Funding Acknowledgements Type of funding sources: Public hospital(s). Main funding source(s): CHU de Rennes Aims Patients presenting heart failure (HF) symptoms with preserved left ventricular ejection fraction (HFpEF) are not a homogenous population. Different phenotypes can differ in prognosis and optimal management strategies. We sought to identify phenotypes of HFpEF by using the medical information database from a large University Hospital Center using machine learning. Methods and results We explored the use of clinical variables from electronic health records (EHR) in addition to echocardiography to identify different phenotypes of patients with heart failure and preserved ejection fraction. The proposed methodology identifies 4 phenotypic clusters based on both clinical and echocardiographic characteristics which have differing prognoses (death and cardio-vascular hospitalization). Conclusion This work demonstrated that AI derived phenotypes could be used as a tool for physicians to assess risk and to target therapies that may improve outcomes.

Patients suffering from heart failure (HF) symptoms and a preserved left ventricular ejection fraction (HFpEF) present very different clinical phenotypes that could influence their survival. This study aims to identify phenotypes of this type of HF by using the medical information database from our hospital center. Methods and results We develop a POC of machine learning model to predict the phenotype of patients suffering from HFpEF from both clinical and echocardiography data from data warehouse of our hospital center. We performed a retrospective observational study. We used the inpatient clinical data warehouse at our hospital center to identify patients suffering from HFpEF and having echocardiography between January 2018 and December 2019. A list of relevant clinical and echocardiographic variables was established by cardiologists. For clinical variables, we extracted information from the data warehouse using either structured data (ICD-10 codes, lab results, ...) or regular expressions for narratives. The echocardiographic parameters consisted only in structured data. We performed a two-step cluster analysis to identify common characteristics among patients. The final sample eligible included 2500 patients with echocardiography and diagnosed for HFpEF. The variables included in the two-step cluster analysis were both clinical (13 variables) and echographic (17 variables). Four clusters were identified from the clustering algorithm. The groups were relatively well balanced with respectively: 753, 744, 519 and 545 patients. After performing survival analysis, we observed that clusters have significative different survival curves in particular for death. The performances of these two models are satisfying with an AUC upper than 0.97 and an accuracy upper than 0.92) (Figure 1). Conclusion In the future, this work could be deployed as a tool for the physician to assess risks and contribute to support better care for patients. The phenotyping of HFpEF could improve the characterization of patients, the definition of the most appropriate treatments, and the care pathways. Funding Acknowledgement Type of funding sources: Foundation. Main funding source(s): Fédération française de cardiologie