Use of genetic programming, logistic regression, and artificial neural nets to predict readmission after coronary artery bypass surgery

Abstract

As many as 14 % of patients undergoing coronary artery bypass surgery are readmitted within 30 days. Readmission is usually the result of morbidity and may lead to death. The purpose of this study is to develop and compare statistical and genetic programming models to predict readmission. Patients were divided into separate Construction and Validation populations. Using 88 variables, logistic regression, genetic programs, and artificial neural nets were used to develop predictive models. Models were first constructed and tested on the Construction populations, then validated on the Validation population. Areas under the receiver operator characteristic curves (AU ROC) were used to compare the models. Two hundred and two patients (7.6 %) in the 2,644 patient Construction group and 216 (8.0 %) of the 2,711 patient Validation group were re-admitted within 30 days of CABG surgery. Logistic regression predicted readmission with AU ROC = .675 ± .021 in the Construction group. Genetic programs significantly improved the accuracy, AU ROC = .767 ± .001, p < .001). Artificial neural nets were less accurate with AU ROC = 0.597 ± .001 in the Construction group. Predictive accuracy of all three techniques fell in the Validation group. However, the accuracy of genetic programming (AU ROC = .654 ± .001) was still trivially but statistically non-significantly better than that of the logistic regression (AU ROC = .644 ± .020, p = .61). Genetic programming and logistic regression provide alternative methods to predict readmission that are similarly accurate.

Genetic Programming for the Downscaling of Extreme Rainfall Events on the East Coast of Peninsular Malaysia

“…Biesheuvel et al [30] compared the performance of GP with logistic regression in diagnosing pulmonary embolism and reported that although the interpretation of a GP model is less intuitive, it is a promising technique for the development of prediction rules for diagnostic and prognostic purposes. Engoren et al [31] also came to a similar conclusion and reported that GP can improve the prediction accuracy of logistic regression. The application of GP-based logistic regression has increased in recent years in different fields of science and technology [61][62][63][64].…”

“…Genetic programming (GP) is a kind of non-parametric regression, which can relate the predictors and predictands and provide a predictive model identical to the analytical optimal solution when interrelationships between variables are poorly understood [23][24][25][26][27][28][29]. Genetic-based statistical logistic regression offers a clear advantage over the standard statistical logistic regression method [30,31]. It generates a set of logical expressions describing the structure of the data through iterative subsumption and probabilistically picks the most appropriate set to allow the system to predict in non-deterministic situations, while the achievement of this is not possible with current statistical logistic regression.…”

Years of Life Lost After Complications of Coronary Artery Bypass Operations

“…A problem with using just the operative mortality rate as the quality benchmark is that it does not account for deaths that occur more than 30 days after discharge or provide insight into the cause of death. Complications will occur in half of patients undergoing cardiac operations, and although most-particularly the most common complications-rarely lead to death during the same hospitalization, they may be associated with increased late death [11]. Consequently, such complications, especially if common (eg, postoperative atrial fibrillation), may have a large effect on the total years of life lost.…”

An Evolutionary Computation Approach for Optimizing Multilevel Data to Predict Patient Outcomes

“…Further, the optimal level of specificity may change for different outcomes or even the same outcome in different populations. For example, there is a substantial body of work involving the prediction of readmission for patients who undergo coronary artery bypass graft (CABG) surgery [13–16]. In much of this work, there are risk factors for comorbidities, medications, and complications that could be defined more generally or more specifically, and minimal rationale was provided about how these modeling decisions affected model performance.…”