Feature extraction and auto-regressive simulation of hepatic disordered states based on time-seriesed clinical laboratory data

To clarify the characteristics of clinical states of liver disease, the principal component analysis, the multi-dimensional Auto-Regression method, the cross-correlation and the cross-covariance method were applied to time-seriesed clinical laboratory data of patients with liver disease by using a time oriented clinical database. 1) Three components, a component of damage of liver parenchymal cells, that of damage of protein synthesis and that of damage of biliary cells, were extracted by the analysis of 88 patient's data. The discrimination of the liver disease entities by the component score was satisfactory to a certain degree. 2) The simulation of the time course of three kinds of serum constituents by the Auto-Regression model well described the actual time course of measurement of these serum constituents, but the prediction failed to follow the time course. 3) The values of cross-covariance among the three constituents adequately characterized the liver disease entity. The feature extraction of time-seriesed data seemed to contribute to clinical diagnosis and prognosis.