# Association between the platelet-to-albumin ratio and 28-day all-cause mortality in critically ill patients with Pulmonary embolism: a retrospective cohort study and predictive model establishment based on machine learning

## TL;DR

This study finds that a low platelet-to-albumin ratio is linked to higher mortality in critically ill patients with pulmonary embolism.

## Contribution

The study introduces a predictive model using machine learning to assess mortality risk in pulmonary embolism patients based on the platelet-to-albumin ratio.

## Key findings

- Lower platelet-to-albumin ratio is significantly associated with increased 7-, 14-, and 28-day ICU mortality in pulmonary embolism patients.
- A predictive model using Conditional Inference Trees achieved an AUC of 0.623 for mortality prediction.
- The relationship between platelet-to-albumin ratio and survival follows an 'L-shaped' pattern, indicating higher risk at lower ratios.

## Abstract

Pulmonary embolism (PE) is a serious condition that is frequently encountered in clinical practice. It has been demonstrated that the body’s physiological responses to platelet activation can lead to significant complications, including pulmonary hypertension, bronchoconstriction, and right heart failure. Albumin is recognized as a composite indicator of acute-phase reactant proteins, which have osmotic and anti-inflammatory properties, as well as nutrient and metabolic imbalance. Albumin demonstrates independent prognostic value in a variety of diseases. The platelet-to-albumin ratio (PAR) has emerged as a reliable predictor of mortality and complications based on systemic inflammation in a number of diseases. However, studies on the relationship between PAR and adverse outcomes in critically ill patients with pulmonary embolism are limited. Thus, this study aimed to investigate whether PAR could be a useful indicator for assessing pulmonary embolism outcomes.

The clinical data of 1163 patients with critical pulmonary embolisms were extracted from the MIMIC-IV (version 2.2) database. The study population was categorized into four groups according to PAR quartiles. The primary regression was 28-day ICU mortality, while the secondary regressions were 7-d and 14-d ICU mortality. Restricted cubic splines, Cox proportional hazards regression, and Kaplan-Meier curves were used to explore the relationship between PAR and adverse outcomes. We assessed the predictive power of PAR using the Boruta algorithm and built predictive models using machine learning algorithms.

Data from 1163 patients diagnosed with pulmonary embolism were analyzed. Lower PAR was significantly associated with an increased risk of 7-d (p < 0.001), 14-d (p < 0.001), and 28-d (p < 0.005) ICU mortality compared with higher PAR. The restricted cubic spline curve revealed an “L-shaped” relationship between PAR and survival, suggesting that an increase in PAR is linked to a reduced risk of adverse events. Patients with lower PAR had a higher risk of death within 7, 14, and 28 days in the ICU compared to those with higher PAR (p < 0.05). Boruta feature selection showed PAR had a higher Z score, and the model built using the Conditional Inference Trees algorithm had the best performance (AUC = 0.623).

PAR showed an “L”-shaped relationship with all-cause mortality at 7, 14, and 28 days in critically ill patients. Low PAR was significantly associated with an increased risk of adverse events, suggesting that PAR may be a predictor of adverse outcomes in patients with pulmonary embolisms.