Machine Learning –Based Development of a Prediction Model of Complications in Patients on Left Ventricular Assist Device Support: the LVAD-LVAD (Left Ventricular Assist Devices provide Loads of Valuable Additional Data) Study

Collectively, the University Medical Center Utrecht, Utrecht University, and Abbott will develop a novel prediction model of LVAD (mechanical heart pump) clinical outcomes based on unique state-of-the-art technology, clinical input, and data mining.

LVAD therapy has greatly improved the survival of patients with end-stage heart failure. However, major adverse events are common (approximately 20% of the patients) and often occur suddenly and unpredictably. It is technically possible to retrieve a wealth of data from the LVADs itself, as well as from patients on LVAD support, but these data are not routinely retrieved because appropriate analysis models are lacking and thus the clinical value of these data are unknown. A prediction model is required that can combine input from multiple sources, handle repeated measurements and missing data. 

By exploiting existing big data from the device itself and the patient, we will develop a prediction model of clinical outcomes. We will retrieve and combine data in a custom-made database, apply data analysis and machine learning, and finally replicate in alternative sample datasets.  

Within the short term objective (this project) we will develop a novel prediction model based on a unique approach using state-of-the-art technology, clinical input, and data mining (enabling technologies); thus delivering an innovative product for health care. On the longer term, the use and comprehension of “big data” from LVADs and patients on LVAD support will lead to better outcomes for LVAD patients, by optimizing settings and early detection of complications. In addition, this knowledge may advance the development of the next generation LVADs.