Exercise Interventions in Polycystic Ovary Syndrome: A Systematic Review and Meta-Analysis

Background: Polycystic ovary syndrome (PCOS) is a common and complex endocrinopathy with reproductive and metabolic manifestations. Exercise training has consistently been found to result in improved clinical outcomes in women with PCOS, but shortfalls with exercise prescription are evident. The aim of this systematic review and meta-analysis was to identify exercise intervention characteristics that provide favourable outcomes in women with PCOS. Methods: A systematic review of published literature was conducted using EBSCOhost and Ovid Medline up to May 2019. The review adheres to the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guidelines as per our PROSPERO protocol (CRD42018088367). Randomised controlled trials, non-randomised controlled trials, and uncontrolled trials that evaluated an exercise intervention of at least moderate intensity in women with PCOS were included. Meta-analyses were performed using general linear mixed modelling and Bayesian inferences about effect magnitudes. Results: Thirty-three articles were identified for systematic review of which 19 were meta-analysed. Intervention duration ranged from 6 to 26 weeks. A total number of 777 women were included in the meta-analysis. The meta-analysis found that improvements in health outcomes are more dependent on exercise intensity rather than dose. Fixed effects analysis reported a moderate increase in VO2peak (24.2%; 90% CL, 18.5–30.1), and small reductions in HOMA-IR (−36.2%; 90% CL, −55.3 to −9.0), and waist circumference (−4.2%; 90% CL −6.0 to −2.3) as a result of vigorous intensity exercise. These results are confirmed in the predicted analysis which reported the greatest improvements in VO2peak, BMI, and waist circumference after vigorous intensity exercise alone or when combined with diet, particularly for women with clinically adverse baseline values. Conclusions: Exercise training in the management of PCOS is becoming more common. Results from our analysis support the use of exercise and suggest that vigorous intensity exercise may have the greatest impact on cardiorespiratory fitness, body composition, and insulin resistance. Our results indicate that, a minimum of 120 min of vigorous intensity per week is needed to provide favourable health outcomes for women with PCOS with studies of longer duration required to evaluate outcomes with sustained exercise.


INTRODUCTION
Polycystic ovary syndrome (PCOS) is one of the most common endocrine conditions, affecting 8-13% of reproductive aged women (Teede et al., 2018a). PCOS is complex with diverse features including reproductive, metabolic, and mental health complications. PCOS is diagnosed via the internationally endorsed Rotterdam criteria, which require the presence of two or more features including clinical or biochemical signs of hyperandrogenism, oligo-or anovulation, and polycystic ovaries on ultrasound, with the exclusion of other aetiologies (Rotterdam EA-SPCWG, 2004). Although not currently recognised in the diagnostic criteria, insulin resistance is a key aetiological factor contributing to the severity of reproductive and metabolic features (Poretsky et al., 1999;Altuntas et al., 2003), with obesity known to exacerbate the severity of clinical symptoms (Lim et al., 2012). Consequently, women with PCOS are at a two to eight times greater risk of developing impaired glucose tolerance and type 2 diabetes mellitus compared to women without PCOS (Moran et al., 2010).
Exercise is well-established as a therapy for preventing and managing chronic diseases in the general population (Booth et al., 2000;Roberts and Barnard, 2005), and in women with PCOS Teede et al., 2018a;Stepto et al., 2019). The beneficial effects of exercise in women with PCOS have been summarised in several recent systematic reviews and meta-analyses (Benham et al., 2018;Kite et al., 2019). In addition, the international evidence-based guidelines for the management of PCOS recommend lifestyle intervention, including exercise training and diet, as the first line of therapy to improve general health, hormonal outcomes, and quality of life (Teede et al., 2018a). However, the studies utilised in the development of the guidelines were limited to a small number of randomised controlled trials (RCTs), resulting in a general consensus recommendation of exercise rather than a clear exercise prescription for the management of PCOS (Teede et al., 2018b;Stepto et al., 2019). In particular, there is uncertainty about suitable intensity, duration, and modality of exercise, and the interaction between exercise and diet. We have addressed this uncertainty by meta-analysing the effects of exercise characteristics on key clinical markers in women with PCOS, with the aim of assisting clinicians with exercise prescription and guiding future research in women with PCOS.

Protocol and Registration
This systematic review and meta-analysis was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and was registered on the International prospective register for systematic reviews (PROSPERO) CRD42018088367.

Search Strategy, Study Selection, and Data Extraction
We performed a systematic search of the literature to May 2019 inclusive using EBSCOhost (MEDLINE, SPORTDiscus, PsycINFO, CINAHL) and Ovid Medline. The search was limited to peer reviewed, published, English language articles from 1980current. The search terms were modified when required for each database and are reported in Supplementary Table 1. The reference lists of other review articles were searched to identify other potential eligible studies. After removal of duplicates, two reviewers (RB and RP) independently screened articles by title and abstract. Subsequently, the same reviewers independently completed full-text screening. Any discrepancies were resolved by consensus or by consultation with a third reviewer (NS). After full-text screening, data extraction of eligible studies was performed independently by RB and RP using a pre-determined extraction form.
Where required, authors were contacted via email using an institutional email address in order to obtain additional or raw data. Following a second email, if no response was received within 14 days, the article was excluded from the meta-analysis. Where multiple publications resulted from the same trial, results were combined, and only one result (largest participant number) for each outcome was used in the analysis.

Eligibility Criteria
The Participant, Intervention, Comparison, Outcomes, and Studies (PICOS) framework was used for this systematic review (Table 1). Briefly, included studies involved women aged 18-45 (pre-menopausal) and with a diagnosis of PCOS via any established diagnostic criteria. The interventions included RCTs, non-randomised controlled trials, and uncontrolled trials that had a pre-post design and reported outcomes of an exercise training intervention greater than two weeks in duration, and of moderate intensity or greater. Exercise intensity was categorised according to Norton et al. (2010), classified as moderate (55 to <70% HR max or 40 to < 60% VO 2max ), vigorous (70 to <90% HR max or 60 to <85% VO 2max ), or high (≥90% HR max or ≥85% VO 2max ) intensity. Two weeks was used as a minimum intervention duration in order to capture the effects of exercise training. We accepted exercise interventions that included aerobic exercise, resistance training, or a combination. Exercise interventions that were combined with a drug therapy that may affect the outcomes measures were excluded. Exercise interventions that were combined with a dietary intervention or dietary advice were included. Comparison groups consisted of a no exercise control group or a diet only group. The primary outcomes specified for the metaanalysis were peak oxygen consumption (VO 2peak ) used to measure cardiorespiratory fitness, homeostatic model assessment of insulin resistance (HOMA-IR) to measure insulin resistance, free androgen index (FAI) to measure androgens, and body mass index (BMI), and waist circumference to assess weight related outcomes. Secondary outcomes were those that were included in the systematic review only and consist of additional reproductive, cardio-metabolic, or anthropometric outcomes ( Table 1).

Assessment of Risk of Bias in Included Studies
The Downs and Black Checklist for the Assessment of Methodological Quality (Downs and Black, 1998) was used to evaluate the included randomised and non-randomised studies. Two reviewers (RB and RP) independently assessed the methodological quality and disagreements were resolved by consensus. Questions regarding blinding of participants were removed from the checklist (Supplementary Table 2). As per our previous meta-analysis (Cassar et al., 2016) publication bias was assessed by examining a scatter plot of t-statistic associated with each study estimate value contributing to the study-estimate random effect versus log of the standard error of the effect. No outliers or publication bias was identified using this approach.

Data Analysis
The meta-analyses were performed with the general linear mixed-model procedure in the Statistical Analysis System (Version 9.4, SAS Institute, Cary, NC, USA A random effect representing the identity of each study estimate was included to allow for differences in the means of study estimates not accounted for by the fixed effects, and an additional random effect representing estimate identity accounted for study clusters of estimates (control and/or one or more experimental estimates). In the mixed-model, each study estimate was weighted by the inverse of the square of its standard error (SE), and the random effects were estimated by setting the residual variance to unity (Yang, 2003). The standard error of each estimate was either derived as the standard deviation (SD) of change scores divided by the square root of the sample size or was computed with a spreadsheet from either the exact p-value or compatibility intervals for the mean change. For estimates where the standard error could not be derived from the data, it was imputed from the error of measurement averaged over other similar estimates (Supplementary Table 3). The number of imputed standard errors represented 0.1-3% of the total number, depending on the meta-analysed measure.
All study-estimates, standard errors, and baseline betweensubject SDs were converted to factor effects by dividing by the group mean and then log-transformed for the metaanalysis. The qualitative magnitudes of meta-analysed mean effects were evaluated via standardisation using magnitude thresholds provided by an appropriate baseline between-subject SD (Hopkins et al., 2009). This SD was derived by combining (via variances) the mean of the study SDs with the betweenstudy SD of the study baseline means, to represent the SD of women drawn randomly from a population of sub-populations. The threshold for the smallest important effect (0.2 SD) was rounded down towards the mean of the study SD to allow for differences in study means to be due partly to differences in assay technique; the thresholds for small, moderate, large, very large, and extremely large were 1, 3, 6, 10, and 20 times the smallest important threshold, respectively (corresponding to 0.2, 0.6, 1.2, 2.0, and 4.0 times the SD, Supplementary Table 4) (Cassar et al., 2016). These thresholds were also used to evaluate the magnitude of the numeric linear modifying effects of baseline and training duration, by multiplying the beta-coefficients (slopes) in the model by two between-subject SD and between-study SD, respectively (Hopkins et al., 2009). The thresholds were halved for evaluation of the magnitude of the random-effect SDs (Smith and Hopkins, 2011). The meta-analysis also provided predicted population and individual-setting effects of various combinations of exercise, diet and baseline values for the dependent variable. The predicted effects for individual settings had the same value as the predicted population mean effects, but their compatibility intervals were wider, owing to the contribution of the betweenand within-study random effects.
Uncertainty in the estimates of effects is presented as 90% compatibility limits. Probabilistic decisions about true (largesample) magnitudes accounting for the uncertainty were based on one-sided hypothesis tests of substantial magnitudes (Lakens et al., 2018). The p-value for rejecting a hypothesis of a given magnitude was the area of the sampling t distribution of the effect statistic with values of that magnitude. For effect modifiers, random-effect SDs, and predicted population mean effects, hypotheses of substantial decrease, and increase were rejected if their respective p-values were <0.05. For predicted effects in individual settings, hypotheses of harm, and benefit were rejected if the respective p-values were <0.005 and 0.25. If one hypothesis was rejected, the p-value for the other hypothesis was interpreted as evidence for that hypothesis, since the p-value corresponds to the posterior probability of the magnitude of the true effect in a reference Bayesian analysis with a minimally informative prior (Hopkins and Batterham, 2018;Hopkins, 2019). The pvalue is reported qualitatively using the following scale: 0.25-0.75, possibly; 0.75-0.95, likely; 0.95-0.995, very likely; >0.995, most likely (Hopkins et al., 2009). This scale was also used to interpret the posterior probability of a true trivial effect, which is given by the area of the sampling distribution in trivial values. If neither hypothesis was rejected, the magnitude of the effect was considered to be unclear, and the magnitude of the effect is shown without a probabilistic qualifier. To reduce inflation of error arising from the large number of effects investigated, effects were considered decisive with more conservative p-value thresholds (p < 0.01 for a substantial decrease or increase; p < 0.001 for harm; p < 0.05 for benefit) and are formatted bold in tables and figures.
The combined searches identified 1,476 papers for review. Of these, 635 were excluded due to duplication, 784 were removed by title and abstract screening, with 57 full-text articles reviewed. Twenty-four of these were excluded due to medication use, lack of data on outcomes of interest and intervention type, with 33 publications deemed suitable for inclusion in the systematic review. Multiple publications were identified for six exercise intervention studies and thereafter amalgamated with the largest reported number (N) for outcome measures of interest carried forward for meta-analysis. The remaining 20 articles were included in the systematic review with one article excluded from the meta-analysis due to using a nonparametric analysis (Brown et al., 2009) (Figure 1). A summary of study and participant characteristics, exercise intervention, and study outcomes is reported in Table 2 and results of the methodological quality are presented in Table 4 (for full results see Supplementary Table 5).

Meta-Analysis
The results from the meta-analysis of the effect of exercise characteristics on cardiorespiratory fitness measured by VO 2peak , body composition (BMI and WC), insulin resistance (HOMA-IR) and hyperandrogenism as measured by FAI are presented as the population mean effects and modifying effects of exercise characteristics (Tables 2, 3) and predicted effects of exercise across various durations and baseline values (Figures 2-4).

Effect of Exercise on VO 2peak
Meta-analysis from 16 studies with a total population of 600 women with PCOS, revealed moderate improvements in VO 2peak after moderate and vigorous intensity aerobic exercise, with the largest increase was seen after vigorous intensity exercise ( Table 3). Across all conditions, the modifying effects of intervention duration and dietary co-intervention on VO 2peak were trivial. The predicted effects analysis showed that irrespective of training dose, vigorous intensity aerobic exercise alone had the most substantial increase in VO 2peak (Figure 2). Moreover, it is clear that baseline value plays a major role in the magnitude of improvements, with lower baseline VO 2peak values resulting in the largest improvements.

Effects of Exercise on BMI
Meta-analysis of 17 studies which included a total of 759 women with PCOS were included to determine the effect of exercise on BMI. The predicted mean results of each intervention were trivial ( Table 3). The largest reductions in BMI were reported for women undertaking vigorous intensity exercise compared to a control group. The modifying effects of baseline BMI, duration and diet were also trivial with the exception of the effect of diet in a control group which resulted in a small decrease in BMI.
In the predicted effects analysis, training dose appears to have a limited effect on BMI outcome. The addition of diet intervention to exercise resulted in clear reductions in BMI. Notably, vigorous intensity exercise combined with a dietary intervention potentiated BMI changes, with small to moderate reductions of BMI across all baseline BMIs and training durations (Figure 3).

Effects of Exercise on Waist Circumference (WC)
Thirteen studies which included 463 women overall were used in this analysis of the fixed effects of exercise on WC. Vigorous intensity exercise when compared to a control group resulted in the greatest reductions in WC. The modifying effect of diet in a control group resulted in a moderate decrease in WC. In contrast, there was a trivial effect of diet in the exercise group ( Table 3).
The predicted effects analysis found the greatest improvement in WC with a combined vigorous intensity aerobic exercise and diet across the range of baseline WCs (Figure 3). Greater improvements were seen in women with a higher baseline WC. It was also apparent that training dose had a clear moderating effect on WC with greater decreases being reported after 50 h of exercise in comparison to 20 h of exercise (Figure 3).

Effects of Exercise on Free Androgen Index (FAI)
Sixteen studies were included in the meta-analysis of exerciseinduced changes in hyperandrogenism as measured by FAI and included a total of 667 women with PCOS. Of the 16 studies, three included a resistance training intervention. Our analysis showed that the greatest improvements in FAI occurred after resistance training ( Table 3). Both moderate and vigorous aerobic exercise resulted in only trivial changes. The effect of diet resulted in a small decrease in FAI in both the exercise and control groups ( Table 3).
The predicted effects analysis also reported trivial changes in FAI after aerobic exercise. Resistance training when combined with diet had the largest effect on FAI, resulting in small to moderate reductions of FAI across all baseline values and training doses, however the results were mostly unclear (Figure 4). It is apparent from the analysis that training duration plays a role in the extent of improvements in FAI, with the largest effects being seen after 50 h of exercise.

Effects of Exercise on Insulin Resistance (HOMA-IR)
Eleven studies (307 women with PCOS) were included in the meta-analysis for the effect of exercise on HOMA-IR. Vigorous intensity aerobic exercise and resistance training both resulted in moderate reductions in HOMA-IR when compared to a control group ( Table 4). The modifying effect of diet on HOMA-IR resulted in a moderate reduction in a no-exercise control group, and a small reduction in an exercise group. The modifying effect of baseline in a control group resulted in moderate increases in HOMA-IR but only trivial effects in an exercise group ( Table 4).
The predicted effects analysis on the effects of exercise on HOMA-IR show that clear improvements in HOMA-IR were only seen after vigorous intensity exercise both alone and vigorous exercise when combined with a dietary intervention, resulting in moderate reductions, irrespective of training dose (Figure 4).

DISCUSSION
This is the first systematic review and meta-analysis to evaluate the effectiveness of varying exercise intensities and the moderating effects of dietary co-intervention, training dose, and baseline values on cardiorespiratory, metabolic, and reproductive health outcomes in women with PCOS. Results from this systematic review demonstrate clear improvements in several of these outcomes following an exercise intervention in women with PCOS. The most consistent improvements were seen with cardiorespiratory fitness (VO 2peak ), BMI, WC, and various markers of metabolic health, including fasting insulin, and HOMA-IR. These results are supported by our meta-analysis, which revealed improvements in VO 2peak , body composition and insulin sensitivity following an exercise intervention, particularly when compared to a no-exercise control group. Vigorous intensity exercise, both alone and when combined with a dietary intervention, resulted in the greatest improvements in health parameters in both the fixed effects and predicted effects analyses. Moderate intensity exercise resulted in clear improvements in VO 2peak , WC, and BMI when combined with diet as seen in the predicted analysis. Interestingly, resistance training showed promising improvements in FAI and HOMA-IR in both fixed effect and predicted analyses, however further research is required to confirm these improvements.
This systematic evaluation of exercise interventions align with identified knowledge gaps in current international evidencebased guidelines (Teede et al., 2018a;Stepto et al., 2019) and meta-analyses (Benham et al., 2018;Kite et al., 2019) recently undertaken in this population of women. There is substantial evidence that supports the effectiveness of aerobic exercise training for improving some health outcomes in women with PCOS. In particular, aerobic exercise of various intensities has consistently been found to result in improvements in VO 2peak in women with PCOS (Haqq et al., 2015;Kite et al., 2019;Stepto et al., 2019). VO 2peak is a measure of cardiorespiratory fitness and is an important indicator of health and mortality (Blair et al., 1989). Individuals with a lower VO 2peak are at an increased risk of all-cause mortality and morbidity with the risk of death being more dependent on cardiorespiratory fitness than BMI (Barry et al., 2014). To illustrate this point, with each 3.5 mL/kg/min increase in VO 2peak , there is an associated 13% risk reduction from all-cause mortality (Kodama et al., 2009). Based on observed improvements from our meta-analysis, women with PCOS and relatively low VO 2peak of 24mL/kg/min are likely to experience a ∼30% risk reduction in all-cause mortality after 30 h of vigorous intensity exercise over 10-12 weeks, irrespective of any dietary co-intervention. An increase in exercise intensity becomes of paramount importance for improving VO 2peak in women with high baseline values. These results expand on existing studies that have reported improvements in VO 2peak after vigorous or high intensity exercise interventions Roessler et al., 2013;Almenning et al., 2015;Costa et al., 2018) and highlights the importance of exercise intensity when prescribing exercise training in clinical practice or as part of a clinical trial.
A large proportion of women with PCOS are overweight or obese, with a recent meta-analysis reporting a pooled prevalence of 61% (Lim et al., 2012). It is therefore not surprising that many exercise and dietary interventions have an ultimate aim of reducing body weight/BMI. Modest weight loss of 5-10% in overweight women with PCOS is encouraged to yield clinical improvements (Teede et al., 2018a). However, it is important to note that health benefits can occur without significant weight loss Sprung et al., 2013b;Covington et al., 2016). The lack of improvement in BMI following an exercise only intervention observed from our analysis is not surprising. However, when exercise is complimented with a dietary intervention, small, but clear, decreases in BMI can be achieved. In addition, our results support the inclusion of diet in order to promote improvements in WC. Research conducted by Thomson et al. (2008) reported reductions in body weight and waist circumference of ∼10% across three different treatments groups (diet alone, diet + aerobic exercise, or diet + combined aerobic and resistance exercise) after 20 weeks. A study conducted by Bruner et al. (2006) reported no significant differences in body weight or BMI following an intervention of either nutritional counselling or a combined resistance training, aerobic exercise and nutritional counselling intervention. They did, however, report significant decreases in waist circumference of 5% in both groups following the intervention period. BMI as a measure of obesity is considered to have its limitations, with changes in BMI not necessarily reflecting changes in body fat (Rothman, 2008). Body composition assessment using direct methods such as dual-energy X-ray absorptiometry (DXA) may provide valuable information on changes in body composition. When deprived of DXA information, measures of WC may provide a better measure of obesity-related health risk than BMI (Janssen et al., 2004). It is possible that exercise training alone 4 | Meta-analysed effects on homeostatic model assessment of insulin resistance (HOMA-IR) and free androgen index (FAI) expressed as population mean effects in control and exercise groups, and as modifying effects of exercise duration, baseline, and dietary co-intervention. may have a limited impact on BMI but positively improves waist circumference or other markers of body composition, including increased lean mass and decreased fat mass, which can occur without changes in total body weight. Insulin resistance is a key aetiological feature in PCOS and underpins the metabolic dysfunction present in women with PCOS (Dunaif et al., 1989;Diamanti-Kandarakis and Papavassiliou, 2006). Although not currently included in the diagnostic criteria, insulin resistance determined from insulin clamps is prevalent in 56-95% of women with PCOS (Stepto et al., 2013;Tosi et al., 2017;Li et al., 2019). It is therefore important to understand the impact of exercise type and intensity and its interaction with diet to explore effective exercise interventions to alleviate insulin resistance in women with PCOS before major complications occur. Resistance training is an effective treatment for improving insulin sensitivity in individuals with diabetes (Ishii et al., 1998;Holten et al., 2004;Ibañez et al., 2005), however, there is limited evidence to support the benefits of such training in PCOS (Thomson et al., 2008). We identified moderate decreases in HOMA-IR after resistance training interventions when compared to a control group. Resistance training is yet to be implemented in the treatment of PCOS, with current knowledge limited to few studies with small numbers of participants. However, there is evidence to support the effects of resistance training for improving insulin sensitivity in diabetic populations and therefore this may be applicable to women with PCOS. Our meta-analysis showed that vigorous intensity aerobic training also resulted in moderate decreases in HOMA-IR in women with PCOS. This is in line with findings from a number of other clinical populations (Pattyn et al., 2014;Weston et al., 2014;Jelleyman et al., 2015;Cassidy et al., 2017). Results from a study conducted by Greenwood et al. (2016) support the superior health benefits of vigorous exercise compared to moderate exercise in women with PCOS. They reported that 60 min of vigorous intensity exercise per week was associated with a 22% reduced odds of metabolic syndrome. In addition, Harrison et al. (2012) reported a 16% improvement in insulin sensitivity in women with PCOS following a 12-weeks vigorous intensity exercise intervention, as determined by the gold-standard euglycaemic-hyperinsulinaemic clamp method (DeFronzo et al., 1979). The use of the clamp method in clinical practice is impractical, however one must be cognisant that using HOMA-IR as a surrogate marker for IR has significant limitations which includes a low sensitivity in identifying IR (Tosi et al., 2017). Despite the pitfalls of using HOMA-IR to measure insulin resistance, most clinical research in PCOS continues to use this method due to its cost-effectiveness and ease of translation into clinical practice.

HOMA-IR
Elevated FAI is the most consistently observed androgenic abnormality in PCOS . Current research that measures FAI prior to and following an exercise intervention show contradictory results Thomson et al., 2008;Stener-Victorin et al., 2012). This may relate to the complex relationship between FAI and insulin resistance, as the latter has profound effects on SHBG. Results from our meta-analysis could not provide any conclusive evidence in support of any type of exercise training or exercise intensity influencing FAI and is consistent with another recent meta-analysis (Kite et al., 2019). Our results suggest that resistance training may be the most likely to induce positive changes in FAI, however, due to the limited number of studies utilising resistance training, more research is required to validate this outcome. One study of 16 week study of progressive resistance training (n = 45) reported decreases in FAI values of 0.82% . In addition, a study comparing a 10-weeks intervention of either resistance training (n = 8) or high intensity interval training (n = 8) to a control group (n = 9), reported the largest decrease in FAI in the strength training group, with a decrease of −0.7% from baseline values (Almenning et al., 2015). Although resistance training shows promising results, reductions in FAI have also been reported after aerobic exercise (Randeva et al., 2002;Giallauria et al., 2008;Covington et al., 2015). Further research is required to determine the effective modality, dose and intensity of exercise for improvements in hyperandrogenism. There is also a need to identify more valid measures of androgen levels in women with PCOS to monitor impacts all interventions (e.g., exercise and/or diet, pharmacotherapies).

Strengths and Limitations
An important strength of our analysis is the inclusion of a variety of study designs with well-characterised participants. This allowed us to go beyond existing systematic reviews and meta-analyses to generate a large dataset that included a nointervention control group. We were also able to explore the modifying effects of diet, exercise intensity, training dose, and baseline values of the outcome measures, according to a particular current health and fitness level, enabling more individualised exercise prescription for women with PCOS. However, the inclusion of studies other than RCTs may be viewed as a limitation due to the possible increase in the risk of bias. However, all studies were assessed for bias and deemed of acceptable quality. It could also be argued that including a noexercise control group in a study design could be considered of no additional use (Jones and Podolsky, 2015;Frieden, 2017) and it is established that in many clinical conditions, most outcomes impacted by exercise remain unchanged or worsen over the course of an intervention in no-exercise controls (Jelleyman et al., 2015). A limitation of this analysis is the large heterogeneity among the included studies with interventions varying greatly in frequency, intensity, and the extent of exercise supervision. Some studies had sparse description of the exercise interventions, further limiting our analysis. The inclusion of unsupervised exercise interventions may have under-estimated the benefits of exercise and future research should aim to document level of supervision to better gauge its effect on clinical outcomes.

CONCLUSIONS
This work considerably expands on previous evidence and advances the knowledge of benefits of exercise prescription in women with PCOS. Our analysis demonstrates that exercise training in women with PCOS improves cardiometabolic outcomes, both in the presence and independent of anthropometric changes, supporting the role of exercise therapy, as the first-line approach for improving health outcomes in women with PCOS. Specifically, for greater health improvements, exercise interventions and/or exercise prescription should aim to achieve and sustain a minimum of 20 h of vigorous intensity exercise over 10-12 weeks, equating to 120 min per week across this timeframe. Once achieving this goal, women should sustain this level of exercise for continued health maintenance. Resistance training also appears to have some health benefits and could be considered for women with PCOS. Adequate reporting of exercise intervention characteristics (i.e., exercise session supervision, exercise intensity, adherence, and compliance), use of gold-standard clinical outcome measures and consideration of long-term intervention sustainability is required through the application of high-quality, large clinical studies of longer duration required to provide definitive exercise prescription recommendations in women with PCOS.

DATA AVAILABILITY STATEMENT
Publicly available datasets were analyzed in this study. This data can be found here: Supplementary Table 3.