A causal model for type 2 diabetes and its comparison with other modelling methods. (10th February 2021)

• Record Type: Journal Article
• Authors: Zhang, Sheng; An, Xiangdong; Wang, Hai
• Is Part Of: International journal of electronic healthcare. Volume 11:Number 2(2020)
• Page Start: 157
• Page End: 169
• Publication Date: 2021-02-10
• Subjects: risk assessment -- Bayesian networks -- diabetes; Medical informatics -- Periodicals; 362.10285
• Languages: English
• ISSNs: 1741-8453

Abstract

In this paper, we investigate probabilistic graphical models for risk modelling and assessment of type 2 diabetes. In particular, we study a new cause-effect model and focus on the impacts of life styles and socioeconomics to type 2 diabetes. The proposed model encodes cause-effect dependencies instead of correlations or conditional independencies among variables, which is different from previous work. Experiments on a large healthcare dataset show that the proposed causal modelling method significantly outperforms the baseline naive Bayesian network (BN) models and performs similarly to the conventional conditional independency modelling BNs and correlation modelling logistic regression models. The proposed model has the advantage of modelling cause-effect relationships over other models.