Fixed-point dynamical modelling of causal/correlation patterns in large clinical databases.

Abstract

In our study, we determine the risk of getting lung cancer that cause by smoking behaviors of populations by mining a clinical database for causal and correlational patterns. In this study we first simulate a large artificial clinical database using Monte Carlo methods. Then, we develop a dynamical modeling framework to describe the clinical data at the population level with the transition between them as well as the corresponding risk of getting lung cancer. This time dependent model involve a lot of unknown parameters, we intend to estimate all these parameters by using recursive Bayesian analysis. However, after some calculation, this method very tedious to implement and required a lot of computation and time. Then, we propose to cast Hidden Markov Model to estimate the unknown parameters by using Baum-Welch Algorithm.By using the data that we extract from the artificial database, we estimated the parameters by running the Baum-Welch Algorithm for several times. Then, we analyze the result by plotting the graph and histogram of the parameters. Finally, we will get preliminaries but encouraging results.