Clinical trials studying multiple sclerosis (MS) — a chronic, inflammatory, progressive, autoimmune disease affecting the central nervous system — employ various common endpoints. These typically target frequency of relapses, progression of disability, and MRI activity, as well as “no evidence of disease activity” (NEDA), which is a concept/composite endpoint combining the prior three components. Analyzing these can present several statistical challenges.

Here, we provide an overview of the common endpoints (including their definitions) in MS clinical trials and key statistical considerations together with the statistical modeling techniques to analyze them, as well as considerations of how to overcome several statistical challenges that we encountered in this indication.

Frequency of relapses

A key clinical feature of MS is the occurrence of relapses, i.e., episodes of new or progressing neurological dysfunction, lasting for a period, followed by periods of remission. Distinguishing a relapse from other clinical conditions may not be straightforward; therefore, an accurate definition should be included in the protocol.  A typical endpoint here is the number of relapses occurring within one year, i.e., the annualized relapse rate (ARR).

From a statistical point of view, this constitutes count data and thus, we analyze it using:

• Poisson regression model, or
• Negative binomial model. This model accounts better for a high number of zero counts (i.e., zero inflation) and overdispersion (i.e., greater variability than expected in terms of variance being larger than the mean).

Both models are often adjusted for MS prognostic factors.

In recent years, we’ve seen a decrease in number of relapses, largely due to earlier MS diagnoses and the widespread use of high-efficacy disease-modifying therapies. In the case of absence of relapses, derivation of ARR might go wrong. When performing quality checks of a sponsor’s analyses, especially if a not-adjusted approach was followed for cases with no relapses, the exposure or observation time in the study is mistakenly not accounted.

Another common method is time to first relapse, applying the survival data analysis methods. At Cytel, our teams have also explored recurrent event analysis using the Mean Cumulative Function (MCF) method, though these are limited by the evolving nature of relapse patterns.

Progression of disability

The most widely used measurement tool to describe disease progression in patients with MS is the Expanded Disability Status Scale (EDSS).1 The EDSS includes a neurological evaluation of seven functional systems (plus “other”) in conjunction with observations and information concerning gait and use of assistive devices to rate the level of disability, resulting in a single score.

While EDSS is a widely accepted measure, it has been criticized for certain limitations. For example, a 1-point increase at lower EDSS levels (e.g., 2.0 to 3.0) may reflect different functional implications than the same increase at higher levels (e.g., 6.0 to 7.0).

To address these limitations, we commonly use Confirmed Disability Progression (CDP), which is based on an increase in the EDSS score (e.g., 0.5, 1.0, or 1.5 points) that is confirmed after a specified period (e.g., 3 or 6 months), depending on the baseline EDSS value.

It’s important to note that neither terminology nor definition are standardized; there are several variations in its application across different sponsors.

When Cytel analyses CDP as a binary endpoint (yes/no), we typically use logistic regression adjusting for relevant MS prognostic factors. One of the challenges encountered with this approach is the presence of incomplete data when attempting to obtain a relevant assessment to confirm the disease progression:

• Some patients withdraw from the study.
• In other cases, the EDSS assessments are not frequent or consistent enough to confirm progression. For example, if the confirmation of CDP is required in 6 months, the study protocol should define the corresponding visits frequency, and the statistical plan should consider the minimum time interval required for the confirmatory assessment, e.g. 6 months x 30 days – 14 days, so that the confirmation would not be missed by few days.

Both scenarios result in patients being classified with “unknown” status, which can complicate the interpretation and robustness of the analysis.

Another approach is to analyze time to first CDP via survival data analysis methods.

Magnetic resonance imaging (MRI)

Relapses or EDSS may not be a sufficient indicator of MS activity. The inflammation caused by MS does not always result in a relapse or any visible symptoms and may only be seen with an MRI.

The most common MRI-related endpoints are T2 lesions count or volume, active (i.e., new or enlarging) T2 lesion count, and T1 (Gd+ / hypointense) lesions count or volume.

The expectation from the treatment with the MS drug is that MRI activity is also “suppressed” (e.g., broadly speaking, we do not observe new or enlarging T2 lesions; new T1 Gd+ do not appear, etc.). This usually happens at the early stage of the study according to the foreseen onset of action for a given drug. However, new lesions may eventually appear, or others may grow in size.

For the statistical analysis, MRI-derived endpoints reflect MRI activity such as counts of lesions that are new or enlarging. The counts can be further classified on:

• binary scale where at least one new or enlarging lesion is present vs none (coded as 1/0 for lesion count: >0/ =0).
• continuous scale, e.g. changes in MRI activity such as counts of lesions that are new or enlarging compared to baseline (or any other relevant visit).

Such data is challenging, but can be handled using parametric approaches, including:

• counts via Poisson regression, or in case of zero inflation via generalized linear model assuming negative binomial distribution adjusted for MS prognostic factors.
• binary scale using the McNemar test since a shift in the number of lesions through visits from present to none is frequently of interest.
• continuous endpoint (change from baseline) can be handled via (mixed; if random effects accounted) linear regression model adjusted for MS prognostic factors.

(The non-parametric methods are not discussed here.)

The analysis is, however, much more complex due to the nature of data collection for lesion count: the assessment of the lesions may be performed multiple times per visit (such as for T1 Gd+ lesions), or in reference to previous MRI scan to detect new or enlarging lesions. This is reflected in the analysis by random effects, standardization, or by using offset in the count models, accounting for the number of scans or time since baseline.

NEDA

No evidence of disease activity (NEDA), also referred to as freedom from disease activity, is one of the composite endpoints taking clinical and imaging endpoints into account.

NEDA is defined by the absence of:

• Relapse
• CDP
• MRI activity

NEDA is typically analyzed as a binary outcome (yes/no), using logistic regression adjusted for MS prognostic factors.

While the goal of most MS DMDs is to reduce relapse frequency, relapses tend to be less common in treated people with MS. As a result, the two remaining components of NEDA, CDP and MRI, may have greater impact on its outcome.

However, there is no crisp definition regarding CDP and MRI endpoints (frequency and parameter selection), and thus the reports on NEDA between different studies might be incomparable:

• We observe that studies present 3-months, 6-months, or even 12-months CDPs,
• MRI activity may be defined in various ways, including
  • frequency of MRI scans,
  • selection of relevant MRI readouts (e.g. presence of T1 Gd+ lesions or new or enlarging T2 lesions), etc.

The more frequently the measures are taken, the higher the likelihood of identifying the progression. In case of post-marketing studies that reflect real-world clinical practice (half-yearly or yearly visits), EDSS measures necessary for CDP definition and MRI scans are often not collected in the necessary frequency.

In addition, for assessments that have rather rare scheduled frequency, each missing assessment may affect NEDA heavily and by experience, there is no harmonization of dealing with such missing across different companies. In such situations, NEDA might be analyzed differently including time to first NEDA using survival analysis methods. Overall, it is important that the potential challenges (whether related to data collection or analytical methods) need to be carefully considered already at the early stages of study planning.

Final takeaways

While these endpoints provide valuable frameworks for assessing disease progression and treatment efficacy of MS patients, statistical challenges remain. Addressing these challenges, in close collaboration with medical experts, is essential to ensure that the analyses remain both scientifically sound and clinically meaningful.

Share Via