Using EHRs and Machine Learning for Heart Failure Survival Analysis

By Maryam Panahiazar, Vahid Taslimitehrani, Naveen L. Pereira, Jyotishman Pathak. 2015

134 citations recorded by the source corpus.

“Heart failure (HF) is a frequent health problem with high morbidity and mortality, increasing prevalence and escalating healthcare costs” [1]. By calculating a HF survival risk score based on patient-specific characteristics from Electronic Health Records (EHRs), we can identify high-risk patients and apply individualized treatment and healthy living choices to potentially reduce their mortality risk. The Seattle Heart Failure Model (SHFM) is one of the most popular models to calculate HF survival risk that uses multiple clinical variables to predict HF prognosis and also incorporates impact of HF therapy on patient outcomes. Although the SHFM has been validated across multiple cohorts [1–5], these studies were primarily done using clinical trials databases that do not reflect routine clinical care in the community. Further, the impact of contemporary therapeutic interventions, such as beta-blockers or defibrillators, was incorporated in SHFM by extrapolation from external trials. In this study, we assess the performance of SHFM using EHRs at Mayo Clinic, and sought to develop a risk prediction model using machine learning techiniques that applies routine clinical care data. Our results shows the models which were built using EHR data are more accurate (11% improvement in AUC) with the convenience of being more readily applicable in routine clinical care. Furthermore, we demonstrate that new predictive markers (such as co-morbidities) when incorporated into our models improve prognostic performance significantly (8% improvement in AUC).

Subfields: Artificial Intelligence, Cardiology and Cardiovascular Medicine, Health Information Management

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Cardiology and Cardiovascular Medicine

In Health Information Management

Citations in the corpus

Natural Language Processing for EHR-Based Computational Phenotyping (2018, background): “matoid arthritis [35], diabetes [36], heart failure [37], [38], neuropsychiatric disorders [39], etc.”
Mining Electronic Health Records: A Survey (2017, methodology): “Panahiazar et al. [Panahiazar et al. 2015] built a heart failure risk prediction model using several machine learning techniques where in they included multiple comorbidities which lead to improvement in prognostic predictive accuracy.”
Machine learning can predict survival of patients with heart failure from serum creatinine and ejection fraction alone (2020, methodology): “Computational intelligence, especially, shows its predictive power when applied to medical records [25, 26], or coupled with imaging [27–29].”