PhD

dissertationThe main motivation for this investigation was to determine if a reliable predictive model of cardiopulmonary arrest could be developed to serve as the basis of a computerized alerting tool. The alerting tool would be used prospectively in clinical practice, so the underlying predictive model could only use computerized data available at the time of patient care. Such an alerting tool would target appropriate patients for do-not-attempt resuscitation (DNAR) discussions. A retrospective matched-case control study was performed. A total of 6956 patients (1446 cases and 5510 controls) who were seen as inpatients at Latter Day Saints (LDS) Hospital during the years 1997-2000 were included in this study. Case patients included those who received a response to a code blue call and/or experienced asystole. Asystole is defined as absence of contractions of the heart. Data were collected on outcome, patient demographics and various clinical findings including clinically relevant vital signs and laboratory results and many nursing observations. All possible predictors were retrieved from LDS Hospital's electronic medical record. In bivariate tests, age, all vitals signs analyzed, all laboratory findings analyzed and most nursing observations analyzed were significantly associated with cardiopulmonary arrest. Three multivariate models were developed: one using logistic regression techniques and two using artificial neural network techniques. The logistic regression model offered the most discrimination with an area under the receiver operating characteristic curve of 0.863. Many factors are independently associated with experiencing cardiopulmonary arrest. Also, reliable predictive models of in-hospital cardiopulmonary arrest can be achieved using data exclusively from an existing electronic medical record. The predictive model produced by this work is part of a larger plan to improve DNAR ordering at LDS Hospital. Computerized alerts will be generated for high-risk patients, targeting them for DNAR discussion. The alert application has great potential to increase appropriate cardiopulmonary resuscitation (CPR), improve health care outcomes, decrease costs and improve patient and family satisfaction