According to new research, deep learning can be used to better predict pediatric hospitalization. And by forecasting certain patients ‘symptoms, the authors of the study stated, health systems can also forecast resource utilization. Deep learning is used in medicine in a variety of contexts of health information engineering, including genetic research and biomedical image analysis, wrote lead author En-Ju D. Lin, Ph.D., Research Institute at the National Children’s Hospital in Columbus, Ohio, and colleagues. A new area of research is the application of deep learning to patient-level hazard forecasting more reliably. The team turned to Skip-Gram, a three-layer neural network which “learns” the complicated relationships between different clinical codes, analyzing data from a single responsible service agency from more than 112,000 pediatric patients. Patients had an average age of eight years. Although data were used in 2014-2015 to build the deep learning system of the researchers, data from 2015-2016 was used to validate it. Ultimately, the authors found that for forecasting next year’s hospitalization, a template utilizing demographic and consumption attributes alone resulted in a region under the ROC curve (AUC) of 73.1 million. Nevertheless, by applying deep learning to this formula, the AUC rose to 75.1%. “Small but meaningful enhancements” were also observed by applying deep learning to demographic and application functionality for the accuracy of the template and negative predictive value. While deep learning computational methods are steadily increasing, the application of these models to hazard stratification in healthcare delivery systems has been slow to gain traction, the authors concluded. Thankfully, several future-oriented accountable care systems continue to support and conduct research which illustrates the’ value-add’ of deep learning approaches that can improve predictive modeling capacities and thus ensure better health and financial results.