Cluster analysis of heart failure patients based on their psychological and physical symptoms and predictive analysis of cluster membership

Abstract

AbstractAimPatients with heart failure experience multiple co‐occurring symptoms that lower their quality of life and increase hospitalization and mortality rates. So far, no heart failure symptom cluster study recruited patients from community settings or focused on symptoms predicting most clinical outcomes. Considering physical and psychological symptoms together allows understanding how they burden patients in different combinations. Moreover, studies predicting symptom cluster membership using variables other than symptoms are lacking. We aimed to (a) cluster heart failure patients based on physical and psychological symptoms and (b) predict symptom cluster membership using sociodemographic/clinical variables.DesignSecondary analysis of MOTIVATE‐HF trial, which recruited 510 heart failure patients from a hospital, an outpatient and a community setting in Italy.MethodsCluster analysis was performed based on the two scores of the Hospital Anxiety‐Depression scale and two scores of the Heart‐Failure Somatic Perception Scale predicting most clinical outcomes. ANOVA and chi‐square test were used to compare patients' characteristics among clusters. For the predictive analysis, we split the data into a training set and a test set and trained three classification models on the former to predict patients' symptom cluster membership based on 11 clinical/sociodemographic variables. Permutation analysis investigated which variables best predicted cluster membership.ResultsFour clusters were identified based on the intensity and combination of psychological and physical symptoms: mixed distress (high psychological, low physical symptoms), high distress, low distress and moderate distress. Clinical and sociodemographic differences were found among clusters. NYHA‐class (New York Heart Association) and sleep quality were the most important variables in predicting symptom cluster membership.ConclusionsThese results can support the development of tailored symptom management intervention and the investigation of symptom clusters' effect on patient outcomes. The promising results of the predictive analysis suggest that such benefits may be obtained even when direct access to symptoms‐related data is absent.ImplicationsThese results may be particularly useful to clinicians, patients and researchers because they highlight the importance of addressing clusters of symptoms, instead of individual symptoms, to facilitate symptom detection and management. Knowing which variables best predict symptom cluster membership can allow to obtain such benefits even when direct access to symptoms‐data is absent.ImpactFour clusters of heart failure patients characterized by different intensity and combination of psychological and physical symptoms were identified. NYHA class and sleep quality appeared important variables in predicting symptom cluster membership.Reporting MethodThe authors have adhered to the EQUATOR guidelines STROBE to report observational cross‐sectional studies.Patient or Public ContributionPatients were included only for collecting their data.