Effectiveness of aerobic exercise for adults living with HIV: systematic review and meta-analysis using the Cochrane Collaboration protocol

Background People with HIV are living longer with the health-related consequences of HIV, multi-morbidity, and aging. Exercise is a key strategy that may improve or sustain health for people living with HIV. Our aim was to examine the safety and effectiveness of aerobic exercise interventions on immunological, virological, cardiorespiratory, strength, weight, body composition, and psychological outcomes in adults living with HIV. Methods We conducted a systematic review using the Cochrane Collaboration protocol. We searched databases up to April 2013. We included randomized controlled trials comparing aerobic exercise with no exercise or another intervention performed at least three times per week for at least four weeks among adults living with HIV. Two reviewers independently determined study eligibility. Data were extracted from studies that met inclusion criteria using standardized forms. We assessed risk of bias using the Cochrane Collaboration’s tool for assessing risk of bias. Outcomes were analyzed as continuous and meta-analyses conducted using random effects models with Review Manager (RevMan) computer software. Results Twenty-four studies met inclusion criteria (n = 936 participants at study completion); the majority of participants were men (73 %) and the majority were taking antiretroviral therapy (19/24 included studies). The exercise intervention included aerobic exercise alone (11 studies) or a combination of aerobic and resistive exercise (13 studies) ranging from 5 to 52 weeks. Fifty-eight meta-analyses were performed. Main results indicated statistically significant improvements in selected outcomes of cardiorespiratory status (maximum oxygen consumption, exercise time), strength (chest press, knee flexion), body composition (lean body mass, percent body fat, leg muscle area), depression symptoms, and quality of life (SF-36 questionnaire) among exercisers compared with non-exercisers. No significant differences in change in CD4 count and viral load were found. Conclusions Performing aerobic exercise or a combination of aerobic and resistive exercise at least three times per week for at least five weeks is safe and can lead to improvements in cardiorespiratory fitness, strength, body composition and quality of life for adults with HIV. Aerobic exercise is safe and beneficial for adults living with HIV who are medically stable. Electronic supplementary material The online version of this article (doi:10.1186/s12879-016-1478-2) contains supplementary material, which is available to authorized users.


Background
Access to combination antiretroviral therapy has transformed HIV into a chronic illness whereby many individuals are living longer and aging with the health-related consequences of HIV, adverse effects of treatment and multi-morbidity [1][2][3][4]. These consequences may be known as "disability" including symptoms and impairments (problems with body function or structure, such as pain or fatigue), activity limitations (difficulties in executing dayto-day activities, such as inability to walk), challenges to social inclusion (problems in life situations, such as inability to work, personal relationships), and uncertainty about future health (worrying about the future) [5,6].
Exercise is a strategy employed by people living with HIV and by rehabilitation professionals to address disability and improve or sustain the health of people living with HIV [7]. Exercise has been shown to improve strength, cardiovascular function, and psychological status in general populations [8,9]. Similar benefits of aerobic exercise were documented in earlier versions of this systematic review among adults living with HIV [10,11]. However, knowledge about the benefits and risks of exercise, and optimal parameters for exercise for adults living with HIV is still emerging. If the risks and benefits of exercise for people living with HIV are better understood, appropriate exercise may be undertaken by people in this population and appropriate exercise prescription may be practiced by healthcare providers. Effective and safe exercise may enhance the effectiveness of HIV management, thus improving overall health outcomes for adults living with HIV.
Our aim was to examine the safety and effectiveness of aerobic exercise interventions on immunological, virological, cardiorespiratory, strength, weight, body composition, and psychological outcomes in adults living with HIV.

Methods
We conducted a systematic review using the Cochrane Collaboration protocol [12].

Inclusion criteria
We included randomized controlled trials (RCTs) comparing aerobic exercise (or combined aerobic and resistance exercise) with no aerobic exercise or another exercise or treatment modality performed at least three times per week for at least four weeks [13]. We included studies of adults (18 years of age and older) living with HIV at all stages of infection with or without comorbidities. We defined aerobic exercise as a regimen containing aerobic interventions performed at least three times per week for at least four weeks. Aerobic interventions included but were not limited to walking, jogging, cycling, rowing, stair stepping, and swimming. Interventions may or may not have been supervised [13].

Outcomes
We assessed immunological (CD4 count, cells/mm 3 ) and virological (viral load, log10 copies) outcomes. Cardiorespiratory measures included but were not limited to maximal oxygen consumption (VO2max), exercise time, oxygen pulse, maximum heart rate, maximum tidal volume, minute ventilation, lactic acid threshold (LAT), maximum work rate, and rate of perceived exertion. Strength measures included amount of weight able to resist in kilograms (1-repetition maximum) for major muscle groups. Weight and body composition measures included any outcome that contributes to the direct or indirect measurement of muscle, fat, bone or other tissues of the body. These included but were not limited to body weight, body mass index (BMI), lean body mass, girth, percent body fat, crosssectional muscle area, and waist and hip circumference. Psychological measures included general measures of psychological status and health-related quality of life.

Search strategy
In the update of this systematic review, we searched databases from 2009 to April 2013 including Medline, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects, PsycINFO, CINAHL, EMBASE, Web of Science: Science Citation Index, SPORTdiscus, Virology and AIDS Abstracts and LILACS. We also searched clinicaltrials.gov and reference lists from pertinent articles. All languages were included. See Additional file 1 for the detailed MEDLINE search strategy which we modified as needed for use with other databases.

Selection of included studies
All abstracts retrieved from the search were reviewed independently by two reviewers (KKO and AMT) who applied the following four inclusion criteria to determine if the abstract warranted further investigation: a) Did the study include human participants who were HIV positive? b) Did the study include adults 18 years of age or older? c) Did the study include an aerobic exercise intervention performed at least three times/week, at least 20 min per session for at least four weeks? d) Was there a randomized controlled comparison group?
When the review based on the abstract alone indicated that one or both raters believed the study met eligibility criteria (i.e., if reviewers answered "yes" or "unsure" to the four questions) then full versions of the article were independently reviewed by the two reviewers to determine article inclusion. In instances where there was a lack of agreement by the two reviewers, a third reviewer was asked to review the full article to determine final inclusion.

Data extraction
Data were extracted onto standard data extraction forms independently by at least two reviewers (KKO and AMT and/or SAN). Data extracted included the study citation, study objectives, study design, length of study, time at which participants were assessed, inclusion and exclusion criteria for participants, characteristics of included participants (i.e., age, gender, stage of disease, comorbidity), description of intervention(s) (i.e., frequency, intensity, duration, type, level of supervision, location of intervention), types of outcome variables assessed and their values at baseline and study completion, and number of participants at baseline and study completion (including number of withdrawals). For the purposes of this review, constant exercise was defined as exercise at a constant intensity for a period of time. Interval exercise was defined as exercise conducted at a varied intensity for a total period of time. The reviewers met to achieve consensus regarding any difference in data interpretation or extraction from included studies that arose during the review process. In the case of missing data, authors were contacted in an attempt to obtain further information.
Two authors assessed the risk of bias in the included studies using the Cochrane Collaboration's tool for assessing risk of bias [14]. Potential biases in studies may have included selection bias (random sequence generation and allocation concealment which may result in systematic differences in the comparison groups), performance bias (blinding of participants and personnel which could lead to systematic differences in the care provided apart from the intervention being evaluated), detection bias (blinding of outcome assessment that may result in systematic differences in outcome assessment), attrition bias (incomplete outcome data), and reporting bias (selective reporting of outcomes) [14].
We assessed the overall quality of evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) method [15]. We rated the quality of evidence for each outcome based on categories of very low, low, moderate and high [16]. We downgraded the evidence from high quality by one level for each of the following: attrition bias (where withdrawal rates were >15 %), performance bias (when participants were not blinded to the intervention), detection bias (when assessors of outcomes were not blinded to group allocation), publication bias (when publication bias was suspected), and inconsistency (when moderate I 2 > 40 % or substantial I 2 > 75 % heterogeneity exists) [17].
We produced a summary of findings (SoF) table for the main comparison of exercise versus no exercise with the following seven outcomes: immunological (CD4 count) and virological (viral load); cardiorespiratory (VO2max); strength (upper and lower body), weight (body weight); body composition (body mass index); and quality of life (SF-36 sub-scale scores). The SoF table was developed to illustrate the confidence in the effect estimates (quality of evidence) and magnitude of effect for seven key outcomes [18].

Analysis
Outcomes were analysed as continuous and dichotomous outcomes whenever possible. Meta-analyses were performed using the random-effects model for outcomes using Review Manager (RevMan) computer software whenever there were sufficient data available in the studies, when similar or comparable outcome measures were used, and when participant comparison groups were similar [19].
For continuous outcomes, the weighted mean difference (WMD) and 95 % confidence intervals for the means were calculated whenever possible. For dichotomous outcomes, the odds ratio, absolute difference in odds, relative risk (RR), risk difference (RD), and the number needed to treat (NNT) and 95 % confidence intervals were calculated whenever possible. A p value of less than 0.05 indicated statistical significance for overall effect. Subgroup analyses were performed whenever possible to estimate whether aerobic exercise interventions were associated with differences among groups using identified outcome measures.
We considered 50 cells/mm 3 to indicate a clinically important change in CD4 count, 5 % to indicate a clinically important change in CD4 percentage, and 0.5 log10 copies to indicate a clinically important change in viral load. For cardiorespiratory outcomes, we considered 2 mL/kg/min to indicate a clinically important change in VO2max, 10 beats per minute to indicate a clinically important change in heart rate maximum (HRmax), and 5 min to indicate a clinically important change in exercise time. For strength outcomes we considered 5 kg to indicate a clinically important change in strength for lower extremities, 2 kg to indicate a clinically important change in strength for upper extremities. For weight and body composition outcomes, we considered 3 kg to indicate a clinically important change in body weight (which equals approximately 5 % of the average baseline body weight of participants), 5 cm to indicate a clinically important change in girth (waist and hip circumference), 3 cm to indicate a clinically important change in waist-to-hip ratio, 5 kg/cm 2 to indicate a clinically important change in body mass index, 5 kg to indicate a clinically important change in fat mass, 5 % to indicate a clinically important change in percent body fat, and 5 cm 2 to indicate a clinically important change in leg muscle area. For psychological outcomes, we considered 10 points to indicate a clinically important change in the sub scales of the SF-36 quality of life questionnaire; and 5.6 to indicate a clinically important change in the sub scales of the Profile of Mood States (POMS) scale [20]. While no established minimal clinically important difference (MCID) exists for the SF-36 questionnaire specifically with people living with HIV, we drew from other literature with clinically important differences for the SF-36 with other chronic conditions which demonstrates a small change is approximately 10 points [21,22]. No other established MCID values exist for the other outcomes. Authors based the a priori estimates based on a combination of clinical experience and interpretations in the individual included studies.
We considered a p value of less than 0.1 as statistical significance for heterogeneity between studies [23]. We considered I 2 ≤ 40 as low heterogeneity, I 2 > 40-75 % moderate and I 2 > 75 % substantial heterogeneity [17]. In instances of lack of statistical significance for an overall effect, confidence intervals were assessed for potential trends that may suggest movement towards an increase or decrease in overall effect. In instances of statistical significance for heterogeneity, we performed sensitivity analyses and explained potential reasons for heterogeneity.

Results
Fourteen studies were included in the previous systematic review [10]. In this update, we retrieved a total of 529 citations, of which 58 were judged to merit scrutiny of the full article. Of the 58 studies reviewed, 11 met the inclusion criteria, one of which was a duplicate study [24] resulting in a total of 10 studies included in this fourth update [25][26][27][28][29][30][31][32][33][34] (Fig. 1-PRISMA Flow Diagram). Hence, a total of 24 studies (14 from the earlier review and 10 from the update) were included in this systematic review (See Table 1-Characteristics of Included Studies). An additional seven articles were identified as duplicate publications of studies included in the review: Kaushik [24] and Fitch [31] (in this update); and; LaPerriere [35] and LaPerriere [36]; Lox [37] and Lox [38]; Neidig [39] and Smith [40]; Fairfield [41] and Grinspoon [42]; Driscoll [43] and Driscoll [44]; and Mutimura [45] and Mutimura [46] (in the earlier review). In these instances, we extracted outcomes from all available sources but refer to the initial citation and/or the citation that included our primary outcomes of interest.                  For the purpose of this review, only the aerobic and control groups were included in meta-analyses c LaPerriere (1990) [36] participants were not included in overall calculation of total number of participants because it was unclear how many participants were HIV positive in the baseline sample

Characteristics of participants
A total of 1242 participants were included in the review (number of participants in included studies at baseline). Participants included adults living with HIV at various stages of HIV infection, with CD4 counts ranging from <100 cells/mm 3 to greater than 1000 cells/mm 3 . Studies included both men and women, with women comprising approximately 22 % of the total number of participants at study completion. The mean age of the participants in the included studies ranged from 30 to 49 years (inclusion criteria ranged from 18 to 65 years of age). Four studies (17 %) were published prior to introduction of combination antiretroviral therapy (prior to 1996) [36,38,47,53] [25,26,28,29,[31][32][33][34]. The majority of participants in 14 studies were taking combination antiretroviral therapy including 72 % [42]; 82 % [51] and 100 % on highly active antiretroviral therapy [25, 26, 28-34, 44, 46, 52]. Smith [40] reported 23 % of participants were taking protease inhibitors and all other participants were on some form of antiretroviral therapy. Five studies included participants who were not on combination antiretroviral therapy; however, others reported including participants taking some form of antiretroviral medication [38,[48][49][50] (Table 1).

Correspondence with authors
We wrote to authors of 11 included studies for clarification and additional data, three of whom responded. Yarasheski provided additional data including mean change and standard deviations of body mass index outcomes and viral load outcomes [25]. Agostini clarified the intervention included a combination of aerobic and resistive exercise and provided more details on the intervention. Authors indicated they were not able to provide raw data on body weight, fat mass, muscle mass, or waist circumference (data were reported as % increase or decrease) [34]. We requested SF-36 Physical Component Scores (PCS) and Mental Component Scores (MCS) from Tiozzo who responded with data on the eight SF-36 sub-scale scores [26].

Risk of bias
An overview of risk of bias of included studies is provided in Fig. 2. We describe details of potential bias of included studies below.

Allocation concealment
Overall an unclear risk of selection bias exists as 18 of the 24 included studies (75 %) did not describe the allocation sequence of participants. Six of the 24 included studies (25 %) had low risk for selection bias describing methods to conceal the allocation sequence of participants [27,42,44,46,50,51] (Fig. 2).

Performance bias
Overall a high risk of performance bias exists across the included studies. Twenty-one of the 24 included studies (88 %) had a high risk for performance bias because participants were not blinded to the exercise intervention.
Of these, five studies reported single-blinding of study personnel who were assessing outcomes to the group allocation [27,29,[31][32][33]. In two studies, participants were blinded to co-interventions including metformin [31] and testosterone [42]. Nevertheless, we considered all of the above studies as high risk for bias given the inability to blind participants to the exercise intervention ( Fig. 2). Unclear risk of performance bias occurred in three studies. In studies that compared aerobic to resistive exercise, different intensities of exercise, or where both comparison groups included some form of exercise, blinding was unclear [30,34]. In MacArthur [53], participants were not told whether they had been randomized to the low-intensity group or the high-intensity group but it is unclear whether study personnel were blinded [53] (Fig. 2).

Detection bias
Overall an unclear risk of detection bias exists as 19 of the 24 included studies (79 %) did not provide sufficient information to determine whether study personnel were blinded to the outcome assessment. Five studies (21 %) had low risk for detection bias reporting blinding of study personnel who were assessing outcomes to group allocation [27,29,[31][32][33] (Fig. 2).

Incomplete outcome data (Attrition Bias)
Overall 303 participants withdrew from the included studies resulting in an overall~24 % withdrawal rate (303/1242 participants at baseline  (Fig. 2). Withdrawal rates were similar between comparison groups for the majority of included studies. Three studies reported differences in the characteristics of participants who withdrew from the study [29,33,50] with more women and/or African Americans withdrawing from two studies [29,50] and older participants with less familial history of diabetes remaining in the other study [33]. Twenty-three of the 24 included studies (96 %) made reference to participants who withdrew from or were non-adherent (or non-compliant) with the intervention. Withdrawal rates were not reported by LaPerriere [36]. Rates of withdrawal for individual studies are provided in Table 1.

Selective reporting (Reporting Bias)
Overall a low risk of reporting bias exists as the majority of included studies; 20/25 (83 %) were free of selective outcome reporting as authors provided data on all prespecified outcomes. Four studies (17 %) had incomplete or inconsistent data [28,33,34,53]. Agostini [34] provided outcome data for body weight, fat mass, muscle mass and waist circumference in % increase or decrease only. Authors responded to our request stating that they did not have access to the raw data. Balasubramanyam [33] did not report all outcomes for body composition and cardiorespiratory fitness. MacArthur [53] only reported on six of the participants who were compliant with the exercise intervention [53]. In Ogalha [28], the data across tables are inconsistent and authors did not report data for all outcomes such as viral load (Fig. 2).

Other potential sources of bias
Overall a low risk for other sources of potential bias exists as the majority of studies (92 %) appeared free from other problems that could place a study at high risk of bias. Two studies possessed unclear risk of additional bias [32,33]. Balasubramanyam [33] reported receiving > $10,000 funding from Abbott and research support from consultancy fees [33]. In Farinatti [32], a higher number of participants were assigned to the exercise group to maintain the sample size at study completion if adherence to exercise was low; it is unclear if this may have skewed results [32].

Group similarity at baseline
Sixteen of the 24 included studies (67 %) reported that comparison groups were similar at baseline [25, 27, 29-32, 34, 36, 40, 44, 46-50, 54]. MacArthur [53] and Grinspoon [42] did not report on group similarity at baseline [42,53]. Lox [38] indicated significant differences between comparison groups for "most" participant characteristics, but variables that were different between the groups were not specified. Dolan [51] indicated that the exercise group had more endurance and knee strength at baseline. Terry [52] indicated that the low lipid diet group had a small but significantly higher haemoglobin level compared with the combined exercise and low lipid diet group [52]. Balasubramanyam [33] indicated that the family history of diabetes was less frequent in the diet and exercise intervention group compared with the non-exercising control group. Ogalha [28] indicated a small but significant difference in baseline fasting total cholesterol and high-density lipoprotein cholesterol closer to normality levels in the control group compared with the exercise group (mean rates were higher in the exercise group at baseline). Tiozzo [26] indicated that a higher number of participants in the exercise group had a smoking history compared with the control group placing them at higher risk of cardiovascular disorders. Faster mean heart rate recovery was evidence among participants in the control group compared with the exercise group at baseline [26].

Meta-analyses -effects of interventions
Seventy-five meta-analyses were completed across eight sub-group comparisons in this review (17 of which included similar studies) resulting in 58 unique metaanalyses to this systematic review. Meta-analyses were performed for immunological and virological outcomes (CD4 count, CD4 percentage, and viral load), cardiorespiratory outcomes (VO2max, HRmax, exercise time), strength outcomes (chest press, leg press, knee extension, knee flexion, upper and lower body strength), weight and body composition outcomes (body weight, body mass index, lean body mass, fat mass, percent body fat, leg muscle area, waist circumference, hip circumference, waist-to-hip ratio), and psychological outcomes (quality of life, depression-dejection symptoms).
Of the 58 unique meta-analyses, 28 were new to this systematic review update, 16 were updated with additional studies, and 14 were the same as in the previous review [10]. Subgroup comparisons of the meta-analyses included 1) constant or interval aerobic exercise or combined aerobic and progressive resistive exercise (PRE) versus no exercise; 2) constant or interval aerobic exercise versus no exercise; 3) constant aerobic exercise versus no exercise; 4) interval aerobic exercise versus no exercise; 5) moderate-intensity aerobic exercise versus heavy-intensity aerobic exercise; 6) constant or interval aerobic exercise combined with PRE versus no exercise; 7) aerobic exercise versus PRE; and 8) combined aerobic exercise and diet and/or nutrition versus diet and/or nutrition only.
The number of meta-analyses was limited due to variability in types of exercise interventions (aerobic exercise vs. combined aerobic and PRE exercise), level of exercise supervision, types of outcomes reported, and methodological quality. Aerobic interventions in the trials varied according to constant compared to interval exercise and moderate compared to heavy intensity exercise, and combined aerobic and resistive exercise compared to aerobic exercise alone. See Table 1 for characteristics of included studies and descriptions of specific interventions in individual studies.

Heterogeneity
Heterogeneity (p < 0.1) was evident in 31 of the 58 unique meta-analyses (53 %). Reasons for heterogeneity may include differences in the types of participants in relation to antiretroviral use, body composition, comorbidity, gender, type and location of intervention, as well as methods of outcome measurement. We conducted sensitivity analyses on 18 of the 31 meta-analyses with heterogeneity (those with greater than two studies included in the meta-analysis. We discuss the specific sensitivity results and reasons for heterogeneity within the analyses below.

Immunological and virological outcomes
Twenty-two of the 24 included studies (92 %) assessed immunological or virological outcomes, or both, in the form of CD4 count or viral load. Fourteen of the studies included a non-exercising control group, seven of which included combined interventions of aerobic and PRE [26,27,31,32,42,47,51]. Seven studies also measured immunological and virological outcomes but did not include a non-exercising control group [25,28,33,44,[52][53][54].

CD4 count (cells/mm 3 )
Seven meta-analyses were performed for CD4 count. The majority (6 out of 7 meta-analyses) demonstrated no statistically significant changes in CD4 count between comparison groups (Table 3). Results demonstrated a non-significant trend towards an increase in change in CD4 count for participants in the aerobic or combined aerobic and PRE intervention group compared with the non-exercising control group; constant or PRE compared with no exercise; and significant increase in CD4 count for interval aerobic exercise compared with no exercise ( Table 3). The point estimate in the latter two meta-analyses was above 50 cells/mm 3 , which suggests a trend towards a potential clinically important improvement in CD4 count among exercisers compared with non-exercisers.
Meta-analyses resulted in no difference in change in CD4 count for constant or interval aerobic exercise compared with no exercise; constant aerobic exercise compared with no exercise; as well as combined aerobic exercise and diet and/or nutrition counselling group compared with diet and/or nutrition counselling alone (Table 3). Similarly, no difference in change in CD4 count was found for participants exercising at moderate compared with heavy intensity (Table 3).

Heterogeneity -CD4 Count
Six of the seven meta-analyses were statistically significant for heterogeneity (p < 0.1). Sensitivity analyses were conducted for five of the meta-analyses with greater than two studies. While removing combinations of studies reduced heterogeneity, sensitivity analyses did not change the overall effect of exercise on CD4 count beyond clinical importance.

CD4 percentage
Three meta-analyses were performed for CD4 percentage, two of which included similar studies. Meta-analyses   (Table 3).

Heterogeneity -CD4 percentage
One meta-analysis was statistically significant for heterogeneity (P < 0.1) ( Table 3). Sensitivity analysis did not reduce heterogeneity, which was likely attributed to differences in characteristics of participants in the included studies.

Viral load (log10copies)
Four meta-analyses were performed for viral load, two of which included the same studies. Meta-analyses demonstrated no difference in change in viral load for participants in the aerobic exercise intervention group compared with the non-exercising control group as well as the constant aerobic exercise group compared with the non-exercising control group; no difference in the combined aerobic and PRE group compared with the non-exercising control group; and no difference for participants in the aerobic or combined aerobic and PRE intervention group compared with the non-exercising control group (Table 3). None of the meta-analyses were significant for heterogeneity.

GRADE rating -viral load
We are moderately confident in the non-significant effect estimate of 0.18 log10copies in viral load demonstrating no difference in change in viral load comparing aerobic exercise (or combined aerobic and PRE). The true effect is likely to be close to the estimate of effect, but there is a possibility that it may be substantially different. This outcome was downgraded from high to moderate GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies were >15 %) (see Additional file 2 -GRADE Summary of Findings Table).

VO2max
Six meta-analyses were performed for VO2max, five of which were significant favouring exercise compared with non-exercise. Meta-analyses showed a significant improvement in change of VO2max of 2.63 mL/kg/min for participants in the aerobic exercise intervention group compared with the non-exercising control group (Table 4); significant improvement in change of VO2max of 2.40 ml/kg/min for participants in the constant aerobic exercise group compared with the non-exercising control group (Table 4); significant improvement in VO2max of 3.71 ml/kg/min for participants in the combined aerobic and PRE group compared with the non-exercising control group (Table 4); significant improvement of 2.87 ml/kg/ min for participants in the aerobic or combined aerobic and PRE group compared with non-exercising control group (Table 4); and a significant trend towards a greater improvement in VO2max of 4.30 mL/kg/min for participants in the heavy-intensity exercise group compared with the moderate-intensity exercise group (Table 4). No significant difference in change in VO2max was found for participants in the combined aerobic exercise and diet or nutrition counselling group compared with the diet or nutrition counselling group only (Table 4). All point estimates were greater than 2 mL/kg/min, which suggest a potential clinically important improvement in VO2max among exercisers and a greater improvement with heavyversus moderate-intensity exercise.

Heterogeneity -VO2max
Four of six meta-analyses were statistically significant for heterogeneity (p < 0.1). Sensitivity analyses (three performed) indicated that removing Mutimura [45] from the aerobic versus non-exercise comparison, constant exercise versus non-exercise comparison, and aerobic or combined aerobic and PRE comparison successfully reduced heterogeneity. Results for overall effect remained statistically significant favouring exercise; however, the point estimate was reduced to 1.64 mL/kg/min (95 % CI: 1.06, 2.22), 1.53 mL/kg/min (95 % CI: 0.94, 2.12), and 1.99 mL/kg/min (95 % CI: 1.25, 2.73) respectively (not shown), which are below the threshold for clinical importance. Reasons for heterogeneity may be due to differences in characteristics of participants in the Mutimura [45] study; they were from Rwanda and all possessed moderate to severe body fat redistribution.

GRADE rating -VO2max
We have very little confidence in the effect estimate demonstrating a significant increase of 2.87 ml/kg/min for VO2max comparing aerobic exercise (or combined  indicates potential clinically important improvement in outcome aerobic and PRE) with non-exercising control. The true effect is likely to be substantially different from the estimate of effect (Table 4). This outcome was downgraded from high to very low on the GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies >15 %), suspected publication bias, substantial heterogeneity (I 2 = 67 %); and because the lower level of the confidence interval did not cross the estimated clinically important change in VO2max (despite the estimate surpassing our hypothesized clinically important change in VO2max of 2 ml/kg/ min) (see Additional file 2 -GRADE Summary of Findings Table).

Maximum Heart Rate (HRmax)
Three meta-analyses were performed and showed a nonsignificant trend towards a decrease in HRmax of −9.81 beats/min, 7.33 beats/min and 4.91 beats/min for participants in the aerobic exercise intervention group compared with the non-exercising control group; aerobic or combined aerobic and PRE group compared with the non-exercising control; and combined aerobic and PRE compared with non-exercising control (Table 4), respectively.

Heterogeneity -maximum heart rate
Heterogeneity was present in all three meta-analyses (p < 0.1). Removing Perna [48] and Perez-Moreno [27] removed heterogeneity from the aerobic exercise or combined aerobic and PRE intervention versus non-exercise control comparison and the overall effect demonstrated a small but significant decrease in maximum heart rate of −19.21 beats per minute [95 % CI: −22.87, −15.55] (not shown). Reasons for heterogeneity may be due to differences in characteristics of participants in the included studies; participants in Perez-Moreno [27] were all in prison and co-infected with Hepatitis C.

Exercise time
Two meta-analyses were performed and significant increases in exercise time of 3.29 min were found for participants in the combined aerobic and PRE group compared with the non-exercising control group (Table 4); and 2.66 min for participants in the aerobic or combined aerobic and PRE group compared with the non-exercising control group (Table 4). Point estimates did not reach the 5 min threshold for clinical importance.

Heterogeneity -Exercise time
Both meta-analyses were statistically significant for heterogeneity (p < 0.1). Removing Rigsby [47] and Smith [40] from the aerobic exercise or combined aerobic and PRE intervention versus non-exercise control comparison and Rigsby [47] from the combined aerobic and PRE versus non-exercise control comparison removed heterogeneity and the overall effect remained significant, but was reduced to 1.72 min [95 % CI: 1.03, 2.42] among exercisers compared with control (not shown). Reasons for heterogeneity may be due to differences in characteristics of participants in the included studies. See Table 2 for individual study results for outcomes unable to be combined in meta-analyses.

Strength outcomes
Eleven of the 24 included studies (46 %) assessed strength outcomes [26, 27, 30-32, 38, 42, 44, 47, 48, 51]. Ten metaanalyses were performed, four of which included duplicate studies. Meta-analyses demonstrated significant improvements in upper and lower body strength as measured by increases in 1-repetition maximum for chest press, and knee flexion; and a non-significant improvement (trend) towards increases in 1-RM for leg press and knee extension for participants in the combined aerobic and PRE group versus non-exercising control group (Table 5). Two metaanalyses were conducted comparing aerobic versus resistive exercise. Significantly greater increases in strength were found among participants in the PRE group compared with participants in the aerobic exercise only group for upper and lower muscle groups (Table 5).
All six point estimates for upper and lower extremity strength are greater than 2 kg and 5 kg respectively indicating a clinically important greater increase in strength for resistive exercisers compared with aerobic exercise.

Heterogeneity -Strength
Heterogeneity was present in four meta-analyses. Removing Grinspoon [42] from the combined aerobic and PRE versus control comparison reduced heterogeneity (p = 0.95) for knee extension and the overall effect became significant for exercise compared with no exercise. Reasons for heterogeneity may be attributed to differences in study participants. Participants in Grinspoon [42] had signs of AIDS-related wasting.

GRADE ratings -Strength
Our confidence is limited in the effect estimate of a significant increase of 11.86 kg for 1-repetition maximum for chest press comparing aerobic exercise (or combined aerobic and PRE) with non-exercising control. The true effect may be substantially different from the estimate of effect (Table 5). This outcome was downgraded from high to low on the GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies were >15 %), publication bias suspected, and moderate heterogeneity (I 2 = 46 %). However, the estimate demonstrated a significant effect for improvement in chest press and the lower limit of the confidence interval surpassed our hypothesized clinically important change   Table).
We have very little confidence in the effect estimate of a non-significant increase of 50.96 kg for 1-repetition maximum for leg press comparing aerobic exercise (or combined aerobic and PRE) with non-exercising control. The true effect is likely to be substantially different from the estimate of effect (Table 5). This outcome was downgraded from high to very low on the GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies were >15 %), publication bias was suspected, and there was considerable heterogeneity (I 2 = 88 %). Furthermore, the confidence intervals cross the clinically important improvement and deterioration for change in lower body strength (see Additional file 2 -GRADE Summary of Findings Table).

Weight
Fourteen studies assessed body weight [25,26,28,30,33,34,38,40,42,44,46,51,52,54]. Five meta-analyses were performed, two of which included the same studies. Meta-analyses demonstrated no difference in change in mean body weight for participants in the aerobic exercise group compared with the non-exercising control group as well as participants in the constant aerobic exercise group compared with the non-exercising control group; no difference in the combined aerobic and PRE group compared with the non-exercising control group; and no difference in the aerobic or combined aerobic and PRE group compared with non-exercising control ( Table 6). Results also demonstrated no significant difference in change in body weight for participants in the combined aerobic and diet/nutrition counselling group compared with diet/nutritional counselling group only ( Table 6).

Heterogeneity -Weight
Heterogeneity was present in two of the five metaanalyses (p < 0.1). Removing Balasubramanyam [33] from the combined aerobic and diet/nutritional counselling versus non-exercise control comparison reduced heterogeneity (p = 0.24) but the overall effect remained nonsignificant (1.34 kg; 95 % CI: −0.24, 2.9) (not shown). Reasons for heterogeneity may be due to differences in the comorbidity of participants in the included studies. In Balasubramanyam [33], participants had dyslipidemia, in Ogalha [28], 54 % of participants had lipodystrophy and in Terry [52] participants had hyperlipidemia.

GRADE rating -Weight
We are moderately confident in the effect estimate of a non-significant increase of 0.38 kg for body weight comparing aerobic exercise (or combined aerobic and PRE) with non-exercising control. The true effect is likely to be close to the estimate of effect; but there is a possibility that it is substantially different ( Table 6). This outcome was downgraded from high to moderate on the GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies were >15 %), and moderate heterogeneity (I 2 = 48 %). The confidence interval limits and effect estimate do not cross our estimated clinically important change in body weight (see Additional file 2 -GRADE Summary of Findings Table).

Body mass index
Results demonstrated no difference in change in body mass index for four comparisons of participants in the aerobic or combined aerobic and PRE group compared with non-exercising control; constant aerobic exercise compared with non-exercising control; combined aerobic and PRE exercise group compared with nonexercising control and combined aerobic exercise and diet/nutrition counselling group compared with diet/nutritional counselling group only (Table 6).

GRADE rating -Body Mass Index
We are very confident with the effect estimate of a nonsignificant increase of 0.07 kg/m 2 for body mass index comparing aerobic exercise (or combined aerobic and PRE) with non-exercising control. The true effect is likely to be close to the estimate of effect; but there is a possibility that it is substantially different ( Table 6). This outcome was not downgraded on the GRADE quality of evidence because this was an objective outcome of interest and publication bias was not suspected (see Additional file 2 -GRADE Summary of Findings Table).

Lean body mass
Meta-analyses demonstrated a significant increase in lean body mass of 1.75 kg for participants in the aerobic or combined aerobic and PRE group compared with participants in the non-exercising control (Table 6), No difference in lean body mass was found for participants in    (Table 6).

Leg muscle area
Results demonstrated a significant increase in change in leg muscle area of 4.79 cm 2 among participants in the combined aerobic and PRE group compared with the non-exercising control group (Table 6).

Percent body fat
Results also found a significant decrease in percent body fat of 1.12 % for participants in the constant aerobic exercise group compared with participants in the nonexercising control group and a significantly greater decrease in percent body fat of 2.35 % among participants in the combined aerobic exercise and diet or nutrition counselling group compared with diet or nutritional counselling group alone (Table 6).

Fat mass
Results demonstrated no difference in change in fat mass for two comparisons of participants in the aerobic or combined aerobic and PRE group compared with non-exercising control, and combined aerobic and PRE exercise group compared with non-exercising control ( Table 6).
Waist and hip circumference and waist-to-hip ratio No significant differences were found in change in waist circumference, hip circumference or waist-to-hip ratio for participants in the aerobic or combined aerobic and PRE group compared with non-exercising control; as well as participants in the constant aerobic versus exercise groups and combined aerobic and PRE exercise groups (Table 6). Results found a slightly greater increase in waist to hip ratio of 0.02 (95 % CI: 0.01 to 0.03) for participants in the combined exercise and diet or nutrition counselling group compared with diet or nutritional counselling group only however these results were not clinically important.

Heterogeneity -Body Composition
Heterogeneity was present in five meta-analyses for body mass index; waist circumference; and waist-to-hip ratio (p < 0.1). Removing Mutimura [45] from the aerobic or combined aerobic and PRE exercise intervention versus non-exercising control comparison reduced heterogeneity for body mass index (p = 0.41) and waist circumference (p = 0.16) but the overall effects remained non-significant (not shown). Removing Balasubramanyam [33] from the combined aerobic and diet/nutritional counselling versus non-exercise control comparison reduced heterogeneity for body mass index (p = 0.48) but the overall effect remained non-significant (not shown). Reasons for heterogeneity may be due to differences in participants in the included studies. The Mutimura [45] study was conducted in Rwanda opposed to the other included studies conducted in developed countries. In Balasubramanyam [33], participants had dyslipidemia, in Ogalha [28], 54 % of participants had lipodystrophy and in Terry [52] participants had hyperlipidemia.

Health-related quality of life
Meta-analyses were performed for the eight sub-scales of the SF-36 questionnaire (2 studies; 59 participants). Results demonstrated statistically significant and clinically important improvements (>10 point change) on sub scales of mental health, role emotional and physical functioning sub scale scores, as well as statistically significant improvements in role physical, general health, and energy/vitality sub-scale scores of the SF36 questionnaire for participants in the aerobic or combined aerobic and PRE group compared with participants in the non-exercising control group. Results represent a clinically important improvement in mental health, role emotional and physical functioning compared to nonexercisers. A statistically significant decrease in pain sub-scale score was found indicating an increase in pain favouring non-exercisers compared with exercisers. No difference in change in social functioning was found between groups (Table 7).

Heterogeneity -Health-related quality of life
Two of the meta-analyses were significant for heterogeneity (p < 0.1) however we were unable to conduct sensitivity analyses given only two studies were included in the meta-analyses.

GRADE rating -Health-related quality of life
We have very little confidence in the effect estimate demonstrating an improvement of 6.47 in SF-36 sub scale scores comparing aerobic exercise (or combined aerobic and PRE) with non-exercising control. The true effect is likely to be substantially different from the estimate of effect (Table 7). This outcome was downgraded from high to very low on the GRADE quality of evidence due to incomplete outcome data (withdrawals of included studies were >15 %), publication bias was suspected, and due to variable heterogeneity on specific meta-analyses of sub scales ranging from (I 2 = 0 to 87 %). Nevertheless, the estimates for mental health, role  Table).

Depression-dejection symptoms
One meta-analysis was performed and demonstrated a significant improvement in the depression-dejection sub scale of the Profile of Mood States Scale (POMS) by a reduction of 7.68 points for participants in the aerobic exercise intervention group compared with the nonexercising control group (Table 7). This represents a clinically important improvement in depressiondejection among exercisers compared to non-exercisers.

Adverse events (Safety)
Safety in the form of monitoring adverse events was reported in 12 of the 24 studies (38 %) [25, 27, 29, 31-33, 40, 42, 47-49, 51] (Table 2). Meta-analysis was not possible due to the scarcity and variability of reporting adverse events. Adverse events were reported in five of the 24 studies, none of which were attributed to exercise or considered serious [31,33,47,48,51]. Rigsby [47] reported one death during the study in the counselling group; however, the death was not attributed to the exercise intervention. Perna [48] reported one hospitalization during the course of the study [48]. Dolan [51] reported that one participant in the exercise group had an exacerbation of asthma and one participant in the nonexercising group experienced chest pain at baseline . Balasubramanyam [33] indicated that adverse events (such as headaches, dizziness, flushing, diarrhea, nausea and vomiting and fatigue) were infrequent and events appeared to be similar across the comparison groups. Fitch [31] reported two participants in the exercise group experienced muscle strains related to the resistance training requiring modification of weights. Otherwise, no serious adverse events were reported and the exercise program was well tolerated. Three studies reported no serious adverse events, health problems, or complications [25,27,32]. Two studies reported that no participants withdrew due to illness or infection [40,49]. Grinspoon [42] reported having no reported deaths, adverse events or side effects during the course of the study. Maharaj [29] reported that none of the participants showed any adverse effects on their clinical status of CD4 counts, viral load, or increase in opportunistic infections, heart, respiratory and blood pressure either during or after the exercise intervention. The other studies did not report on any outcomes of adverse events (Table 2).

Discussion
Ten additional studies were incorporated into this systematic review, eight of which were able to be included in meta-analyses [26][27][28][29][30][31][32][33]. This enabled us to perform 28 new meta-analyses for outcomes of CD4 count, VO2max, maximum heart rate, strength (chest press, leg press, knee extension and knee flexion, upper and lower muscle groups), weight, body mass index, percent body fat, lean body mass, waist circumference, hip circumference, waist-to-hip ratio, and quality of life. We were also able to incorporate additional studies into 16 metaanalyses from the previous review.
Meta-analyses suggest that performing constant or interval aerobic exercise, or a combination of constant aerobic and PRE for at least 20 min three times per week for at least five weeks appears to be safe and may lead to improvements in selected outcomes of cardiorespiratory fitness (maximum oxygen consumption, exercise time), body composition (lean body mass, leg muscle area, percent body fat), strength (chest press, knee flexors), and psychological status (quality of life, depression-dejection symptoms). We also found a trend towards potential clinically important improvements in cardiorespiratory fitness and psychological status. Results of this review suggest that aerobic exercise appears to be safe for adults living with HIV who are medically stable. This finding is based on the absence of reports of adverse events among exercisers within individual studies and the stability of CD4 count and viral load. Findings should be interpreted cautiously because results are based on participants who completed the exercise interventions and for whom there were adequate follow-up data.
Results of meta-analyses showed statistically significant improvements for outcomes of cardiorespiratory fitness (maximum oxygen consumption (VO2max); exercise time), strength (chest press, knee flexors), body composition (increase in lean body mass, leg muscle area, decrease in percent body fat), and psychological status (quality of life, depression-dejection symptoms).
Cardiorespiratory status results demonstrated potentially clinical important improvements in VO2max among aerobic exercisers compared with non-exercisers, combined aerobic and resistive exercisers compared to non-exercisers, and even greater improvements among participants doing heavy-versus moderate-intensity exercise.
This systematic review is the first to include metaanalyses for strength and quality of life, both of which demonstrated significant improvements among exercisers compared with non-exercisers. Strength results demonstrated potentially clinically important improvements in chest press and knee flexion, among combined progressive resistive and aerobic exercisers compared with nonexercisers. Greater clinically important improvements were found among resistive compared with aerobic exercisers suggesting a combination of aerobic and PRE may be ideal to maximize cardiovascular and strength benefits of exercise [55].
Weight and body composition results reached statistical significance but not clinical importance for lean body mass, leg muscle area, and percent body fat. Interpretations of changes in weight and body composition have shifted since the original review since the widespread use of combination antiretroviral therapy. Increases in body weight and lean body mass, and reductions in body fat may be interpreted as favorable outcomes as a reflection of increase in muscle mass and strength for adults living with HIV. Future updates of this review may be able to conduct sub-group analyses for studies conducted in the pre versus post-combination antiretroviral era.
Psychological status results demonstrated significant and clinically important improvements in quality of life and depression symptoms among exercisers compared to non-exercisers. Results demonstrated significant improvements for six of the eight SF-36 sub-scale scores, three of which demonstrated potential clinically important improvements for physical function, role emotional, and mental health. A significant decrease in pain subscale score on the SF-36 questionnaire for exercisers versus non-exercisers raised the query of whether exercisers may have experienced pain with exercise. This requires further study. Overall, with minimal adverse effects and few attributed to the interventions, exercise appears to be safe for adults living with HIV.
No significant differences were found in all but one meta-analysis for CD4 count or viral load outcomes, suggesting that aerobic exercise has little impact on immunological or virological status. These results were similar to those in previous versions of this review [10].
Results of this systematic review should be interpreted cautiously for a variety of reasons. Meta-analyses were limited due to variation in outcome measures used, comparison groups and types of interventions, enabling only two studies to be combined in the majority of meta-analyses [56]. Studies included in this review demonstrated a high risk of performance bias due to the inability to blind participants to the exercise intervention. This subsequently resulted in low GRADE ratings for quality of evidence (Additional file 2). The inability to blind participants to the aerobic exercise intervention may have resulted in a Hawthorne effect, whereby participants might perceive greater benefits associated with exercise based on the expectation that exercise should be linked to positive outcomes. In addition, the lack of assessor blinding may have resulted in assessor bias whereby assessors may have measured outcomes in favor of the exercise intervention. High risk of attrition bias was also evident due to incomplete data as many studies had withdrawal rates >15 %. Individual studies in this review included small sample sizes and high withdrawal or non-adherence rates (0-76 %). Participants who withdrew from the exercise program were often excluded from the results by the authors, resulting in a per protocol approach to the analysis. Thus, the overall findings among those who continued to exercise might not reflect the general experience of exercise among adults living with HIV. Nevertheless, authors often reported withdrawal rates were similar across comparison groups, and characteristics of participants who withdrew were similar to those who remained in the study, minimizing the potential for migration bias.
The majority of study participants were men between the ages of 18-65 years. This limits the external validity and ability to generalize results to women and older adults living with HIV and comorbidity. Furthermore, the maximum duration of aerobic exercise intervention was 52 weeks. Thus the long-term sustainable effects of aerobic exercise remain less clear. Next, despite combination antiretroviral therapy now increasingly reaching developing countries, this review only includes one study that assessed the impact of aerobic exercise in the developing country context [45,46]. More recently, Ezema [57] similarly reported improvements in VO2max and CD4 count among aerobic exercisers compared with non-exercisers living with HIV in Nigeria. Recent evidence from South Africa also includes an investigation of the impact of home-based and pedometer walking interventions to improve physical activity among adults living with HIV [58,59], demonstrating evidence on the role of exercise and physical activity for adults living with HIV in developing countries. Finally, recent evidence has also described the potential benefits of physical activity and exercise on mental health, neurocognitive outcomes, and activities of daily living for adults living with HIV [60][61][62].
Results of this systematic review are consistent with findings from previous iterations of this review concluding exercise is safe and beneficial for people living with HIV [10,11]. Results are consistent with a systematic review that assessed the effect of combined twice weekly aerobic and resistive exercise on cardiorespiratory status, quality of life, physiologic and functional outcomes for adults living with HIV and similarly concluded that exercise is safe and beneficial for medically stable adults living with HIV [63,64]. Gomes-Neto and colleagues concluded that aerobic exercise demonstrated benefits to cardiorespiratory status and quality of life, specifically improving body composition and aerobic capacity. Meta-analyses were not performed as part of this review; hence findings from our systematic review provide additional details on the pooled effect of aerobic exercise for adults living with HIV. Fillipas [65] conducted a systematic review and meta-analysis to investigate the effect of aerobic and resistive exercise on metabolic and body composition outcomes, including randomized controlled trials involving at least twice weekly exercise [65]. Results indicated that exercise resulted in decreased body mass index, triceps skinfold thickness, body fat percentage, waist circumference and waist-to-hip ratio [65]. We did not perform meta-analyses on metabolic outcomes as this was beyond the scope of this review; however the impact of exercise on metabolic health is increasingly important to consider in the era of combination antiretroviral therapy.

Implications for research
Evidence about the safety and effectiveness of aerobic exercise for adults living with HIV is increasing, including narrative and systematic reviews, and clinical guidelines that support aerobic exercise interventions in the context of HIV [63][64][65][66][67][68][69][70][71]. Recent evidence-informed guidelines also highlight the potential benefits and role of aerobic and resistive exercise for older adults living with HIV [55,72,73].
Interpretations of weight and body composition outcomes should be considered relative to the publication dates of the included studies. Prior to the advent of combination antiretroviral therapy, studies on exercise tended to include participants with AIDS-wasting, whereas recent evidence includes participants on combination antiretroviral therapy with lipodystrophy, body fat redistribution, or hyperinsulinemia. These studies commonly assessed the impact of exercise on weight, body composition, and metabolic outcomes, and reflect our inclusion of weight and body composition in this review. Furthermore, an increasing number of studies include interventions with a combination of resistive and aerobic exercise, which led us to consider the effectiveness of combined PRE and aerobic exercise with non-exercise. We observed an increase in the number of studies that assessed interventions such as tai chi [74,75] or cointerventions with exercise such as diet and/or nutritional counselling [28,33,34,52] metformin [31,44] and pioglitazone [25]. Finally, this review included only 22 % of women participants reflecting a largely under-represented population in the HIV and exercise literature.
Future studies should make efforts to include all participants in an "intent-to-treat" analysis, blind assessor of outcomes, and include clinically meaningful and standardized outcomes to strengthen existing meta-analyses. As the population living with HIV ages, future research should include older adults and those living with multiple concurrent health conditions such as cardiovascular disease, liver disease, kidney disease, and bone and joint disorders. As the HIV and exercise literature expands, future updates of this review should assess the impact of exercise co-interventions, the impact of exercise on outcomes including functional status and metabolic or inflammatory markers and make an attempt to conduct sub-group analyses to acknowledge the changing healthrelated consequences associated with the pre versus post-combination antiretroviral era. Finally, given the majority of studies (18/24) included exercise interventions in clinical settings supervised by health care or research personnel, future research may consider evaluating the effect of non-supervised exercise interventions or community-based fitness programmes that may reflect a self-management model of community-based exercise for people with HIV [76].

Conclusions
Performing constant or interval aerobic exercise, or a combination of constant aerobic exercise and progressive resistive exercise three times per week for at least five weeks appears to be safe and can lead to significant improvements in outcomes of cardiorespiratory fitness (maximum oxygen consumption, exercise time), strength (chest press, knee flexors), body composition (lean body mass, percent body fat, leg muscle area), and psychological status (quality of life, depression-dejection symptoms). Greater increases in strength were found with resistive exercise compared with aerobic exercise. Interpretations of weight and body composition outcomes should be considered relative to the pre versus post combination antiretroviral therapy era. These findings are limited to participants who continued to exercise and for whom there were adequate follow-up data. Aerobic exercise is safe and beneficial for adults living with HIV who are medically stable.

Availability of data and materials
Data supporting the findings can be found in the Tables. Data supporting the GRADE ratings can be found in Additional file 2. Additional data extracted from included studies may be shared upon request.