Database searches yielded 5394 manuscripts, and after removing duplicates (n = 1469) and screening of titles and abstracts, 90 articles remained for full-text assessment (Fig. 1). Sixteen manuscripts (presenting findings from 15 studies) of varying quality (Supplementary Table 2) were identified as eligible [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31] (Fig. 1), consisting of two randomised controlled trials (RCT; findings from one of these trials were presented in two manuscripts) [20, 29, 31], seven prospective cohorts [16, 18, 19, 24, 26,27,28], two cross-sectional [22, 23] studies, one case–control study  and three case-reports [21, 25, 30].
Sample sizes ranged from 1 (case-reports) to 243 (a prospective cohort study ), and included participants with an age range of 20–82 years. Seven studies almost exclusively included patients with high-grade glioma (HGG) (WHO grade III/IV) [19, 21, 22, 25,26,27,28, 30]. Other studies included mixed samples of low-grade glioma (LGG; WHO grade I/II) and HGG [16,17,18, 20, 23, 24, 29, 31], involving newly diagnosed brain cancer [19, 21, 24,25,26,27, 30] or recurrent brain cancer [22, 25, 28] (three articles did not specify [20, 29, 31]).
Aim one: PA levels in people with primary brain cancer
Level of evidence: III-2. PA levels were reported in six studies [19, 22,23,24, 27, 28] (Table 2), including two cross-sectional (n = 171,243) [22, 28] and four longitudinal (range n = 15 to 106) [19, 23, 24, 27] studies. Recruitment rates reported in four of six studies were < 51% (range: 28  to 71% ), with common reasons for not participating being uninterested or time. Attrition rates ranged from 10  to 61% , with four from six studies involving HGG reporting higher attrition rates due to disease progression and deaths (range: 38  to 61% ). All studies used self-reported measures of PA (mostly the Godin Leisure Time Exercise Questionnaire [19, 22,23,24, 28]), with five studies reporting total PA in minutes per week (mins/wk) [19, 22,23,24], or MET-h/wk , and two of the four longitudinal studies involved retrospective collection of pre-diagnosis PA levels [23, 27]. One study categorised participants according to PA levels (i.e. almost completely inactive, some PA < 3 h/week, regular PA, or regular hard physical training > 4 h/week)  and four others categorised participants according to meeting PA guidelines (≥ 150-mins of moderate aerobic exercise per week) [19, 22, 23, 28]. Timing of PA measurement varied from pre-diagnosis, during treatment and post-treatment (Table 2).
Mean PA during or post-treatment ranged from 134 ± 123  to 177.2 ± 164.9  mins/wk. Between 20 and 71% of participants pre- or at diagnosis [19, 23, 27], and 22–41% during or post-treatment met recommended PA guidelines. Longitudinal findings suggested that the proportion of ‘regularly active’ patients more than halved between pre- (59%) and post-diagnosis (25%) . Participants reporting “no exercise” ranged from 24 to 44% [22, 23, 27, 28] (during or post-treatment), and overall, most participants did not meet PA guidelines at any time from diagnosis to follow-up (approximately 60% categorised as insufficiently active or sedentary).
Aim two: PA and health outcomes
Level of evidence for any given outcome: III-3 to III-2. Five studies assessed the association between PA levels and cancer-related outcomes [19, 22, 24, 27, 28] (Table 2), including: survival , QoL FACT-G [19, 24], side-effects relating to brain cancer (FACT-Brain cancer subscale) [19, 24], physical function (6-min walk test , Karnofsky Performance Status ), anxiety (Hospital Anxiety Depression Scale ), muscular strength (lower-limb dynamometry ), and cardiopulmonary fitness (VO2peak ).
Baseline PA (during or post-treatment) was shown to be an independent predictor of survival (p = 0.008) among patients with recurrent grade III/IV brain cancer in a cohort study (n = 243) . Median survival was 22 months (95% CI 13.32–∞) for patients reporting ≥ 9 MET-h/wk compared to 13 months (95% CI 11.25–17.37) for patients reporting < 9 MET-h/wk. Two prospective cohort studies [19, 24] showed positive associations between total weekly PA levels and QoL (that is, higher PA levels were associated with higher QoL and fewer brain cancer specific concerns), but only one was supported statistically . In a mixed (LGG and HGG) sample from a small (n = 35), prospective cohort study, increases in PA levels (from pre- to post-diagnosis) were associated with improvements in muscular strength, body composition, and cardiopulmonary function, although associations were not supported statistically . Other studies failed to show an association between PA levels and physical function  or anxiety .
Aim three: effect of exercise interventions on cancer related outcomes
Nine studies (seven involving newly diagnosed patients [16,17,18, 21, 26, 30, 31]) evaluated the effect of an exercise intervention on cancer-related outcomes in patients with brain cancer (Table 3). These included three case-reports [21, 25, 30], three pre-post intervention studies (n: 5–24) [16, 18, 26], one case–control study (n = 43)  and two RCTs (n: 20–34) [20, 29, 31]. The case-reports related to three patients on treatment [21, 25, 30] and one post-treatment [25, 30]. Two studies evaluated exercise post-surgery (during inpatient rehabilitation) [16, 17], three studies during radiation and/or concurrent chemotherapy [18, 26, 31], and one evaluated exercise a minimum of 6-months post-treatment (surgery and/or adjuvant chemotherapy and/or radiotherapy) [20, 29]. The number of studies that evaluated any given outcome (objectively-assessed or patient-reported) ranged between one and six studies (Table 4).
Exercise intervention details
A summary of the intervention details is presented in Supplementary Table 3. Intervention studies included patients with grade I–IV disease (although most [17, 21, 25, 26, 30] involved patients with HGGs only), five studies included mixed Gliomas (e.g. astrocytoma, glioblastoma) [16, 18, 20, 25, 29, 31] and four studies involved glioblastoma only [17, 21, 26, 30]. Except for one case-report describing an 87-week intervention, the intervention period ranged from 4 to 24 weeks. Most investigated either aerobic exercise only (40%) or mixed-mode (aerobic and resistance exercise; 40%), while two studies (20%) evaluated yoga (Table 3). Frequency of sessions and session duration ranged between one to six days per week and 15–60 mins, respectively. Exercise intensity was moderate or higher as measured by rating of perceived exertion or age-predicted heart rate maximum. Most studies involved some degree of supervision with a qualified exercise professional. Only one home-based study evaluated a completely unsupervised intervention [20, 29]. The remaining studies were conducted either in a class/clinic setting, in-patient or combination of clinic and home-based.
Feasibility, safety and acceptability
Recruitment and retention rates (not including case-reports) ranged from 25  to 83%  and 58  to 100% [16, 17, 26], respectively (Supplementary Table 3). Reasons for not enrolling included being uninterested, lack of motivation, disease progression, physical limitations, and being too busy [16, 18, 20, 31]. Reasons for withdrawal or lost to follow-up included illness progression, travel, returning to work, and lack of motivation or time [16, 18, 20, 29, 31]. Safety was reported in all except two studies [17, 31], with one adverse event (participant lost balance and fell, reporting soreness to the head) recorded . Intervention adherence (average number of sessions attended/sessions planned × 100%) ranged from 61  to 100% [21, 25, 26]. Exercise adherence was reported in two studies as ≥ 75% of participants meeting exercise intensity and duration (70  to 100% ), mean distance cycled per session (6.27 ± 1.29 km) , and mean MET-hours completed per week (43.7 MET-h/wk) . The most common reason for session absence was illness/disease progression [18, 20]. The one study that assessed acceptability, patient-reported satisfaction was rated as “good” to “excellent” by the majority (84%) .
Summary of exercise intervention outcomes
Level of evidence for any given outcome: III-4 to III-2. The effect of exercise on objectively-assessed outcomes and patient-reported outcomes are presented in Table 4. Evidence from one RCT supports clinically- and statistically-significant changes in overall symptoms severity . Statistically-significant differences (p < 0.05) in aerobic capacity, body composition and PA levels were supported by individual RCTs [20, 31]. Outcomes that were found to have a clinically-significant improvement (although p > 0.05) included neurocognitive domains (particularly attentional inhibition, attention span and auditory select attention) [29, 31], mental health-related QoL and mood disturbance, all of which were supported by two, small sample (n = 20–34), RCTs [29, 31]. Symptom interference with daily life was measured in a single RCT and had a clinically-significant change . Within the RCTs no consistent change was observed in self-reported physical functioning [29, 31]. Two RCTs reported improvements in fatigue and cognition, however these changes were only supported clinically in single studies [29, 31]
Preliminary evidence from case–control, pre-post intervention studies, and case-reports suggest that clinically-relevant improvements were observed in lower-body strength, balance, QoL , symptom severity and interference total score , symptom severity related to brain cancer , brain tumour symptoms interference to daily life [17, 31], fatigue , and sleep  following exercise. While upper-body strength, physical functioning, and shortness of breath have also been assessed, no changes were observed [17, 18, 21, 25, 26, 29, 31].