Leveraging Artificial Intelligence to Predict and Improve Health Outcomes, Maximize Quality Improvement, and Reduce Costs

HealthEC, LLC, United States

Keywords: AI, Machine Learning, Health Outcomes

We consider the problems of predicting patients’ unplanned hospital and skilled nursing facility (SNF) admission and adverse events within 30 days. We formulate each of these problems as a classification problem. To solve each of these problems, we develop and implement AI-based models: i) a single-layer NN, ii) a multilayer feedforward DNN, and iii) a recurrent NN. We verify the utility of the AI-based models and compare their performance to that of a set of baseline machine learning models: logistic regression, random forest, and support vector machines -- for each of the three cases (unplanned hospital and SNF admissions, and 30-day adverse events) for both binary and multilabel classification models. Our performance metrics extend beyond the model accuracy and include precision, recall, F1-score, sensitivity, specificity, kappa statistic, area under the ROC curve, and Akaike information criterion. Our results confirm better performance of the AI-based models as compared to the baseline models. HealthEC’s solution is a testament to the fact that clinicians and medical professionals are essential to the intelligent use of AI in medical practice. Our AI-based modeling is designed to augment and complement clinicians’ capabilities rather than replace them, empowering physicians by supporting rather than replacing their decision-making practices.