Figure 1 presents the PRISMA flow chart for study selection. Screening of abstracts left twenty one studies for review of the full text. After analysis of these articles, seven studies were excluded because they did not meet the criteria for inclusion [23, 27,28,29,30,31,32]. A table of these excluded studies can be found in Additional file 2.
Characteristics of studies comparing exercise vs waiting list control groups
There were ten randomised controlled trials that compared exercise groups with non-exercise control groups with a total of 422 patients (Table 1).
The numbers of participants ranged from twenty four in the smallest  to seventy four in the largest study . Five trials were included in the meta-analysis [33,34,35,36,37] and five were eligible but the data presentation did not allow them to be included in the meta-analysis [23, 38,39,40,41]. Of the ten studies, five trials included patients with formally diagnosed anxiety disorders [23, 34, 35, 38, 39] and five included patients with clinically raised anxiety, or anxiety sensitivity, on a recognised rating scale [33, 36, 37, 40, 41].
Exercise interventions included running, walking, treadmill training and supervised aerobic training sessions. Intensity of the exercise was not recorded in all of these trials as some of the programmes included exercise undertaken outside of the lab. In these cases the exercise was recorded with pedometers and accelerometers worn by the participants. In those trials were intensity was recorded, the high intensity level ranged from 60 to 90% of HR max.
The duration of the programmes were as follows; four trials were ten weeks long [34, 37,38,39], two were of eight weeks duration [23, 35], three programmes were of two weeks duration [33, 36, 41] and one lasted for six months . Frequency of exercise sessions varied from five times per week in one study , four times a week one study , three times per week in six studies [33, 34, 36, 38, 40, 41], twice a week in one study  and one study did not have a defined number of sessions .
There was a non-active waiting list control group in six studies [33,34,35,36, 40, 41], two studies used psychological therapy in both exercise and control groups [23, 37], and two studies used a non-active control group with placebo pills [38, 39]. The intervention and control groups were only matched for time spent with a therapist in one of the trials .
A range of outcome measures were used in the trials and are given in Table 1. The broad range of rating scales used to measure anxiety outcomes in the trials reflect the spectrum of anxiety disorders represented in the review including conditions such as panic disorder, General Anxiety Disorder, raised anxiety sensitivity and also more generally raised anxiety levels.
Characteristics of studies of high intensity vs low intensity exercise
There were five studies with a combined number of 253 participants which compared high intensity exercise to low intensity exercise [42,43,44,45,46], (Table 2).
One study identified by the search was eligible for the review but could not be included in the meta-analysis due to the data presentation . Three studies included patients with formally diagnosed anxiety disorders [42, 44, 46] and two studies included patients with general anxiety scores above the clinical mean [43, 45].
High intensity intervention groups undertook aerobic exercise such as jogging, treadmill exercise or walking which was performed to a minimum of 60% HR max or 70% VO2 max. Low intensity control groups undertook less strenuous aerobic exercise such as comfortable walking or stretching and flexibility exercises.
The longest programme was of ten weeks duration , three programmes were eight weeks long [42, 44, 46] and one trial was just two weeks long . All of the studies comparing high and low intensity exercise were matched for time spent with the therapist. No trials utilised psychological therapies such as CBT in intervention or control groups.
Risk of Bias in studies
The overall risk of bias for all the different studies included in the review is demonstrated in Fig. 2.
The main source of bias in the studies was that of the lack of blinding of participants and personnel. This was true of the majority of trials and is a common problem where participants engage in an active intervention. Only three studies addressed this problem; in the study by Gaudlitz et al. , participants in exercise groups of differing intensity were blinded as to the group they were in and so were the personnel assessing outcome. This study was judged overall to be the best quality for minimising the risk of bias in all categories. Control participants in the study by Brookes et al were given placebo medication and outcome assessors were also blinded . Participants in the trial by Broman-Fulks et al were unaware of the intervention used in the other group . Attention bias, where the groups were not matched for the time spent with a supervisor, was also a difficulty in some trials. In these instances, the improvement could potentially be due to the time spent with the trainer, rather than the exercise itself. Of the fifteen studies, only eight had matched the time spent with a therapist between the groups [23, 38, 39, 42,43,44,45,46]. Evaluation of risk of bias in each trial is given in Table 3.
The quality of evidence was moderate for all outcomes and is demonstrated in Table 4. The GRADE score was downgraded by one in all outcomes due to the lack of blinding of participants in most of the trials and also for the possibility of attention bias due to time spent with the supervisor being therapeutic in itself.
Exercise vs waiting list control group
The results from the meta-analysis of Exercise group vs Waiting list control group gave an effect size of − 0.41 (95% CI = − 0.70 to − 0.12), where a negative effect size denotes an improvement in anxiety scores. The heterogeneity given by the I2 test was 0% (95% CI = 0 to 61%)) (Fig. 3).
Of the studies that were excluded from the numerical meta-analysis due to data presentation, three of the trials also found that exercise was significantly more effective than placebo in reducing anxiety symptoms [38,39,40]. One study found that exercise did not lead to a statistically significant improvement in symptoms compared to patients in a relaxation group . Table 5 shows the results for the Exercise vs Waiting List Control Group.
High intensity vs low intensity exercise
Figure 4 shows the meta-analysis of the results of four trials which each included a comparison of high intensity exercise with low intensity exercise. The pooled estimate of differences found in those four studies gave an effect size of − 0.38 (95% CI = − 0.68 to − 0.08), suggesting that high intensity exercise training is more effective at lowering anxiety levels than lower intensity training (Fig. 4). The heterogeneity given by the I2 test was 0% (95% CI (0 to 67.9%).
Drop-out rates were greater in the high intensity group in the study by Sexton and equal in the studies by Broman-Fulks and Steptoe et al. [43,44,45]. In the study by Gaudlitz et al. there more non-completers in the low intensity groups than in the high intensity group . Of the trials that were excluded from the meta-analysis, Martinsen et al found that there was no difference in anxiety reduction between walkers and joggers, but that there was a higher drop-out rate in the jogging group (P > 0.1) . Table 6 shows the results for the High Intensity Exercise vs Low Intensity Exercise Groups.
Results from long term follow up scores
Three studies from the high intensity vs low intensity group gave measurements for long term follow up scores [44,45,46]. They all found that the reduction in anxiety from exercise was maintained several months after the training in both high intensity and low intensity groups. Two studies found that high intensity exercise lead to bigger reductions in the long term [45, 46] and one study reported a similar reduction between high intensity and low intensity groups . The combined effect size from these three studies was − 0.30 (95% CI = − 0.72 to 0.12) (Table 7). The heterogeneity given by the I2 test was 18.2% (95% CI = 0 to 77.5%).
Subgroup analysis of results from patients with raised anxiety levels and those with diagnosed anxiety disorders
There were seven trials identified where patients had raised anxiety levels based on a validated rating scale [33, 36, 37, 40, 41, 43, 45], five of these trials gave data which could be included in the meta-analysis [33, 36, 37, 43, 45]. The overall effect size in this group of participants was − 0.46 (− 0.74 to − 0.17) (Fig. 5). Both of the trials not included in the meta-analysis also showed a reduction in anxiety scores in the exercise group compared with the control group [39, 40]. There were eight trials identified were participants had formally diagnosed anxiety disorders [23, 34, 35, 38, 39, 42, 44, 46], of which four were included in the meta-analysis [34, 35, 44, 46]. The overall effect size in this group of patients was a smaller reduction in anxiety symptoms of − 0.32 (0.62 to − 0.01) There was no significant difference in the mean reduction of symptoms between the two groups of trials P = 0.24 (− 0.39 to 0.19).