Can machine learning methods accurately estimate NYHA class from EHR data for heart failure patients when it is not explicitly documented?
Machine learning models can accurately estimate NYHA class from EHR data, potentially improving clinical documentation and research for heart failure patients.
New York Heart Association (NYHA) Class is an important measure of functional status for heart failure (HF) patients used for clinical documentation, treatment decisions, as well as for eligibility criteria and outcome measures in clinical studies. Electronic health records (EHRs) possess the potential to more efficiently access NYHA class information to measure effectiveness of treatments such as cardiac resynchronization therapy (CRT) and to monitor disease progression. However, our previous study has shown that the percentage of encounters of HF patients with a CRT implant that have explicit NYHA class documentation is low. In the present study, we examine if NYHA class can be estimated from EHR data for HF patients when it is not explicitly documented. Using machine learning methods, we constructed a model that estimated NYHA class for HF patient encounters with high quality, demonstrated by an AUC of 0.87.
Ma et al. (Sat,) studied this question.