Benefits and harms of exercise therapy in people with multimorbidity: A systematic review and meta-analysis of randomised controlled trials

Objectives To investigate the benefits and harms of exercise therapy on physical and psychosocial health in people with multimorbidity. Design Systematic review of randomised controlled trials (RCTs). Data sources MEDLINE, EMBASE, CENTRAL and CINAHL from 1990 to April 20th, 2020 and Cochrane reviews on the effect of exercise therapy for each of the aforementioned conditions, reference lists of the included studies, the WHO registry and citation tracking on included studies in Web of Science. Eligibility criteria for study selection RCTs investigating the benefit of exercise therapy in people with multi-morbidity, defined as two or more of the following conditions: osteoarthritis (of the knee or hip), hypertension, type 2 diabetes, depression, heart failure, ischemic heart disease, and chronic obstructive pulmonary disease on at least one of the following outcomes: Health-related quality of life (HRQoL), physical function, depression or anxiety. Summary and quality of the evidence Meta-analyses using a random-effects model to assess the benefit of exercise therapy and the risk of non-serious and serious adverse events according to the Food and Drug Administration definition. Meta-regression analyses to investigate the impact of pre-specified mediators of effect estimates. Cochrane ‘Risk of Bias Tool’ 2.0 and the GRADE assessment to evaluate the overall quality of evidence. Results Twenty-three RCTs with 3363 people, testing an exercise therapy intervention (mean duration 13.0 weeks, SD 4.0) showed that exercise therapy improved HRQoL (standardised mean difference (SMD) 0.37, 95 % CI 0.14 to 0.61) and objectively measured physical function (SMD 0.33, 95 % CI 0.17 to 0.49), and reduced depression symptoms (SMD -0.80, 95 % CI -1.21 to -0.40) and anxiety symptoms (SMD -0.49, 95 % CI -0.99 to 0.01). Exercise therapy was not associated with an increased risk of non-serious adverse events (risk ratio 0.96, 95 % CI 0.53-1.76). By contrast, exercise therapy was associated with a reduced risk of serious adverse events (risk ratio 0.62, 95 % CI 0.49 to 0.78). Meta-regression showed that increasing age was associated with lower effect sizes for HRQoL and greater baseline depression severity was associated with greater reduction of depression symptoms. The overall quality of evidence for all the outcomes was downgraded to low, mainly due to risk of bias, inconsistency and indirectness. Conclusions Exercise therapy appears to be safe and to have a beneficial effect on physical and psychosocial health in people with multimorbidity. Although the evidence supporting this was of low quality, it highlights the potential of exercise therapy in the management and care of this population.


Introduction
Multimorbidity, the coexistence of two or more chronic conditions, is as a priority for global health research (Anon., 2018). Compared to people living with only one chronic condition, people with multimorbidity are more likely to die prematurely, to be admitted to and have an increased length of stay at the hospital, (Menotti et al., 2001;Vogeli et al., 2007) in addition to having poorer physical and psychosocial health, higher intake of multiple drugs and increased health care utilization (Bayliss et al., 2004;Fortin et al., 2006Fortin et al., , 2004. The increasing burden and complexity of multimorbidity challenge current standards of care, which focus on single-disease management approaches rather than individualised care (Pefoyo et al., 2015;Sinnott et al., 2013). One approach to deal with the complexity of multimorbidity is to focus on specific combination of conditions, linked by physiological factors and risk factors (Smith et al., 2018(Smith et al., , 2013. Osteoarthritis of the knee or hip, hypertension, type 2 diabetes, depression, heart failure, ischemic heart disease, and chronic obstructive pulmonary disease are among the leading causes of global disability, affect hundreds of millions of people around the world, and often coexist (Global Burden of Disease Study 2013 Collaborators, 2015). As an example, two out of three people with osteoarthritis have one or more other chronic conditions (Wesseling et al., 2013), with heart failure, ischemic heart disease, hypertension, type 2 diabetes, depression and chronic obstructive pulmonary disease being some of the most common (Reeuwijk et al., 2010). These conditions share a common risk factor (physical inactivity) and pathogenesis (systemic low-grade inflammation) which may trigger a cascade of reactions resulting in the development of a 'vicious cycle' of chronic diseases and poor outcomes (Gleeson et al., 2011;Pedersen and Saltin, 2015).
Several systematic reviews investigating the effect of exercise therapy have demonstrated the safety and benefits of exercise therapy on each of the aforementioned individual conditions, (Bartels et al., 2016;Cornelissen and Smart, 2013;Fransen et al., 2015;Long et al., 2019;Puhan et al., 2016) and clinical guidelines recommend exercise therapy as the cornerstone of treatment (National Institute for Health and Care Excellence, 2013). On the contrary, the few existing multimorbidity guidelines, summarised in a systematic review (Muth et al., 2019), encourage a healthy lifestyle, including regular physical activity, but do not include specific recommendations for exercise therapy, (Muth et al., 2019) which is among the top 10 research priorities for people with multimorbidity (Parker et al., 2019). Moreover, the interventions for people with multimorbidity evaluated so far have shown minor and negligible effects on important outcomes for people with multimorbidity such as health-related quality of life and function Smith et al., 2016). One of the key features of exercise therapy is its anti-inflammatory effects at cellular, tissue and organ level as well as its positive psychosocial and physiological effects such as increase in muscle strength, improved blood pressure regulation and insulin sensitivity (Gleeson et al., 2011;Pedersen and Saltin, 2015). The positive effects of exercise therapy may hence disrupt the 'vicious cycle' of systemic inflammation and improve physical and psychosocial health in people with multimorbidity. This highlights the importance of investigating whether exercise therapy can be recommended as a cornerstone treatment also for people with multimorbidity. No previously published systematic reviews exist on this topic.
We therefore performed a systematic review of randomised controlled trials (RCTs) on the benefits and harms of exercise therapy in people with multimorbidity defined as at least two of the following conditions: osteoarthritis of the knee or hip, hypertension, type 2 diabetes, depression, heart failure, ischemic heart disease, and chronic obstructive pulmonary disease on health-related quality of life, physical function, depression and anxiety.

Methods
This systematic review was guided by the recommendations for performing systematic reviews in the Cochrane Handbook (Higgins et al., 2019) and the reporting was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Moher et al., 2009).

Protocol
The protocol for this systematic review has been published, (Bricca et al., 2020) and it was registered at PROSPERO (CRD42020150628). The statistical analysis plan was made publicly available on the Open Science Framework website (Foster and Deardorff, 2017) (https://osf. io/38fcg/) before the title and abstract (TIAB) screening phase was initiated.

Study design and participants
RCTs published in peer-reviewed journals. We included studies, with adult (>18 years old), reporting at least 80 % of participants with two or more of the following conditions: osteoarthritis of the knee or hip, heart failure, ischemic heart disease, hypertension (systolic blood pressure >90 and diastolic blood pressure >140), type 2 diabetes mellitus, chronic obstructive pulmonary disease and depression as defined by the studies or calculated based on baseline participants characteristics. This pragmatic approach was pre-specified and adopted in an effort to capture all the studies which included people with multimorbidity, given the expected inconsistency of reporting of the conditions across trials.

Intervention
We included exercise therapy interventions with or without additional pharmacotherapy or other adjuvant interventions (e.g. weight loss), based in all settings, including at least 80 % of participants with multimorbidity as defined in our study. Exercise therapy is defined as 'a regimen or plan of physical activities designed and prescribed for specific therapeutic goals with the purpose of restoring normal physical function or to reduce symptoms caused by diseases or injuries' (Caspersen et al., 1985).
We excluded interventions not delivering a structured exercise therapy programme such as providing only a pedometer or booklet to the participants without a specific plan for general physical activity.

Comparator
We included studies comparing exercise therapy interventions to usual/standard care (e.g. counselling from their health care provider), at the time the trial was conducted, and comparator groups non-exposed such as wait-and-see and placebo treatments.

Outcomes
We included studies assessing at least one of the following outcomes: • Physical health: Objectively measured and self-reported physical function (e.g. 6-minute walking test, 36-item Short-Form Health Survey (SF-36)) • Psychosocial health: Health-related quality of life (e.g. EQ-5D questionnaire), depression symptoms and anxiety symptoms (e.g. Hospital Anxiety and Depression Scale), • Adverse events according to the FDA definition (Anon., 2020) that is any unfavourable and unintended sign, symptom or disease temporally associated with the use of a medicinal (investigational) product whether or not considered related to the medicinal (investigational) product and grouped in serious adverse events such as death, hospitalization, disability or permanent damage, and non-serious adverse events such as pain, falls and fatigue.
The rationale for including these outcomes is based on a consensus study which identified outcomes for multimorbidity intervention studies (Smith et al., 2018) and the fact that they are generic and widely used across the conditions of interest. Additionally, the choice of these outcomes was supported by the patient partner of MOBILIZE who were made aware of the systematic review and outcome measures included.

Literature search
We obtained information from searching MEDLINE via PubMed, EMBASE via Ovid, CINAHL (including preCINAHL) via EBSCO and the CENTRAL with no restriction on language. We only included RCTs published from 1990 since the reporting as well as the treatment of multimorbidity has changed substantially within the last years, up to October 12 th , 2019. The search was repeated for the period from October 2019 to April 20 th , 2020 in the same databases to identify additional studies published before manuscript submission. We also screened the reference lists of the latest Cochrane reviews investigating the effect of exercise therapy on the following conditions: osteoarthritis, hypertension, type 2 diabetes, depression, heart failure, ischemic heart disease, chronic obstructive pulmonary disease, the reference lists of included RCTs. Furthermore, we screened for completed trials in The World Health Organization's International Clinical Trials Registry Platform (ICTRP) http://apps.who.int/trialsearch/ which comprises the 16 primary registries of the WHO registry network and ClinicalTrials. gov, and searched Web of Science (WoS) for studies citing the RCTs included in this systematic review (citations tracking).

Search method and study selection
The search strategy was developed for MEDLINE (https://osf. io/84vzn/) and was customised for EMBASE, CINAHL and CENTRAL. All terms were searched both as keywords (Mesh) and as text words in title and abstract, if possible. To identify RCTs, we used the Cochrane sensitive search strategy for identifying RCTs. The results of the literature search were uploaded to EndNote X9.3.1. Two reviewers (AB and LKH) independently screened titles and abstracts and all studies deemed eligible by at least one of the reviewers were checked independently in full text by the same two reviewers. Disagreement between the reviewers about inclusion of individual studies was discussed until consensus was reached. When disagreement persisted, a third reviewer (CBJ) was contacted to resolve the disagreement. We checked whether multiple reports from the same study were published by juxtaposing author names, treatment comparisons, sample sizes and outcomes. When multiple reports of the same study provided different study characteristics (e.g. number of participants and presence of one or more chronic conditions) we used the primary publication. We recorded the reasons for excluding full text RCTs.

Data collection
We used a modified version of the Cochrane Collaboration data collection form for intervention reviews: RCTs only (Higgins et al., 2019). When a study had two intervention groups with two different exercise therapy interventions (A and B) and one usual care comparator group (C), we split the comparator group in two groups of smaller sample size and compared A and B versus C, and reported the results as two separate study comparisons. This procedure is in accordance with the Cochrane handbook (Higgins et al., 2019).
We extracted the following data, from the follow-up immediately after the end of the intervention and the follow-ups closest to 12 months, for continuous outcomes (number of participants, mean and standard deviation, standard error or 95 % Confidence Interval) and for categorical outcomes, we extracted adverse events (cases and total number of participants): • Trial characteristics: location of the trial (e.g. Country), number of patients allocated to the exercise and comparator groups, respectively, number of patients in the intention to treat (ITT) and per protocol analysis, in the intervention and comparator groups, respectively. • Participant characteristics: Age, % female, body mass index (BMI), socioeconomic status (SES) (studies were labelled as low socioeconomic status when the majority of the participants were described as having low education levels, low income, being unemployed or sample otherwise labelled as 'low SES'), baseline severity and diagnosis of the conditions, and number, type and frequency of other conditions. • Intervention and comparator characteristics: Components of intervention (e.g. exercise therapy + education), type of intervention/ comparator interventions (aerobic, neuromuscular, strengthening or a combination of those), frequency of the sessions (times per week), intensity of the session (% of maximum pulse, or % of 1 Repetition Maximum), primary mode of delivery (individual, group, self-help), setting (home-based, hospital-based, or clinic/facility/rehabilitation centres based), duration of the interventions (in weeks), supervision (yes, no or a combination), tailoring (intervention developed according to guidelines and individual patients' needs), adherence to intervention (number of sessions attended out of the total number of planned sessions). • Outcome characteristics: time points assessed and the magnitude of objectively and subjectively measured changes (e.g. change in health-related quality of life, number of adverse events in the intervention and comparator groups). To avoid multiplicity, we used a hierarchy of selection rules for the outcomes. We prioritised data extraction of outcome measures important for the participants (Smith et al., 2018) and generic over disease-specific measures (Bricca et al., 2020). For objectively measured physical function, we prioritised: 1) the 6-minute walking test, 2) Incremental Shuttle Walking Test, 3) any other outcome measure related to daily function (e.g. Chair stand test). For self-reported physical function, we prioritised outcomes in the following order: 1) the 36-item Short-Form Health Survey (SF-36) Physical Function subscale, 2) the SF-36 Role Function subscale, 3) any other self-reported measure of physical function. For health-related quality of life outcomes, we prioritised: 1) the EQ-5D questionnaire, 2) any other health-related quality of life questionnaires, 3) disease-specific health related quality of life questionnaires (e.g. The Minnesota living with heart failure questionnaire). For depression symptoms, we prioritised: 1) The Beck Depression Inventory (BDI), 2) any other depression questionnaire (e.g. the Hospital Anxiety and Depression Scale (HADS depression). For anxiety symptoms, we prioritised: 1) State Trait Anxiety Inventory questionnaire, 2) any other anxiety questionnaire (e.g. HADS anxiety).
If the data could not be extracted from the published studies, we emailed the corresponding author of the study with a checklist including the data we aimed to obtain. If the email we sent bounced back, we contacted the second author and so forth. After three days, we sent a reminder including the last author of the paper. After seven days of the first email, we re-sent the email to the corresponding and last author. A second reminder followed ten days after the first email. We considered the data as missing after not receiving any communication from the authors fifteen days after we sent the first email.

Risk of bias assessment and overall evaluation of the quality of the evidence
Two reviewers (AB and LKH) independently assessed the risk of bias of the included studies using the Cochrane 'Risk of Bias Tool 2.0' (Higgins et al., 2019). Bias was assessed in five distinct domains: Bias arising from the randomisation process or lack of allocation concealment, Bias due to deviations from intended interventions, Bias due to missing outcome data, Bias in measurement of the outcome or delivery of the intervention (blinding), Bias in selection of the reported result. Within each domain, the two reviewers answered one or more signalling questions (e.g. Was the allocation sequence random? Were participants aware of their assigned intervention during the trial?) which lead to judgments of "low risk of bias," "some concerns," or "high risk of bias". The judgments within each domain lead to an overall risk-of-bias judgment for the outcome being assessed (Higgins et al., 2019). The overall quality (or certainty) of evidence for the estimates were evaluated using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach (Schünemann et al., 2013). GRADE is a systematic approach to rate the certainty of evidence across studies for specific outcomes. It is based on five domains that involve the methodological flaws of the studies (i.e., risk of bias), the heterogeneity of results across studies (i.e., inconsistency), the generalisability of the findings to the target population (i.e., indirectness), the precision of the estimates and the risk of publication bias.

Main analyses
We performed meta-analyses using a random-effects model as heterogeneity was expected in participant, intervention and outcome characteristics. For physical function, we performed two separate metaanalyses for objectively and self-reported outcomes, respectively. Standardised mean differences (SMD) with 95 % CIs were calculated for outcome measures of continuous data and adjusted to Hedges g. The magnitude of the effect size of the pooled SMD was interpreted as 0.2 representing a small effect, 0.5 a moderate effect, and 0.8 a large effect (Cohen, 1988). Effect estimates above 0.5 were considered clinically relevant (Cohen, 1988). The relative risk (RR) of serious and non-serious adverse events was estimated separately as the rate of adverse events in the intervention group divided by the rate of adverse events in the control group. A random effects meta-analysis was applied to estimate the overall RR of adverse events in the exercise therapy groups compared with comparator groups (Higgins et al., 2019). Heterogeneity was examined as between-study variance and calculated as the I-squared statistic measuring the proportion of variation in the combined estimates due to between study variance. An I-squared value of 0% indicates no inconsistency between the results of individual trials, and an I-squared value of 100 % indicates maximal inconsistency. Meta-analyses were performed in STATA (16.1) using the 'meta' command.

Analysis of subgroups and meta-regression analyses
For the pre-planned sub-group analyses of studies reporting objectively measured physical function with the 6-minute walk test (6MWT), we estimated the weighted mean difference (WMD) between the intervention and comparator groups. Meta-regression analyses were also performed to identify factors (covariates) which predicted better health outcome when more than ten studies provided data for the covariates of interest, in accordance with the Cochrane Handbook recommendations (Higgins et al., 2019). Relevant study-level covariates were defined as those able to decrease the between-study variance Tau-square (and thus inconsistency measured as the I-squared statistic) (Higgins et al., 2019). Pre-identified factors were extracted based on a systematic screening of the latest Cochrane systematic reviews investigating the effect of exercise therapy on each of the individual conditions of interest and input from members of the study team. These factors were pre-specified and made publicly available prior to the title and abstract screening (htt ps://osf.io/zx9vw/), and are related to participant, intervention and outcome measure characteristics and listed in the 'data collection' above. Additionally, we investigated the impact of risk of bias on the estimates for the outcomes of interest by classifying studies at "low risk of bias", "some concerns", or "high risk of bias" according to the Cochrane Risk of Bias tool 2.0.

Patients' involvement
The MOBILIZE project is committed to patient involvement and has so far included patients living with multimorbidity in all aspects of the decision-making process in the project. Their experiences, needs and preferences play an important role in the development of a novel intervention (Collaborate level on the IAP2 Spectrum of Public Participation).

Study selection and characteristics
The literature search identified a total of 17,547 unique publications, of which 336 individual RCTs were identified and full texts screened for potential eligibility. We ultimately included 23 RCTs from 24 papers involving 41 study comparisons (Fig. 1). One study was reported in two different papers (Edelmann et al., 2011;Nolte et al., 2015). We included them all since they report different outcome measures in different papers and counted them as one study. The number of study comparisons exceed the number of studies because when several intervention groups were included in a study, the between-group difference was reported for each possible comparison, at different follow-ups.

Participants
The most common conditions reported in the 23 RCTs (3363 people) The most common conditions reported in the 23 RCTs (3363 people) were heart failure in 16 studies, depression and type 2 diabetes in 15 studies each, hypertension in 14 studies, chronic obstructive pulmonary disease in 6 studies and osteoarthritis of the knee or hip in 4 studies each. The number of conditions reported per study varied from two to seven and the most common combinations of conditions were heart failure and depression, (Koukouvou et al., 2004;Kulcu et al., 2007;Blumenthal et al., 2012a;Gary et al., 2010;Blumenthal et al., 2012b;Keihani et al., 2014;Abdelbasset et al., 2019;Bernocchi et al., 2018b) (Table 1).

Effect of exercise therapy on health-related quality of life
Thirteen of the 23 studies (including 15 study comparisons) were included in the meta-analysis. Exercise therapy, average duration 13.3 weeks (SD 4.8), promoted a small improvement in HRQoL (SMD 0.37,    A*= Arthritis, not specified which form ( 95 % CI 0.14 to 0.61; I 2 = 63.35 %) (Fig. 2). Meta-regressions indicated that increasing age was associated with lower effect sizes (slope -0.03, 95 % CI -0.05 to -0.01), suggesting that for each additional year in the study, the effect size was reduced by 0.03 SDs (Supplementary Table 1). When including follow-up closest to 12-month post randomisation, mean 39.7-week (SD 14.2), exercise therapy promoted a small effect on HRQoL (mean n = 5, SMD 0.35, 95 %CI 0.18 to 0.52; I 2 = 22.41 %) ( Supplementary Fig. 1).
Meta-regression showed that studies including participants with higher levels of depression were associated with higher reduction in depression symptoms (SMD -0,04, 95 % CI -0.07 to -0.02) Supplementary Fig. 3), suggesting that for each additional increase in depression score, the effect size increases by 0.04 SDs (Supplementary Table 1).

Effect of exercise therapy on anxiety symptoms
Six of the 24 studies (including six study comparisons) were included in the meta-analysis investigating the effect of exercise therapy on anxiety. Exercise therapy (mean 13.2-week, SD 13.4) appeared to have a moderate effect on anxiety (SMD -0.49, 95 % CI -0.99 to 0.01; I 2 = 71.02 %) (Fig. 7). Only one study assessed long term effects (52-week followup) of exercise on anxiety symptoms reporting no effect (SMD 0.03, 95 % CI -0.58 to -0.64).

Adverse events after exercise therapy in people with multimorbidity
Fourteen of the 23 RCTs reported non-serious adverse events data, of which thirteen were included in separate meta-analyses for serious and non-serious adverse events. One study (de Groot et al., 2019) was not included in the meta-analysis due to insufficient data and this study reported no difference in adverse events in the intervention vs the comparator group. The non-serious adverse events reported were knee, arm or back pain, falls, arrythmias, syncope, fatigue and sexual problems. The serious adverse events reported were hospitalisation, pneumonia, cardiac decompression and uncontrolled ventricular arrythmia, sepsis, and extreme fatigue. Meta-analysis showed no difference of non-serious adverse events (RR 0.96, 95 %CI 0.53-1.76; I 2 = 57.49 %) (Fig. 8). However, exercise therapy reduced the risk of serious adverse A. Bricca et al. events, including hospitalisation, death, pneumonia and cardia disorders (RR 0.62, 95 %CI 0.49 to 0.78; I 2 = 0.00 %) between the intervention and comparator groups (Fig. 9).

Risk of bias assessment and overall evaluation of the quality of the evidence
Forty-three percent of the included studies were deemed as 'low' risk of bias, while 57 % as 'some concerns' or 'high' risk of bias (Supplementary Fig. 5). The majority of the RCTs applied a proper randomisation process and reported and assessed the outcomes of interest correctly. Also, there was no clear sign of publication bias in the funnel plots of each meta-analysis (Supplementary 6). However, all the studies were judged as 'some concerns' for the 'measurement of the outcome' item since it is not possible to blind participants to exercise interventions. Sensitivity analyses indicated that studies judged as 'some concerns' or 'high' risk of bias reported more favourable results in favour of the intervention groups than studies judged as 'low' risk of bias for the outcomes depression. We therefore downgraded the quality of evidence for depression due to risk of bias. The overall quality of the  evidence assessed using GRADE, including reasons for downgrading the quality of the evidence, is summarised in Table 2.

Discussion
This systematic review included 23 RCTs using various types of exercise therapy administered in different settings and including more than 3300 people with multimorbidity from 17 countries. Although based on low quality of evidence, we found that exercise therapy is safe and effective to improve physical and psychosocial health. The greatest benefits from exercise therapy were reported for depression, anxiety, health-related quality of life and objectively measured physical function. While exercise therapy characteristics had no impact on physical and psychosocial health, younger people may expect greater improvements in health-related quality of life, and people with higher baseline depression may expect greater reduction in depression following an exercise therapy intervention.

Results in context
The exercise therapy types used included aerobic exercise (47 %), strengthening exercise (9%), a combination of these (35 %), and Tai Chi (9%), and were performed at home (45 %), at a fitness centres or clinical medical centre based (45 %) or at a hospital (10 %). The greatest benefits from exercise therapy were observed for depression, followed by anxiety, health-related quality of life and objectively measured physical function. This is in line with the effect of exercise therapy for each of the chronic conditions investigated (Bartels et al., 2016;Fransen et al., 2015;Puhan et al., 2016;Cooney et al., 2013;Stonerock et al., 2015;McCarthy et al., 2015), and exercise RCTs including people with multimorbidity (de Rooij et al., 2017;Martinez-Velilla et al., 2019). When comparing exercise therapy to primary or community care setting  interventions, which include consultations with nurses, physicians and pharmacists aiming at promoting self-management, we observed that for HRQoL and physical function, the improvements promoted by exercise therapy were greater Smith et al., 2016). However, the indirect comparisons of these interventions, in different multimorbid populations, need to be interpreted with caution.
Exercise therapy appeared to be safe with no risk difference between the exercise therapy and comparator groups on non-serious adverse events. By contrast, exercise therapy reduced the risk for serious adverse events. Although, we were unable to distinguish whether the adverse events were directly related to the interventions, this is promising as adverse events are common in people with chronic conditions (Marengoni and Onder, 2015;Nunes et al., 2016). A systematic review investigating adverse events across all conditions in exercise therapy trials found no risk for an increase of serious adverse events but a 19 % increase in non-serious adverse events (Niemeijer et al., 2019). Differences in risk of non-serious adverse events may be partially explained by the population included and the reduced length of the exercise interventions (on average 13 weeks vs 20 weeks).

Implication for clinical practice
High quality evidence on the effectiveness of different treatments are limited for people with multimorbidity. While future studies are carried out to improve the confidence in the effect of exercise in this population, given that exercise seems safe and beneficial, it can be recommended in clinical practice. People with multimorbidity engaging in an exercise therapy intervention may experience reduced depression and improved health-related quality of life and physical function. These improvements ranged from being not clinically to clinically relevant, with younger people and people with higher depression scores experiencing greater improvements in HRQoL and depression, respectively. For physical function, the improvement observed corresponded to an average increase of 43 m in the 6MWT in favour of the exercise therapy interventions. This improvement is greater than the 30 m cut off for clinical relevance generally used in people with chronic conditions such as chronic obstructive pulmonary disease, lung cancer, coronary artery disease and adults with fear of falling (Bohannon and Crouch, 2017).  A. Bricca et al.

Implications for future research
This systematic review highlights a need for future, high quality, RCTs evaluating the effects of well-designed exercise therapy interventions in people with multimorbidity. Based on the result of this systematic review, we suggest that future studies should also include people with different combinations of conditions. For example, osteoarthritis, while being the most common joint disease and often associated with more chronic conditions (Wesseling et al., 2013), has been investigated in only four of the included studies (Hinrichs et al., 2016;Gary et al., 2004;Gretebeck et al., 2019;Leung et al., 2013).
In the design of future exercise interventions, given the fact that multimorbidity is heterogeneous, focusing on individualised treatments tailored to people's goals and preferences might also help increase the effect estimates and exercise therapy adherence, which is required for expecting greater improvements on health outcomes (Smith et al., 2013;Fransen et al., 2015). In fact, the improvements observed for short-term follow-ups (i.e. immediately after the end of the intervention) were not  maintained for depression and anxiety at long-term follow-ups (i.e. the follow-ups closest to 12 months), presumably due to a decrease in exercise adherence. Also, particular attention should be given to the choice and reporting of comparator groups (Freedland et al., 2019). This may help explain the inconsistency of effect estimates. Regarding the selection of the outcome measures, we suggest future studies to select primary outcome measures important to the patients (Smith et al., 2018(Smith et al., , 2013, as highlighted by a consensus agreement by a group of international researchers working to improve care for people with multimorbidity.

Strengths and limitations
This systematic review with meta-analysis has been conducted and reported according to recommended international guidelines (Higgins et al., 2019;Moher et al., 2009) and it followed a pre-specified protocol (Bricca et al., 2020). Additionally, we have contacted study authors to retrieve missing information from the included RCTs which allowed us to conduct meta-regression analyses with complete data for intervention characteristics such as mode of delivery and setting of the intervention.
This systematic review has limitations. First, among all the conditions included, the majority of the RCTs included people with depression and heart failure and not all studies included 100 % of people with multimorbidity. Additionally, most people of the included studies reported mild to moderate depression severity, limiting the generalisability of the findings to people with other combinations of conditions or more severe depression. However, the combination of the conditions (studies including participants with heart failure and depression vs. studies including participants with different combination of conditions) and the inclusion criteria (80 % of participants with multimorbidity vs. 100 % with multimorbidity) of this systematic review had no impact on effect estimates. Second, the heterogenous nature of the multimorbidity definition and interventions tested reflected the large inconsistency in the effect estimates of the meta-analyses and that positive results may be attributed to combined effects instead of exercise therapy alone. Third, the quality of evidence for the outcomes of interest was low, suggesting that the true effect of exercise therapy may differ from the one reported in this systematic review. Finally, reporting of adverse events was nonconsistent across studies. This is common in exercise trials and highlights the need for harmonising reporting of adverse events (Niemeijer et al., 2019) so that all adverse events are reported regardless of whether they are considered related to the intervention or not.

Conclusions
While the low quality of the evidence limits the confidence in our results, exercise therapy appeared to be a safe and beneficial intervention to improve physical and psychosocial health in people with multimorbidity, highlighting its potential in the management and care of this population.

Author contributions
STS procured the funding for this systematic review. AB and CBJ designed the search strategy. AB and LKH independently screened title/ abstracts, full texts, extracted data from the included studies, assessed risk of bias and contacted study authors to retrieve additional information. AB and CBJ performed the GRADE assessment. AB did the data analysis, and CBJ provided statistical expertise and SM provided clinical expertise about multimorbidity. AB wrote the first draft of the manuscript and STS provided feedback. All authors read, provided feedback and approved the study design, methods, protocol and manuscript drafts as well as approved the final manuscript. AB is the guarantor and drafted the manuscript.

Data statements
The dataset and statistical script necessary to reproduce the analyses presented in the systematic review will be made available online at the Open Science Framework page of the MOBILIZE project (https://osf. io/qk6yg/).

Dissemination
The findings of this systematic review will be presented at national and international conferences and made available to end-users via infographics, podcasts, press releases and videos.

Role of funder
The funding source was not involved in any aspect of this systematic review protocol.