Ting Huang1, Siyang Liu2, Jian Huang2, Jiarong Li1, Guixiong Liu2, Weiru Zhang3, Xuan Wang3. 1. Department of Rheumatology, Xiangya Hospital, Central South University, Changsha, Hunan, China. 2. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou, China. 3. Department of General Medicine, Xiangya Hospital, Central South University, Changsha, Hunan, China.
Abstract
OBJECTIVES: The prevalence of hypothyroidism in systemic lupus erythematosus (SLE) is significantly higher than that in the common public. While SLE itself can affect multiple organs, abnormal thyroid function may exacerbate organ damage in patients with SLE. We aimed to predict abnormal thyroid function and to examine the associated factors with multiple machine learning approaches. METHODS: In a cross-sectional study, 255 patients diagnosed with SLE at the rheumatology department in Xiangya Hospital between June 2012 and December 2016 were investigated. Feature engineering was used for filtering out principle clinical parameters, and five different machine learning methods were used to build prediction models for SLE with hypothyroidism. RESULTS: Feature engineering selected 11 variables with which to build machine learning models. Among them, random forest modelling obtained the best prediction performance, with an accuracy rate of 88.37 and an area under the receiver operating characteristic curve of 0.772. The weights of anti-SSB antibody and anti-dsDNA antibody were 1.421 and 1.011, respectively, indicating a strong association with hypothyroidism in SLE. CONCLUSIONS: Random Forest model performed best and is recommended for selecting vital indices and assessing clinical complications of SLE, it indicated that anti-SSB and anti-dsDNA antibodies may play principal roles in the development of hypothyroidism in SLE patients. It's feasible to build an accurate machine learning model for early diagnosis or risk factors assessment in SLE using clinical parameters, which would provide a reference for the research work of SLE in China.
OBJECTIVES: The prevalence of hypothyroidism in systemic lupus erythematosus (SLE) is significantly higher than that in the common public. While SLE itself can affect multiple organs, abnormal thyroid function may exacerbate organ damage in patients with SLE. We aimed to predict abnormal thyroid function and to examine the associated factors with multiple machine learning approaches. METHODS: In a cross-sectional study, 255 patients diagnosed with SLE at the rheumatology department in Xiangya Hospital between June 2012 and December 2016 were investigated. Feature engineering was used for filtering out principle clinical parameters, and five different machine learning methods were used to build prediction models for SLE with hypothyroidism. RESULTS: Feature engineering selected 11 variables with which to build machine learning models. Among them, random forest modelling obtained the best prediction performance, with an accuracy rate of 88.37 and an area under the receiver operating characteristic curve of 0.772. The weights of anti-SSB antibody and anti-dsDNA antibody were 1.421 and 1.011, respectively, indicating a strong association with hypothyroidism in SLE. CONCLUSIONS: Random Forest model performed best and is recommended for selecting vital indices and assessing clinical complications of SLE, it indicated that anti-SSB and anti-dsDNA antibodies may play principal roles in the development of hypothyroidism in SLE patients. It's feasible to build an accurate machine learning model for early diagnosis or risk factors assessment in SLE using clinical parameters, which would provide a reference for the research work of SLE in China.
Entities:
Keywords:
Feature engineering; SLE; machine learning; random forest modelling