Real-world algorithm development and analysis for recognition of advanced heart failure (RADAR-HF): a multicenter, electronic health record study
M Wosten, F Ver Donck, C L Oeste, L W Van Laake, A B Gevaert, G P J Van HoutAbstract
Introduction
Advanced heart failure (AdHF) remains underrecognized in real-world populations, often delaying assessment of eligibility for AdHF therapies such as left ventricular assist devices (LVAD) and heart transplantation (HTx). We developed an algorithm with which we were able to identify AdHF and candidates for AdHF therapy using multicenter real-world electronic health records (EHR).
Methods
A retrospective real-world cohort study was conducted using both structured and unstructured EHR data from 2018-2023. All patients ≥ 18 years with possible heart failure (HF) based on their administrative diagnosis code were included. Demographic data, laboratory results, medication administration, hospitalizations and imaging parameters were analyzed. We used a natural language processing pipeline to extract clinical information from unstructured data. Based on the current ESC guidelines, we first created a cohort with HF patients. Next, we developed a scoring system based on the current ESC AdHF criteria to identify AdHF patients and added a time window to isolate AdHF probably requiring AdHF therapy. Figure 1 shows the algorithm flowchart from initial identification to the final cohort (Cohort 9). The final cohort was then manually reviewed by experienced AdHF cardiologists to determine the indication and eligibility for AdHF therapy screening.
Results
18,180 patients from two hospitals from different countries were included in the analysis. Patients with AdHF were significantly older, had more comorbidities and left sided valvular diseases, more frequently used device therapy and more often had ischemic cardiomyopathy compared to HF patients without AdHF (p-value <0.001 for all). The AdHF cohort detected by the algorithm consisted of 4,690 patients (Figure 1). Of these, 172 (4%) were in the final cohort with probable indication for AdHF therapy and recent AdHF. Main exclusion criteria for AdHF therapy were age >75 years (22%), LVEF >35% (21%), and renal failure (18%). Patients no longer in follow-up at the center (n=51) were excluded from further review. Manual chart review of the final cohort confirmed AdHF in 44 (36%) of the remaining patients, in 29 (24%) action or follow-up was recommended by AdHF cardiologists (Figure 2). Algorithm performance was assessed in a cohort of patients who received HTx or LVAD implantation during the study period (n=26), with AdHF correctly identified in 81% of cases.
Conclusion
In a real-world multicenter, multinational cohort of HF patients, we were able to identify AdHF patients eligible for LVAD or HTx using an algorithm based on natural language processing of EHR data and the ESC AdHF definition. This EHR-based identification could serve as a clinically usable safety net and enable timely referral for AdHF therapy.For image description, please refer to the figure legend and surrounding text.For image description, please refer to the figure legend and surrounding text.