5 minute lightning oral presentation (and poster) ESA-SRB 2023 in conjunction with ENSA

Hierarchical Cluster Analysis and Principal Components Analysis confirm metabolic and reproductive subtypes in PCOS   (#220)

Kharis Burns 1 2 , Alexander Stuckey 3 4 , Scott Wilson 5 6 , Gerald Watts 2 7 , Bronwyn Stuckey 2 6 8
  1. Department of Endocrinology and Diabetes, Royal Perth Hospital, Perth, WA, Australia
  2. Medical School, University of Western Australia, Crawley, WA, Australia
  3. School of Agriculture and Environment, University of Western Australia, Crawley, WA, Australia
  4. Pink Lake Analytics, Nedlands, WA, Australia
  5. School of Biomedical Sciences, University of Western Australia, Perth, WA, Australia
  6. Department of Endocrinology and Diabetes, Sir Charles Gairdner Hospital, Nedlands, WA, Australia
  7. Lipid Disorders Clinic, Department of Cardiology, Royal Perth Hospital, Perth, WA, Australia
  8. Keogh Institute for Medical Research, Nedlands, WA, Australia
  • .

Polycystic Ovary Syndrome (PCOS), characterised by hyperandrogenism and oligomenorrhoea, is an umbrella term encompassing notable heterogeneity.

We aimed to explore the heterogeneity of PCOS and potential subtypes by statistical analysis of biochemical and anthropometric data using two statistical methods.

We studied 1035 women with NIH-defined PCOS including data on BMI, serum LH:FSH, testosterone, DHEAS, androstenedione,17-OH-progesterone, SHBG, HOMA_IR, lipids and blood pressure. Unsupervised agglomerative hierarchical cluster analysis was used to group a) phenotypic variables and b) patients into clusters. Principal components analysis (PCA) was used to resolve correlated variables (excluding BMI) into independent factors. The relationship between resultant components and BMI was then explored.

Study subjects had a median age 29.0y (23.0, 37.0). 28% were lean, 20.5% overweight, and 51.5% obese. Analysis of phenotypic variables revealed two main clusters - one characterised by blood pressure, BMI, HOMA_IR and triglycerides, and a second by LH:FSH, androgens, SHBG, and lipids. There were 3 separate patient clusters: Cluster A (40.4% of women) demonstrated lower BP, BMI, HOMA_IR, triglycerides, testosterone, FAI, and higher LH:FSH, DHEAS, androstenedione, 17-OH-progesterone, SHBG and HDL.  In contrast, cluster C (45.4%) had higher BP, BMI, HOMA_IR and triglycerides and lower LH:FSH, androgens, SHBG and lipids. Cluster B (14.2%) was intermediate.

PCA, excluding BMI, found two components that aligned with the cluster analysis. Variables with greatest weight in principal component 1 (PC1) included HOMA_IR, triglycerides, systolic and diastolic BP, FAI, and SHBG. PC1 demonstrated positive correlation with BMI (R2=0.26) and aligned with cluster C. Principal component 2 (PC2) was strongly influenced by LH:FSH, testosterone, FAI, DHEAS and androstenedione, with loadings in the opposite direction from LDL and cholesterol, and aligned with cluster A. There was little relationship between BMI and PC2 (R2=0.028).

Our analysis revealed “metabolic” and “reproductive” subtypes. BMI was influential in the metabolic subtype but not the reproductive subtype.