Neural net-bootstrap hybrid methods for prediction of complications in patients implanted with artificial heart valves.

A novel hybrid methodology for prediction of valve related complications in patients with implanted artificial heart valves is discussed. Artificial neural networks provided a mechanism for prediction of postoperative valve-related deaths based on preoperative patient information and valve parameters. Then bootstrap methodology was applied for estimating prediction errors and maximizing prediction accuracy. Data from a clinical trial with 10 years of follow-up on 789 patients implanted with Carpentier-Edwards Pericardial Bioprosthesis were used. A random subset of the data was reserved for validation of the final outcome. The remaining patients' records were repeatedly divided into two groups, using resampling strategy provided by the bootstrap methodology. One of the groups was used for training the neural net and the other one for testing the trained network and determining error rates. Patient information, such as sex, age, NYHA class and anticoagulation therapy, as well as valve parameters, such as size and the date of implant were used as the network inputs. Calculated error rates were then used for assessing the distribution of the error, further optimization of the neural network, and constructing confidence intervals for the error rates. Thus, reliable statistical estimation was obtained on the prediction accuracy. Additionally this new hybrid methodology allowed us to optimize the neural network even further, raising the accuracy of prediction to 78%.