Predicting outcome for hospitalized cardiac patients using a combined neural network and rough set approach

Describes a hospitalization prediction system based on neural network technology. The paper focuses on predicting the following output variables identified as crucial for the purposes of the project: the length of stay in hospital, the length of stay in the intensive care unit, and the outcome of hospitalization defined as a transferred discharged or deceased patient. The general approach adopted for solving the problem consists of first applying inductive learning based on the techniques of rough sets to generate a reduced set of input data and a small set of rules specific to the output variable. The results serve to define and structure the architecture of the neural system in terms of the number of neural networks and their input variables, as well as to dynamically select the networks that best fit the type of information describing the current patient. A database of over 1000 cardiac patients, admitted to the Ottawa General Hospital over a period of 4 years was used