Clinical data preprocessing and case studies of POMDP for TCM treatment knowledge discovery

Abstract

Partially Observable Markov Decision Processes (POMDP) has been applied to induce sequential treatment scheme from Traditional Chinese Medicinal (TCM) clinical data. The data required by POMDP should be of rich structure and with heterogeneous variables. But sometimes there is large number of missing values in the real-world TCM clinical data set. This makes it difficult for data preprocessing. This paper designs a data preprocessing framework of TCM clinical data for POMDP applications. It significantly facilitates the process of sequential treatment scheme discovery through POMDP when applying the framework on TCM clinical cases of coronary heart disease and lung cancer. We also systematically analyze the sequential treatment scheme.