The effect of metformin on fat distribution and the metabolic syndrome in women with polycystic ovary syndrome—a randomised, double-blind, placebo-controlled trial

Authors


Dr J Lord, Department of Obstetrics and Gynaecology, Royal Cornwall Hospital, Truro, Cornwall TR1 3LJ, UK. Email jonathan.lord@pms.ac.uk

Abstract

Objective  To establish whether metformin has a significant action in reducing visceral fat and improving other metabolic parameters in women with polycystic ovary syndrome (PCOS).

Design  Randomised, double-blind, placebo-controlled trial.

Setting  Reproductive medicine clinic.

Population  Forty women with anovulatory PCOS.

Methods  Participants were randomised into receiving metformin 500 mg three times a day or placebo for 3 months.

Main outcome measures  Fat distribution was measured by computed tomography scan. Secondary outcome measures included serum indices of the metabolic syndrome and evidence of ovulation.

Results  We found no significant differences in any of the measures of fat distribution between the placebo and metformin groups. The metformin group had significantly lower total cholesterol (P= 0.02), low-density lipoprotein cholesterol (P= 0.02) and cholesterol:high-density lipoprotein cholesterol ratio (P= 0.05), but there was no statistically significant treatment effect on androgens, insulin, insulin resistance, triglycerides, ovulation or pregnancy.

Conclusions  Metformin has no clinically significant effect in reducing visceral fat mass, although it does have a beneficial effect on lipids. This trial lends support to the growing evidence that metformin is not a weight loss drug. Metformin might therefore be used as an adjunct to lifestyle modification in women with PCOS, but not as a substitute for it.

Introduction

The pathophysiology of polycystic ovary syndrome (PCOS) is complex but there is strong evidence that hyperinsulinaemia plays a major role for many with the syndrome, probably as a result of peripheral insulin resistance.1 As such, PCOS is one of a number of metabolic disturbances that are associated with cardiovascular disease and diabetes, referred to as the metabolic syndrome.2

Fat of intra-abdominal or visceral distribution is believed to be crucial in the development of insulin resistance,3 and although there are exceptions, PCOS is associated with obesity.4 Our aim was to investigate whether metformin has a significant action in reducing visceral fat in women with anovulatory PCOS. If it does, then metformin therapy should have important metabolic advantages over symptomatic treatment like clomifene, even if it were no more effective in inducing ovulation.

Methods

Participants

Women with anovulation and PCOS were recruited from those attending outpatients at the South West Centre for Reproductive Medicine, Derriford Hospital, Plymouth, UK. Anovulation was defined as oligomenorrhoea (<6 periods in preceding 12 months) or a luteal-phase progesterone of <20 nmol/l. PCOS was defined as anovulation and a raised free androgen index (FAI) >5.0. Participants with diabetes mellitus, thyroid disease, raised prolactin and late-onset adrenal hyperplasia (i.e. raised 17α-hydroxyprogesterone) were excluded. Other exclusion criteria were the use of ovulation-inducing agents or drugs that could affect insulin metabolism within the previous 2 months, age outside the range of 18–40 years and pregnancy (excluded by high sensitivity urine βhuman chorionic gonadotrophin). Measures of weight and fat distribution did not form part of the recruitment criteria.

Figure 1 shows the characteristics of the trial participants. Forty-four subjects were randomised, 22 in each arm. Four subjects were subsequently excluded for not fulfilling the entry criteria at the time of recruitment: two had FAI <5.0 at trial entry (both randomised to placebo arm) and two were ovulating at trial entry (one in placebo arm and one in metformin arm). Five participants became pregnant (two in the placebo and three in the metformin arm) and did not therefore have exit investigations. Three subjects were lost to follow up (two in metformin arm and one in placebo).

Figure 1.

Trial characteristics.

Ethical approval was obtained from the local research ethics committee, and informed written consent was obtained from each participant.

Protocol

The trial was randomised, double-blind and placebo-controlled. After recruitment, participants were randomised centrally by the hospital pharmacy using a block allocation and received for 12 weeks either metformin 500 mg three times a day or placebo in the same regimen. Treatment was commenced on day 1 of the menstrual cycle or immediately following a negative pregnancy test if they had oligomenorrhoea. Both participants and investigators were blinded to the treatment allocation. Participants were given general preconceptional counselling and advised to eat a healthy diet and to take regular exercise. However, they were not given any specific lifestyle modification or weight loss advice.

Measurements were taken on recruitment and repeated after 3 months of treatment. The primary outcome measure was visceral fat mass. Secondary outcome measures are described in detail below and include other measures of fat distribution (subcutaneous fat mass and anthropometric measurements), serum indices of the metabolic syndrome and evidence of ovulation.

Measurements—fat distribution

Visceral and subcutaneous fat mass were measured by areal planimetry using a single 5-mm collimation slice through the midpoint of L2 at full inspiration, with the radiation dose reduced as much as possible. This was performed using a Siemens Somatom (Siemens PLC, Bracknell, Berkshire, UK) Plus 4, with a preliminary limited topogram of the lower abdomen to identify the level. Postprocessing was performed on a Siemens Virtuoso Workstation and all results were reported by one of two radiologists (B.F. and R.T.), with inter- and intra-observer variations recorded.

Anthropometric measurements were taken using standard techniques.5 Circumferences were measured to within 5 mm using a tape measure in the standing position. Waist circumference was taken at the midpoint between the lowest rib margin and the iliac crest at the end of normal expiration. Hip circumference was measured at the widest level of the greater trochanters. The body mass index (BMI) was calculated using the standard formula (kg/m2) using one set of scales and height measure which were regularly checked and calibrated by the hospital’s medical physics department. Inter- and intra-observer variations were recorded.

Measurements—serum

Serum was taken after an overnight fast of at least 10 hours on the same day as the measurements of fat distribution. Serum was centrifuged for 5 minutes and frozen to −80°C within 30 minutes of being taken. It was analysed in batches by the hospital’s clinical chemistry department using their standard operating procedures within 30 minutes of being thawed.

  • Insulin (Immulite®; DPC, Los Angeles, CA, USA): Insulin resistance was estimated by homeostasis model assessment, a computerised model of the physiological loop that controls the concentration of blood glucose.6–8 It allows an estimation of insulin resistance from the fasting levels of insulin and blood glucose. The insulin assay cross reacts at <1% with proinsulin.

  • Glucose: COBAS® Integra glucose HK; Roche Diagnostic, Basel, Switzerland.

  • Testosterone: ACS Centaur®; Chiron Diagnostics Corporation, East Walpole, MA, USA.

  • Sex hormone binding globulin (SHBG): Immulite 2000®; DPC.

  • FAI: serum testosterone/SHBG × 100.

  • Cholesterol (COBAS® Integra cholesterol; Roche Diagnostics), low-density lipoprotein (LDL) cholesterol and high-density lipoprotein (HDL) cholesterol (COBAS® Integra HDL-cholesterol direct; Roche Diagnostics).

  • Triglycerides: COBAS® Integra triglycerides; Roche Diagnostics.

  • Urate: COBAS® Integra uric acid, Roche Diagnostics.

  • Glycated haemoglobin (HbA1C): Biomen 8160 (Menarini Diagnostics, Florence, Italy).

  • Progesterone (ACS Centaur®; Chiron Diagnostics): It was taken at recruitment in those with oligomenorrhoea in order to exclude ovulation or the possibility of the participant being in the luteal phase. A serum progesterone level of ≥30 nmol/l was taken as evidence for ovulation. Measurements were taken on days 14, 21, 28 and 35 or until menstruation occurred. If menstruation had not occurred by day 35, then progesterone was checked fortnightly. Following menstruation, the testing cycle was repeated for 3 months in total.

Statistical analysis

SPSS version 11.5 and SamplePower 2.0 (SPSS Inc., Chicago, IL, USA) were used for analysis. Parameters were tested for normality and transformed logarithmically where skew was evident. The t test was used in the comparison of means for parametric data and the Mann–Whitney U test for nonparametric data. Pearson’s correlation coefficient was used to define the correlations. The paired t test was used for subgroup analysis where means were compared at baseline and exit. Analysis was on an intention-to-treat basis. Participants who became pregnant did not have repeat investigations at trial exit and were recorded as drop-outs (although they were recorded as having ovulated in the analysis).

Power calculation

When the trial was planned there were minimal data to permit an accurate power calculation based on the measure of visceral fat. Therefore, a retrospective calculation was going to be required to assess the power of this study for the primary outcome measure. A provisional power calculation was based on ovulation, which was more widely reported. Other trials had demonstrated that ovulation occurred following a weight loss of 5–10%9,10 which correlated with a reduction in visceral fat mass of 30%.11 This degree of reduction, together with the resumption of ovulation, was taken as the clinically significant parameters for this trial. To demonstrate an effect on ovulation of metformin versus placebo, each arm would need seven participants for a 90% power at a significance of 0.05, assuming that metformin has the same efficacy as described in the trial by Nestler et al.12 We aimed for a recruitment of 20 participants in each arm, which would give the same power if metformin was 15% less effective in our patients.

Trial validity

Intra- and inter-observer variability were checked for visceral fat measurement (r2= 0.97 and 0.99, respectively), subcutaneous fat measurements (r2= 0.99 and 0.99), waist circumference (r2= 0.98 and 0.98), hip circumference (r2= 0.98 and 0.94) and test/re-test variability for insulin (r2= 0.82).

Results

Baseline characteristics

The characteristics of trial participants at recruitment are given in Table 1. The inclusion criteria did not include measures of weight or fat mass. As would be expected in a population with PCOS, while some participants were lean, overall 89% had BMI >25 kg/m2 with mean weight 95 kg (SD 16.4, range 50–127), mean BMI 35 kg/m2 (SD 7.1, range 20.7–52.9), mean waist circumference 101.2 cm (SD 14.5, range 71–131) and mean waist:hip ratio 0.8 (SD 0.07, range 0.7–1.0). There were no significant differences between the placebo and metformin groups, with the exception of systolic and mean arterial blood pressure where the metformin group had significantly lower values (P= 0.01).

Table 1.  Baseline characteristics of trial participants
 PlaceboMetformin 
nMeanSDnMeanSD
  1. DHEAS, dehydroepiandrosterone sulphate; HOMA, homeostasis model assessment.

  2. Values are mean ± SD.

Age (years)1930.634.842127.764.89NS
Testosterone (nmol/l)192.740.65212.600.78NS
SHBG (nmol/l)1930.588.412136.7444.31NS
FAI199.563.212111.075.26NS
DHEAS (μmol/l)194.862.26196.113.31NS
Fasting insulin (miu/l)1918.856.042121.5715.54NS
Glucose (mmol/l)195.170.42215.060.55NS
HOMA β-cell19243.88115.2021294.75203.06NS
HOMA insulin resistance194.351.51214.913.72NS
Cholesterol (mmol/l)195.291.13215.180.92NS
Triglycerides (mmol/l)191.340.56211.660.79NS
HDL cholesterol (mmol/l)191.250.23211.290.28NS
LDL cholesterol (mmol/l)193.431.12213.130.98NS
Cholesterol:HDL ratio194.361.18214.171.07NS
HbA1C (%)185.390.35205.320.35NS
Urea (mmol/l)194.280.63214.120.93NS
Urate (mmol/l)190.330.08210.320.08NS
Weight (kg)1997.8416.182192.4316.49NS
Height (cm)19164.595.8021165.907.27NS
BMI (kg/m2)1936.377.462133.746.74NS
Waist circumference (cm)19104.8715.882197.9012.53NS
Hip circumference (cm)19122.1312.7221119.7412.73NS
Waist:hip ratio190.860.08210.820.06NS
Systolic blood pressure (mmHg)18133.2812.5120122.0014.10P= 0.01
Diastolic blood pressure (mmHg)1878.787.262074.259.25NS
Mean arterial blood pressure (mmHg)1896.946.632090.179.40P= 0.01
Subcutaneous fat (mm2)1940252.9316676.482131303.6712860.96NS
Visceral fat (mm2)1911417.034322.572111007.355694.01NS
Visceral:subcutaneous fat ratio190.310.11210.360.16NS

Effect of metformin on fat distribution

There were no significant differences in any of the measures of fat distribution between the placebo and metformin groups, whether by computed tomography (CT) or by anthropometric measurement (Table 2). When analysis was by effect size, the metformin group had a 4.6% reduction in visceral fat (not significant [NS]) and a 7.5% reduction in subcutaneous fat (NS). The placebo group had reductions of 4% (NS) and 2% (NS), respectively. In the metformin group, weight, BMI and waist:hip ratio all had nonsignificant increases of 4, 4 and 1%, respectively, while waist circumference was reduced by 1%. In the placebo group, weight, BMI and waist circumference showed changes of <1%, while waist:hip ratio increased by 2%.

Table 2.  Treatment effect on fat distribution
 PlaceboMetformin 
nMeanSDnMeanSD
  1. Values are mean ± SD.

Subcutaneous fat (mm2)1637834.1315626.011629803.7511099.16NS
Visceral fat (mm2)1611165.474336.481610653.385136.58NS
Visceral:subcutaneous fat ratio160.310.08160.360.12NS
Weight (kg)1594.9015.511694.6627.13NS
Height (cm)15164.344.7616165.236.78NS
BMI (kg/m2)1535.266.531634.609.13NS
Waist circumference (cm)15103.3314.441697.0613.02NS
Hip circumference (cm)15117.7010.5616117.4713.67NS
Waist:hip ratio150.880.07160.830.06NS

Effect of metformin on other variables

The metformin group had significantly lower total cholesterol (P= 0.02), LDL cholesterol (P= 0.02) and cholesterol:HDL cholesterol ratio (P= 0.05). The metformin group also had significantly lower systolic blood pressure (P= 0.007), although a difference was also present at baseline. Urea was significantly lower in the metformin group (P= 0.007). There was no evidence of treatment effect on androgens, insulin, insulin resistance, triglycerides, ovulation or pregnancy (Table 3).

Table 3.  Treatment effect on metabolic parameters
 PlaceboMetformin 
nMeanSDnMeanSD
  1. DHEAS, dehydroepiandrosterone sulphate; HOMA, homeostasis model assessment.

  2. Values are mean ± SD.

Testosterone (nmol/l)152.260.61162.510.64NS
SHBG (nmol/l)1530.279.351627.419.98NS
FAI157.942.731610.364.75NS
DHEAS (μmol/l)144.842.40167.043.92NS
Fasting insulin (miu/l)1515.366.301617.358.90NS
Glucose (mmol/l)155.050.48165.030.53NS
HOMA β-cell15221.79159.5316261.52179.38NS
HOMA insulin resistance153.441.29163.861.92NS
Cholesterol (mmol/l)155.651.15164.780.82P= 0.02
Triglycerides (mmol/l)141.340.62161.440.71NS
HDL cholesterol (mmol/l)141.270.19161.260.25NS
LDL cholesterol (mmol/l)143.841.15162.870.85P= 0.02
Cholesterol:HDL ratio144.581.03163.880.74P= 0.05
HbA1C (%)155.440.36165.370.37NS
Urea (mmol/l)144.580.49143.790.87P= 0.007
Urate (mmol/l)140.330.07150.300.09NS
Systolic blood pressure (mmHg)15138.4011.3116122.6913.51P= 0.001
Diastolic blood pressure (mmHg)1579.4712.891576.7310.91NS
Mean arterial blood pressure (mmHg)1599.119.101592.2910.83NS

When analysis was performed by effect size, the metformin group had significant reductions in fasting insulin (P= 0.01), insulin resistance (P= 0.01) and cholesterol (P= 0.03), while the placebo group showed reductions only in fasting insulin (P= 0.01) and insulin resistance (P= 0.01).

Subanalysis by ovulation

Overall, 49% of participants ovulated, with similar proportions in the placebo and metformin groups (Table 4). Those who ovulated had significantly lower testosterone (r=−0.66, P < 0.0001), FAI (r=−0.52, P= 0.003), triglycerides (r=−0.43, P= 0.007), HbA1C (r=−0.33, P= 0.05), urate (r=−0.35, P= 0.04), weight (r=−0.35, P= 0.04) and BMI (r=−0.43, P= 0.01). While there was a trend towards the ovulation group having lower visceral and subcutaneous fat and lower insulin levels, these did not reach significance. There was no evidence that metformin influenced ovulation (r= 0.05, P= 0.4).

Table 4.  Ovulation and pregnancy rates
 PregnancyOvulation
Metformin (n= 19)39
Placebo (n= 18)29
 NSNS

Discussion

This study did not find evidence that metformin has a clinically or statistically significant effect on reducing visceral fat in women with PCOS. However, metformin did have modest beneficial effects on lipids with statistically significant reductions in total cholesterol, LDL cholesterol and cholesterol:HDL cholesterol ratio when compared with placebo.

Modern imaging techniques enable visceral fat to be distinguished from subcutaneous fat at waist level, something a tape measure or BMI calculation is unable to do (Figure 2). Indeed, our results showed that despite absolute reductions in fat mass as measured by CT scan in both the metformin and placebo arms, the traditional measures of weight, BMI and waist:hip circumference actually showed increases, with only waist circumference correlating with changes in fat mass.

Figure 2.

CT scan of the abdomen at the level of L2 showing compartmentalisation of fat into visceral (V) and subcutaneous (SC) components.

We have previously outlined arguments as to why visceral fat deposition is either causative or a very early effect of PCOS; therefore, why interventions that reduce it should be beneficial.13 Visceral fat is metabolically active with high rates of lipogenesis and lipolysis resulting in a high turnover of fatty acids and lipoproteins.14,15 It is visceral fat that largely accounts for the variation in insulin sensitivity seen in various populations16 and for the differences in lipid levels normally seen between obese men and women.17 Those with visceral fat obesity have the highest risk of cardiovascular mortality, insulin resistance, dyslipidaemia, hypertension, left ventricular enlargement and developing diabetes.11,15,18–20

For women with PCOS, studies have shown excess abdominal fat compared with controls as measured by anthropometric measurement.21–23 Where more advanced imaging techniques have been used, nonobese women with PCOS were shown to have a significantly higher amount of total body and upper body fat with no significant difference in lower body fat mass when compared with controls.24–26 The difference in insulin resistance between women with PCOS and weight-matched controls is completely accounted for by abdominal fat, but not by androgens.27 The increasing differences in insulin resistance at high BMI between obese controls and women with PCOS can be accounted for by adjusting for abdominal fat mass.28

Exercise and lifestyle modifications are thought to be the most effective methods in reducing visceral fat and the associated metabolic syndrome.29 However, metformin has been shown to be beneficial in populations with impaired glucose tolerance30 and it is now used widely in women with PCOS. Although metformin gave few real advantages to our population of women with PCOS, meta-analysis does report some benefit.31 Our results concur with the published diabetes prevention programme trial which failed to show benefit in preventing the development of the metabolic syndrome in women who were treated with metformin, although they did report that lifestyle modification was effective.32 There is some evidence that metformin may enhance the degree of visceral fat loss when taken during a hypocalorific diet.33 Further randomised controlled trials are needed on the individual and combined effectiveness of lifestyle modification and metformin in women with PCOS.

This study does not have the sensitivity to detect small reductions in fat mass. In order to limit the radiation dose from the CT scan, we used a single-slice technique which introduces an estimated error of 4.6% when compared with multiple slices.34 A retrospective power analysis shows that this study has a 75% power to detect a 30% reduction of visceral fat in the metformin group at a significance of 0.05. This loss of visceral fat corresponds to an overall weight loss of 5–10% which is the degree of weight loss that results in resumption of ovulation and clinically significant improvements in metabolic parameters.9–11 It is possible that a longer term study may have detected changes in fat mass, but other trials have shown that where metformin is effective on metabolic and clinical parameters, this effect is seen within 2 months.31 Therefore, while this study may have missed minor variations in visceral fat, it is likely to have detected clinically significant changes.

A Cochrane review had previously shown a modest reduction in LDL cholesterol in the metformin arm, although none of the individual trials showed a significant effect.31 Other metabolic parameters were not significantly different in the metformin group, although this trial lacked the power to detect the subtle variations that might be expected in this population without diabetes.

Metformin has been shown to be effective in inducing ovulation in women with PCOS.35 Our population had a high rate of ovulation in both placebo and treatment arms. This is similar to other randomised controlled trials where high rates of ovulation in the placebo arm have been reported.36,37 In our population, the label of anovulation is too simplistic and a concept of oligo-ovulation is more appropriate, especially when follow up is over several months. The fact that five participants were found to have either resumed ovulation or had a normal FAI at the time of randomisation despite having had previously documented anovulation and raised FAIs confirms that PCOS is a condition that is variable.

Conclusion

This trial lends support to the growing evidence that metformin is not a weight loss drug.35 We have also shown that a 12-week course of metformin has no clinically significant effect in reducing visceral fat mass, although it does have a beneficial effect on lipids. Metformin has been shown to be an effective treatment for ovulation induction, and the small but beneficial metabolic improvements may be an advantage over traditional treatments like clomifene. There is strong evidence to suggest that visceral fat is an important contribution to, or possibly cause of, metabolic disturbance in women with PCOS and that reducing it has beneficial clinical and metabolic effects in this population.29 We therefore conclude that metformin may be used as an adjunct to lifestyle modification in women with PCOS but not as a substitute for it.

Ancillary