CLINICAL DECISION SUPPORT SYSTEM FOR OPTIMAL VAD WEANING

Abstract

Ventricular Assist Devices (VADs) have demonstrated their therapeutic role in cardiac rehabilitation. However, due to the complexities of caring for these patients and the relatively limited clinical experience, identifying candidates for weaning remains challenging. This study proposes the use of a Clinical Decision Support System (CDSS) to both aid in the identification of VAD weaning candidates, and as a tool for predicting patient outcome. Based upon the UPMC VAD weaning experience, three CDSS models were developed: an expert model, a data model, and an expert/data hybrid model. The decision structures of the expert model were elicited from an 11 member, multi-disciplinary panel through a series of structured interviews and polls. Pattern recognition through Artificial Neural Networks and Natural Language Processing was used to analyze patient data and acquire the decision structures for the data model; all patients receiving a Thoratec VAD which were considered for weaning between 1996 and 2004 (n=19), regardless of outcome, were included in this study. Decision structures were modeled using Bayesian Belief Networks and their predictive values were assessed. A user interface, based on a pocket-PC, was developed to anticipate the translation of this system to clinical practice. The hybrid model, consisting of a 21-parameter health screening and a 3-tier evaluation of cardiac recovery, was the best predictor of outcome, predicting 90% true weans, 100% true transplants, 0% false weans and 10% false transplants. By objectively combing knowledge from experts and data, this study illustrates how a CDSS can facilitate the decision making processes for identifying VAD weaning candidates and promote responsible and widespread use of VADs for cardiac rehabilitation.