Effect of lifestyle intervention on the reproductive endocrine profile in women with polycystic ovarian syndrome: a systematic review and meta-analysis

Polycystic ovarian syndrome (PCOS) affects 18–22% of women at reproductive age. We conducted a systematic review and meta-analysis evaluating the expected benefits of lifestyle (exercise plus diet) interventions on the reproductive endocrine profile in women with PCOS. Potential studies were identified by systematically searching PubMed, CINAHL and the Cochrane Controlled Trials Registry (1966–April 30, 2013) systematically using key concepts of PCOS. Significant improvements were seen in women receiving lifestyle intervention vs usual care in follicle-stimulating hormone (FSH) levels, mean difference (MD) 0.39 IU/l (95% CI 0.09 to 0.70, P=0.01), sex hormone-binding globulin (SHBG) levels, MD 2.37 nmol/l (95% CI 1.27 to 3.47, P<0.0001), total testosterone levels, MD −0.13 nmol/l (95% CI −0.22 to −0.03, P=0.008), androstenedione levels, MD −0.09 ng/dl (95% CI −0.15 to −0.03, P=0.005), free androgen index (FAI) levels, MD −1.64 (95% CI −2.94 to −0.35, P=0.01) and Ferriman–Gallwey (FG) score, MD −1.01 (95% CI −1.54 to −0.48, P=0.0002). Significant improvements were also observed in women who received exercise-alone intervention vs usual care in FSH levels, MD 0.42 IU/l (95% CI 0.11 to 0.73, P=0.009), SHBG levels, MD 3.42 nmol/l (95% CI 0.11 to 6.73, P=0.04), total testosterone levels, MD −0.16 nmol/l (95% CI −0.29 to −0.04, P=0.01), androstenedione levels, MD −0.09 ng/dl (95% CI −0.16 to −0.03, P=0.004) and FG score, MD −1.13 (95% CI −1.88 to −0.38, P=0.003). Our analyses suggest that lifestyle (diet and exercise) intervention improves levels of FSH, SHBG, total testosterone, androstenedione and FAI, and FG score in women with PCOS.


Introduction
Polycystic ovarian syndrome (PCOS) is a heterogeneous endocrine disorder, affecting 18-22% of reproductive-age women (1). PCOS was first reported in 1935 by Stein & Leventhal (2) and is characterised by clinical or biochemical hyperandrogenism (clinical manifestations are hirsutism, android alopecia and acne), oligo/amenorrhoea (infrequent or no menstruation), polycystic ovaries and infertility or reduced fertility (3,4). Often women with PCOS are obese, which contributes to insulin resistance and hyperinsulinaemia, but these two features are also present in lean women with PCOS (5,6). Hormonal manifestations include elevated levels of androgens (testosterone, DHEA and androstenedione), oestrogens and prolactin. Occasionally, thyroid-stimulating hormone levels are also lower leading to hypothyroidism (7). Most women with PCOS have elevated luteinising hormone (LH) levels and reduced follicle-stimulating hormone (FSH) levels particularly during the follicular phase of the menstrual cycle (8). The elevated LH level probably increases the follicular androgen concentrations leading to follicular arrest and the reduced FSH concentrations lead to an accumulation of small follicles (9). The resultant oestrogen environment alters the hypothalamic release of gonadotrophin-releasing hormone and leads to an increase in LH secretion and suppression of FSH secretion by the pituitary (8,10). This altered LH:FSH ratio is used as a diagnostic criterion for this condition, but it is not universally present (11).
The levels of sex hormone-binding globulin (SHBG), the primary plasma transport system which controls the availability of androgens, are reduced in women with PCOS leading to an increase in free testosterone levels contributing to the free androgen index (FAI) (12). Owing to the effects of insulin on hepatic SHBG production, insulin insensitivity may affect ovulation and fertility. Dyslipidaemia, increased insulin levels, obesity, hypertension, impaired glucose tolerance and insulin-induced metabolic syndrome are also the risk factors that can predispose women with PCOS to cardiovascular disease and type 2 diabetes mellitus (6).
A systematic review was completed in 2011 by Harrison et al. (13), but presumably as insufficient data were available at that time, these authors did not conduct data pooled analyses. A systematic review and subsequent meta-analyses were conducted by Moran et al. (14), but these analyses included only six published studies with slightly different inclusion/exclusion criteria. However, our work provides a greater number of hormonal analyses when compared with Moran's work (14). We therefore conducted a systematic review and meta-analysis, and the primary aim was to evaluate the expected benefits of exercise training and dietary interventions on a range of endocrinal outcomes in women with PCOS.

Search strategy
Potential studies were identified by conducting a systematic search using PubMed (www.ncbi.nlm.nih.gov/ pubmed; 1966- April 30, 2013

Inclusions
Randomised, controlled trials of exercise-alone or lifestyle (exercise plus diet) intervention in people with PCOS were included. There were no language restrictions.

Exclusions
Animal studies, review papers and non-randomised controlled trials were excluded. Studies that did not have desired outcome measures or used non-PCOS participants in exercise or lifestyle (exercise plus diet) interventions or usual care groups were excluded. Several authors were contacted and they provided missing data, these data were used in the analyses. Incomplete data or data from an already included study were excluded. Studies using interventions other than lifestyle (e.g. electro-acupuncture and ultrasound) were excluded.

Studies included in the review
Our initial search identified 201 manuscripts, and looking into the latest editions of relevant journals, a further 32 manuscripts were identified. Out of 233 studies, 28 were excluded at first inspection as duplicates, and 182 were removed after reading titles or abstracts, leaving 23 studies of which 16 studies were excluded for various reasons (including five studies which did not provide data that could be included in our analysis), leaving seven studies for final analysis (see consort statement, Fig. 1).

Data synthesis and outcome measures
Our lifestyle intervention groups were defined as exercise-alone or exercise-plus-diet groups. Our definition of usual care (comparator) groups could include sedentary control, placebo, diet only or medication. Analyses were conducted only on intervention vs comparator 1 (see Table 1). Information regarding all outcome measures was archived in a database. The outcome measures comprised hormones and hormone ratios, including LH, FSH, SHBG, total testosterone, measured free testosterone, androstenedione, FAI, LH:FSH ratio, oestradiol (E 2 ) and Ferriman-Gallwey (FG) score. Table 4 Excluded randomised, controlled trials.

Reference
Reason for exclusion Thomson (4) Data reported unsuitable for analysis Bruner (11) No data reported were included in analysis Stener-Victorin (22) Data reported unsuitable for analysis Brown (24) Data reported as median values so unsuitable for analysis Galletly (25) Both intervention groups exercised, no control group Karimzadeh (26) Data reported unsuitable for analysis Ladson (27,28) Both intervention and comparator groups underwent similar exercise training, no control group Ladson (27,28) Duplicate of previous study Ma (29) Both intervention and comparator groups underwent similar exercise training, no control group Moran (30) Both intervention and comparator groups underwent similar exercise training, no control group Orio (31) No randomised, matched controls Otta (32) Both intervention and comparator groups underwent similar exercise training, no control group Palomba (33) Study conducted for 6 weeks, but data reported for 2 weeks only Thomson (34) Data reported unsuitable for analysis Thomson (36) Data reported unsuitable for analysis Nybacka (37) No data reported were included in analysis    Meta-analyses were completed for continuous data using the change in the mean and S.D. of outcome measures, as we did not wish to assume that randomisation would adjust for baseline imbalance. Change in post-intervention mean was calculated by subtracting baseline from post-intervention values. Change in the S.D. of post-intervention outcomes was calculated using RevMan 5.0 (Nordic Cochrane Centre, Copenhagen, Denmark). Data required were either i) 95% CI data for pre-post intervention change for each group or, when this was unavailable, ii) actual P values for pre-post intervention change for each group or, if only the level of statistical significance was available and iii) we used default P values (e.g. P!0.05 becomes PZ0.049, P!0.01 becomes PZ0.0099 and PZnot significant becomes PZ0.05).
A random-effects inverse variance was used with the effects measure of mean difference (MD). Heterogeneity was quantified using Cochran's Q test (15). Sensitivity analyses were conducted by removing studies of lifestyle intervention, leaving studies of exercise only, for the outcomes, such as FSH, SHBG, total testosterone, androstenedione, FAI, E 2 , LH:FSH ratio and FG score. The purpose of sensitivity analyses was to compare the effect sizes of exercise alone with exercise plus diet. Egger plots (16) were produced to assess the risk of publication bias. Study quality was assessed using a modified PEDro (17) score (out of 9 maximum scores) as blinding participants is difficult in lifestyle studies. We used a 5% level of significance and 95% CIs; figures were produced using RevMan 5.0.

Results
Our analyses included data from seven studies (3,18,19,20,21,22,23), which yielded data on 206 women with PCOS. In three studies, the mean BMI indicated that the participants were obese, three studies indicated that women were overweight and in one study this was unclear. The mean age of participants in all but one study was 21-32 years of age. Details about the number of participants, duration of studies and withdrawals for included studies can be seen in Table 1. Table 2

Hormonal parameters
LH levels were not significantly different for women in lifestyle intervention vs usual care groups, with a MD of 0.99 IU/l (95% CI K0.11 to 2.09, PZ0.08). Moreover, LH levels were also not significantly different for women in exercise vs usual care groups, with an MD of 0.51 IU/l (95% CI K1.11 to 2.13, PZ0.54). FSH levels were found to be improved (higher) in women in lifestyle intervention vs usual care groups, with an MD of 0.39 IU/l (95% CI 0.09 to 0.70, PZ0.01), see Fig. 2.
When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, FSH levels were found to be improved in exercisealone group, with an MD of 0.42 IU/l (95% CI 0.11 to 0.73, PZ0.009), see Fig. 3. SHBG levels were found to be improved in women in lifestyle intervention vs usual care groups, with an MD of 2.37 nmol/l (95% CI 1.27 to 3.47, P!0.0001), see Fig. 4.
When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, SHBG levels showed greater improvement in exercise-alone groups, with an MD of 3.42 nmol/l (95% CI 0.11 to 6.73, PZ0.04), see Fig. 5.
When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, total testosterone levels were found to be improved significantly (lowered) in exercise-alone group, with an MD of K0.16 nmol/l (95% CI K0.29 to K0.04, PZ0.01), see Fig. 7.
Measured free testosterone levels were found to be slightly higher in women in exercise-alone vs usual care groups, with an MD of 0.66 nmol/l (95% CI 0.13 to 1.19, PZ0.01).

Figure 5
Change in SHBG levels in exercise-alone vs usual care groups.  When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, androstenedione levels were found to be similar in exercise-alone group and the lifestyle intervention group, with an MD of K0.09 ng/dl (95% CI K0.16 to K0.03, PZ0.004), see Fig. 9.
FAI levels were found to be improved significantly (lower) in women in the lifestyle intervention vs usual care groups, with an MD of K1.64 (95% CI K2.94 to K0.35, PZ0.01), see Fig. 10. When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, FAI levels were not significantly decreased in exercise-alone group, with an MD of K1.38 (95% CI K2.98 to 0.23, PZ0.09).
When studies using exercise plus diet were removed to distinguish between exercise-alone and exercise-plus-diet groups, FG score was found to be improved in women in exercise-alone vs usual care groups, with an MD of K1.13 (95% CI K1.88 to K0.38, PZ0.003), see Fig. 12.

Study quality
In terms of study quality, the median score was 7, with one study scoring 6, three studies scoring 7, two studies scoring 8 and one study scoring 9, using a modified PEDro scale (out of 9). Details about the scores and PEDro scale are given in Table 5. Egger plots showed little or no evidence of publication bias.

Discussion
This study is an update from a 2011 meta-analysis examining the effectiveness of lifestyle intervention on PCOS (14). We examined the effects of lifestyle Change in total testosterone levels in exercise-alone vs usual care groups.  intervention on endocrinal parameters and the FG score. Unlike the previous meta-analysis (14), we took the approach of conducting analyses on combined exercise and dietary (lifestyle) interventions and also sensitivity analyses consisting of exercise-alone intervention groups; the comparator for both main and sensitivity analyses were usual care groups. Our findings suggest that levels of FSH, SHBG, total testosterone, androstenedione and FAI, and FG score were found to be improved in response to lifestyle intervention, while levels of FSH, SHBG, total testosterone and androstenedione, and FG score were found to be improved with exercise alone. Our findings suggest that lifestyle intervention is beneficial for improving FSH profile. Previous work has not clearly defined the effects of exercise or diet on FSH levels in women with PCOS. The most significant improvement (increase) in SHBG levels (narrowest CI) was observed in the combined exercise and dietary intervention vs usual care. Although the effect size of improvement in SHBG was larger for exercise-alone group compared with usual care groups, the wider CI limits the inferences that can be drawn from this finding. This leads us to conclude that exercise, either in combination with dietary treatment or in isolation, improves the SHBG profile. The hypothesis that, in combination with appropriate dietary intervention, there is an improvement in SHBG levels has been suggested previously (11).
Total testosterone levels were reduced in both the exercise-alone and lifestyle intervention vs usual care groups in our analyses. Our findings reassert the results of previous work which suggest that reductions in total testosterone levels are achieved by lifestyle intervention (3,23).
Our analyses demonstrate that lifestyle intervention vs usual care groups showed improvement (reduced) in FAI, while there was no change in the exercise-alone vs usual care groups. Our findings therefore suggest that lifestyle intervention is the optimal therapy for eliciting beneficial effects; previous work has also suggested this to be the case (38). FAI is the total testosterone level divided by the SHBG level. In our analyses, levels of FAI decreased, SHBG increased and total testosterone decreased after exerciseplus-diet intervention. While these effects are consistent with the expectations and are self-explanatory, the timing of sample collection could have moderated the findings due to peaks and troughs in these hormones during the ovulatory cycle.
Lifestyle intervention showed a significantly favourable reduction in androstenedione levels. Our findings Change in androstenedione levels in exercise-alone vs usual care groups.

Figure 10
Change in FAI levels in lifestyle intervention vs usual care groups.

Endocrine Connections
Research showed almost identical reductions in androstenedione levels in lifestyle intervention and exercise-alone groups. These findings support the results of previous work suggesting that exercise reduces androstenedione levels (3), while our analyses suggested limited evidence of additional benefit from diet. Lifestyle intervention or exercise alone showed a significantly favourable improvement in hirsutism (FG score). These findings support those of previous work suggesting that both lifestyle intervention and exercise improve hirsutism (14).

Limitations
The sample size of this analysis may be underpowered. Analyses in Figs 6 and 8 exhibit moderate to high evidence of between-study heterogeneity, but the majority of our analyses have little or no heterogeneity. As always with exercise training studies, there are small variations in study duration and exercise modality, although these appear to be minimal in this analysis. While not necessarily a limitation, we chose to modify the PEDro scale to assess the study quality as all studies would have found it impossible to blind participants and investigators to the allocation of exercise training or sedentary control. The Egger plot for testosterone showed minimal evidence of publication bias, while the other Egger plots showed no such evidence of publication bias. It is therefore not probable that unpublished datasets exist for the majority of our outcome measures and the level of significance for testosterone suggests that unpublished data would not change the findings presented here.

Recommendations for future research
Measures such as testosterone, FSH and LH vary with the stage of the menstrual cycle (stage of follicular development) and additionally are secreted in a pulsatile manner, so caution must be exercised in interpreting these present findings. Not all studies stated whether the timing of blood draws was standardised to each individuals' menstrual and ovulatory cycles; moreover, we believe that multiple samples should be collected to determine true baseline measures of FSH and LH precisely. We suspect that LH has a central role in PCOS management and the failure to detect changes with exercise and dietary interventions is probably a result of poor study design. Future studies should employ more rigorous interpretation of the monthly variations in LH that occur naturally. Other hormones such as growth hormone and  Figure 11 Change in FG score in lifestyle intervention vs usual care groups.

Figure 12
Change in FG score in exercise-alone vs usual care groups.

Endocrine Connections
Research insulin-like growth factor 1 together with their binding proteins are influenced by exercise and have an impact on ovarian function and we recommend that these should be included in future studies. Future study should also attempt to compare and evaluate the relative effects of exercise alone vs exercise plus diet.

Conclusions
Our analyses suggest that lifestyle (diet and exercise) intervention improves the levels of FSH, SHBG, total testosterone, androstenedione and FAI, and FG score in women with PCOS. Exercise alone improved all of these outcomes except FAI and LH; however, given the uncertainty of when the blood samples were taken and the fact that LH levels increase dramatically towards ovulation, further studies are required to elucidate this finding.