Deep Learning for Improved Risk Prediction in Surgical Outcomes

Abstract

The Norwood surgical procedure restores functional systemic circulation in neonatal patients with single ventricle congenital heart defects, but this complex procedure carries a high mortality rate. In this study we address the need to provide an accurate patient specific risk prediction for one-year postoperative mortality or cardiac transplantation and prolonged length of hospital stay with the purpose of assisting clinicians and patients' families in the preoperative decision making process. Currently available risk prediction models either do not provide patient specific risk factors or only predict in-hospital mortality rates. We apply machine learning models to predict and calculate individual patient risk for mortality and prolonged length of stay using the Pediatric Heart Network Single Ventricle Reconstruction trial dataset. We applied a Markov Chain Monte-Carlo simulation method to impute missing data and then fed the selected variables to multiple machine learning models. The individual risk of mortality or cardiac transplantation calculation produced by our deep neural network model demonstrated 89 ± 4% accuracy and 0.95 ± 0.02 area under the receiver operating characteristic curve (AUROC). The C-statistics results for prediction of prolonged length of stay were 85 ± 3% accuracy and AUROC 0.94 ± 0.04. These predictive models and calculator may help to inform clinical and organizational decision making.

Figure 1

Demographics of patients in the PHN SVR dataset. The patient’s sex, race and age on the day of their Norwood surgical procedure is shown. Gestational age is the gestational age at birth reported in weeks, indicating presence and degree of prematurity with full term ≥ 37 weeks. % below federal poverty level is an indication of socioeconomic status. The turquoise color represents patients who survived to one year, red represents those who died.

Figure 2

Histogram distribution and box plot of of the LOS data for patients who survived the Norwood procedure.

Figure 3

ROC curves afor each machine learning model after testing using all of the 50 MCMC MI datasets. (left) Risk of mortality prediction (right) Prolonged LOS prediction.

Figure 4

The calculator display of the mortality or cardiac transplantation risk and cluster for an example patient. Section labeled 1 contains two columns that allows the user to input a new patient’s data values such as age, sex, race, anatomic diagnosis etc. These values are used by the DNN model to calculate the patient specific risk score and provide a prediction for one-year transplant free survival (section 2). Section 3, a stacked bar graph, is used to depict cluster segments for the registry’s population risk scores and allow clinicians to evaluate a patient’s risk as low, medium or high.