Sequential classifications of patient recovery trajectories after heart surgery

A new application of Bayesian technique to classify the recovery trajectories of patients after Aortic-coronary (AC) bypass surgery is presented. Patients are classified into three types by the physicians: rapid recovery, satisfactory recovery and marginal recovery. The objective of this study was to develop an analytical classification method which is more repeatable than the clinical classification and which can be used to predict a patient's recovery type. Eleven selected clinical measurements were observed before, immediately after, one and two days after surgery. The feature vector used for the trajectory classification is based on the normalized distances to five shock states defined in terms of these clinical measurements, at each epoch. The existing data base of 55 patients is divided into two groups: 32 for analysis and the remaining 23 for test. For analysis, the time dependent and multi-dimensional characterizing parameters such as the mean and covariance matrics for these three types of patients were estimated from the history of the 32 AC bypass patients. A two level non-linear recognition algorithm, using Chow's dependence tree and other procedures, was developed for this study. Very good agreement was found between the physician's and the computer classifications: 100% on analysis data and 95% on the test data. The results also indicate that earlier prediction is feasible, 89% agreement has been found between the predicted computer classification for the 23 test patients immediately after surgery and the physicians' judgement based on the clinical data up to two days after surgery.