confidence interval [CI]). The Cochrane Risk of Bias tool was used to appraise the quality of included RCTs [23], performed by one reviewer and validated by a second reviewer. Disagreements between reviewers pertaining to study selection, data collection, and quality assessment were resolved by discussion, or by a third reviewer.

ITC feasibility assessment

The validity of ITCs relies on whether there are systematic differences among the included studies, especially in patient characteristics that are treatment effect modifiers [24,25,26,27]. Failure to account for these differences can lead to biased comparisons of treatment effect [28, 29]. The feasibility of conducting ITCs was assessed qualitatively by comparing key study and patient characteristics across trials included in the SLR that reported BVL data. Assessment of cross-trial differences was conducted for various study design elements (e.g., trial phase, blinding, sample size) using summary tables, and for patient characteristics (e.g., age, gender, baseline Expanded Disability Status Scale [EDSS], prior DMT exposure) using bar charts.

The measurement timepoints and outcome definitions were also compared between studies. Any measurement techniques for BVL were considered eligible for inclusion. The use of different measurement techniques was not expected to introduce bias to relative effect estimates. This assumption is aligned with published meta-analyses of BVL [7, 30]. Notably, the relevance of BVL data measured at or prior to 24 weeks of DMT treatment was considered limited, due to potential confounding caused by pseudoatrophy. As such, only BVL measurements collected beyond 24 weeks were considered eligible for inclusion in ITCs.

Selection of study data for ITCs

Mean change-from-baseline data was preferred for incorporation in ITCs, however median changes were considered if mean estimates were not reported. Reported measures of variance were either converted to SD, or where no such measure was available, SD values were imputed using the following equation:

$${SD(\%BVL)}_{ij}=\beta_0+\beta_1\cdot{\left|\%BVL\right|}_{ij}$$

(1)

Where ${\beta }_{0}$ and ${\beta }_{1}$ were estimated based on the linear relationship between BVL and its SD, weighted by sample size, in the j^th arm of the i^th trial.

MBMA methods

MBMA methods aligned with practices published by Mandema et al. [31, 32]. All MBMAs were conducted using the R programming language version 3.6.3 (R Core Team, Vienna, Austria) with the nlme() library version 3.1–144. All data points after 24 weeks and all tested doses of each included drug were utilized in the MBMA.

As multiple time points and multiple doses were included in the MBMA, a longitudinal, dose–response model was utilized to explore the influence of time and dose on relative treatment effect (i.e., BVL relative to placebo). Both fixed effect and random effect approaches were explored to determine the most appropriate model based on an ANOVA test. A dose response power function was explored separately for all DMTs with data at multiple doses. Time was incorporated as a covariate on relative effects using a power function anchored to 104 weeks: [time/104]^k. Additionally, baseline mean number of T1 gadolinium (Gd +) enhancing lesions was explored as an influential covariate on relative treatment effect. For this specific investigation, a subset of trials that did not report any relevant T1 Gd + lesion information were excluded.

Additional MBMA methodology details (including a description of the model with equation) are provided in Online Resource 2.1.

NMA methods

Bayesian NMAs were performed using the R programming language version 3.5.3 and Just Another Gibbs Sampler (JAGS) version 4.3.0. NMA models were based on code adapted from the National Institute of Health and Care Excellence (NICE) Evidence Synthesis Decision Support Unit (DSU) Technical Support Document (TSD) Series [33]. Only DMT dosages that are currently authorized for use in the US were included. Given the progressive decline of BVL over time, only data collected at approximately the same timepoint, i.e., two years, were included in the NMA. This was aligned with the MBMA, which used two years (104 weeks) as the reference timepoint for predicted BVL. Additional details regarding the NMA methods, model fit, and inconsistency analysis are provided in Online Resource 2.2.

Results interpretation for indirect treatment comparisons

For both MBMA and NMA, effect estimates reported within Forest plots reflected the probability of a DMT regimen being better than placebo, based on being to the right or left of the Y axis. Where the interval pertaining to an effect estimate did not cross the Y axis, this indicated significant outperformance of the DMT regimen over placebo. The relative distances of effect estimates from the Y axis also indicated the relative rank of DMT regimens. For the NMA, a league table reporting all possible pairwise comparisons is provided in the Online Resource 5.0.

Results

Evidence identified in the systematic literature review

The SLR identified 40 RCTs, across 158 individual reports (Online Resource 3.1 and 3.2). More than 30% of these reports were congress proceedings, trial registry records, and HTA or regulatory agency documents. In total, 31 trials reported data on BVL. Quality assessments broadly indicated low risks of bias for these trials (Online Resource 3.3). A summary of the main characteristics of these trials is provided in Table 1, including details regarding the reporting of BVL.

Table 1 Characteristics of RCTs reporting BVL identified in the SLR

Table 1 considers all RCTs reporting BVL in any form, for the DMTs in the scope of this review. The Phase 2 trials of ponesimod and ocrelizumab were limited to 24 weeks in duration [34, 35]. Similarly, the only BVL results reported from baseline in the CLARITY trial were collected at 24 weeks [36]. Following the exclusion of these three trials, 28 trials reporting BVL between 36 and 156 weeks were considered eligible for ITCs. Missing standard deviation values were imputed as described above (Eq. 1), with estimated parameters: ${SD(\%BVL)}_{ij}=0.558+0.713\cdot{\left|\%BVL\right|}_{ij}$.

Feasibility of conducting indirect treatment comparisons

The ITC eligible trials were found to be comparable with regards to study design, patient characteristics, and measurement/reporting of total BVL (see Online Resource 4.1 through 4.4 for supporting tables and plots). The majority of studies were multi-national, phase III, double-blinded, and included follow-up of approximately two years.

Moreover, the measurement technique used to collect BVL varied across trials; most trials indicated the use of either SIENA or BPF calculation, with SIENA being more frequently utilized in recent trials.

Indirect treatment comparison findings

The full network of evidence included in the ITCs is provided in Fig. 1. All trial data leveraged for the ITCs is provided in Online Resource 5.0. For the MBMA, the investigation of dose–response considered all drugs for which multiple doses were included: alemtuzumab, dimethyl fumarate, fingolimod, interferon β-1b, ozanimod, peginterferon β-1a, and teriflunomide. No significant difference was found between dimethyl fumarate 240 mg (twice daily) and dimethyl fumarate 240 mg (three times daily), thus these doses were pooled, and a single treatment effect was estimated for dimethyl fumarate. Peginterferon β-1a and interferon β-1b had a negligible dose–response relationship (p > 0.75), and a pooled effect estimate was incorporated for each respective drug. Although not statistically significant (p > 0.05), a dose–response relationship was incorporated for the remaining four drugs with data at multiple doses (alemtuzumab, fingolimod, ozanimod, teriflunomide) to ensure that any potential dose-related variability was accounted for.

Fig. 1

Evidence network for ITCs of BVL^a

^aTreatment nodes are sized proportionally to sample sizes, and connections between treatment nodes are indicated with line thickness proportional to the number of trials informing the connection. All connections were incorporated in the MBMA analysis, whereas only the nodes and trials shown in boldface were included in the NMA (since DMT dosages that were not of interest were excluded from the NMA)

Abbreviations: 24W = every 24 weeks, 2W = every two weeks, 4W = every four weeks, ALE = alemtuzumab, BID = twice daily, BVL = brain volume loss, CI = confidence interval, DMF = dimethyl fumarate, FIN = fingolimod, GA = glatiramer acetate, IFNβ = interferon β, NAT = natalizumab, OFA = ofatumumab, OZA = ozanimod, PBO = placebo, PEG = peginterferon β-1a, PON = ponesimod, Q2W = every two weeks, QD = once daily, QOD = every other day, QW = once weekly, TER = teriflunomide, TID = three times daily

Upon incorporation of time as a covariate on relative effects, results demonstrated that a relative treatment effect at 52 weeks would be approximately 38% smaller than the same treatment effect at 104 weeks [95% CI: 17%—53%] based on the time exponent (k) estimation of 0.70 with 95% CI [0.27 – 1.1]. Compared to a model excluding any influence of time on relative effect, the final MBMA described the data significantly better (p < 0.01 from ANOVA test).

A model adjusting for the number of T1 gadolinium enhancing lesions (T1 Gd +) at baseline was also considered. However, this model was not significantly better fitting than the final model, which excluded the influence of baseline T1 Gd + lesions (p > 0.45).

A random effects model was also explored. Using an ANOVA test, there was a negligible difference in how well this model described the data versus the fixed effects model.

Considering the final MBMA incorporating adjustment for dosage and timepoint, the DMTs predicted to significantly outperform placebo on BVL in order of decreasing relative efficacy included: ponesimod, alemtuzumab, teriflunomide, ozanimod, and fingolimod (Fig. 2a). That is, statistically significant benefits versus placebo were noted in each case. Interferons and natalizumab appeared to perform the most poorly in this analysis. A league table reporting the full results of the MBMA can be found in Online Resource 6.0.

Fig. 2

MBMA (A) and NMA (B) results for differences in brain volume loss at two years, versus placebo^a

^aMBMA and NMA used fixed effect models. Measurement timepoint and dosage were covariates in the MBMA. Round parentheses indicate the probability of being better than placebo. Peginterferon could not be incorporated in the NMA due to a lack of eligible BVL data reported at two years

Abbreviations: 24W = every 24 weeks, 2W = every two weeks, 4W = every four weeks, ALE = alemtuzumab, BID = twice daily, BVL = brain volume loss, CI = confidence interval, DMF = dimethyl fumarate, FIN = fingolimod, GA = glatiramer acetate, IFNβ = interferon β, NAT = natalizumab, OFA = ofatumumab, OZA = ozanimod, PEG = peginterferon β-1a, PON = ponesimod, Q2W = every two weeks, QD = once daily, QOD = every other day, QW = once weekly, TER = teriflunomide

Although the confirmatory NMA utilized data from the same clinical trials as the MBMA, only data collected at approximately two years, for treatment dosages that are currently authorized for use in the US, were included (Fig. 1).

The effect estimates from the confirmatory NMA were generally associated with greater uncertainty (i.e., wider 95% credible intervals) as compared to the MBMA (Fig. 2b). However, the main findings were consistent. In order of decreasing relative efficacy, the DMTs predicted to significantly outperform placebo included: ponesimod, fingolimod, teriflunomide, and ofatumumab. That is, statistically significant benefits versus placebo were noted in each case. The NMA results broadly aligned with the MBMA results in that most DMT regimens, namely the first-generation injectable therapies, were not predicted to significantly reduce BVL compared with placebo. The effect estimate for ofatumumab was significant in the NMA but not the MBMA, whereas the effect estimates for ozanimod and alemtuzumab were significant in the MBMA but not the NMA. A league table reporting the full results of the NMA can be found in Online Resource 7.0.

Discussion

To our knowledge, this study represents the first attempt to use ITC methods for the purpose of comparing BVL outcomes across various DMTs in patients with RMS. The MBMA was the primary approach for ITCs, as it allowed for adjustment of key variables such as dosage and timepoint, whereas the NMA was chosen to serve as a confirmatory analysis given its wide use and familiarity among clinicians and health care decision-makers. Across both analyses, the DMTs which significantly outperformed placebo included fingolimod, teriflunomide, and ponesimod, where the latter always ranked first.

BVL is not currently considered a core outcome in clinical trials of RMS, although it has become more commonly investigated in recent years. Traditionally, lesion burden assessed through MRI has been used as a surrogate marker for disease activity in RMS [37]; however, the link between lesion burden and other clinical findings has been weak to modest [38]. For example, although the appearance of T2 hyperintense lesions is correlated with disability accumulation in MS, this relationship plateaus at higher disability levels [39]. Nevertheless, it is well understood that available DMTs reduce new lesion activity in RMS, and lesion activity is routinely evaluated in clinical trials. In contrast to lesion counts, BVL is a measure of whole-brain atrophy. Notably, both lesion load and atrophy (including that of white matter and grey matter) have been correlated with disability status in MS. A study comparing MS patients and controls conducted by Tedeschi et al. concluded that amongst lesion load, white matter atrophy, and grey matter atrophy, the latter is the most significant MRI variable in determining the final disability level of MS patients [40]. Ghione et al. used a similar study design to ascertain a correlation between BVL and disability progression [4]. Furthermore, Sprenger et al. have highlighted the potential predictive value of BVL earlier in the disease course on long-term disability outcomes based on clinical data from the TEMSO extension study [5]. Finally, Sormani et al. have described the prediction of disability accumulation by brain atrophy through a meta-analysis of RCTs.(3) Given the correlation between BVL and disability, a greater understanding of how various treatments may impact BVL is likely to be valuable to clinicians and decision makers.

It is notable that of the three sphingosine-1-phosphate receptor subtype 1 (S1P1) modulators included in our analyses, ponesimod and fingolimod were found to significantly outperform placebo across analyses. Additional research is warranted to understand if a mechanistic link exists between S1P1 regulation and the rate of BVL in RMS. Ozanimod was found to significantly outperform placebo in the MBMA, but the same comparison did not reach statistical significance in the NMA.

In general, monoclonal antibody DMTs are noted for their efficacy on disability outcomes such as 3-month and 6-month disability progression, as demonstrated in published ITCs [41]. However, monoclonal antibody DMTs did not appear to outperform most other agents in the ITCs of BVL. Natalizumab in particular performed poorly overall in comparison to several DMT comparators. Compared with the MBMA results, the effect estimates for the monoclonal antibodies shifted slightly in the NMA, but overall conclusions were similar. Statistically significant benefits versus placebo were not consistently noted between analyses for alemtuzumab, ofatumumab or ocrelizumab.

Our study has several strengths. Broadly, the SLR conforms to guidance from the Cochrane Collaboration and PRISMA reporting guidelines. The search strategy was comprehensive in that it included several database and grey literature sources and was peer-reviewed by a second medical information librarian in advance of conducting searches. ITCs were informed by RCTs identified in the systematic literature review, which were deemed to be of high quality overall using the Cochrane Risk of Bias tool. The methodology for conducting ITCs was aligned with best practices outlined by major HTA agencies such as NICE (NMA), or found in the published literature (MBMA). Notably, the MBMA is able to simultaneously account for multiple potential sources of variability through adjustment, and incorporate data reported at different timepoints and for different dosages. Hence, the current MBMA allows for incorporation of all trial data regardless of dosage or timepoint, whereas some data was necessarily excluded from our NMA (e.g., timepoints other than approximately two years) to ensure that effect estimates were sufficiently comparable to yield a valid analysis. To ensure the general alignment of these strategies, the MBMA was contrasted against an NMA which leveraged a more limited data set.

Some limitations of this study should also be considered. Heterogeneity across studies may have had unanticipated impacts on the validity of ITCs. Though patient populations were considered adequately similar for incorporation in ITCs based on the feasibility assessment, there was substantial variation in prior DMT use across trials. If this trait (or other unmeasured traits) are effect modifiers, then variation therein threatens the validity of presented ITCs. Unfortunately, network meta-regression was not feasible to assess the potential impact of prior DMT variation due to the number of single-trial connections in the network, an issue that has been noted in previous ITCs of MS therapies [42]. Future studies could consider the use of more detailed techniques such as propensity score matching or matching-adjusted indirect comparisons to correct for cross-trial differences, where data availability permits (eg, thorough reporting of patient baseline characteristics and access to individual patient data).

Differences over time were also noted in regards to the techniques used for BVL measurement, with the SIENA technique being more predominant in the most recently conducted studies. However, the heterogeneity in measurement techniques between studies was not considered to threaten the validity of ITCs, given that they rely on relative differences between treatment arms per trial, which are expected to be less prone to confounding due to measurement technique than absolute changes in a single trial arm. Fisher et al. reported that BPF and SIENA methods of BVL measurement have robust correlation at baseline and for brain volume change in MS patients, though different absolute results are generated [43]. This high degree of correlation supports the validity of including both measures in NMAs of BVL. The phenomenon of pseudoatrophy was also considered to potentially impact the validity of these analyses. While it is generally understood that pseudoatrophy mostly occurs in the first few months after DMT initiation, there is uncertainty regarding the specific timepoint beyond which the impacts of pseudoatrophy might be negligible [44,45,46]. Transient volume changes could also result in confounding of BVL measurements, for example, if pseudoatrophy is more likely with one drug class versus another, however limited data are available to inform this issue. In an effort to reduce the potential for bias due to pseudoatrophy, we excluded BVL measurements collected at or before 24 weeks from the start of DMT treatment. Further research is needed to fully understand this phenomenon, and the extent to which it may confound BVL data collected for various DMT regimens. Finally, it should be noted that additional clinical data is needed to ascertain the longer-term impacts of BVL prevention, as most data points identified as eligible for incorporation in this study were collected at or before two years of treatment.

Conclusion

The MBMA indicated that S1P1 agents fingolimod, ponesimod and ozanimod, as well as teriflunomide and alemtuzumab significantly outperformed placebo on the BVL outcome, after adjustment for measurement timepoint and DMT dosage. These results were confirmed using NMA, which yielded results that were broadly aligned. Some limitations of these analyses included the potential for confounding introduced by pseudoatrophy, and a lack of long-term clinical data. Nevertheless, the differentiation of various DMTs substantiates the importance of evaluating BVL as a clinical endpoint in future trials, as a complement to traditional measures of disability. Further investigations regarding the mechanisms by which various DMTs may reduce BVL rate are also warranted, as are studies that improve our understanding of the long-term relationship between BVL rate and disability progression.