Development of a novel risk prediction and risk stratification score for polycystic ovary syndrome

Predicting polycystic ovary syndrome with machine learning algorithms from electronic health records

Introduction

Predictive models have been used to aid early diagnosis of PCOS, though existing models are based on small sample sizes and limited to fertility clinic populations. We built a predictive model using machine learning algorithms based on an outpatient population at risk for PCOS to predict risk and facilitate earlier diagnosis, particularly among those who meet diagnostic criteria but have not received a diagnosis.

Methods

This is a retrospective cohort study from a SafetyNet hospital's electronic health records (EHR) from 2003-2016. The study population included 30,601 women aged 18-45 years without concurrent endocrinopathy who had any visit to Boston Medical Center for primary care, obstetrics and gynecology, endocrinology, family medicine, or general internal medicine. Four prediction outcomes were assessed for PCOS. The first outcome was PCOS ICD-9 diagnosis with additional model outcomes of algorithm-defined PCOS. The latter was based on Rotterdam criteria and merging laboratory values, radiographic imaging, and ICD data from the EHR to define irregular menstruation, hyperandrogenism, and polycystic ovarian morphology on ultrasound.

Results

We developed predictive models using four machine learning methods: logistic regression, supported vector machine, gradient boosted trees, and random forests. Hormone values (follicle-stimulating hormone, luteinizing hormone, estradiol, and sex hormone binding globulin) were combined to create a multilayer perceptron score using a neural network classifier. Prediction of PCOS prior to clinical diagnosis in an out-of-sample test set of patients achieved an average AUC of 85%, 81%, 80%, and 82%, respectively in Models I, II, III and IV. Significant positive predictors of PCOS diagnosis across models included hormone levels and obesity; negative predictors included gravidity and positive bHCG.

Conclusion

Machine learning algorithms were used to predict PCOS based on a large at-risk population. This approach may guide early detection of PCOS within EHR-interfaced populations to facilitate counseling and interventions that may reduce long-term health consequences. Our model illustrates the potential benefits of an artificial intelligence-enabled provider assistance tool that can be integrated into the EHR to reduce delays in diagnosis. However, model validation in other hospital-based populations is necessary.

Progress of the application clinical prediction model in polycystic ovary syndrome

Clinical prediction models play an important role in the field of medicine. These can help predict the probability of an individual suffering from disease, complications, and treatment outcomes by applying specific methodologies. Polycystic ovary syndrome (PCOS) is a common disease with a high incidence rate, huge heterogeneity, short- and long-term complications, and complex treatments. In this systematic review study, we reviewed the progress of clinical prediction models in PCOS patients, including diagnosis and prediction models for PCOS complications and treatment outcomes. We aimed to provide ideas for medical researchers and clues for the management of PCOS. In the future, models with poor accuracy can be greatly improved by adding well-known parameters and validations, which will further expand our understanding of PCOS in terms of precision medicine. By developing a series of predictive models, we can make the definition of PCOS more accurate, which can improve the diagnosis of PCOS and reduce the likelihood of false positives and false negatives. It will also help discover complications earlier and treatment outcomes being known earlier, which can result in better outcomes for women with PCOS.

Predicting polycystic ovary syndrome (PCOS) with machine learning algorithms from electronic health records

Introduction

Predictive models have been used to aid early diagnosis of PCOS, though existing models are based on small sample sizes and limited to fertility clinic populations. We built a predictive model using machine learning algorithms based on an outpatient population at risk for PCOS to predict risk and facilitate earlier diagnosis, particularly among those who meet diagnostic criteria but have not received a diagnosis.

Methods

This is a retrospective cohort study from a SafetyNet hospital's electronic health records (EHR) from 2003-2016. The study population included 30,601 women aged 18-45 years without concurrent endocrinopathy who had any visit to Boston Medical Center for primary care, obstetrics and gynecology, endocrinology, family medicine, or general internal medicine. Four prediction outcomes were assessed for PCOS. The first outcome was PCOS ICD-9 diagnosis with additional model outcomes of algorithm-defined PCOS. The latter was based on Rotterdam criteria and merging laboratory values, radiographic imaging, and ICD data from the EHR to define irregular menstruation, hyperandrogenism, and polycystic ovarian morphology on ultrasound.

Results

We developed predictive models using four machine learning methods: logistic regression, supported vector machine, gradient boosted trees, and random forests. Hormone values (follicle-stimulating hormone, luteinizing hormone, estradiol, and sex hormone binding globulin) were combined to create a multilayer perceptron score using a neural network classifier. Prediction of PCOS prior to clinical diagnosis in an out-of-sample test set of patients achieved AUC of 85%, 81%, 80%, and 82%, respectively in Models I, II, III and IV. Significant positive predictors of PCOS diagnosis across models included hormone levels and obesity; negative predictors included gravidity and positive bHCG.

Conclusions

Machine learning algorithms were used to predict PCOS based on a large at-risk population. This approach may guide early detection of PCOS within EHR-interfaced populations to facilitate counseling and interventions that may reduce long-term health consequences. Our model illustrates the potential benefits of an artificial intelligence-enabled provider assistance tool that can be integrated into the EHR to reduce delays in diagnosis. However, model validation in other hospital-based populations is necessary.

Identifying risk factors for polycystic ovary syndrome in women with epilepsy: A comprehensive analysis of 248 patients

To assess the risk factors for polycystic ovary syndrome (PCOS) in women with epilepsy (WWE) and develop a practical approach for PCOS screening based on clinical characteristic, blood indicator, and anti-seizure medication (ASM) profiles. This cross-sectional study was conducted with 248 WWE who were consecutively enrolled from the Epilepsy Center of West China Hospital between April 2021 and March 2022. The epilepsy characteristics, blood indicators, and use of ASMs were compared between WWE with and without PCOS. Multivariate logistic regression was used to identify the factors independently associated with PCOS. The differential analysis showed that younger age at onset of epilepsy (<13 years), a history of birth hypoxia, obesity (BMI ≥25 kg/m² ), use of levetiracetam (LEV) (≥1 year), higher levels of cholesterol, luteinizing hormone (LH) and anti-Müllerian hormone (AMH), and lower levels of sex hormone-binding globulin were associated with PCOS (p < .05). Multivariate logistic regression identified that obesity (BMI ≥25 kg/m² ), use of LEV (≥1 year), and higher levels of AMH and LH were independently associated with PCOS in WWE (p < .05). Obesity (BMI ≥25 kg/m² ), LEV use (≥1 year), and elevated AMH and LH levels suggest an increased in the probability of occurrence of PCOS in WWE. The combination of these profiles provides a practical approach for screening PCOS in WWE.

Assessing the androgenic and metabolic heterogeneity in polycystic ovary syndrome using cluster analysis

INTRODUCTION

Some but not all women with polycystic ovary syndrome (PCOS) develop the metabolic syndrome (MS). The objective of this study was to determine if a subset of women with PCOS had higher androgen levels predisposing them to MS and whether routinely measured hormonal parameters impacted the metabolic syndrome score (siMS).

METHODS

We included data from a discovery (PCOS clinic data) and a replication cohort (Hull PCOS Biobank) and utilized eight routinely measured hormonal parameters in our clinics (free androgen index [FAI], sex hormone-binding globulin, dehydroepiandrosterone sulphate (DHEAS), androstenedione, luteinizing hormone [LH], follicular stimulating hormone, anti-Müllerian hormone and 17 hydroxyprogesterone [17-OHP]) to perform a K-means clustering (an unsupervised machine learning algorithm). We used NbClust Package in R to determine the best number of clusters. We estimated the siMS in each cluster and used regression analysis to evaluate the effect of hormonal parameters on SiMS.

RESULTS

The study consisted of 310 women with PCOS (discovery cohort: n = 199, replication cohort: n = 111). The cluster analysis identified two clusters in both the discovery and replication cohorts. The discovery cohort identified a larger cluster (n = 137) and a smaller cluster (n = 62), with 31% of the study participants. Similarly, the replication cohort identified a larger cluster (n = 74) and a smaller cluster (n = 37) with 33% of the study participants. The smaller cluster in the discovery cohort had significantly higher levels of LH (7.26 vs. 16.1 IU/L, p < .001), FAI (5.21 vs. 9.22, p < .001), androstenedione (3.93 vs. 7.56 nmol/L, p < .001) and 17-OHP (1.59 vs. 3.12 nmol/L, p < .001). These findings were replicated in the replication cohort. The mean (±SD) siMS score was higher in the smaller cluster, 3.1 (±1.1) versus 2.8 (±0.8); however, this was not statistically significant (p = .20). In the regression analysis, higher FAI (β = .05, p = .003) and androstenedione (β = .03, p = .02) were independently associated with a higher risk of SiMS score, while higher DHEAS levels were associated with a lower siMS score (β = -.07, p = .03) CONCLUSION: We identified a subset of women in our PCOS cohort with significantly higher LH, FAI, and androstenedione levels. We show that higher levels of androstenedione and FAI are associated with a higher siMS, while higher DHEAS levels were associated with lower siMS.

The Relationship Between Insulin Resistance and Obesity and Serum Anti-Mullerian Hormone Level in Chinese Women with Polycystic Ovary Syndrome: A Retrospective, Single-Center Cohort Study

Background

Anti-Mullerian hormone (AMH) is vital in the pathophysiological process of polycystic ovary syndrome (PCOS). The exact relationship between obesity and insulin resistance (IR) with AMH levels remains unclear.

Methods

A retrospective, single-center cohort study of 220 women with PCOS who underwent physical, endocrine, and metabolic assessments were performed. Patients were grouped by age, body mass indices (BMI), Homeostatic Model Assessment of Insulin Resistance (HOMA-IR), and different phenotypes. Pearson correlation analysis assessed the correlation between AMH and HOMA-IR, BMI, and other PCOS indicators, and multiple linear regression analysis was performed to determine factors influencing AMH.

Results

In 220 patients with PCOS, serum AMH levels decreased with age and were significantly higher in the IR group than in the non-IR group (P < 0.01). AMH increased significantly in anovulatory patients with hyperandrogenemia and/or polycystic ovary, with no significant difference between obese and non-obese individuals. AMH levels correlated positively with luteinizing hormone (LH), LH/follicular stimulating hormone (FSH), testosterone, fasting insulin (FINS), and HOMA-IR levels; negatively with age and BMI levels (P < 0.05) and weakly with fasting plasma glucose in the classical PCOS phenotype (r=0.148, P < 0.05). Regression analysis showed that age, testosterone, FINS, LH, LH/FSH, and BMI influenced AMH levels (P < 0.05).

Conclusion

Chinese women with PCOS-IR showed associations with greater AMH levels. AMH levels correlated positively with HOMA-IR levels and negatively with BMI. AMH combined with BMI and HOMA-IR levels may help determine PCOS severity.

Prediction model of persistent ovulatory dysfunction in Korean women with polycystic ovary syndrome

AIM

There is no validated tool to predict persistent ovulatory dysfunction after medication with oral contraceptives in women with polycystic ovary syndrome (PCOS), which is the most severe subtype of PCOS. We aimed to build a model to predict persistent ovulatory dysfunction after medication of oral contraceptives in women with PCOS.

METHODS

A total of 286 patients with PCOS were treated with and without oral contraceptives at a tertiary academic medical center. Data were obtained from the electronic medical record system between January 2016 and March 2019. A risk prediction model was developed using multivariable logistic regression. Model 1 was based on age and chief complaints and Model 2 further included predictors evaluated during a clinic visit. Model 3 additionally included laboratory findings.

RESULTS

Of the study population, ovulatory dysfunction was persistent in 117 patients (40.9%). Compared with the simple model (Models 1 and 2), the full prediction model (Model 3) had better Akaike's information criterion (286, 244 vs. 225) and the area under the curve (AUC) increased from 0.74 and 0.79 to 0.84. The full model included 7 covariates measured during the evaluation of PCOS, and two covariates were significant predictors of persistent ovulatory dysfunction in PCOS: age (OR 0.91; 95% CI 0.84-0.97), and anti-Müllerian hormone (OR 1.17; 95% CI 1.09-1.26). This model demonstrated good discrimination (AUC, 0.84) and calibration (Hosmer-Lemeshow goodness of fit test, p = 0.74).

CONCLUSIONS

This prediction model was shown to be a useful method for predicting persistent ovulatory dysfunction after oral contraceptive medication in patients with PCOS.

The Effect of Free Androgen Index on the Quality of Life of Women With Polycystic Ovary Syndrome: A Cross-Sectional Study

Purpose: Free androgen index (FAI) and anti-Mullerian hormone (AMH) are independently associated with polycystic ovary syndrome (PCOS). This study aimed to describe the relationship between these two markers and health-related quality of life (HR-QoL) in women with PCOS. Methods: This cross-sectional study consisted of 81 women in the Hull PCOS biobank, who fulfilled the Rotterdam consensus criteria for the diagnosis of PCOS. The primary outcome was to measure the various domains of the QoL in the modified polycystic ovary syndrome questionnaire (MPCOSQ). Results: Mean age of the study participants was 28 ± 6.0 years, mean body mass index (BMI) 33.5 ± 7.8 kg/m², mean FAI (6 ± 5.5), free testosterone (2.99 ± 0.75) and mean AMH (3.5 ± 0.8 units). In linear regression analysis, the FAI was associated with overall mean MPCOSQ score (Beta = 0.53, P-value = 0.0002), and with depression (Beta = 0.45, P-value = 0.01), hirsutism (Beta = 0.99, P-value = 0.0002) and menstrual irregularity (Beta = 0.31, P-value = 0.04). However, with adjustment for age and BMI, FAI was only associated with the hirsutism domain (Beta = 0.94, P-value = 0.001) of the MPCOSQ. FAI was also associated with the weight domain (Beta = 0.63 P-value = 0.005) of MPCOSQ. However, AMH was not associated with the overall mean MPCOSQ score or with any of its domains. Conclusion: FAI but not AMH was associated with QoL in women with PCOS, and this effect was mediated by BMI.

A Polygenic and Phenotypic Risk Prediction for Polycystic Ovary Syndrome Evaluated by Phenome-Wide Association Studies

CONTEXT

As many as 75% of patients with polycystic ovary syndrome (PCOS) are estimated to be unidentified in clinical practice.

OBJECTIVE

Utilizing polygenic risk prediction, we aim to identify the phenome-wide comorbidity patterns characteristic of PCOS to improve accurate diagnosis and preventive treatment.

DESIGN, PATIENTS, AND METHODS

Leveraging the electronic health records (EHRs) of 124 852 individuals, we developed a PCOS risk prediction algorithm by combining polygenic risk scores (PRS) with PCOS component phenotypes into a polygenic and phenotypic risk score (PPRS). We evaluated its predictive capability across different ancestries and perform a PRS-based phenome-wide association study (PheWAS) to assess the phenomic expression of the heightened risk of PCOS.

RESULTS

The integrated polygenic prediction improved the average performance (pseudo-R2) for PCOS detection by 0.228 (61.5-fold), 0.224 (58.8-fold), 0.211 (57.0-fold) over the null model across European, African, and multi-ancestry participants respectively. The subsequent PRS-powered PheWAS identified a high level of shared biology between PCOS and a range of metabolic and endocrine outcomes, especially with obesity and diabetes: "morbid obesity", "type 2 diabetes", "hypercholesterolemia", "disorders of lipid metabolism", "hypertension", and "sleep apnea" reaching phenome-wide significance.

CONCLUSIONS

Our study has expanded the methodological utility of PRS in patient stratification and risk prediction, especially in a multifactorial condition like PCOS, across different genetic origins. By utilizing the individual genome-phenome data available from the EHR, our approach also demonstrates that polygenic prediction by PRS can provide valuable opportunities to discover the pleiotropic phenomic network associated with PCOS pathogenesis.