Cognitive impairments in relapsing-remitting multiple sclerosis: a meta-analysis

Abstract

There is debate in the literature regarding the magnitude, nature, and influence of cognitive impairment in individuals with relapsing-remitting multiple sclerosis (RRMS), Therefore, we conducted a meta-analysis that quantified the overall magnitude of cognitive impairment in individuals with RRMS and identified the domains of cognition and clinical/demographic variables that were moderators of the overall effect. We included 57 studies with 3891 participants that yielded a total of 755 effect sizes. Overall, there was a moderate decline in cognitive functioning in individuals with RRMS compared with healthy controls. Larger effects were observed in cognitive domains of motor functioning, mood status and memory and learning. Regarding demographic and clinical variables, age and gender were moderators of cognitive impairment in all cognitive domains, whereas neurological disability and disease duration primarily moderated performance on tasks assessing memory and learning.

Introduction

Multiple sclerosis (MS) is the most prevalent chronic disabling neurological disease among adults in the United States. The National Multiple Sclerosis Society estimates that there are approximately 400,000 patients with MS in the United States with an incidence of nearly 200 new patients each week (www.nationalmssociety.org). The majority of people with MS are diagnosed between 20 and 50 years of age, and women are affected two to three times as often as men (www.nationalmssociety.org). MS is more common among people with northern European ancestry than people of African, Asian, and Hispanic backgrounds (www.nationalmssociety.org).

Multiple sclerosis involves progressive and unpredictable episodes of axonal demyelination and transection, resulting in lesions along axons of nerve fibers in the brain, brain stem, spinal cord, and optic nerves. The demyelination and transection of the nerve fibers interferes with the smooth and rapid conduction of electrical potentials along neuronal pathways in the central nervous system (CNS). This interference has been associated with motor, sensory, psychological, visual, bladder, and sexual symptoms commonly experienced by people with MS [1,2]. Although researchers have long been interested in the symptoms associated with MS, the study of cognitive deficits as a symptom of MS is still in its infancy [3–5]. Indeed, Figure 1 presents a graphical comparison of MS research in general vs. cognition-related MS research based on the number of publications across the period of 1983 through 2007.

Figure 1.

A graphical comparison of MS research in the last two decades. Blue line graphs research related to Multiple Sclerosis in general, whereas the pink line represents cognition-related MS research.

As can be seen in the figure, over the years, there has been an increased interest by researchers and clinicians in the study of cognitive deficits associated with MS. Cognitive deficits have been reported in broad domains of memory [6–10, 98], attention and executive functioning [11–15], and verbal fluency [16–20]. Indeed, two previous meta-analyses [21,22] have quantitatively examined the nature of neurocognitive deficits in MS. Results from both meta-analysis provided evidence for the presence of significant cognitive impairments in patients with MS. Importantly, one of the meta-analysis did not report differences in cognitive impairment as a function of disease subtype [21], whereas the other meta-analytic review reported that individuals with chronic-progressive MS had greater impairments on tasks of executive control, whereas those with relapsing-remitting multiple sclerosis (RRMS) have significant memory-related dysfunction [22].

To date, researchers have not performed a meta-analysis for quantifying the magnitude and nature of cognitive deficits for individuals with RRMS. We believe that there is significant merit in systematically and quantitatively investigating the nature of cognitive decline in this subtype of MS. RRMS is the most common type of MS [97] and there is still some degree of uncertainty regarding the magnitude and nature of cognitive deficits in those with RRMS. We further note that there has been a sizable increase in the number of published papers on cognitive deficits in RRMS allowing for a better characterization of the magnitude, type, and influences of cognitive deficits in this sub-population of individuals with MS. Such an analysis would underscore the degree of cognitive deficits even in non-progressive types of MS and highlight the importance of therapeutic regimens for mitigating cognitive deficits in those with RRMS.

The primary goal of this study involved extending previous meta-analytic reviews by specifically quantifying the overall magnitude of deficits in cognition associated with RRMS. The secondary goal of this meta-analysis involved identifying types or domains of cognitive impairments in RRMS and the influence of clinical and demographic variables as moderators of cognitive impairment in RRMS. Such an analysis can identify types of cognitive processes that demonstrate the largest degree of cognitive impairment and identify clinical and demographic aspects of individuals with RRMS that are associated with larger cognitive deficits.

Method

Literature search

This meta-analysis was conducted according to the recommendations provided by the Meta-Analysis of Observational Studies group [23]. Two of the investigators (RSP and JL) conducted the literature search using the following online databases: PsychInfo, MEDLINE, Current Contents. We searched across the period from 1983 (the year Poser Criteria were published. Poser, et al. [24]) to July, 2007 using the keywords “multiple sclerosis” in conjunction with “cognition”, “attention”, “executive functioning”, “information processing speed”, “language”, “memory”, “motor functioning”, “fMRI”, “EEG”, and “neuroimaging”, We located a total of 650 studies using those keywords. We further undertook a manual search of pertinent journals: Archives of Clinical Neuropsychology; NeuroImage, Neuropsychologia, Neuropsychology, Multiple Sclerosis, The journal of International Neuropsychological Society, The Clinical Neuropsychologist, Brain and Cognition, Archives of Neurology, The journal of Clinical Neuropsychology, Neuropsychiatry, Neuropsychology and Behavioral Neurology. We identified additional articles from reference lists of previous meta-analyses, review articles and book chapters. Finally, we contacted study authors in an effort to obtain information on additional publications.

Inclusion criteria

Studies were included in the final meta-analysis that met the following criteria: (a) available in the English language; (b) included a healthy control group and a group of individuals with RRMS for whom the neuropsychological results were available independent of other MS types; and (c) results were reported in statistics convertible to effect size g (e.g., M, SD, F, t, chi-square: Roberts, et al. [25]). Statistics of studies that met the above reported criteria were coded based on the classification provided below to compute an overall effect size.

Moderator coding

The coding of moderator variables was undertaken by two of the investigators (RSP and JL). One of the primary moderators was type of cognitive domain. We used the classification system developed in a previous meta-analysis [21] along with referencing to the standard textbooks used in neuropsychological assessments [26,27] for categorizing effects into different cognitive domains. The categorization of cognitive domains adopted in this meta-analysis was done solely for ease of presentation and does not imply that there is universal agreement among researchers on the specific constructs tapped by these measures. The individual results were first assigned to the broad cognitive domain (General Cognitive Ability, Attention and Executive Ability, Language, Memory and Learning, Construction, Perception, Motor functioning) followed by assignment to specific sub-domains within the broader cognitive domain. In addition to the cognitive domains, we coded for the effect of RRMS on mood status, followed by assignment to either the sub-domain of depression or anxiety. For ease of presentation, all tables and figures presented in the paper include the domain of mood status along with other cognitive domains.

In addition, participant characteristics (EDSS level. Duration of MS, age, gender, and education of participants) were coded to examine their role in moderating the impact of MS on cognition. Specifically we coded for the level of neurological disability, assessed usually through the Kurtzke Disability Status Scale (EDSS, [28]). EDSS is a measure of ambulation with scores ranging from 0 (no neurological abnormality) to 10 (death from MS). We used two main categories of classification for the EDSS scores: below 4 or above 4. A cut-off of 4 is usually used in studies to classify participants who can walk with no rest or aid for >500 m [29]. In addition to the level of neurological disability, we also coded for the duration of MS (>10 years or ≤10 years), sex [all female, predominantly female (F:M::2:1), or mixed], age (<30 years, between 30 and 39 years, or more than 40 years) and education of participants (<12 years, or more than 12 years).

Computation of effect sizes and data analyses

We computed effect sizes as Hedges g [30]. This statistic calculates effect size by subtracting the mean of the control group (M_Control) from the mean of the RRMS group (M_MS) divided by the pooled standard deviation (SD_P) such that:

$$g=\frac{M_{MS}-M_{\mathit{Control}}}{{SD}_P}$$

The aggregated or overall mean effect size was computed using a random effects model and adjusted for sample size (Hedges adjusted g) using Comprehensive Meta-Analysis, Version 2.0. [31]. The random effects model operates under the assumption that samples are drawn from populations with different effect sizes and the true effect differs between studies (i.e., the true effect within studies might vary based on differences in sample characteristics, the treatment itself, or the outcome measure) [31]. Effect sizes were interpreted using Cohen’s (1992) guidelines [32], such that:

• Small effects: 0.1–0.3

• Medium effects: 0.4–0.6

• Large effects: >0.7

We further used the Comprehensive Meta-analysis software for testing the heterogeneity of the mean effect size, computing the 95% confidence interval (CI) around the mean effect size, and examining categorical moderators of the mean effect size. Heterogeneity was indicated if Q (sum of squares of each effect size about the weighted mean effect size) was statistically significant (p< 0.05). The effect of the categorical moderators was determined by partitioning the variance into that accounted for by the moderator variable (i.e., Q_B or Q between statistic) and the residual variance (i.e., Q_W or Q within statistic) [25]. The Q_B statistic indicates differences between levels of the moderator variable (i.e., index of differences in group means within a categorical moderator variable) such that a significant Q_B statistic indicates that effect size variability is explained by the categorical variable [31]. Importantly, a significant Q_W statistic indicates that the categorical moderator variable does not account for all of the effect size variability [31].

Although meta-analyses are considered to be an important improvement over the traditional qualitative reviews, an important limitation is that they aggregate results of available studies such that the included articles may just constitute a subset of all studies that might have been published on the topic [25]. We estimated the possibility of publication bias in our meta-analysis using the funnel plot and Fail-safe k. The funnel plot includes a graphical representation of effect sizes vs. standard errors (based on sample size) and suggests against publication bias (i.e., smaller studies are published if there are larger effect sizes) based on its shape as a funnel where no sections of the funnel are missing or asymmetrical. We further calculated Rosenthal’s fail-safe number (k) [33]. The Fail-safe k indicates that the number of null effects from unpublished or unretrieved studies would be necessary to make the overall effect non-significant (i.e., bring the P-value to be greater than alpha of 0.05).

Results

Participant characteristics

Fifty-seven studies met the inclusion criteria and were included in the meta-analysis. There were 3891 participants in the studies and roughly half had RRMS (2042) and the other half were healthy controls (1849). Demographic characteristics of those with RRMS and the controls along with the clinical characteristics of those with RRMS are given in Table 1. The proportion of males in both the RRMS group and the control sample group is roughly 30% (28.72% for patients and 32.87% for controls). The age and education of those with RRMS was comparable to that of healthy controls (mean age of patients and controls = 38.66 and 38.50 years, respectively).

Table 1.

Descriptive characteristics for the RRMS patients and healthy controls

Variables	MS sample		Controls
	Mean	SD	Mean	SD
Sample size	37	32	34	25
Age	38.66	5.95	38.5	7.07
Education	13.52	1.59	14.05	1.31
Percentage of females	71.28		67.13
Disease duration (years)	7.76	3.81	n/a	n/a
Neurological disability (EDSS)	2.49	0.97	n/a	n/a

Main effect of RRMS on cognition

The impact of RRMS on overall cognitive functioning was statistically significant and moderate in magnitude (g = −0.585, 95% CI = −0.619, −0.551). The average effect size was heterogeneous under a random effects model [Q(755) = 3083.21, P<0.05]. Further, the obtained effect was robust to publication bias as suggested by the Funnel Plot presented in Figure 2 and 868,413 null effects would be needed to render the current findings non-significant at the 0.05 level.

Figure 2.

A funnel plot diagram of the 755 effect sizes calculated In the study.

We further calculated an average effect size per study, presented in Table 2. Of the 57 studies, the P-value only one effect was >0, suggesting that for nearly all studies there was cognitive deterioration seen in those with RRMS compared with healthy controls. The distribution of the 57 effects had slight negative skewness (g1 = −1.27) and leptokurtosis (g2 = 2.39).

Table 2.

Average effect size per stud

Study name	Statistics for each study
	Hedge’s g	Standard error	z-value	P-value
Andrade [34]	−0.498	0.047	−10.596	0.000
Andrade, et al. [35]	−0.623	0.073	−8.534	0.000
Archibald and Fisk [36]	−0.293	0.075	−3.907	0.000
Arnett, et al. [37]	−0.19	−0.057	−3.333	0.001
Arnott, et al. [38]	−0.278	0.106	−2.623	0.009
Barkhof, et al. [39]	−0.555	0.154	−3.604	0.000
Beatty, et al. [40]	−0.601	0.056	−10.732	0.000
Birnboim and Miller [41]	−1.118	0.111	−10.091	0.000
Blum, et al. [42]	−0.686	0.197	−3.482	0.000
Bobholz, et al. [43]	−0.821	0.086	−9.547	0.000
Deloire, et al. [44]	−0.474	0.052	−9.115	0.000
Denney, et al. [45]	−0.648	0.068	−9.513	0.000
Denney, et al. [12]	−1.006	0.015	−67.067	0.000
Drake, et al. [46]	−1.04	0.044	−23.636	0.000
Dujardin, et al. [47]	−0.62	0.158	−3.810	0.000
Fisk and Archibald [48]	−0.409	0.197	−2.076	0.038
Forn, et al. [49]	−0.28	0.101	−2.772	0.006
Forn, et al. [50]	−0.478	0.097	−4.928	0.000
Friend, et al. [51]	−0.57	0.057	−10.000	0.000
Gaudino, et al. [9]	−0.403	0.104	−3.875	0.000
Gilad, et al. [52]	−0.873	0.108	−8.083	0.000
Gonzalez-Rosa, et al. [53]	−0.712	0.141	−5.050	0.000
Grigsby, et al. [54]	−0.574	0.157	−3.858	0.000
Haase, et al. [55]	−0.519	0.065	−7.933	0.000
Haase, et al. [56]	−0.651	0.104	−6.243	0.000
Heaton, et al. [57]	−0.475	0.026	−18.269	0.000
Huijbregts, et al. [58]	−0.341	0.052	−6.558	0.000
Kolonoff, et al. [59]	−0.383	0.029	−13.207	0.000
Kotterba, et al. [60]	−0.751	0.213	−3.526	0.000
Kraus, et al. [61]	−0.347	0.059	−5.881	0.000
LaPointe, et al. [13]	−0.725	0.082	−8.841	0.000
Lethlean and Murdoch [62]	−1.415	0.074	−19.122	0.000
Liguori, et al. [63]	−0.416	0.039	−10.754	0.000
Ling and Selby [64]	−0.586	0.049	−11.551	0.000
Mainero, et al. [65]	−0.981	0.143	−6.860	0.000
Manson, et al. [66]	−1.117	0.396	−2.821	0.005
Moore, et al. [67]	−0.763	0.078	−9.782	0.000
Morgen, et al. [68]	−0.805	0.165	−4.879	0.000
Nagy, et al. [69]	−0.817	0.126	−6.484	0.000
Olivares, et al. [70]	−0.454	0.034	−13.353	0.000
Oliveri, et al. [7]	−0.834	0.133	−6.271	0.000
Parmenter, et al. [71]	−0.481	0.093	−5.172	0.000
Parmenter, et al. [72]	−0.707	0.091	−7.769	0.000
Pelletier, et al. [73]	−1.647	0.183	−9.000	0.000
Pozzilli, et al. [74]	−0.513	0.057	−4.981	0.000
Randolph, et al. [75]	−0.532	0.103	−3.224	0.001
Rocca, et al. [76]	−0.226	0.165	−0.883	0.377
Rosti, et al. [77]	−0.669	0.256	−7.194	0.000
Rosti, et al. [15]	−0.669	0.036	−18.526	0.000
Ruggieri, et al. [78]	−0.901	0.075	−12.013	0.000
Ryan, et al. [79]	−0.314	0.036	−8.722	0.000
Selby, et al. [80]	−0.58	0.026	−22.308	0.000
Stablum, et al. [81]	−0.698	0.019	−36.737	0.000
Staffen, et al. [82]	−0.536	0.098	−5.469	0.000
Staffen, et al. [83]	−0.598	0.104	−5.750	0.000
Sweet, et al. [84]	−0.298	0.1	−2.980	0.000
Wishart, et al.	−0.598	0.255	−2.345	0.000

Effect sizes categorized by cognitive domain

The average effect sizes associated with each of the cognitive domains along with the overall effect size are presented in Figure 3.

Figure 3.

Effect sizes of all cognitive domains along with the overall effect size.

With the exception of Motor functioning and Mood Status, most effect sizes were in the moderate range. The Q_B statistic was statistically significant [Q_B (8) = 169.93, P<0.05] and indicated that some domains yielded significantly greater effect sizes than others. The effect sizes of “General Cognitive Ability”, “Attention and Executive Ability”, and “Memory and Learning” were significantly higher than the effect sizes associated with “Verbal functions and Language”, and “Concept formation and Reasoning”. The Q_W statistic was significant for each of the eight cognitive domains and thus, we did additional intra-domain analyses to determine the sources of variability [85].

Table 3 presents a list of all the cognitive domains, the respective sub-domains and the effect sizes for each of these sub-domains. The mean effect sizes were significant for all cognitive domains. With the exception of motor functioning, all cognitive domains were comprised of more than one category and thus, we did additional heterogeneity analyses to examine the additional sources of variability. The Q_B statistic was significant for the cognitive domains of “General Cognitive Ability”, “Memory”, “Attention and Executive functioning”, “Verbal functions and Language”. These domains will be discussed below. Differences between sub-domains were not statistically significant for “Perception”, “Concept formation and Reasoning”, “Construction”, and “Mood Status”.

Table 3.

Effect sizes of all cognitive domains with their respective sub-domains. Also shown are the confidence interval and the overall Q-Statistic

Cognitive domain	Sub-cognitive domain	Number of effect sizes	Average effect sizes	95% CI		Q
				Lower limit	Upper limit
General cognitive ability		35	−0.539*	−0.610	−0.467	139.672*
	Intellectual ability	8	−0.551*	−0.720	−0.381	12.955
	Verbal intellectual ability	9	−0.304*	−0.426	−0.182	8.075
	Non-verbal intellectual ability	6	−0.603*	−0.778	−0.428	1.208
	Cognitive screening tests	12	−0.560*	−0.740	−0.379	28.05*
Attention and executive functioning		237	−0.555*	−0.588	−0.522	580.474*
	Information processing speed	13	−0.646*	−0.807	−0.484	18.169
	Vigilance/sustained attention	10	−0.565*	−0.757	−0.374	27.866*
	Short-term storage capacity	27	−0.430*	−0.515	−0.345	87.395*
	Working memory	85	−0.515*	−0.576	−0.454	157.175*
	Selective/focused attention	63	−0.647*	−0.710	−0.584	187.574
	Executive functioning	29	−0.514*	−0.600	−0.427	43.517*
Perception		15	−0.440*	−0.552	−0.328	84.167*
	Visual perception	11	−0.553*	−0.705	−0.402	74.826*
	Auditory perception	3	−0.291*	−0.485	−0.097	4.481
Memory		215	−0.607*	−0.636	−0.577	1131.03*
	Verbal-immediate recall	75	−0.588*	−0.636	−0.540	381.965*
	Verbal-delayed recall	44	−0.775*	−0.837	−0.714	375.536*
	Verbal-recognition	17	−0.486*	−0.595	−0.377	41.88*
	Visual-immediate recall	39	−0.522*	−0.592	−0.451	177.177*
	Visual-delayed recall	14	−0.548*	−0.666	−0.429	27.355*
Verbal functions and language		64	−0.442*	−0.496	−0.388	233.686*
	Comprehension	3	−0.510*	−0.750	−0.269	42.823*
	Verbal expression	20	−0.340*	−0.432	−0.248	19.78
	Discourse	6	−0.278*	−0.487	−0.070	1.407
	Verbal fluency	26	−0.689*	−0.778	−0.601	41.242*
	Verbal academic skills	13	−0.081	−0.186	0.025	28.904*
Construction		24	−0.529*	−0.613	−0.444	36.45*
	Visual construction	22	−0.535*	−0.621	−0.450	35.697*
	Visual spatial	2	−0.329	−0.797	0.138	0.032
Concept formation and reasoning		68	−0.310*	−0.355	−0.264	184.961*
	Concept formation_Verbal	7	−0.283*	−0.405	−0.160	11.366
	Concept formation_Visual	5	−0.381*	−0.563	−0.198	8.939
	Sorting and shifting	25	−0.346*	−0.43	−0.262	28.575
	Reasoning-verbal and visual	18	−0.367*	−0.456	−0.278	55.806*
	Mathematical reasoning	9	−0.315*	−0.429	−0.202	48.027*
Motor functioning		30	−0.728*	−0.804	−0.652	160.928*
Mood status		31	−0.700*	−0.795	−0.605	126.636*
	Depression	27	−0.701*	−0.804	−0.599	104.699*
	Anxiety	4	−0.688*	−0.956	−0.421	21.929*

^*P-value ≤ 0.05.

General Cognitive Ability deficits

Effect sizes of tests assessing General Cognitive Ability of those with RRMS suggested moderate differences between patients and healthy controls (g = −0.539, P < 0.05). Consistent with the current literature assessing general cognitive deficits and the classifications used in previous meta-analyses [21,86], we used the following sub-domains to classify measures of general cognitive functioning: intellectual ability, verbal-intellectual ability, non-verbal intellectual ability, and cognitive screening tests. The effect size for verbal intellectual ability was small (g = −0.300, P < 0.05), whereas effect sizes for other domains were in the medium range (Table 3). We also found a significant difference between measures assessing verbal intellectual deficits as opposed to non-verbal intellectual deficits such that MS patients showed greater impairment on tests of non-verbal intellectual ability.

Attention and executive functioning deficits

The impact of MS on the broader domain of attention and executive functioning was significant (g = −0.555, P < 0.05). For this cognitive domain, we used the following categories: processing speed, sustained attention/vigilance, short-term storage capacity, working memory, selective/focused attention, and executive control. Patients with MS showed the greatest cognitive impairment on measures of processing speed and selective/focused attention. Selective/focused attention assessed through measures like Symbol Digit Modalities Test (SDMT), Trail Making, Stroop word and color reading was significantly more impaired than measures of working memory and short-term storage capacity.

Memory and Learning deficits

Tasks assessing the broader domain of Memory and Learning found significant performance differences between patients with RRMS and healthy controls (g = −0.607, P < 0.05). For the domain of memory and learning, the following categories were used: verbal-immediate recall, verbal-delayed recall, verbal recognition, visual immediate recall, visual delayed recall, and visual recognition. Although our classification included visual recognition, there were not enough effect sizes to code for this sub-domain and thus was dropped from the final analyses. With the exception of verbal-delayed recall, effect sizes of all other sub-domains were found to be in the medium range. The effect size of verbal-delayed recall was −0.775 (P< 0.0001) and MS patients showed significantly greater impairment on tests of verbal-delayed recall than on any other measure.

Verbal functions and language deficits

The difference between MS patients and healthy controls on this domain was found to be moderate (g = −0.44, P<0.01). Measures assessing this domain were further divided into: comprehension, verbal expression, discourse, verbal fluency, and verbal academic skills. Measures assessing verbal academic skills were not found to be significantly different between MS patients and healthy controls (g = 0.081, P = ns). Consistent with the literature [21,22], we found that tests assessing verbal fluency skills in MS (g = −0.689, P<0.001) were significantly more impaired than tests assessing Comprehension (g = −0.51, P<0.001), verbal expression (g = −0.341, P<0.01) and discourse (g = −0.278, P<0.01).

Effect sizes categorized by specific tests

In addition to examining the impact of MS on specific cognitive domains, we were interested in the sensitivity of different tests used in the MS literature to assess neuropsychological functioning. Out of the 755 effects examined, the seven most commonly used tests were analyzed. These included the color-word Stroop test, Verbal Fluency, Symbol Digit Modalities Test (SDMT), Paced Auditory Serial Addition Test (PASAT), Wisconsin Card Sorting Test (WCST), Selective Reminding Test, and the N-Back test. The Q_B statistic was significant, suggesting that some tests used yielded significantly higher effect sizes than others. We found that the Stroop task. Verbal fluency measures and the SDMT had higher effect sizes than the PASAT, WCST and the Selective Reminding test.

Impact of moderator variables on the relationship between RRMS and cognition

Given the heterogeneity in the mean effect of RRMS on cognition, we examined the influence of a number of moderating variables first on the overall mean effect size and then secondly within each of the cognitive domains. Only statistically significant tests of moderators based on the Q_B statistic are reported below, with the complete list presented in Table 4.

Table 4.

Lists clinical and demographic variables that were used as moderator variables in the analyses

Moderator	Level of moderator	Number of effect sizes	Average effect size	95% CI		Q
				Lower limit	Upper limit
Age-matched	Yes	710	−0.550	−0.566	−0.533	3065.338*
	No	36	−0.485	−0.557	−0.413	77.011*
Education-matched	Yes	658	−0.548	−0.565	−0.53	2959.515*
	No	26	−0.385	−0.482	−0.288	15.776
Gender-matched	Yes	501	−0.554	−0.575	−0.534	1834.717*
	No	154	−0.495	−0.534	−0.457	645.449*
Age of MS patients	<30 years	86	−0.450	−0.499	−0.402	176.987*
	Between 30 and 39	322	−0.492	−0.518	−0.465	813.408*
	More than 40 years	291	−0.640	−0.666	−0.614	1660.93*
Education of MS patients	<12 years	76	−0.515	−0.57	−0.46	247.85*
	More than 12 years	455	−0.562	−0.573	−0.531	2239.7*
Gender of MS patients	All female	24	−0.593	−0.695	−0.491	21.893*
	Predominantly female(F:M::2:1)	374	−0.592	−0.617	−0.567	1497.364*
	Mixed	141	−0.482	−0.519	−0.444	316.606*
EDSS Level	<4	425	−0.512	−0.535	−0.489	1330.144*
	≥4	89	−0.547	−0.586	−0.508	782.54*
Duration level	<10 years	339	−0.546	−0.571	−0.521	1105.481*
	More than 10 years	150	−0.570	−0.604	−0.536	1149.96*

We found that the effect of RRMS on cognition differed when studies matched the gender [Q_B(1) = 6.9, P<0.05] and education [Q_B(1) = 10.44, P < 0.05] of patients to that of healthy controls. We further found that this relationship was moderated by the age of MS patients such that patients who were 40 years or older showed a significant deterioration of cognitive skills as opposed to younger patients. We also found a main effect of gender such that studies that recruited primarily females demonstrated greater cognitive deficits as opposed to studies using mixed samples [Q_B(2) = 23.43, P<0.05]. Interestingly, we did not find a main effect of EDSS level, disease duration and education levels on the relationship between MS and cognition.

For these three moderator variables, we did additional analyses to examine the impact of these moderating variables on specific cognitive domains. We found that for all three variables (EDSS level, disease duration and education levels), the Q_B statistic was significant for the domain of memory and learning. MS patients with an EDSS score of 4 or more performed worse on tasks of memory and learning (g = −0729, P < 0.05) than patients with an EDSS of <4. Similarly, patients with more than 10 years of disease duration performed worse (g = −0.69, P < 0.05) than patients with <10years of education. Moreover, those with <12 years of education (g = −0.674, P < 0.05) performed worse than patients with 12 years of education or more. This suggests that EDSS levels, disease duration and education specifically moderate the relationship between MS and cognition, such that the impact was primarily observed on tasks of memory and learning.

Discussion

The primary goal of this meta-analysis involved the quantification of the overall magnitude of cognitive impairment in persons with RRMS compared with healthy controls. Indeed, one of the main findings was the overall moderate degree of impairment in cognitive functioning in those with RRMS. We further note that the individuals with RRMS demonstrated significant impairments compared with healthy controls across all domains of cognitive function. Our results extend the findings in the extant literature by providing a quantitative metric of the magnitude [87–90] of cognitive problems in this population. Such findings further underscore that the degree of cognitive impairment in RRMS is not trivial, but rather moderate in magnitude. This is important as cognitive deficits interfere with everyday functioning and are associated with feelings of depression, low self-esteem and impaired social functioning [91–95]. Therefore, there is a clear need for identifying strategies for mitigating cognitive impairment in RRMS.

The cognitive deficits in RRMS differed in magnitude based on the domain of cognitive function. The largest detriments in cognition were observed for motor functioning and mood status, followed by memory and learning, and attention and executive functioning. Motor functioning assessed through the grooved pegboard, finger tapping, nine-hole peg test was severely affected (g = −0.728, P < 0.05) and this is consistent with findings of a previous meta-analysis that included all types of MS [22]. Such a finding is not surprising given the profound motor disturbances that are a defining feature of MS. The domain of mood status further demonstrated large cognitive deficits and the largest effect size was seen for measures of depression (g = −0.701, P < 0.05), although the Q_B statistic was not significant for this domain and the effect of RRMS on depression was not significantly greater than the effect on anxiety. Such findings are particularly important for interventions designed to reduce cognitive deficits in this population by suggesting that emotional variables may either be influenced and/or influence the cognitive functioning of persons with RRMS. These results underscore the importance of investigating the efficacy of current psychotherapeutic approaches with this population as a target of both emotional and cognitive symptoms (See ref. [96] for such an intervention).

An important goal for future research, as mentioned above will be to design intervention studies that simultaneously target cognitive and emotional symptoms experienced by patients with MS. In order to successfully design such studies, it is important to delineate variables that might affect or moderate the relationship between MS and cognition. In our search for such moderating variables, we found that age and gender of MS participants had a significant bearing on the relationship between MS and cognition. Although, at first we did not find either neurological disability or disease duration to significantly contribute to the variability in effect sizes, domain-specific analyses suggested that MS patients with an EDSS score of 4 or more and a disease duration of more than 10 years performed poorly on tasks of memory and learning. This finding thus suggests that when the effect sizes for different cognitive domains are averaged, neurological disability or disease duration fail to show an impact, however, for the domain of memory and learning, both of these variables impact performance. Given that this particular domain shows the largest effect sizes, the results are interesting in that they suggest that the disability progression in MS is rather divergent. In the meta-analysis of memory impairments in MS [6], such a similar finding was noted for chronic-progressive MS course. This meta-analysis, which examines RRMS patients, provides evidence for the impact of clinical variables on the domain of memory and language.

In this meta-analysis, we were also interested in the sensitivity of the different measures used in the literature to assess declines in cognitive functioning. We found that the classic color-word Stroop task and measures of verbal fluency showed the largest effect sizes (g = −0.785 and g = −0.76, P < 0.05), suggesting that these measures might be more sensitive in detecting cognitive deficits as opposed to other measures. This finding is important as it suggests that these two measures, if included, in the brief neuropsychological batteries developed to assess neurocognitive functioning in patients with MS, would be more efficient at detecting subtle declines in cognitive functioning.

To summarize, the results of the current meta-analytic review suggest that the effects of MS on cognition is both general, such that all cognitive domains show a differential performance effect as well as specific, such that the effects are largest for the domains of mood, motor functioning and memory and learning. We further found evidence for the association between several demographic and clinical variables on selective aspects of cognition.