Non-invasive bioimpedance analysis predicts NT-proBNP changes during hospitalization in a cohort of acutely decompensated heart failure patients

Heart failure (HF) patients are at high risk of re-hospitalizations and cardiovascular mortality. NT-proBNP is utilized during hospitalization to assess the trend in recompensation during therapy. Further, bioimpedance analysis (BIA) offers a non-invasive method to assess the volume status of HF patients and has been shown to predict early decompensation events and re-hospitalizations after hospital discharge. This study explores the role of BIA-derived parameters and a newly developed hydration status questionnaire (HSQ) in predicting changes in NT-proBNP during hospitalization, aiming to enhance HF management.

In this single-center observational study, patients (n=49) admitted to the cardiology ward for acute decompensated HF (ADHF) underwent a BIA-derived volume status assessment. BIA was assessed daily, while biomarkers and vital parameters were obtained from standard care. Median hospital stay was 7 [4-10] days. NT-proBNP was measured at admission and discharge, and all patients filled out a “hydration status questionnaire” (HSQ) designed for this study, including three sections to subjectively determine hydration status, diet adherence and therapy.

We performed a neural network analysis through Bayesian models utilizing age, biomarkers, BIA parameters and HSQ to predict changes in NT-proBNP during hospitalization. Three models were constructed. In Table 1, the detailed results and the marginal posterior inclusion probability of regressors are displayed. 

The first two models share the most important predictors such as BIA derived ΔResistance and ΔTotal body water (TBW) or NT-pro-BNP at admission. They differ in sections used from the HSQ, TQ1 including diet and medication adherence. Notably, NT-pro-BNP at admission, ΔResistance, and ΔPotassium had high posterior probabilities (>0.9) in models 1 and 2, and thus high importance for predicting ΔNT-pro-BNP. In Model 2, TQ1 showed a high prediction accuracy. Model 3, excluding NT-pro-BNP at admission, showed lower overall prediction accuracy (51%) compared to models 1 (96%) and 2 (97%), with ΔPotassium and ΔResistance showing the highest importance. Notably, ΔWeight consistently exhibited low probability across all models. 

Our models demonstrate an effective prediction of ΔNTproBNP during hospitalization using BIA-derived parameters and a novel HSQ  throughout the hospital stay. The study suggests benefits in using a questionnaire and BIA to assess HF patients, identifying those at higher medium term risk and highlighting the the value of these non-invasive predictors in the management of HF in in and out of hospital settings. Further studies are needed to assess the role of BIA in preventing hospital admissions.