The use of optimal partitionings for multiparameter data analysis in clinical trials

Abstract

A predictive model is presented which allows estimating six-month-risk of cardiovascular disease in patients discharged from hospital after acute coronary syndrome. A database, that has been collected from 16 medical centers in seven Russian cities during seven years, was used to create the model. The database contains a wide range of clinical, biochemical and genetic characteristics. The approaches based on the use of optimal partitioning, such as the method of optimal valid partitioning (OVD) and the modified method of statistically weighted syndromes (MSWS), were used in order to create the predictive model. The accuracy of the model is quite well and is estimated by the value of AUC=0.72. This model shows the better predictive ability in comparison with the most widely used methods such as logistic regression, usage of decision trees, neural networks etс.