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.

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.

Multimodal Recruitment to Study Ovulation and Menstruation Health: Internet-Based Survey Pilot Study.

BACKGROUND: Multimodal recruitment strategies are a novel way to increase diversity in research populations. However, these methods have not been previously applied to understanding the prevalence of menstrual disorders such as polycystic ovary syndrome. OBJECTIVE: The purpose of this study was to test the feasibility of recruiting a diverse cohort to complete a web-based survey on ovulation and menstruation health. METHODS: We conducted the Ovulation and Menstruation Health Pilot Study using a REDCap web-based survey platform. We recruited 200 women from a clinical population, a community fair, and the internet. RESULTS: We recruited 438 women over 29 weeks between September 2017 and March 2018. After consent and eligibility determination, 345 enrolled, 278 started (clinic: n=43; community fair: n=61; internet: n=174), and 247 completed (clinic: n=28; community fair: n=60; internet: n=159) the survey. Among all participants, the median age was 25.0 (SD 6.0) years, mean BMI was 26.1 kg/m2 (SD 6.6), 79.7% (216/271) had a college degree or higher, and 14.6% (37/254) reported a physician diagnosis of polycystic ovary syndrome. Race and ethnicity distributions were 64.7% (176/272) White, 11.8% (32/272) Black/African American, 7.7% (21/272) Latina/Hispanic, and 5.9% (16/272) Asian individuals; 9.9% (27/272) reported more than one race or ethnicity. The highest enrollment of Black/African American individuals was in clinic (17/42, 40.5%) compared to 1.6% (1/61) in the community fair and 8.3% (14/169) using the internet. Survey completion rates were highest among those who were recruited from the internet (159/174, 91.4%) and community fairs (60/61, 98.4%) compared to those recruited in clinic (28/43, 65.1%). CONCLUSIONS: Multimodal recruitment achieved target recruitment in a short time period and established a racially diverse cohort to study ovulation and menstruation health. There were greater enrollment and completion rates among those recruited via the internet and community fair.

Cognitive testing of a survey instrument for self-assessed menstrual cycle characteristics and androgen excess.

BACKGROUND: In large population-based studies, there is a lack of existing survey instruments designed to ascertain menstrual cycle characteristics and androgen excess status including hirsutism, alopecia, and acne. Our objective was to cognitively test a survey instrument for self-assessed menstrual cycle characteristics androgen excess. METHODS: Questions to assess menstrual characteristics and health were designed using existing surveys and clinical experience. Pictorial self-assessment tools for androgen excess were also developed with an experienced medical illustrator to include the modified Ferrimen-Galway, acne and androgenic alopecia. These were combined into an online survey instrument using REDCap. Of the 219 questions, 120 were selected for cognitive testing to assess question comprehension in a population representative of the future study population. RESULTS: Cognitive testing identified questions and concepts not easily comprehended, recalled, or had problematic response choices. Comprehension examples included simplifying the definition for polycystic ovary syndrome and revising questions on historic menstrual regularity and bleeding duration. Recall and answer formation examples include issues with recalling waist size, beverage consumption, and interpretation of questions using symbols (> or <). The survey was revised based on feedback and subsequently used in the Ovulation and Menstruation (OM) Health Pilot study. CONCLUSION: We present a cognitively tested, novel survey instrument to assess menstrual cycle characteristics and androgen excess.

Cardiovascular risk factors among women with self-reported infertility.

BACKGROUND: Amongst women with certain types of ovulatory disorder infertility, the studies are conflicting whether there is an increased risk of long-term cardiovascular disease risk. This paper evaluates the associations of several CVD risk factors among Framingham women with self-reported infertility. METHODS: Women who completed the Framingham Heart Study Third Generation and Omni Cohort 2 Exam 2 (2008-2011), and reported on past history of infertility and current cardiovascular disease status were included in this cross-sectional study. Directly measured CVD risk factors were: resting blood pressure, fasting lipid levels, fasting blood glucose, waist circumference, and body mass index (BMI). Multivariable models adjusted for age, smoking, physical activity, and cohort. Generalized estimating equations adjusted for family correlations. We performed sensitivity analyses to determine whether the association between infertility and CVD risk factors is modified by menopausal status and menstrual cycle length. RESULTS: Comparing women who self-reported infertility to those who did not, there was an average increase in BMI (ß = 1.03 kg/m2, 95% CI: 0.18, 1.89), waist circumference (ß = 3.08 in., 95% CI: 1.06, 5.09), triglycerides (ß = 4.47 mg/dl, 95% CI:-1.54, 10.49), and a decrease in HDL cholesterol (ß = -1.60 mg/dl, 95% CI: -3.76, 0.56). We estimated that infertile premenopausal women have an increased odds of obesity (BMI ≥ 30 kg/m2) (OR = 1.56, 95% CI: 1.11, 4.49) and diabetes (OR = 1.96, 95% CI: 0.86, 4.49). CONCLUSIONS: BMI and waist circumference were the most strongly correlated CVD risk factors amongst women reporting a history of infertility.