# New models to predict survival of patients with difficult ventilator-weaning diagnosed with community-acquired pneumonia

## TL;DR

This study developed new models to better predict survival in patients with community-acquired pneumonia who struggle to wean off ventilators.

## Contribution

The paper introduces three new prediction models that outperform existing tools like CURB-65 for predicting survival in ventilated pneumonia patients.

## Key findings

- Three new models achieved higher AUROC scores (75-78%) than CURB-65 in predicting survival for CAP patients.
- Models 1 and 3 showed statistically significant improvements in predictability compared to CURB-65.
- The new models performed well in validation samples with AUROCs ≥ 80%.

## Abstract

High mortality is common in mechanically ventilated patients with severe community-acquired pneumonia (CAP). This study sought to create prediction models and determine factors for the pneumonia patients under difficult ventilator weaning in a respiratory care center.

In total, 353 CAP and hospital-acquired pneumonia (HAP) patients admitted to a respiratory care center (RCC) from January 1, 2015 to December 31, 2017 were included in this retrospective cohort study. Mortality and weaning risks factors were collected and analyzed for validating the prediction models. The study focuses primarily on CAP patients, with HAP data used to external validation.

Among 270 CAP patients in model testing and validation, CURB-65 (Confusion, Uremia, Respiratory rate, Blood pressure, aged ≥65 years) and CUB-65 similarly predicted RCC survival (AUROCs ~65%). Three new RCC survival prediction models incorporated age ≥65, hypotension, BUN >19 mg/dl, ventilator type (replacing respiratory rate), and GCS ≤ 8 (replacing confusion), and either white blood cells (WBCs) or hemoglobin (Hb) were additionally included in model 1 or model 2. In CAP test samples, AUROCs with CAP test sample were 77.51% (model 1), 75.99% (model 2), and 77.97% (model 3). All three new models showed higher AUROCs than CURB-65, with significantly improved predictability in models 1 (p = 0.0354) and 3 (p = 0.0383). All three models demonstrated satisfactory performance (AUROC ≥ 80%) for predicting RCC survival in the CAP validation sample.

These new models more accurately predicted survival during RCC admission. Furthermore, they offer clinicians a better predictive tool for survival in CAP patients facing difficult ventilator weaning.