Is aerobic or resistance training the most effective exercise modality for improving lower extremity physical function and perceived fatigue in people with multiple sclerosis? A systematic review and meta-analysis

Conclusions: AT and RT appear equally highly effective in terms of improving lower extremity physical function and perceived fatigue in pwMS. Clinicians can thus use either modality to target impairments in these outcomes. In a future perspective, head-to-head exercise modality studies are warranted. Future MS exercise studies are further encouraged to adapt a consensus ‘core battery’ of physical function tests to facilitate a detailed comparison of results across modalities.


Introduction
Multiple sclerosis (MS) is a chronic, autoimmune, and inflammatory disease of the central nervous system, exemplified through demyelination and axonal loss 1 . As a consequence, multiple symptoms can appear 1-3 , with fatigue and walking limitations reported to be among the most debilitating [4][5][6][7] . Moreover, an estimated 50% of persons with MS (pwMS) will require a walking aid within 15-25 years after disease onset 8,9 . Since physical function is associated with lowered quality of life at the individual level along with a greater economic burden at a health service and societal level 10,11 , it is crucial to diminish progression of disability 12 .
While pharmacological treatments appear to have limited beneficial effect on fatigue and walking limitations 13 , exercise has proven to be a potent non-pharmacological treatment option, being both safe and eliciting numerous beneficial effects in pwMS 14,15 . Specifically, exercise is an effective way of reducing fatigue 16,17 and improving walking performance 18,19 , with the latter often considered to be clinically meaningful 20,21 .
Exercise constitutes a number of different modalities known to elicit different physiological adaptations (such as neuromuscular function or cardiovascular function) that in most cases are paralleled by (and perhaps even translated into) improved physical function 22 . A recent review investigating randomized controlled trials (RCTs) of exercise interventions in pwMS reported that the two most applied exercise modalities were aerobic training (AT) and resistance training (RT) 23 . Several studies have reported positive effects of both AT 24-26 and RT 27,28 on parameters directly related to lower extremity physical function (e.g. walking performance, chair rise, stair negotiation) as well as on parameters indirectly related to lower extremity physical function, such as perceived fatigue. However, based on the existing literature it currently remains unknown which of these two common exercise modalities is the most effective in terms of improving physical function and perceived fatigue in pwMS. Despite the somewhat impossible task of matching AT and RT on traditional exercise parameters such as duration, frequency, and intensity, understanding the specific effectiveness of the two different exercise modalities is an important factor for consideration in optimizing exercise prescription in pwMS.
Therefore, the objectives of this systematic review were to investigate which of the two exercise modalities (AT or RT) are the most effective in terms of improving lower extremity physical function and reducing perceived fatigue in pwMS.

Methods:
The present systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) guidelines on systematic reviews of RCTs 29 . Search strategy, study selection, eligibility criteria, methodology assessment, data extraction and analysis were performed in accordance with a protocol pre-registered in PROSPERO (CRD42020189855).

Definitions:
In this review the following definitions were applied: Exercise: A form of physical activity that is planned, structured and repetitive, and is undertaken with the objective of improving or maintaining at least one aspect of physical fitness, comprising strength, flexibility or aerobic endurance 30 .
Physical activity: Any bodily movement produced by skeletal muscles that requires energy expenditure above resting levels 30 .
Physical function: The ability of an individual to perform physical activities of daily living. For the purposes of this systematic review, this particularly relates to lower extremity tasks (e.g. simple/complex/endurance walking, chair rise, stair negotiation) 31 .
Perceived fatigue: Subjective sensations of weariness, increasing sense of effort, mismatch between effort expended and actual performance or exhaustion 32 .
Resistance training: Performed with external resistance of varying degrees relative to maximal strength provided by either free weights, machines, bodyweight, or some other implements (e.g., resistance bands), either with single or multiple sets of repetitions which may or may not be performed to momentary failure (but are often performed to a relatively high effort) 33 .
Aerobic training: Performed using locomotor or ergometer tasks (e.g., walking, jogging, running, cycling, rowing, etc.) in a continuous or intermittent fashion with respect to duration at submaximal intensities of effort, commonly determined relative to either maximal heart rate, heart rate reserve, VO 2max , or sometimes using ratings of perceived effort scales 33 .

Data sources and search strategy
In brief, the search strategy was based on the key terms "multiple sclerosis" OR MS AND exercise OR "physical activity". For full search strategy please see Dennett et al. 2020 23 .
The original search was carried out in October 2018 and updated in April 2020.
Two reviewers (LM and RD) conducted the original search in the electronic databases MEDLINE, EMBASE, CINAHL, AMED, PEDro, SPORTdiscus, PsycINFO, Web of Science and SCOPUS limited to scientific research papers being published between January 1993 and October 2018. The same databases were searched from September 2018 to March 2020 by two reviewers (LM and LC) in April 2020. All searches were supplemented by hand searches of reference lists.

Study selection
The following PICO (population, intervention, comparison, outcome) question guided the search and inclusion strategy. "Which exercise modality, AT or RT, is most effective in improving physical function (specifically lower extremity tasks such as simple/complex/endurance walking, chair rise, stair negotiation) and perceived fatigue in pwMS?"

Eligibility criteria
RCT studies involving adults over the age of 18 with a definite diagnosis of MS, regardless of gender, disease duration, MS phenotype or level of disability were considered eligible for inclusion. While all identified studies could be included regardless of location, group/ individual structure, level of supervision, intervention duration, session duration, intensity, progression, frequency, the content had to be either AT or RT; with or without a follow-up period.
Control interventions had to include non-training controls only or active control conditions having no expected effects on the cardiovascular system or the musculoskeletal system, for example stretching were accepted.
Studies had to include at least one objective or self-reported measure of lower extremity physical function (such as simple/complex/endurance walking, chair rise, stair negotiation) and/or perceived fatigue. If reported, measures of cardiovascular function (i.e. maximal oxygen uptake) and neuromuscular function (i.e. maximal muscle strength or muscle power) were also extracted, as these outcomes could, (1) help verify the effectiveness of interventions, and (2) are likely mediators of adaptations in lower extremity physical function.

Data management and selection process
The original search resulted in 93 papers included in the previous review, all of which were considered for inclusion in the present review (see figure 1).
Results from the updated search were exported to EndNote, where duplicates were removed.
The remaining papers were imported into Rayyan data management system (rayyan.qcri.org) where titles and abstracts were independently screened for eligibility by two reviewers (LC and LTM). If papers were included at this stage, a full-text reading by the same two reviewers was performed, and any discrepancies were discussed with a third party (LGH). Reasons for excluding full text RCTs were recorded.

Data extraction
Data was extracted using the same spreadsheet as the previous review 23 , which included detailed information on participant characteristics (age, gender, disease duration, MS phenotype, disability level, and fatigue as a symptom); modality of the intervention (setting, group/individual structure, level of supervision, intervention duration, session duration, intensity, frequency); content of the intervention (aerobic or resistance); report of adverse events, % drop-out, and adherence during the intervention period and at any follow up.
Furthermore, an additional customized spreadsheet was made to extract information on all outcomes of lower extremity physical function, perceived fatigue and measures of cardiovascular and neuromuscular function. Data extraction was completed by two reviewers (LC and LTM).

Quality assessment
The methodological quality of the included studies was independently assessed by two reviewers (LTM and LC) using the 'Tool for assessment of study quality for reporting on exercise' (TESTEX) rating scale 37 . Any discrepancies were discussed and resolved between the two reviewers.

Synthesis of results
In addition to the qualitative analysis (summary of identified studies and their data), we also performed quantitative analysis by calculating sample-size weighted averages across selected studies. A minimum of two studies was required in order to conduct a meta-analysis. Random effects meta-analyses comprising data on physiological adaptations, short walking tests, long walking tests and perceptions of fatigue were conducted by using Meta-Essentials version 1.5 designed for Excel. 38 Intervention effect sizes (ES) (between-group differences) for different outcomes at post-treatment, were calculated using Hedges' g statistic, along with 95% confidence intervals (CIs) around the estimated effect-size. Also, if data was available and adequate, we performed a weighted regression of all study ES as a function of intervention duration and frequency (weeks and number of sessions) as well as intervention intensity, as these factors were hypothesized to impact the outcomes 39 . Of note, this approach was done to establish specific within-modality information only. ES were interpreted as follows: small = 0.14, moderate = 0.31, large = 0.61 based on empirical data from 99 meta-analyses examining the effects of rehabilitation/exercise 41 . Statistical heterogeneity was quantified using Higgins' I 2 statistic, and was interpreted as follows: heterogeneity: > 50%, no or limited heterogeneity: < 50% 42 .
If studies reported on more than one outcome in each domain (e.g. physiological adaptations such as knee extensor and knee flexor muscle strength as well as perceptions of fatigue using different questionnaires), an average was calculated and used for the meta-analyses.

Study characteristics
As depicted in figure 1, the search yielded 2117 hits. After removal of duplicates, 1538 papers remained for the screening process, with 12 of these assessed for full-text reading. Five papers were included, which with the addition of 22 papers from the previous review, resulted in a total of 27 papers being included in the qualitative and quantitative synthesis. The median TESTEX score of the included studies was nine out of 15. Detailed information on the scores can be found in Table 2. [INSERT
Six RT studies 28,40,56-59 reported a between-group change in any short walking test, with one of these reporting a significant change ( Table 3

Performance on long walking tests
Of the long walking tests, the Six minute walk test (6MWT) was the most used in AT studies. Four RT studies 28,55,57,58 reported a between-group change in any long term walking test, with one of these reporting a statistically significant finding and the meta-analysis showing a moderate effect of RT on long walking test performance, ES=0.36 [-0.35: 1.08], p=0.11, I 2 =48% ( Figure 4). Long walk ES was positively associated with RT intervention duration (weeks: slope 0.07, r 2 =0.87, p=0.025; number of sessions slope 0.07, r 2 =0.87, p=0.025).

Performance on functional mobility tests
Only one 52 of the AT studies investigated effects on the performance of a functional mobility test, and reported a statistically significant change between groups.
Five 28,56-58,60 of the RT studies investigated the performance on a functional mobility test between groups, with two 56,58 of these changes being reported as statistically significant.
As the aim of this present review was to evaluate differences between modalities, we were not able to conduct a meta-analysis on this outcome.

Self-reported walking performance
Two of the AT 50,52 studies reported a between group change in self-reported walking

Comparison between modalities
While both interventions were shown to elicit adaptations in favor of exercise, we were not able to detect differences in any outcomes between the two different exercise modalities as evidenced by the comparable effect sizes and overlapping confidence intervals.

Discussion
Based on our findings, AT and RT present themselves as broadly equivalent modalities in terms of improving lower extremity physical function (walking performance) and reducing perceived fatigue, with meta-analyses revealing moderate-large effect sizes. Of note, only 14 out of 23 studies reported physiological adaptations thereby limiting the in-depth understanding of the potential mechanistic effect(s) leading to an improvement in physical function (i.e. the translational potential).

Physiological adaptations
Although only seven out of 14 26, [45][46][47][48]50,52  Overall, the physiological adaptations observed by the present systematic review verify that AT and RT interventions overall work as intended, thereby establishing the potential for a translation into improvements in mobility aspects of lower extremity physical function along with reduction in perceived fatigue.

Physical function -walking tests
The identified AT studies predominantly focused on the longer walk tests, with only three studies 43 Previously, the effect of RT on the performance on a short walk test has been summarized in a review 64 and in a meta-analysis based on only one study 19 . However, to our knowledge, this is the first systematic review to perform a meta-analysis on RT studies alone, examining the effects on short walk tests (and walking performance in general). inconclusive, they should be interpreted cautiously. Speculatively, they may indicate that adaptations in objectively measured outcomes precede self-reported outcomes, which is somehow contradictory to what has been shown previously 65 , and/or that adaptations in selfreported outcomes are limited due to a potential ceiling effect.

Physical function -functional measurements
While walking performance is an essential aspect of lower extremity physical function, our sparse and inconclusive findings reveal an existing knowledge gap in terms of how the two exercise modalities (AT in particular) might impact other measures such as chair rise, six spot step test (SSST) and stair negotiation. This is problematic, since complex walking tests such as the SSST 66 along with highly physically demanding walking tests such as stair negotiation 67 , have the potential to give a more in depth picture of patients walking ability. Such tests incorporate not only acceleration and endurance, but also other components such as coordination and balance which are recognized as being important for general physical function. Hence, future AT as well as RT studies should incorporate such complex functional tests in their test battery.

Comparison between modalities
We did not detect any apparent differences in the magnitude of effect on physiological adaptations in the two exercise modalities. Many components such as duration, frequency and intensity should be taken into account when comparing the two modalities. The average frequency and duration was somewhat comparable between the two exercise modalities (AT: 3 days/week*11 weeks (range 3-26 weeks), 28 sessions (range 9-48 sessions); RT: 2 days/week*11 weeks (range 8-24 weeks), 25 sessions (range 15-48 sessions)), along with the intensity being moderate-to-high in both AT and RT. A plausible explanation for the lack of association between intervention duration (weeks and number of sessions) and meta-analysis ES is that the majority of interventions had durations of 8-12 weeks involving 16-24 sessions.
The only exceptions showing positive associations were for RT on muscle strength and long walk test, respectively, although likely driven by one study only 58 having a much longer intervention duration (24 weeks, 48 sessions) compared to the remaining RT studies.
Unfortunately the quantity and quality of the reported exercise intensity data (missing information, use of divergent scales of exercise intensity) did not allow us to examine the associations between exercise intensity and meta-analysis ES within each modality. Since factors such as duration, frequency and intensity are crucial for the extent of adaptations 39 , further studies seem warranted to help advance our understanding of any potential doseresponse association between general exercise parameters (e.g. duration, frequency and intensity) and physiological as well as functional adaptations in pwMS.
To our knowledge, only one pilot study 69 has previously performed a head-to-head comparison of the two modalities, finding no difference in either lower extremity physical function as measured by the six minute walk test and the timed up and go, or in perceived fatigue measured by the Modified Fatigue Index Scale. However, only n=19 participants finished this cross-over study having an eight week wash-out period. Adaptations from exercise interventions may last as long as 12 24 or 24 58 weeks, hence, one must be cautious when interpreting results from this pilot study 69 .
Resembling the observations in physiological adaptions, no difference was observed in the magnitude of change on short or long walking tests with AT or RT. All meta-analyses on the walking tests had comparable moderate ES, although data -based on CIs -appeared most robust for short walk with RT and for long walk with AT, respectively. While this is likely influenced by the number of studies for each meta-analyses, it may also be due to physiological adaptations that are intuitively associated with certain aspects of walking (AT: increment in aerobic capacity associated with walking endurance; RT: increment in muscle strength associated with walking acceleration) 70 . While the present findings are aligned with previously reported findings in systematic reviews and meta-analyses 18,19 , these were based on a limited number of RCT studies (as the search was performed March 2014) 19 or a combination of RCT and non-RCT studies, different exercise modalities, and different measures of walking performance (self-reported as well as clinician-rated short and long walking performance) 18 .
The novel approach of the present systematic review, apart from updating existing evidence, was to include RCTs only, clearly separate study findings across the two most common exercise modalities, and uphold a clear distinction between the selected walking performance outcome measures.
Both modalities were found to be effective in terms of reducing perceived fatigue, with a large ES observed for AT and a moderate ES for RT. While Andreasen et al. 71 in their systematic review previously reported RT to be slightly more effective than AT in terms of reducing

Translational or parallel improvements?
Assessment of physiological adaptations are important due to two aspects. First, it is a simple way of validating exercise efficacy as effects on these basic primary (sensitive) physiological targets are expected (i.e. AT expectedly improve aerobic capacity while RT expectedly improve muscle strength). Second, physiological adaptations may be a prerequisite for improvements in physical function, thereby having a translational effect. Interestingly, the findings from the present systematic review and meta-analyses suggest that improvements in lower extremity physical function can be achieved via different physiological pathways (i.e. cardiovascular system or neuromuscular system). At least, we observed parallel improvements in physiological adaptions and in physical function. However, since only a limited number of studies reported parallel data of both physiological parameters and physical function of the same outcome (see Table 3) and since even fewer studies report associations between changes in these outcomes, we were unable to perform any analysis of association. A small number of studies have reported data supporting an exercise-induced translational link, i.e. between improvements in muscle strength and Fatigue Severity Scale (FSS) 61 , aerobic capacity and FSS 46 , as well as muscle strength and Timed 25-Foot Walk, two minute walk test, five repetition sit-to-stand and stair climb 58 . This is nevertheless challenged by the fact that lower extremity physical function relies on different physiological systems, and adaptations in just one system may elicit little translational response. Also, in high-functioning pwMS the ceiling effect of many commonly used walking measures may mean that changes in performance are not detectable.

Clinical and research implications
The present study findings emphasize the importance of providing structured intensive AT and/or RT when aiming to improve lower extremity physical function (along with physiological adaptations). While many different exercise modalities exist, AT and RT have consistently been shown to be among the most effective in terms of positively affecting numerous different domains 22 . As the two modalities proved somewhat comparable (based on magnitude of ESs), it implies that clinicians could use either modality to target impairments in lower extremity physical function -we suggest patient preference be central to this decision to optimize the likelihood of them sustaining exercise over long term. The inconsistency in reporting across studies, emphasize the need for using a "core battery" of physical function tests, as previously proposed 73 . This would enable comparability of findings across studies and facilitate generation of more robust evidence, which is essential for clinicians' decision-making. Moreover, exercise studies should report data for the physiological outcomes they are targeting. This would advance our understanding of potential translational links between physiology and function.
Finally, future studies should compare the modalities directly by performing a head-to-head study to establish whether differences in outcomes exist.

Study limitations
The present systematic review and meta-analyses, provides a detailed and comprehensive overview of the RCTs investigating the effect of AT and RT on lower extremity physical function and perceived fatigue. However, some methodological considerations deserve mentioning.
First, the majority of identified studies included patients with mild-moderate disease severity, making the results applicable for this subgroup of patients only. Second, more studies are needed to elucidate effects of AT and RT in pwMS with higher levels of disability, including those who are non-ambulatory (EDSS ≥ 7.0), which is a problem that has been exposed previously 74 . Third, this systematic review provides an overview of existing studies evaluating the two modalities, and hence is not able to provide a direct comparison. To provide such information, a well-considered head-to-head study of the two modalities, designed to diminish the difference in intensity and volume, is needed. Finally, our review focused on either solely AT or RT. As such, we cannot comment on the effectiveness of interventions which combine these two exercise modalities or use other exercise modalities (for example pilates, yoga, balance).

Conclusions
Based on knowledge from existing RCTs, aerobic training (AT) and resistance training (RT) appear comparable in improving lower extremity physical function (walking performance in particular) and perceived fatigue. Although substantial physiological adaptations were observed, conclusions about the underlying mechanisms for the improvement are yet to be determined. Future studies should adapt a 'core battery' of physical function tests to facilitate a detailed comparison of results across exercise modalities. This will enable evidence-based treatment selection according to the defined purpose of training.

Funding/Conflict of interest:
The authors Laurits Taul