TOPLINE:

A predictive model incorporating anti-Müllerian hormone (AMH), ovarian volume, hirsutism score, and maximum menstrual cycle length accurately predicts polycystic ovary syndrome (PCOS) diagnosis with performance comparable to models using all ultrasonographic, biochemical, and clinical features. AMH alone predicted PCOS diagnosis with relatively good diagnostic accuracy, with area under the receiver operating characteristic curve (AUROC) of 0.884.

METHODOLOGY:

• Researchers conducted a multicenter, retrospective pilot study using transvaginal ultrasound images and clinical data from 101 participants with PCOS and 50 control individuals recruited from two clinical facilities in the US (University of Rochester and Weill Cornell Medicine) and one community-based academic research center (Cornell University).
• Participants were assigned female at birth, aged 18-45 years, had at least one ovary, and were off hormonal contraceptives for ≥ 8 weeks; PCOS diagnosis was defined by the 2023 International Evidence-Based Guideline requiring at least two of the three criteria: hyperandrogenism, irregular menses, and polycystic ovarian morphology.
• Analysis included ultrasonographic (ovarian volume, follicle number per ovary, follicle number per single cross section), biochemical (sex hormone-binding globulin, total testosterone, free androgen index, and bioavailable testosterone), and clinical (age, BMI, menstrual cycle length, Ferriman-Gallwey hirsutism score, follicle stimulating hormone, luteinizing hormone, estradiol, and AMH) variables.
• A logistic regression model with L2 regularization was trained to predict PCOS diagnosis using subsets of variables, with model performance evaluated using AUROC and F1 score measures; values > 0.8 were considered high performing.
• Phenotyping was performed at the ovary level, with 170 ovaries from 101 PCOS participants and 79 ovaries from 50 control individuals included in the final analysis after excluding 113 ovaries with only one diagnostic feature of PCOS.

TAKEAWAY:

• AMH alone predicted PCOS diagnosis with an AUROC of 0.884 (95% CI, 0.836-0.932), F1 score of 0.807 (95% CI, 0.788-0.827), and accuracy of 0.677 (95% CI, 0.649-0.705).
• Integration of AMH and ovarian volume improved model performance with an AUROC of 0.906 (95% CI, 0.868-0.944), F1 score of 0.811 (95% CI, 0.744-0.873), and accuracy of 0.686 (95% CI, 0.600-0.770).
• Combining ovarian volume with AMH and all ultrasonographic, biochemical, and clinical variables yielded an AUROC of 0.991 (95% CI, 0.973-1.000), F1 score of 0.811 (95% CI, 0.740-0.879), and accuracy of 0.683 (95% CI, 0.588-0.777).
• A refined model using AMH, ovarian volume, modified Ferriman-Gallwey score, and maximum self-reported menstrual cycle length demonstrated produced an AUROC of 0.982 (95% CI, 0.958-1.000), F1 score of 0.805 (95% CI, 0.759-0.850), and accuracy of 0.674 (95% CI, 0.610-0.738).

IN PRACTICE:

“A minimum combination of ovarian volume with AMH and examination/history-derived end points can accurately predict PCOS diagnosis with performance comparable to that of a combination of all ultrasonographic, biochemical, and clinical features. This may streamline diagnostic workflows, thereby reducing clinical burden,” the authors of the study wrote.

SOURCE:

The study was led by Adam T. Evans, MD, Division of Reproductive Endocrinology and Infertility, University Hospitals/Case Western Reserve University School of Medicine in Beachwood, Ohio. It was published online in O&G Open.

LIMITATIONS:

This is a retrospective pilot study with a small sample size, requiring larger prospective studies to validate the findings. Ultrasound images were obtained exclusively with GE Voluson systems, and AMH levels were generated with a specific commercial laboratory, which may have limited generalizability. Each ovary and its associated clinical features were considered a single unit for phenotyping, with both ovaries analyzed separately in most cases; further studies with larger sample sizes including the average of both ovaries should be performed to validate findings. Cases and control individuals were not matched for age due to the retrospective nature of the analysis, and this difference across groups could have affected model performance.

DISCLOSURES:

Financial support for this research was obtained through funding from the Richard W. and Mae Stone Goode Foundation Research Award to Evans and Kathleen Hoeger, Cornell University Multi-Investigator Seed Grant to Marla Lujan and Iman Hajirasouliha, and Cornell CALS Moonshot Seed Grant to Lujan and Hajirasouliha. Parent protocols received funding through grants from the National Institutes of Health and the US Department of Agriculture to Lujan. Hajirasouliha disclosed serving as a consultant and on the advisory board of Noor Sciences. The other authors reported having no potential conflicts of interest.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.