Decipher Cognitive Impairment in Heart Failure by Text Mining

Abstract

Introduction

Heart failure (HF) and cognitive impairment (CI) are prevalent and often co-occurring conditions, affecting a considerable portion of the population. Hemodynamic disturbances are important but do not fully explain CI. Therefore, novel etiological insights are needed to clarify the heart-brain axis. This study focuses on unraveling mechanisms underlying cognitive decline in HF patients. We hypothesized that we could identify symptom commonalities extracted from real- world clinical anamnesis texts of patients with HF and CI using text-mining procedures. This may lead to novel mechanistic clues on cognitive decline in patients with HF and could provide potential leads for future treatment trials.

Methods

In this study, we analyzed anamnesis texts of HF and CI patients between 2011 and 2023 using electronic health records (EHRs) from the Utrecht Patient-Oriented Database (UPOD). Patients with HF and CI were identified through text mining of echocardiogram reports and the presence of clinical neurological and geriatric reports from outpatient memory clinics, respectively. To investigate noted symptomatic overlap in HF and CI, we utilized Named Entity Detection and Linking (NER+L) strategies to link noted symptoms in anamnesis texts with three Dutch biomedical ontologies.

Results

In total, anamnesis texts of 5597 unique HF and 961 CI patients were extracted and analyzed. Top-3 reported symptoms in HF patients were dyspnea (71.3%), hydrops (47.1%), and dizziness (43.6%) (Figure 2). Top-3 reported symptoms in CI patients were memory impairment (55.5%), cognitive problems (54.2%), and getting lost (21.1%). The noted symptoms that overlapped between HF and CI patients were fear, dizziness, tiredness, uncertain behavior, appetite, headache, and lack of energy. We did not find any noted cognitive symptoms in HF patients' anamnesis texts, nor did we find any noted cardiac symptoms in CI patients.

Discussion

Based on real-world evidence, we identified noted overlap in reported symptoms in HF and CI. These symptoms could be potential leads for future treatment trials.