A protocol was registered prospectively with PROSPERO, Centre for Reviews and Dissemination, University of York: CRD42021286429.

Eligible studies were identified based on SPIO criteria – Study design, Population, Intervention, and Outcome (43). To be eligible for inclusion, studies had to be RCTs comparing an MBI to an active comparator or care as usual. Participants had to be PwMS of any age and phenotype. The intervention(s) being tested had to include core practices of Mindfulness-based stress reduction (MBSR) and/or Mindfulness-based cognitive therapy (MBCT), namely mindful breathing, mindful body awareness, and mindful movement. Only validated outcome measures (subjective or objective) of cognitive functioning were considered.

We employed a comprehensive search strategy for use in six major electronic databases, including the Allied and Complementary Medicines Database (AMED), Cumulative Index of Nursing and Allied Health Literature (CINAHL), Cochrane Central Register of Controlled Trials, ExcerptaMedicadataBASE (EMBASE), Medical Literature Analysis and Retrieval System Online (MEDLINE), and PsycINFO. The initial search was in April 2021, and was updated in May 2023. We searched from 1980 to the date of search, given that MBIs were first developed and piloted in the 1980s. We included only studies published in English, among human subjects, in the peer-reviewed academic literature.

Search results were first imported into COVIDENCE, a systematic review data storage software package. Three independent reviewers (AK, KD, JS) screened study titles/abstracts for potential eligibility using the keywords ‘mindfulness’ and ‘multiple sclerosis’. The same three independent reviewers further assessed selected studies against SPIO criteria to determine definitive eligibility. A senior party reviewer adjudicated any disagreements (RS).

Data from the final list of included studies were extracted by three independent reviewers (AK, KD, JS), guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), with intervention programming documented using the Template for Intervention Description and Replication (TIDieR) (44).

The Cochrane Collaboration’s tool for assessing the risk of bias (RoB) was used to summarize risk for individual outcomes in selected studies, graded as high, unclear, or low risk (45). This assessed generation of sequence, concealment of allocation, blinding of participants, outcome assessors and personnel, incomplete outcomes, selective reporting of outcomes, and any other source of bias. Finally, an overall RoB within each trial was determined based on the number of individual outcomes falling into the high, unclear, and low risk categories:

Low = Low RoB for all key domains.

Unclear = Low or unclear RoB for all key domains.

High = High RoB for one or more key domains.

Four separate meta-analyses were conducted to determine the overall mean difference between mindfulness-based interventions (MBIs) and cognitive function in PwMS, which is defined as PASAT-2, PASAT-3, Perceived Deficits Questionnaire (PDQ) and SDMT. The heterogeneity among the studies was evaluated using the chi-square test and the I² statistic, which quantifies the proportion of variation in the effect estimates attributable to heterogeneity rather than random chance. When the heterogeneity test showed statistical significance (I² > 50% and p < 0.05), a random effects model was used; otherwise, a fixed effects model was employed. The meta-analyses were performed with the ReviewManager (RevMan) software (Version 5.4.1, Nordic Cochrane Centre, Cochrane Collaboration, 2011), with statistical significance set at p < 0.05. Effect sizes and standard mean differences were calculated using RevMan software.

The main objective for this study was to determine the impact of MBI on cognitive functioning. Main outcome measures were all reported as continuous with mean and standard deviation (SD) values, plus the number of participants for each treatment group extracted.

Throughout the study, we adhered to the PRISMA guidance (46).

MBIs effectively improved scores on the PASAT-2 (SMD 0.38; 95% CI 0.06-0.71; p=0.02) though heterogeneity was moderate (I² 63%) ( Figure 2 ) (47, 50, 55). There was a trend towards improvement on the PASAT-3 with borderline significant results (SMD=0.32; 95% CI -0.01-0.64; I² 65%; p=0.06) ( Figure 3 ) (47, 50, 55). Benefits on the PDQ (SMD=0.34; 95 CI -0.05-0.74; I² 0%; p=0.09) ( Figure 4 ) (54, 58) and SDMT (SMD=0.14; 95% CI -0.18-0.47; I² 0%; p=0.38) ( Figure 5 ) (47, 55, 58) following MBI training were not statistically significant.

While two studies (48, 50) used the Wisconsin Card Sorting Test (WCST), one of these studies (48) did not report data that could be utilized in a meta-analysis. Additionally, two studies (57, 58) used the Brief Visuospatial Memory Test revised (BVMT-R), but one of these studies (57) did not report data that could be used in the meta-analysis. Another two studies (55, 57) used the Controlled Oral Word Association Test (COWAT), but one study (57) did not report data utilizable for the meta-analysis. One study reported data on information processing speed from the SDMT and Stroop Color-Word Test combined, and data could not be individually extracted for the SDMT (57).

Study quality was highly variable. The assessment was frequently made challenging by incomplete reporting. Eight studies outlined random sequence generation (47, 50–54, 56, 57). Six studies were adjudged low risk for allocation concealment, with the remainder unclear (47, 51–54, 56). Blinding of assessors was outlined in seven studies (47, 51–54, 56, 57), with one study being assessed as high risk (50). Blinding of outcome assessment was outlined in six studies (47, 51–54, 56). Six studies were deemed low risk when assessing reporting of outcomes as incomplete (48, 50, 52, 54, 55, 58), and one study was high risk (47). Two studies were assessed as at high risk for selective reporting of outcomes (48, 53). In terms of overall within trials RoB assessments, two studies were deemed low risk (52, 54), two unclear (55, 56), and eight high risk (47–51, 53, 57, 58).

Two studies reported adverse events associated with MBI exposure (52, 54). In one study that used MBSR, a participant reported an episode of increased spasticity during mindful body awareness (52). In the same study another participant described increased anxiety following the MBSR day retreat (52). In another study using MBSR, one participant reported increased severity of chronic neuropathic pain following the ‘raisin exercise’ (54).

Twelve RCTs that assessed the effects of an MBI on cognitive functioning in PwMS were eligible for inclusion in our systematic review. Out of these studies, eight reported cognition as the primary endpoint (47–51, 55, 57, 58), with four studies reporting cognition as a secondary measure (52–54, 56). From the 12 included studies, five had data extractable for use in our meta-analyses. In the meta-analyses, significant and borderline significant improvements were noted in the PASAT-2 and PASAT-3, respectively. Although trending to positive, no significant improvements were found on the PDQ or SDMT. The pooled result for SDMT differed from the two studies that both reported a significant change in the SDMT scores after MBI, as this meta-analysis compared the mean change from baseline between the control and MBI groups whereas the two studies reported on a within-group comparison. In our narrative synthesis, additional beneficial effects were reported in individual studies, pertaining to a wide range of cognitive functions, both fundamental (attention, memory), and higher order (executive function), suggesting a potential role for MBIs in improving information processing speed, attention, cognitive flexibility, calculation, emotion regulation, and meta-cognition. However, in making sense of these preliminary findings, it is necessary to highlight some important limitations identified in reviewing the studies included in this review.

Firstly, only one study recruited PwMS with baseline evidence of cognitive impairment as an eligibility criterion (50). This raises the risk for a ‘ceiling’ effect, whereby significant improvement may not reasonably be expected if participants are cognitively ‘intact’ according to scale criteria. Secondly, subjective self-report measures of cognitive function are notorious for not demonstrating robust correlations with objective measures. In PwMS, having a lower level of education, progressive phenotype, greater physical disability, and comorbid mood impairment are all known to lead to subjective overestimations of cognitive deficits (28). Thirdly, very few studies used composite assessments of cognitive functioning – this limits interpretation beyond the scope of individual tests, and whilst individual screening tests can have important predictive values [i.e., SDMT predicts cognitive relapses, employment (59)], such findings taken out of context can have limited ecological validity (60), or, more importantly, ability to inform patient need. Although not as comprehensive as a standardized clinical assessment with a neuropsychologist or specialist occupational therapist, comprehensive batteries (MACFIMS, BICAMS) provide the basis for a neuropsychological assessment, overview of impairments that can be captured psychometrically, and can predict task performance in activities of daily living (59).

The above limitations notwithstanding, MBIs may have a role to play in cognitive rehabilitation for PwMS, who value increased awareness of cognitive impairments, simple strategies for addressing difficulties, and group formatting which facilitates the normalization of symptoms and instrumental peer support processes (61). However, before a recommendation can be made for MBIs as part of comprehensive cognitive rehabilitation programming, more high-quality research is necessary. Indeed, MBIs are arguably better suited to addressing affective impairments, where existing evidence is strong for stress, anxiety, and depression – all are frequently comorbid and known to exacerbate cognitive dysfunction in PwMS. Stress has complex, bidirectional relationships with cognition in PwMS, increasing subjective sense of cognitive impairment, whereas executive dysfunction can predict greater comorbid stress, increased reading span predicting less comorbid stress. Executive dysfunction also predicts comorbid anxiety and depression in PwMS, whilst lower scores for anxiety correlate with better nonverbal memory, and better scores for depression are associated with improvements in attention and information processing speed (62–64).

While there is no other previous synthesis that has systematically explored the impact of MBIs on cognitive function in PwMS, a 2022 scoping review on third wave cognitive behavioural therapies in PwMS reviewed the impact of MBSR, MBCT, Acceptance and Commitment Therapy (ACT), and Dialectical Behaviour Therapy (DBT) separately (65). They found that MBSR was the most commonly studied approach. In addition, a 2022 systematic review and meta-analysis (66) examined Mindfulness and Acceptance-Based Interventions (MABIs) on a range of outcomes, including cognition, in PwMS. The interventions included MBIs, ACT, and DBT. This study demonstrated a moderate effect on attention (SMD 0.49; 0.19-0.80) and a large effect on memory (SMD 1.12; 0.06-2.17) (66); however, only one study assessing DBT was included (67) and only three (47, 54, 55) of the 12 RCTs identified in our current review were included, likely reflecting the earlier search cut-off date in their study. By comparison, our meta-analysis indicates that the effectiveness of MBIs at improving cognitive functioning in PwMS is inconsistent at best and no MBI can be said to be optimal in this context.

In non-MS populations, MBIs have been found to improve cognitive functioning generally, with a small but significant pooled effect size (g=0.15; CI 0.05 - 0.24), and small but significant pooled effects on executive function (g=0.15; CI 0.02 – 0.27) and working memory (g=0.23; CI 0.11 – 0.36). The clinical significance of this small effect is unclear, as the clinical populations included were diverse; only 14% were described as ‘individuals with neurocognitive disorders’, but analyses pooled these participants with people with other ‘psychiatric’ and neurological’ disorders. Without overt reporting of clinical ‘case-ness’, we cannot be sure about the degree of cognitive impairment, interpret response (or ‘remission’) in relation to baseline cognitive function. Overall, MBIs outperformed usual care, but not active comparators. Outcomes were moderated most by population (clinical vs non-clinical), comparator intervention type (active vs usual care), session duration and frequency (68). The latter finding makes sense intuitively, in that ‘dose’ has been identified as a mediating factor in beneficial outcomes associated with MBI in other meta-analyses (38).

Guided by the PRISMA checklist (46), the TIDieR checklist (44), and the Cochrane Collaboration RoB tool (45), our multidisciplinary team of reviewers used robust search, appraisal, and analysis techniques for extracting and analyzing data in this systematic review and meta-analysis.

Although we assessed quality using a reference standard, the Cochrane Collaboration RoB tool, we did not estimate the strength of any recommendation for the use of MBIs in PwMS. Future studies could do so by applying the Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria (69).

The standard deviations reported in the meta-analysis are to be considered an estimate as they were calculated based on a formula used for independent measures; however, they are dependent measures. The standard deviation of the mean differences was calculated based on the reported standard deviation across different populations, when it should be paired standard deviation provided by the data calculation.

Meta-analyses of RCTs by design exclude other potentially relevant data, such as that deriving from observational or qualitative research. When considering intervention feasibility, such as acceptability, accessibility, and implementability, as well as perceived effects from the point of view of participants, these alternate study designs can provide important insights into how and why interventions succeed or fail in a given context (61), and how they might be optimized to best meet patient needs and preferences.

Two studies were adjudged low, two studies as unclear, and eight studies as high RoB. Although all MS phenotypes were represented, by far the most participants had relapsing-remitting MS (74.3%), very few had primary progressive (3.4%), some had secondary progressive (21.9%), some had non-specified MS (1.3%), and none had progressive relapsing. This is similar to findings from a large international cohort study (n=2599) (70), which found that ~73% of their study population had relapsing-remitting MS. Furthermore, the mean sample age was relatively low at 31.4-52.9 years, as, arguably, was disability according to EDSS, whilst ethnicity, SES, and comorbidity were poorly covered, limiting the generalizability of the findings. Only three compared an MBI against an active comparator condition. Observed effects were mostly small, with a wide range of confidence intervals.

Given the well documented high levels of cognitive comorbidity in PwMS, it is notable that our meta-analysis has only been able to quantify the effects of MBI training on three commonly used but limited measures of cognition in PwMS. Other cognitive assessments were utilized in individual studies, where beneficial effects were reported, but meta-analysis was not possible. Future studies could address this evidence gap by measuring the impact of MBI training on cognition using a variety of subjective and objective assessments.

More, well-designed, high quality RCTs are needed to assess more definitively the impact of MBIs on cognitive functioning in PwMS. Although RCT methodology normally specifies a single primary outcome, this may have limited value in practice and instead it might make the most sense to study a composite cognitive ‘outcome’ using the BICAMS or MACFIMS. In addition, or alternatively, it may be prudent to consider MBIs as a ‘preventative’ strategy, rather than remediative, along the lines of building cognitive reserve/enrichment (71). Furthermore, MBIs have been reported to have a positive impact on biological markers of inflammation and aging, including telomere length, which hints at a potential neuroprotective effect (72). To the best of our knowledge, MBI impact on inflammatory markers in PwMS remains untested.

Currently, MBIs cannot be recommended as a mode of cognitive rehabilitation for PwMS. However, they do effectively improve common confounders such as stress, anxiety, depression, and fatigue so should be considered for these common comorbidities that often exacerbate cognitive difficulties in this population.