Applying Machine Learning to Construct a Model of Risk of Depression in Patients Following Cardiac Surgery with the Use of the SF-12 Survey

Abstract

Background: Depression is a common problem in patients with cardiovascular diseases. Identifying a risk factor model of depression has been postulated. A model of the risk of depression would provide a better understanding of this disorder in this population. We sought to construct a model of the risk factors of depression in patients following cardiac surgery, with the use of machine learning. Methods and Measures: Two hundred and seventeen patients (65.4% men; mean age 65.14 years) were asked to complete the short form health survey-12 (SF-12v.2), three months after hospital discharge. Those at risk of depression were identified based on the SF-12 mental component summary (MCS). Centroid class principal component analysis (CCPCA) and the classification and regression tree (CART) were used to design a model. Results: A risk of depression was identified in 29.03% of patients. The following variables explained 82.53% of the variance in depression risk: vitality, limitation of activities due to emotional problems (role-emotional, RE), New York Heart Association (NYHA) class, and heart failure. Additionally, CART revealed that decreased vitality increased the risk of depression to 45.44% and an RE score > 68.75 increased it to 63.11%. In the group with an RE score < 68.75, the NYHA class increased the risk to 41.85%, and heart failure further increased it to 44.75%. Conclusion: Assessing fatigue and vitality can help health professionals with identifying patients at risk of depression. In addition, assessing functional status and dimensions of fatigue, as well as the impact of emotional state on daily functioning, can help determine effective intervention options.