# Red blood cell distribution width to albumin ratio predicts mortality in heart failure patients with pneumonia

## TL;DR

This study finds that a blood test ratio (RAR) can predict death risk in heart failure patients with pneumonia, and builds a machine learning model to help doctors assess risk more accurately.

## Contribution

RAR is introduced as a novel biomarker for mortality prediction in ICU heart failure patients with pneumonia, paired with a machine learning model and web tool for clinical use.

## Key findings

- RAR independently predicts 13% higher mortality risk per unit increase in heart failure patients with pneumonia.
- The LightGBM model using RAR and 13 features achieved strong predictive accuracy (AUC 0.735) in internal and external validation.
- A web-based tool was developed to enable real-time risk assessment for clinicians.

## Abstract

Heart failure (HF) patients with pneumonia admitted to the intensive care unit (ICU) face high mortality risks, yet current risk stratification tools lack precision. This study aims to evaluate the prognostic value of the red blood cell distribution width to albumin ratio (RAR) and develop a machine learning model for predicting 31-day in-hospital mortality in HF patients with pneumonia in ICU.

We included the MIMIC-IV cohort and an external validation cohort from the First Affiliated Hospital of Zhengzhou University. The restricted cubic spline (RCS), Kaplan–Meier, and multivariable Cox regression analyses were used to estimate the RAR's prognostic value for 31-day in-hospital mortality. Boruta-selected features were used to build eight machine learning models.

In the Medical Information Mart for Intensive Care -IV (MIMIC-IV) cohort (n = 3,158), RCS revealed a linear relationship between RAR and all-cause mortality. Patients were subsequently stratified into high- and low-risk groups based on the median RAR value. Kaplan–Meier analysis showed higher mortality in patients with above-median RAR (P < 0.05), and each unit increase in RAR independently predicted a 13% higher mortality risk (HR 1.13, 95% CI 1.06–1.21; P < 0.001). Subgroup and sensitivity analyses confirmed RAR's prognostic consistency across demographic strata, comorbidities, and medication regimens. The LightGBM model, trained on Boruta-selected 13 optimal features, was identified as the best prediction model and demonstrated strong generalizability in internal (AUC = 0.735) and external validation cohort (n = 1,110 patients) (AUC = 0.733). SHapley Additive exPlanations analysis ranked RAR as the second most critical predictor in the model. A web-based tool (https://mengzhaoyang.shinyapps.io/PneumoHF-RAR_Predictor/) was developed for real-time risk assessment.

This study identifies RAR as an effective prediction biomarker for HF patients with pneumonia and provides a clinical tool for precise risk profiling.