Heart Failure (HF) poses a challenge for our health systems, and early detection of Worsening HF (WHF), defined as a deterioration in symptoms and clinical and instrumental signs of HF, is vital to improving prognosis. Predicting WHF in a phase that is currently undiagnosable by physicians would enable prompt treatment of such events in patients at a higher risk of WHF. Although the role of Artificial Intelligence in cardiovascular diseases is becoming part of clinical practice, especially for diagnostic and prognostic purposes, its usage is often considered not completely reliable due to the incapacity of these models to provide a valid explanation about their output results. Physicians are often reluctant to make decisions based on unjustified results and see these models as black boxes. This study aims to develop a novel diagnostic model capable of predicting WHF while also providing an easy interpretation of the outcomes. We propose a threshold-based binary classifier built on a mathematical model derived from the Genetic Programming approach. This model clearly indicates that WHF is closely linked to creatinine, sPAP, and CAD, even though the relationship of these variables and WHF is almost complex. However, the proposed mathematical model allows for providing a 3D graphical representation, which medical staff can use to better understand the clinical situation of patients. Experiments conducted using retrospectively collected data from 519 patients treated at the HF Clinic of the University Hospital of Salerno have demonstrated the effectiveness of our model, surpassing the most commonly used machine learning algorithms. Indeed, the proposed GP-based classifier achieved a 96% average score for all considered evaluation metrics and fully supported the controls of medical staff. Our solution has the potential to impact clinical practice for HF by identifying patients at high risk of WHF and facilitating more rapid diagnosis, targeted treatment, and a reduction in hospitalizations.
An explainable model for predicting Worsening Heart Failure based on genetic programming
Visco V.;Robustelli A.;Loria F.;Rispoli A.;Bramanti A.;Carrizzo A.;Vecchione C.;Palmieri F.;Ciccarelli M.;D'Angelo G.
2024-01-01
Abstract
Heart Failure (HF) poses a challenge for our health systems, and early detection of Worsening HF (WHF), defined as a deterioration in symptoms and clinical and instrumental signs of HF, is vital to improving prognosis. Predicting WHF in a phase that is currently undiagnosable by physicians would enable prompt treatment of such events in patients at a higher risk of WHF. Although the role of Artificial Intelligence in cardiovascular diseases is becoming part of clinical practice, especially for diagnostic and prognostic purposes, its usage is often considered not completely reliable due to the incapacity of these models to provide a valid explanation about their output results. Physicians are often reluctant to make decisions based on unjustified results and see these models as black boxes. This study aims to develop a novel diagnostic model capable of predicting WHF while also providing an easy interpretation of the outcomes. We propose a threshold-based binary classifier built on a mathematical model derived from the Genetic Programming approach. This model clearly indicates that WHF is closely linked to creatinine, sPAP, and CAD, even though the relationship of these variables and WHF is almost complex. However, the proposed mathematical model allows for providing a 3D graphical representation, which medical staff can use to better understand the clinical situation of patients. Experiments conducted using retrospectively collected data from 519 patients treated at the HF Clinic of the University Hospital of Salerno have demonstrated the effectiveness of our model, surpassing the most commonly used machine learning algorithms. Indeed, the proposed GP-based classifier achieved a 96% average score for all considered evaluation metrics and fully supported the controls of medical staff. Our solution has the potential to impact clinical practice for HF by identifying patients at high risk of WHF and facilitating more rapid diagnosis, targeted treatment, and a reduction in hospitalizations.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.