Prediction of survival and complications after percutaneous endoscopic gastrostomy in an individual by using clinical factors with an artificial neural network system

Abstract

The demand for percutaneous endoscopic gastrostomy (PEG) has increased because it is safe and a technically easy method, but it has risks of severe complications including death and a high mortality rate within 30 days. At present, we cannot predict survival or the incidence of complications before tube placement in an individual. Earlier studies have used traditional statistical analysis by assuming a linear relationship between clinical features, but most phenomena in the clinical situation are not linearly related. We predicted the survival and complications before PEG placement in an individual by using artificial neural network (ANN) system, which can assess the nonlinear relationship. We studied 100 patients who underwent PEG at the Kitasato Medical Institute Hospital from 1997 to 2005. Clinical data and laboratory data were used as input data. Complications related to PEG placement and survival dates were historically and prospectively measured. From the clinical data and laboratory data, we examined the prediction of outcome in individual patients using multiple logistic regression analysis and an ANN. The correct answer rate of survival by multiple logistic regression analysis was 67.9%. In contrast, using the ANN, we correctly predicted the survival date and aspiration pneumonia in 75 and 89% of patients, respectively. There was a nonlinear relationship among input factors and survival and complications. We correctly predicted the outcome and complications of individual patients with PEG with a high correct answer rate. Our data show the potential of an ANN as a powerful tool in daily clinical use to individualize treatment ('tailor-made medicine') for PEG and reduce costs.