Improving risk predictions in heart failure using multivariate analysis techniques from high energy physics

Predicting mortality in heart failure (HF) is critically important to patients, their providers, healthcare systems, and third-party payers alike. The ability to accurately assess outcomes in patients with HF, however, has proven to be difficult. We have used multivariate machine learning techniques from high energy physics to create MARKER-HF, a risk score for heart HF patients, based on eight commonly available clinical variables. The algorithm was developed on de-identified patient data extracted from the electronic medical records at the UC San Diego medical center. We found that MARKER-HF outperforms existing HF risk scores built through more conventional algorithms. The performance of MARKER-HF was also validated on databases of patients from UC San Francisco and from a large multi-national European registry (BIOSTAT-CHF). The clinical ramifications of our study are numerous. One of the central challenges in HF management is the reliable identification of high-risk patients to allow for the timely deployment of additional resources. MARKER-HF could also assist in the evaluation of patients for potentially life-saving interventions, such as cardiac implantable electronic devices, mechanical circulatory support and heart transplantation