Harnessing AI for heart failure surveillance in the population: The Carpediem approach

Early identification of patients at risk for heart failure (HF) is a challenging endeavor from a preventive and therapeutic perspective. The CarpeDiem project has the aim of validating an innovative approach to identify patients presenting with or at risk of HF by using pre-existing clinical and health-related information. Data of 153,393 subjects (socio-demographic, hospital discharge diagnosis, drug prescriptions, co-pay exemption and outpatient data) from the regional archiving system were analyzed. Validation of the HF algorithm was achieved by comparing it against two standards: the GPs' pathology registers of patients with HF or diabetes (10,149 patients), and the classification of a subset of 389 patients accomplished by blinded cardiologists using EHR. By applying Social Network Analysis to the individual characteristics of 389 patients (sex, age, NHYA score, presence/absence of AF/hypertension/ischemia, HF severity index -HFss-, morbidity severity by clinicians) we identified a graph where the spheres represent nodes (patients/characteristics) connected by arcs representing relations between nodes. Four communities of more densely connected nodes were detected by applying a community detection algorithm: the Yellow community refers to younger patients without diabetes and HF, HFss=0, and NHYA score <3. The Green one represents older patients (>78 years) with cardiometabolic comorbidity, with the highest Hss and NHYA 3-4. The Blue one includes males aged 66-78 years with diabetes, past ischemia and HFss=2. The Pink community includes women with HF, AF and hypertension, having HFss of 1. Our results highlight some clinical profiles needing specific attention. Furthermore, comparing the patient’s distribution by communities with the patient’s classification by (a) the Carpediem algorithm (b) the cardiologists and (c) the GPs, we found a better concordance between the algorithm’s classification and that of blinded specialists compared to that of the GPs. This suggests that CarpeDiem could serve as an effective surveillance tool in the field of General Medicine.