Predicting Heart Failure Outcomes With Biomarkers and Imaging

Overview

This study aims to develop a better model to predict one-year risk of death in patients with heart failure. We will test whether combining information from routine blood tests (like NT-proBNP) and heart scans (measuring features like epicardial fat density) improves risk prediction compared to using either type of data alone. This is a retrospective study using existing medical records of patients treated for chronic heart failure at Xinjiang Medical University First Affiliated Hospital between 2012 and 2024. No new patient contact or interventions are involved. The goal is to enable more accurate, personalized risk assessment across different types of heart failure (HFrEF, HFmrEF, HFpEF).

Background and Rationale: Accurate prognosis in heart failure (HF) remains challenging due to phenotypic heterogeneity across the spectrum of left ventricular ejection fraction (LVEF). While biomarkers like N-terminal pro-B-type natriuretic peptide (NT-proBNP) and imaging parameters like LVEF are standard prognostic tools, each has limitations. Emerging imaging parameters, such as epicardial adipose tissue (EAT) density (reflecting fat inflammation/fibrosis) and left ventricular global longitudinal strain (LVGLS), offer potential incremental prognostic value but are not yet integrated into routine clinical models. This study aims to systematically evaluate whether a multi-parameter model combining established blood biomarkers and advanced imaging metrics improves the prognostic stratification of patients with HFrEF, HFmrEF, and HFpEF compared to traditional approaches. Detailed Methodology: This is a single-center, retrospective cohort study. The study population consists of consecutive adult patients (≥18 years) with a confirmed diagnosis of chronic HF who had both qualifying blood biomarker assessment (NT-proBNP and/or high-sensitivity cardiac troponin) and cardiac imaging (transthoracic echocardiography and/or cardiac computed tomography) performed within a ±3-month window around an index encounter between January 1, 2012, and December 31, 2024, at Xinjiang Medical University First Affiliated Hospital. Key data to be extracted from electronic health records include: 1) Clinical variables: demographics, comorbidities (e.g., ischemic etiology, diabetes, hypertension), medications, and NYHA class; 2) Blood biomarkers: NT-proBNP, hs-cTnT/I, hs-CRP, and renal function (eGFR); 3) Imaging parameters: LVEF, LVGLS, left atrial volume index (LAVI), E/e' ratio, and EAT volume/density (from CT, if available). The primary endpoint is all-cause mortality at one year from the index date. Follow-up data will be obtained from hospital records. Statistical Analysis Plan: The incremental prognostic value will be assessed by constructing and comparing nested Cox proportional hazards models: Model 1 (Base Clinical): Includes age, sex, BMI, ischemic etiology, diabetes, and hypertension. Model 2 (Biomarker-Enhanced): Model 1 + NT-proBNP + eGFR. Model 3 (Imaging-Enhanced): Model 2 + key imaging parameters (e.g., EAT density or LVGLS). Model performance will be compared using Harrell's C-statistic, the Akaike Information Criterion (AIC), Net Reclassification Improvement (NRI), and Integrated Discrimination Improvement (IDI). Pre-specified subgroup analyses will be conducted for HFrEF, HFmrEF, and HFpEF phenotypes. Multiple imputation will be used for variables with low rates of missing data (\<10%).