Multicenter Prospective Blind External Validation of a Machine Learning Model for Predicting Heart Failure Decompensation: A 3-Hospital Validation Study
Authors: Esperança Lladó Nicola Goodfellow Karina Anahi Ojanguren Marco Manso Bárbara Guerra Stanke Ladislav Vohralík Tomás Esteban Fabello Tatiana Silva Michael Scott Glenda Fleming Manuel Graña
Date: 04.06.2024
Abstract
This paper presents the results of a multicenter prospective blind external validation study aimed at validating a machine learning model, HFPred, for predicting heart failure decompensation. The model, initially developed using self-reported daily questionnaires and health monitoring data, was trained on a cohort of 242 patients from Basurto Hospital, Bilbao. The validation study spanned three European cohorts, each with distinct objectives and patient demographics, providing a comprehensive assessment of the model’s applicability. While the model accurately identified instances of decompensation, it also generated false alarms, primarily attributed to measurement errors and uncontrolled external factors. Despite these challenges, patient compliance was commendable, underscoring the potential benefits of the model. Future improvements include incorporating personalized alert thresholds and conducting non-blind pilot studies for enhanced predictive capabilities.
BIB_text
title = {Multicenter Prospective Blind External Validation of a Machine Learning Model for Predicting Heart Failure Decompensation: A 3-Hospital Validation Study},
pages = {368-377},
keywds = {
Cardiac Decompensation; Heart Failure; Machine Learning; Monitoring; Validation
}
abstract = {
This paper presents the results of a multicenter prospective blind external validation study aimed at validating a machine learning model, HFPred, for predicting heart failure decompensation. The model, initially developed using self-reported daily questionnaires and health monitoring data, was trained on a cohort of 242 patients from Basurto Hospital, Bilbao. The validation study spanned three European cohorts, each with distinct objectives and patient demographics, providing a comprehensive assessment of the model’s applicability. While the model accurately identified instances of decompensation, it also generated false alarms, primarily attributed to measurement errors and uncontrolled external factors. Despite these challenges, patient compliance was commendable, underscoring the potential benefits of the model. Future improvements include incorporating personalized alert thresholds and conducting non-blind pilot studies for enhanced predictive capabilities.
}
isbn = {978-303161136-0},
date = {2024-06-04},
}
