Methodology

How we develop our models:Technical Details

Our models use machine learning to predict six-month and one-year mortality risk and can identify with confidence the 1-1.8 million patients each year who need end of life care. The random forest model derives from nearly 100,000 hospitalisations, is validated on almost 20,000 hospitalisations without focusing on a specific disease, and has an AUC of 0.93 for predicting death within six months. The algorithm uses commonly obtained EMR data (e.g., last set of vital signs, complete blood count, basic and complete metabolic panel, demographic information, ICD codes, etc.), physiological readings (e.g., blood pressure, pulse heart rate, etc.) and basic demographic information (e.g., patient age and length of hospital stay). The algorithm can deal with missing data and allows for customisation.

 Variable importance in the model based on the GINI Score

Figure 1: Variable importance in the model based on the GINI Score

Development and Validation of Machine Learning Models for Prediction of 1-Year Mortality Utilizing Electronic Medical Record Data Available at the End of Hospitalization in Multicondition Patients: a Proof-of-Concept Study.Sahni N, Simon G, Arora R.J Gen Intern Med. 2018 Jun;33(6):921-928. doi: 10.1007/s11606-018-4316-y.