Improved risk assessment in parvovirus B19-positive patients with heart failure by multiparametric analysis of...

The analysis of endomyocardial biopsies (EMB) is a prerequisite for a definitive diagnosis in patients with unexplained heart failure (HF). The use of machine learning (ML) methods may help to identify high-risk patients and to initiate therapy. In this study, we develop ML models for risk stratification of parvovirus B19 (B19V) positive patients with HF based on key features from multiparametric EMB analyses.

We retrospectively enrolled 263 B19V-positive patients with HF (mean age 51 ± 15 years, 62% male) and followed them over a median period of 22 months (interquartile range 4-35 months). All-cause mortality, left ventricular (LV) deterioration and persistent LV systolic dysfunction were used as clinical combined endpoint. EMB were analysed for a variety of inflammatory and infectious markers, and patient prognosis was assessed using ML methods (logistic regression, random forest, support vector machines and gradient boosting). Detection of intramyocardial inflammation and B19V viral activity was associated with poor clinical outcome (hazard ratio 3.50, 95% confidence interval 1.96-6.23, p < 0.001). Linear combination of demographic and clinical data with multiparametric EMB markers increased the prognostic performance (area under the curve [AUC] = 0.724) compared to using single features (AUC = 0.667). The use of gradient boosting ML methods significantly improved the accuracy of risk prediction (AUC = 0.926).

Intramyocardial inflammation and B19V viral activity were detected more frequently in patients with poor clinical outcome, suggesting that they are risk factors for death and LV dysfunction. Using multiparametric EMB data, we present an ML-based prognostic tool that can determine the clinical outcome of patients with a high degree of accuracy. This could be helpful in assessing the progression of the disease and making appropriate treatment decisions.