Clinical Endocrinology (2015) 83, 879–887
doi: 10.1111/cen.12830
ORIGINAL ARTICLE
The association between Polycystic Ovary Syndrome (PCOS) and metabolic syndrome: a statistical modelling approach S. Ranasinha*,a, A.E. Joham*,†,a, R.J. Norman‡, J.E. Shaw§, S. Zoungas*,†, J. Boyle*, L. Moran*,‡,b and H.J. Teede*,†,b *Women’s Reproductive Health Research, Monash Centre for Health Research and Implementation, School of Public Health and Preventive Medicine, Monash University, †Diabetes and Vascular Medicine Unit, Monash Health, Clayton, Vic, ‡Robinson Institute, School of Paediatrics and Reproductive Health, University of Adelaide, North Adelaide, SA and §Baker IDI Heart and Diabetes Institute, Melbourne, Vic, Australia
Summary Objective Polycystic ovary syndrome (PCOS) affects 12–21% of women. Women with PCOS exhibit clustering of metabolic features. We applied rigorous statistical methods to further understand the interplay between PCOS and metabolic features including insulin resistance, obesity and androgen status. Design Retrospective cross-sectional analysis. Patients Women with PCOS attending reproductive endocrine clinics in South Australia for the treatment of PCOS (n = 172). Women without PCOS (controls) in the same Australian region (n = 335) from the Australian Diabetes, Obesity and Lifestyle Study (AusDiab), a national population-based study (age- and BMI-matched within one standard deviation of the PCOS cohort). Measurements The factor structure for metabolic syndrome for women with PCOS and control groups was examined, specifically, the contribution of individual factors to metabolic syndrome and the association of hyperandrogenism with other metabolic factors. Results Women with PCOS demonstrated clustering of metabolic features that was not observed in the control group. Metabolic syndrome in the PCOS cohort was strongly represented by obesity (standardized factor loading = 095, P < 0001) and insulin resistance factors (loading = 092, P < 0001) and moderately by blood pressure (loading = 062, P < 0001) and lipid factors (loading = 067, P = 0002). On further analysis, the insulin resistance factor strongly correlated with the obesity (r = 070, P < 0001) and lipid factors (r = 068, P < 0001) and moderately with the blood pressure factor (loading = 043,
Correspondence: Professor Helena J. Teede, School of Public Health and Preventive Medicine, Monash University, Locked bag 29, Monash Medical Centre, Clayton, Victoria 3168, Australia. Tel.: + 613 9594 7545; Fax: + 613 9594 7550; E-mail: [email protected] a
Both authors contributed equally to the manuscript. These authors made equal senior contribution to the manuscript.
b
© 2015 John Wiley & Sons Ltd
P = 0002). The hyperandrogenism factor was moderately correlated with the insulin resistance factor (r = 038, P < 0003), but did not correlate with any other metabolic factors. Conclusions PCOS women are more likely to display metabolic clustering in comparison with age- and BMI-matched control women. Obesity and insulin resistance, but not androgens, are independently and most strongly associated with metabolic syndrome in PCOS. (Received 24 March 2015; returned for revision 14 April 2015; finally revised 7 May 2015; accepted 1 June 2015)
Introduction Polycystic ovary syndrome (PCOS) is a common condition affecting up to 12–21% of reproductive-aged women, depending on the diagnostic criteria applied and population studied.1,2 PCOS has a heterogeneous clinical presentation comprising reproductive (hyperandrogenism, menstrual irregularity, anovulation, infertility, pregnancy complications), metabolic [insulin resistance (IR), increased type 2 diabetes mellitus (T2DM) and cardiovascular disease (CVD) risk factors] and psychological features (worsened quality of life and increased anxiety and depression).3 Given the prevalence, metabolic and clinical complications, PCOS is a major public health challenge.3,4 PCOS is essentially a hormonal disorder underpinned by IR and hyperandrogenism.3,5 Obesity is common in PCOS and increases the risk and severity of the condition through further exacerbation of IR.2,6,7 In this setting, with both inherent PCOSrelated IR8 and obesity-related IR, the majority of women with PCOS exhibit metabolic features9 including glucose intolerance, dyslipidaemia, hypertension and visceral obesity.10,11 While there is some debate on the concept of the metabolic syndrome (MS) and the key criteria that define this syndrome, it is well established that these metabolic features do tend to cluster in populations who are at high risk for CVD. The National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) defined practical criteria for the diagnosis of the MS10 that 879
880 S. Ranasinha et al. requires clustering of three or more of five metabolic criteria. The World Health Organization (WHO) criteria require the presence of IR for the diagnosis of MS in addition to at least two other criteria.10 The International Diabetes Federation (IDF) criteria require the presence of abdominal obesity specific to ethnicity in addition to at least two other criteria.12 The harmonized MS criteria attempt to unify the other criteria and require clustering of three or more of five metabolic criteria including population and country-specific waist circumference.13 Regardless of the criteria used, clustering of metabolic factors is well established and all criteria independently predict risk of CVD.14 Given the high prevalence of MS, the increased risk of T2DM15 and high prevalence of cardiovascular risk factors, women with PCOS represent a high risk group. Studies in PCOS have shown a high prevalence of MS9 and suggested exacerbation by obesity. However, the relationship between IR and MS has rarely been studied independently of obesity in PCOS. More importantly, the impact of hyperandrogenism on metabolic features in PCOS remains controversial with prior studies showing differing relationships with different androgens.16 A greater understanding of the relationship between PCOS, IR, obesity, hyperandrogenism and metabolic features will inform appropriate screening and preventive interventions for women with PCOS at high metabolic risk. To progress this understanding, we aimed to apply rigorous and sophisticated statistical methods to test these relationships. Conventional statistical methods are unsuitable for the proposed analysis as MS does not represent a unified construct reflecting a single underlying pathway. In this study, we aimed to construct models using confirmatory factor analysis (CFA), an extension of generalized estimating equations (GEE) that delineate the clustering among variables characterizing the MS and the interaction with both IR and obesity. Using this model, we can create composite variables which make up the five key components of MS (for example, dyslipidaemia). Each of these composite variables represents a metabolic factor derived from several measured subcomponents of that factor (e.g. the subcomponents for the composite variable dyslipidaemia are HDL, LDL and triglycerides). This method takes into account the interplay between different factors in the model compared to a traditional multivariable model where other variables are held constant. The advantage of using this method over conventional methods is that it provides insight into the magnitude of each of the individual components and subcomponents of MS. This method has not previously been used to study metabolic syndrome in PCOS. Using this method, we can compare differences between PCOS and non-PCOS groups. We also aimed to conduct a separate analysis on the PCOS cohort to examine the relationship between androgen status and metabolic features including IR, obesity, dyslipidaemia and hypertension.
Materials and methods Participants An observational study was conducted using existing records from a clinic recruited sample of PCOS women (n = 172) and
an age- and body mass index (BMI)-matched sample of control women recruited from the community (n = 335). The PCOS population has been previously characterized17 and includes a retrospective data set, based on case records from women presenting to reproductive endocrine clinics in South Australia. Women were not receiving medication at the time of study. Diagnosis of PCOS was made on the National Institutes of Health (NIH) endorsed, internationally accepted Rotterdam criteria of two of three features of chronic anovulation: clinical or biochemical evidence of hyperandrogenism and the presence of polycystic ovaries on ultrasound, in the absence of other disorders which may contribute to these features.4 Ethics approval was received from the Human Research Ethics Committee of the Women’s and Children’s Hospital, Adelaide, and all subjects gave informed consent. The comparison group of women from the general population was selected from the community-based Australian Diabetes, Obesity and Lifestyle Study (AusDiab). The overall cohort and methodology has also been well described.18 In brief, AusDiab is a national population-based study of 11 247 predominantly Caucasian (>98%) adults aged 25 or above in 1999–2000.18,19 Ethics approval was received from the International Diabetes Institute Ethics Committee, and all subjects gave informed consent. Selection of cohorts for analysis On case review, 371 of women attending the reproductive endocrine clinic were diagnosed with PCOS. Women with data on all metabolic risk factors were included in this study (n = 172). Within this cohort, measures of IR had been completed in 139 women. For controls, a subset of female participants from South Australia (n = 335) was selected from the overall baseline AusDiab population, based on age and BMI comparability with the 172 women with PCOS (within 1 standard deviation). Clinical and biochemical measurements For the PCOS cohort, weight, BMI, waist circumference, testosterone, sex hormone binding globulin (SHBG), dehydroepiandrosterone sulphate (DHEAS), androstenedione, fasting and 2 h after 75 g oral glucose tolerance test (OGTT) glucose and insulin, blood pressure (BP), total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol and triglycerides were analysed as previously described.17 The free androgen index (FAI) was calculated as total testosterone/SHBG 9 100. For the AusDiab control cohort, fasting weight, BMI, waist circumference, fasting and 2 h after 75 g OGTT glucose and insulin, BP, total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides were analysed as previously described.19,20 Insulin levels were measured in 139 women with PCOS, but in the controls only in those aged 35 years or more (n = 60). MS was defined according to the NCEP ATP criteria comprising dysglycaemia, hypertension, hypertriglyceridaemia, low HDL cholesterol and abdominal obesity.10 © 2015 John Wiley & Sons Ltd Clinical Endocrinology (2015), 83, 879–887
Clustering of metabolic syndrome in PCOS 881 Statistical methods Descriptive and univariable analyses were performed using STATA v11.2 (StataCorp, TX, USA). Prior to analysis, the normality assumptions of the data were examined, and the variables with a high degree of skewness or kurtosis were transformed with the natural log function. Physiologic and anthropometric variables were assessed between women with PCOS and control women. Continuous normally distributed data were compared using two-sample t-test, and mean and standard deviation (SD) were reported. Mann–Whitney U-tests were performed for skewed variables, and median and interquartile range (IQR) was reported. P < 005 was considered statistically significant. CFA was performed using analysis of moment structures (AMOS) v19. The CFA models were evaluated in three ways. Firstly, the chi-square test was used to assess the congruency between the hypothesized models and empirical data from the PCOS and control samples. Secondly, as the chi-square test is sensitive to sample size, two model fit indices were used to evaluate the models: the Comparative Fit Index (CFI) and rootmean-square error (RMSE). The CFA model was deemed a good model if the chi-square was small enough for the P -value to be greater than 005, the CFI value was greater than 095 and the RMSE value was less than 005. Then Bayesian estimation methods were utilized to test the robustness of the above hierarchical models, for the purpose of comparing the PCOS and controls model parameters derived from maximum likelihood estimations. In the PCOS cohort alone, we examined the relationship between the factor hyperandrogenism, composed of DHEAS, androstenedione and FAI measurements, with the other four factors: IR, obesity, lipids and BP (model 4).
Results Characteristics of the cohort The demographic characteristics and metabolic variables for PCOS and control women are shown in Table 1. Age and BMI were similar across the groups given the matching control selection process (Table 1). Systolic BP was higher (P < 0001) and HDL cholesterol was lower (P =
doi: 10.1111/cen.12830
ORIGINAL ARTICLE
The association between Polycystic Ovary Syndrome (PCOS) and metabolic syndrome: a statistical modelling approach S. Ranasinha*,a, A.E. Joham*,†,a, R.J. Norman‡, J.E. Shaw§, S. Zoungas*,†, J. Boyle*, L. Moran*,‡,b and H.J. Teede*,†,b *Women’s Reproductive Health Research, Monash Centre for Health Research and Implementation, School of Public Health and Preventive Medicine, Monash University, †Diabetes and Vascular Medicine Unit, Monash Health, Clayton, Vic, ‡Robinson Institute, School of Paediatrics and Reproductive Health, University of Adelaide, North Adelaide, SA and §Baker IDI Heart and Diabetes Institute, Melbourne, Vic, Australia
Summary Objective Polycystic ovary syndrome (PCOS) affects 12–21% of women. Women with PCOS exhibit clustering of metabolic features. We applied rigorous statistical methods to further understand the interplay between PCOS and metabolic features including insulin resistance, obesity and androgen status. Design Retrospective cross-sectional analysis. Patients Women with PCOS attending reproductive endocrine clinics in South Australia for the treatment of PCOS (n = 172). Women without PCOS (controls) in the same Australian region (n = 335) from the Australian Diabetes, Obesity and Lifestyle Study (AusDiab), a national population-based study (age- and BMI-matched within one standard deviation of the PCOS cohort). Measurements The factor structure for metabolic syndrome for women with PCOS and control groups was examined, specifically, the contribution of individual factors to metabolic syndrome and the association of hyperandrogenism with other metabolic factors. Results Women with PCOS demonstrated clustering of metabolic features that was not observed in the control group. Metabolic syndrome in the PCOS cohort was strongly represented by obesity (standardized factor loading = 095, P < 0001) and insulin resistance factors (loading = 092, P < 0001) and moderately by blood pressure (loading = 062, P < 0001) and lipid factors (loading = 067, P = 0002). On further analysis, the insulin resistance factor strongly correlated with the obesity (r = 070, P < 0001) and lipid factors (r = 068, P < 0001) and moderately with the blood pressure factor (loading = 043,
Correspondence: Professor Helena J. Teede, School of Public Health and Preventive Medicine, Monash University, Locked bag 29, Monash Medical Centre, Clayton, Victoria 3168, Australia. Tel.: + 613 9594 7545; Fax: + 613 9594 7550; E-mail: [email protected] a
Both authors contributed equally to the manuscript. These authors made equal senior contribution to the manuscript.
b
© 2015 John Wiley & Sons Ltd
P = 0002). The hyperandrogenism factor was moderately correlated with the insulin resistance factor (r = 038, P < 0003), but did not correlate with any other metabolic factors. Conclusions PCOS women are more likely to display metabolic clustering in comparison with age- and BMI-matched control women. Obesity and insulin resistance, but not androgens, are independently and most strongly associated with metabolic syndrome in PCOS. (Received 24 March 2015; returned for revision 14 April 2015; finally revised 7 May 2015; accepted 1 June 2015)
Introduction Polycystic ovary syndrome (PCOS) is a common condition affecting up to 12–21% of reproductive-aged women, depending on the diagnostic criteria applied and population studied.1,2 PCOS has a heterogeneous clinical presentation comprising reproductive (hyperandrogenism, menstrual irregularity, anovulation, infertility, pregnancy complications), metabolic [insulin resistance (IR), increased type 2 diabetes mellitus (T2DM) and cardiovascular disease (CVD) risk factors] and psychological features (worsened quality of life and increased anxiety and depression).3 Given the prevalence, metabolic and clinical complications, PCOS is a major public health challenge.3,4 PCOS is essentially a hormonal disorder underpinned by IR and hyperandrogenism.3,5 Obesity is common in PCOS and increases the risk and severity of the condition through further exacerbation of IR.2,6,7 In this setting, with both inherent PCOSrelated IR8 and obesity-related IR, the majority of women with PCOS exhibit metabolic features9 including glucose intolerance, dyslipidaemia, hypertension and visceral obesity.10,11 While there is some debate on the concept of the metabolic syndrome (MS) and the key criteria that define this syndrome, it is well established that these metabolic features do tend to cluster in populations who are at high risk for CVD. The National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) defined practical criteria for the diagnosis of the MS10 that 879
880 S. Ranasinha et al. requires clustering of three or more of five metabolic criteria. The World Health Organization (WHO) criteria require the presence of IR for the diagnosis of MS in addition to at least two other criteria.10 The International Diabetes Federation (IDF) criteria require the presence of abdominal obesity specific to ethnicity in addition to at least two other criteria.12 The harmonized MS criteria attempt to unify the other criteria and require clustering of three or more of five metabolic criteria including population and country-specific waist circumference.13 Regardless of the criteria used, clustering of metabolic factors is well established and all criteria independently predict risk of CVD.14 Given the high prevalence of MS, the increased risk of T2DM15 and high prevalence of cardiovascular risk factors, women with PCOS represent a high risk group. Studies in PCOS have shown a high prevalence of MS9 and suggested exacerbation by obesity. However, the relationship between IR and MS has rarely been studied independently of obesity in PCOS. More importantly, the impact of hyperandrogenism on metabolic features in PCOS remains controversial with prior studies showing differing relationships with different androgens.16 A greater understanding of the relationship between PCOS, IR, obesity, hyperandrogenism and metabolic features will inform appropriate screening and preventive interventions for women with PCOS at high metabolic risk. To progress this understanding, we aimed to apply rigorous and sophisticated statistical methods to test these relationships. Conventional statistical methods are unsuitable for the proposed analysis as MS does not represent a unified construct reflecting a single underlying pathway. In this study, we aimed to construct models using confirmatory factor analysis (CFA), an extension of generalized estimating equations (GEE) that delineate the clustering among variables characterizing the MS and the interaction with both IR and obesity. Using this model, we can create composite variables which make up the five key components of MS (for example, dyslipidaemia). Each of these composite variables represents a metabolic factor derived from several measured subcomponents of that factor (e.g. the subcomponents for the composite variable dyslipidaemia are HDL, LDL and triglycerides). This method takes into account the interplay between different factors in the model compared to a traditional multivariable model where other variables are held constant. The advantage of using this method over conventional methods is that it provides insight into the magnitude of each of the individual components and subcomponents of MS. This method has not previously been used to study metabolic syndrome in PCOS. Using this method, we can compare differences between PCOS and non-PCOS groups. We also aimed to conduct a separate analysis on the PCOS cohort to examine the relationship between androgen status and metabolic features including IR, obesity, dyslipidaemia and hypertension.
Materials and methods Participants An observational study was conducted using existing records from a clinic recruited sample of PCOS women (n = 172) and
an age- and body mass index (BMI)-matched sample of control women recruited from the community (n = 335). The PCOS population has been previously characterized17 and includes a retrospective data set, based on case records from women presenting to reproductive endocrine clinics in South Australia. Women were not receiving medication at the time of study. Diagnosis of PCOS was made on the National Institutes of Health (NIH) endorsed, internationally accepted Rotterdam criteria of two of three features of chronic anovulation: clinical or biochemical evidence of hyperandrogenism and the presence of polycystic ovaries on ultrasound, in the absence of other disorders which may contribute to these features.4 Ethics approval was received from the Human Research Ethics Committee of the Women’s and Children’s Hospital, Adelaide, and all subjects gave informed consent. The comparison group of women from the general population was selected from the community-based Australian Diabetes, Obesity and Lifestyle Study (AusDiab). The overall cohort and methodology has also been well described.18 In brief, AusDiab is a national population-based study of 11 247 predominantly Caucasian (>98%) adults aged 25 or above in 1999–2000.18,19 Ethics approval was received from the International Diabetes Institute Ethics Committee, and all subjects gave informed consent. Selection of cohorts for analysis On case review, 371 of women attending the reproductive endocrine clinic were diagnosed with PCOS. Women with data on all metabolic risk factors were included in this study (n = 172). Within this cohort, measures of IR had been completed in 139 women. For controls, a subset of female participants from South Australia (n = 335) was selected from the overall baseline AusDiab population, based on age and BMI comparability with the 172 women with PCOS (within 1 standard deviation). Clinical and biochemical measurements For the PCOS cohort, weight, BMI, waist circumference, testosterone, sex hormone binding globulin (SHBG), dehydroepiandrosterone sulphate (DHEAS), androstenedione, fasting and 2 h after 75 g oral glucose tolerance test (OGTT) glucose and insulin, blood pressure (BP), total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol and triglycerides were analysed as previously described.17 The free androgen index (FAI) was calculated as total testosterone/SHBG 9 100. For the AusDiab control cohort, fasting weight, BMI, waist circumference, fasting and 2 h after 75 g OGTT glucose and insulin, BP, total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides were analysed as previously described.19,20 Insulin levels were measured in 139 women with PCOS, but in the controls only in those aged 35 years or more (n = 60). MS was defined according to the NCEP ATP criteria comprising dysglycaemia, hypertension, hypertriglyceridaemia, low HDL cholesterol and abdominal obesity.10 © 2015 John Wiley & Sons Ltd Clinical Endocrinology (2015), 83, 879–887
Clustering of metabolic syndrome in PCOS 881 Statistical methods Descriptive and univariable analyses were performed using STATA v11.2 (StataCorp, TX, USA). Prior to analysis, the normality assumptions of the data were examined, and the variables with a high degree of skewness or kurtosis were transformed with the natural log function. Physiologic and anthropometric variables were assessed between women with PCOS and control women. Continuous normally distributed data were compared using two-sample t-test, and mean and standard deviation (SD) were reported. Mann–Whitney U-tests were performed for skewed variables, and median and interquartile range (IQR) was reported. P < 005 was considered statistically significant. CFA was performed using analysis of moment structures (AMOS) v19. The CFA models were evaluated in three ways. Firstly, the chi-square test was used to assess the congruency between the hypothesized models and empirical data from the PCOS and control samples. Secondly, as the chi-square test is sensitive to sample size, two model fit indices were used to evaluate the models: the Comparative Fit Index (CFI) and rootmean-square error (RMSE). The CFA model was deemed a good model if the chi-square was small enough for the P -value to be greater than 005, the CFI value was greater than 095 and the RMSE value was less than 005. Then Bayesian estimation methods were utilized to test the robustness of the above hierarchical models, for the purpose of comparing the PCOS and controls model parameters derived from maximum likelihood estimations. In the PCOS cohort alone, we examined the relationship between the factor hyperandrogenism, composed of DHEAS, androstenedione and FAI measurements, with the other four factors: IR, obesity, lipids and BP (model 4).
Results Characteristics of the cohort The demographic characteristics and metabolic variables for PCOS and control women are shown in Table 1. Age and BMI were similar across the groups given the matching control selection process (Table 1). Systolic BP was higher (P < 0001) and HDL cholesterol was lower (P =
