Predicting the interaction between treatment processes and disease progression by using hidden Markov model

This study aims to leverage the hidden Markov model (HMM) to simulate the interaction between treatment processes and disease progression for clinical decision support. In this model, clinical orders in each treatment step are considered as hidden states while changes of patient states are observation states. First, Clinical Order Model (COM) and Patient State Model (PSM) are introduced in the HMM to reduce the complexity of clinical orders and clinical notes in real world. Then, the Baum-Welch algorithm with consideration of multiple observation sequences is used to train the parameters of the model in this scenario. Finally, an analysis utilizing 988 treatment processes of 229 patients examines the superiority of the proposed model in medical scenes. The experimental results indicated that the addressed complexity of interaction between 174,601 clinical orders and 7,602 clinical notes proves the feasibility of the model. To sum up, the HMM is suitable to describe the interaction between the treatment process and the patient state; thusly, we can utilize the proposed HMM model to predict critical treatment steps according to patients’ clinical notes