Georgia Tech have Inventors developed an effective predictive model for outcome prediction using readily available objective data from EMR records as input. A proof of concept tool has been developed to predict regression. Variables included demographic, financial, pre-arrival/triage, ER assessment, ER treatment, and time-interval/process data. The models achieve excellent area under the curve for predicting pediatric asthma admission. In particular, the lasso-regularized model outperforms existing models relying on subjective disease-specific symptoms. The models can also be used over time to predict outcome as new data is made available. A web application that ER physicians could access at the point of care for predicting which of their patients will be admitted or discharge has been developed.
- Real-time measure of admission likelihood
- Outperforms other predictive models that rely on subjective disease data
- Utilizes objective, readily available data in EMR
- Optimal variable selection based on statistical measures
- Fast and easy to use
- Clinical practice and workflow in the ER
Outcome prediction in the emergency room (ER) department has the potential to improve resource allocation and efficiency in hospitals, leading to a decrease in healthcare costs. The rise of big data in medicine and other industries has fueled interest in the use of machine learning for effective prediction in healthcare. However, existing predictive models rely on clinical symptom checklists which can be subjective, time consuming, and resource-intensive. The large amount of readily available, objective variables available in today’s electronic medical record (EMR) systems can be used to build more effective predictive patient outcome models.