Heart failure (HF) and atrial fibrillation (AF) are widely spread among the global population, especially in elderly patients. Usually, they co-exist and they are characterized by a complex cause-effect mechanism, responsible for worsening in the patient's clinical status, which can lead to recurrent hospitalisations. Therefore, the main goal of this retrospective study was to reduce the clinical and economic impact of HF hospitalisations on the national healthcare system, by implementing a predicting model to timely identify HF exacerbations, before they require to be managed in a hospital set-up.
The predicting algorithm combines the daily data trends of parameters remotely collected by ICD and CRT devices, to build-up day-by-day a risk index, which is compared to an upper nominal threshold in order to activate an alarm, indicating the risk to experience a hospital admission in a short time. The model for risk score computation was developed in two versions, including: the daily average frequency, the transforms of the daily and nightly average heart rates, physical activity and the addition of atrial fibrillation burden in one version. The two models were developed and tested on a large population, including only patients implanted with devices carrying a working atrial catheter if AF burden was included in the model.
Both versions of the model showed good results (sensitivity = 56%/57%, specificity = 82%/75.5%, median alerting time = 43/47.5 days, false-positive rate per patient-year = 1.16/0.83, PPV = 6.7%/10% and NPV = 99%/99% without and with AF burden, respectively).
Our model for HF prediction showed comparable results and performances compared to the multi-parametric risk-score indices for the prediction of HF-related hospitalisations currently available on the market, implemented by the main biomedical devices companies. On note, our approach is device-independent and it can be applied to remote monitoring data available from all devices.