Machine Learning Approach to find the relation between Endometriosis, benign breast disease, cystitis and non-toxic goiter

Abstract

The exact mechanism of endometriosis is unknown. The recommendation system (RS) based on item similarities of machine learning has never been applied to the relationship between diseases. The study aim was to identify diseases associated with endometriosis by applying RS based on item similarities to insurance data in South Korea. Women aged 15 to 45 years extracted from the Korean Health Insurance Review & Assessment Service National Inpatient Sample (HIRA-NIS) 2009–2015. We used the RS model to extract diseases that were correlated with an endometriosis diagnosis. Among women aged 15 to 45 years, endometriosis was defined as a diagnostic code of N80.x and a concurrent treatment code. A control group was defined as women who did not have the N80.x code. Benign breast diseases, cystitis, and non-toxic goitre were extracted by the RS. A total of 1,730,562 women were selected as the control group, and 11,273 women were selected as the endometriosis group. In logistic regression analysis adjusted for age per 5 years, data year, and socioeconomic status, benign neoplasm of breast (odds ratio (OR): 2.58; 95% confidence interval (CI): 1.90–3.50), other cystitis (OR: 2.63; 95% CI: 1.56–4.44), and non-toxic single thyroid nodule (OR: 1.62; 95% CI: 1.14–2.32) were statistically significant. Endometriosis was associated with benign breast disease, cystitis, and non-toxic goitre. © 2019, The Author(s).