The Minimal Clinically Important Difference and Its Use in Neuromuscular Disorders

The minimal clinically important difference (MCID) has emerged as a promising alternative to statistical significance for determination of clinically important and relevant outcome measures in clinical trials and clinical applications.¹ Different outcome measures may have different MCID cutoff values depending on the calculation method and study design. It is important to recognize the common methodology and major limitations associated with the use of MCID in neuromuscular diseases.

The selection of outcome measures and their thresholds are a crucial part of the study design process which dictates sample size and the ability to show clinically relevant changes on an individual or group level. Over the past decade, the concept of MCID has begun to facilitate translational medicine in clinical research.²

What Is the MCID?

In 1989, Jaeschke et al³ introduced the concept of MCID which is defined as “the smallest difference in score, within the domain of interest, which patients perceive as beneficial and which would mandate, in the absence of troublesome side effects and excessive costs, a change in the patient’s management.” Any amount of change greater than the MCID threshold would be considered clinically meaningful or important from the patient’s or provider’s perspective, or both. Several other terms including “minimally important difference,” “minimal clinically important change,” and “meaningful change,” are used interchangeably with MCID.⁴

MCID was initially proposed as a metric for assessing patient-reported outcomes and is now being increasingly used in clinical trials spanning many different fields of medicine including neurology.^5,6 An important advantage of MCID over a statistically significant difference is the clinical relevance of the determined cutoffs. MCID is a context-specific concept and the clinical importance of the same change may not be perceived the same by different subgroups or during different interventions.⁷ For example, we can consider a case of 2 individuals with myasthenia gravis (MG). If the first individual is a surgeon in her 30s and the second is a retired teacher in his 60s, they will not experience “improvement or resolution of ocular symptoms” as a result of a hypothetical treatment to the same degree, and will have different treatment goals. Similarly, patient and clinician perspectives may diverge with regard to what constitutes an important enough clinical change to decide for or against a treatment change.

How Is MCID Determined?

The 2 main methods of calculating MCID are anchor-based and distribution-based approaches. The Delphi method is also occasionally used, which relies on systematic reviews and expert consensus⁸ (Figure).

Figure. Approaches to estimate minimal clinically important difference.
Abbreviations: SD, standard deviation; SEM, standard error of measurement.

Anchor-Based Approaches

Anchor-based approaches compare the change in patient-reported outcomes with an external measure of change, referred to as an “anchor.” This external criterion is mostly chosen as an individual’s own assessment of change after an intervention. Commonly used anchors include visual analog scale (eg, pain score), Patient Global Impression of Change, Clinician Global Impression of Change, and Global Rating of Change scores.⁹ Based on the anchor, patient’s conditions are grouped as unchanged, better, or worse. In an anchor-based analysis, an anchor is generally considered valid if the correlation coefficient between the anchor and outcome measure is between 0.3 and 0.5.¹⁰

Several variations of MCID calculation with the anchor-based method include assessing for within-patient or between-patients score changes, and sensitivity- and specificity-based or social comparison approaches. Within-patient change assesses the change in a single patient over time, typically from subgroups reporting at least some level of improvement based on a selected assessment scale.¹¹ Between-patients score change compares the change in PRO scores for 2 groups of patients with responses categorizable as 2 adjacent levels of an assessment scale (eg, not at all impaired or very mildly impaired).¹² Another variation uses receiver operating characteristic curves to identify a cutoff value with equal sensitivity and specificity that can discriminate between individuals with improvement or no change.¹³

Distribution-Based Approaches

Distribution-based methods are determined from the statistical characteristics of a study’s results. This approach compares changes in PRO scores with various measures of variability to determine a threshold MCID value that is larger than the expected variation or error in measurement.⁵ Distribution-based MCID methods include the standard deviation (SD), standard error of measurement (SEM), and effect size.

SD quantifies the amount of variation or dispersion of a set of data values for a specific outcome measure based on the theoretical range and available normative data. It is universally accepted that MCID is equal to 0.5 SD.¹⁴ SEM is defined as a measure of variability of the observed scores due to measurement error and is calculated as follows: SD times the square root of (1 minus the reliability coefficient). In contrast to SD, SEM also takes test–retest reliability into consideration. A change of 1 SEM is typically accepted to correspond with the original MCID criterion of an outcome measure.¹⁵ Effect size assesses the mean change in scores of an individual divided by the SD of the baseline scores of the whole group or a subset of stable individuals.¹⁶

MCID in Neuromuscular Diseases

Various MCID cutoffs have been evaluated and reported in the literature for neuromuscular diseases, including chronic inflammatory demyelinating polyneuropathy (CIDP), MG, amyotrophic lateral sclerosis (ALS), spinal muscular atrophy, and genetic or inflammatory myopathies. Key findings from the selected studies are summarized in the following and in the Table.

Merkies et al¹⁷ evaluated various MCID threshold values to assess the clinical importance of changes in outcome data from the first period of the randomized, placebo-controlled ICE clinical trial (NCT00220740) evaluating intravenous immunoglobulin (IVIg) for the treatment of chronic inflammatory demyelinating polyneuropathy (CIDP).¹⁸ This was the first reported use of MCID methods to evaluate treatment response in neuromuscular disease treatment trials. 

One anchor-based and 3 distribution-based methods were used to determine MCID cutoffs for EMG measures, Medical Research Council sum score, grip strength, Inflammatory Neuropathy Cause and Treatment (INCAT) sensory sum score, disability (ie, INCAT scale and Rotterdam handicap scale), and quality of life. All MCID analyses showed that improvements in CIDP associated with IVIg treatment were clinically meaningful, although the 3 methods used resulted in variable MCID cutoff values. The INCAT disability score consistently demonstrated a higher percentage of responders reaching the calculated MCID threshold in the IVIg-treated group compared with the placebo group, regardless of the MCID method used.

Katzberg et al¹⁹ studied 51 participants from a published IVIg vs placebo study, using 1 anchor-based and 3 distribution-based techniques to identify MCID cutoffs for quantitative MG score as well as repetitive nerve stimulation and single-fiber EMG results. To avoid MCID variability based on baseline values, tertiles were calculated for the baseline quantitative MG score, and the anchor-based MCID was calculated for each tertile. Anchor-based and effect-size MCID cutoffs did not show a difference between IVIg treatment and placebo, and MCID methods did not produce meaningful repetitive nerve stimulation cutoffs. This study also found different MCID cutoff points depending on baseline MG severity, which may have implications for defining responders and calculating sample size in future MG clinical trials.

Fournier et al²⁰ investigated clinically meaningful changes in the commonly used Revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-R) as well as a novel Rasch-built Overall ALS Disability Scale (ROADS) using a patient-reported global impression of change scale recorded at the time of a follow-up assessment. MCID was calculated on the basis of ALSFRS-R and ROADS for people with ALS evaluated at a multidisciplinary ALS clinic. Based on the difference between mean change in score for participants who rated their disease as a little worse compared with unchanged or better, MCID was found to be 5.81 for ROADS and 3.24 for ALSFRS-R, indicating that changes below these threshold values may not be clinically meaningful for people with ALS. These findings underlined that treatment effects will not be observed in study participants whose results do not progress by an amount exceeding the MCID on the primary outcome measures, which has important implications for the design of ALS trials.

Gupta et al²¹ assessed MCID thresholds for the North Star Ambulatory Assessment (NSAA) in boys with Duchenne muscular dystrophy aged 7 to 10 years using distribution-based estimates of one-third SD and SEM, an anchor-based approach with 6-minute walk distance as the anchor, and participant-tailored questionnaires to evaluate the patients’ and parents’ perceptions of clinically important change. In Duchenne muscular dystrophy and similar pediatric neuromuscular conditions, higher MCID thresholds with distribution-based approaches are commonly encountered due to the increasing heterogeneity of symptomatic involvement after a certain age, as also noted in this study. The MCID for NSAA was found to be the lowest with the one-third SD method and highest with the anchor-based method which used the 6-minute walk distance as the external criterion. The use of patient- or parent-reported functional changes provided additional insight into the perceived differences between complete or partial loss of function on the NSAA scale. 

Pitfalls and Limitations of MCID

Various factors can confound the MCID including the severity of illness, demographic and socioeconomic factors, and types and burdens of treatment, which can affect individuals’ perceptions of health and improvement.⁹ For instance, people with autoimmune neuromuscular disease receiving a stem cell transplant, which has high risk and involves a long recovery period, will likely expect more improvement from the treatment to consider it clinically relevant compared with people receiving a less-intensive treatment. This heterogeneity hinders the ability to use a single, fixed MCID cutoff measure for all individuals or research studies. The use of MCID also does not account for the cost–benefit considerations of treatments which can be crucial for clinical decisions involving new therapy options, particularly in MG, CIDP, and ALS.

There may be inherent limitations for the type of MCID approach chosen. Ordinal scales used for the assessment of anchor criteria may impede proper clinical interpretation because differences between consequent levels on the scales do not always change in a linear fashion (eg, on the Guillain-Barré syndrome disability scale).²² The Rasch model has emerged as a technique to transform ordinal data to interval data.²³ For the distribution-based approach, if the variability of a measurement is too wide, a subtle but clinically relevant change that is within the range of measurement error may not be detected. Therefore, it is important to integrate multiple approaches (eg, combining anchor-based and distribution-based methods) for MCID estimation when possible.²⁴

Conclusion

MCID is an important consideration in modern clinical trials. The prioritization of clinically important outcome measures will pave the way to a better understanding of the scope over which a disease and proposed intervention can affect a person’s life. Researchers utilizing MCID should recognize its limitations and optimize its use for the specific neuromuscular disorder and population being studied.