A temporal Bayesian network for modeling the temporal relation among multiple chronic conditions

In the sector of bio-informatics Bayesian Network has gained increasing interest of the researchers specially for modelling the hierarchical disease data. As the graphical model encodes the probabilistic relationships among the variables of interest, it can be used to find the causal relationships. An in depth understanding about the problem domain can be achieved from this relationship. One of the big advantage of using Bayesian Network is that, once the model is learned & programmed to encode the information, it can readily handle situations with missing entries. Before performing the analysis of the chronic conditions, it is important to learn the structure of the network for domain of analysis. Learning the structure of the Bayesian Network is a NP-hard problem. The Chow-Liu tree model has been applied to get an initial structure which was modified to work with the sequential time series data in a hierarchical form. To learn the complete directed structure the modeled tree is passed through a scoring based method for retrieving the optimized network structure with maximum likelihood. The attained network is called the Temporal Bayesian Networks (TBN). The proposed Network (TBN) is able to provide a compact representation of complex stochastic processes which captures the temporal interactions among chronic conditions. TBN can be used to reveal hidden patterns of co-occurring disorders (COD) and unknown correlations of risk factors in big and pervasive datasets of Electronic Health Record (EHR). This learned structure was then compared with an expert given structure to check how well it was able to capture the causalities in the search space. The progression of chronic conditions is investigated and correlated to patient related variables using both the proposed methodology and expert given structure. A large longitudinal dataset of comorbid patients monitored over 10 years is used to validate the proposed approach.