Can cluster analysis be used to differentiate between normo-ovulatory women with normal ovaries and normo-ovulatory women with polycystic ovarian morphology (PCOM) in a non-subjective manner?

Cluster analysis can be used to accurately and non-subjectively differentiate between normo-ovulatory women with normal ovaries and normo-ovulatory women with PCOM.

Currently, PCOM is diagnosed using a fixed threshold level, i.e. 12 or more follicles per ovary, and is one of the diagnostic criteria of polycystic ovary syndrome (PCOS). However, PCOM is also encountered in normo-ovulatory women, suggesting that it could just represent a normal variant. On the other hand, recent studies have shown subtle endocrine abnormalities in women with isolated PCOM that resemble those found in women with PCOS. Because of the strong correlation between anti-M�llerian hormone (AMH) and follicle number, a high serum AMH level has been proposed as a surrogate marker for PCOM and could, therefore, be integrated in the diagnostic classifications for PCOS.

This was a retrospective observational cohort study. Original cohorts had been recruited for previous studies between 1998 and 2010. Two hundred ninety-seven regularly cycling women and 700 women with PCOS were eligible for inclusion.

Cluster analysis was performed in 297 regularly cycling women. After exclusion of ‘PCOM’ clusters, each ‘non-PCOM’ cluster (young, n = 118 and old, n = 100) was included in the construction of a receiver operating characteristics curve to test the diagnostic performance of follicle number per ovary (FNPO) and serum AMH in discriminating similarly aged full-blown PCOS patients (n = 411 and 237, respectively) from normal regularly cycling non-PCOM women.

The optimal number of clusters was four; age was the most important classifying variable, followed by the FNPO and serum AMH. Two distinct clusters of normo-ovulatory women with PCOM were isolated and differed solely by age, i.e. ‘young’ and ‘old’. Both ‘PCOM’ clusters had their similarly aged counterpart of ‘non-PCOM’ clusters. Likewise, two clusters comprised women younger than 30 years, with (n = 28, ‘PCOM regularly cycling women’) or without (n = 118, ‘normal regularly cycling women’) features of PCOM (increased FNPO and/or serum AMH). The two other clusters in older women could be labelled ‘normal regularly cycling women’ or ‘PCOM regularly cycling women’ (n = 100 and 51, respectively). The prevalence of PCOM was significantly greater in old than in young regularly cycling women controls. In the young population, after exclusion of the ‘PCOM regularly cycling women’, the diagnostic performance of AMH, expressed by area under the curve (AUC) (AUC = 0.903; CI (0.876–0.930)) to differentiate PCOS women from normal regularly cycling women was similar to that using the FNPO (AUC = 0.915, CI (0.891–0.940)) (P = 0.25), confirming results from earlier studies. In the old population, the diagnostic performance of AMH was greater than that of FNPO (AUCs = 0.948 (0.927–0.970) vs 0.874 (0.836–0.912), respectively, P = 0.00035). Cut-off levels of AMH and antral follicle count distinguishing regularly cycling non-PCOM women from PCOS women were higher in young women than in older women.