Development and validation of a machine learning model for predicting stroke-associated pneumonia in older patients with acute ischemic stroke

 What fucking stupidity; predicting pneumonia rather that creating a protocol to prevent it! You're all fired!

You've known about this problem for a long time. SOLVE IT! 

Just maybe this vaccine!

Pneumonia Vaccine (3 posts to July 2020)

• 11% Stroke-associated pneumonia (17 posts to October 2020)

Development and validation of a machine learning model for predicting stroke-associated pneumonia in older patients with acute ischemic stroke

• W

  Wen-Jie Chu 1

• S

  Si-Ran Zhang 2*

• Q

  Qi-Lun Lai 2

• J

  Jing-Ying Yu 2

• Y

  Yi-Qian Xu 2

• 1. Department of Hospital Infections, Zhejiang Hospital, Hangzhou, China

• 2. Department of Neurology, Zhejiang Hospital, Hangzhou, China

Abstract

Objective: 

Stroke-associated pneumonia (SAP) is a common and serious complication in older patients with acute ischemic stroke (AIS). However, early identification of high-risk patients remains challenging. This study aimed to develop and validate an interpretable machine learning model for predicting SAP risk in older AIS patients.

Methods: 

This retrospective study included 1,011 eligible patients (aged ≥65 years) with AIS who were consecutively admitted to Zhejiang Hospital in China from September 1, 2018, to December 31, 2023. A total of 1,011 patients were randomly divided into training and testing sets (7:3 ratio). Demographics, comorbidities, laboratory test results, and admission assessments were collected to evaluate the risk of SAP. The synthetic minority oversampling technique (SMOTE) was used to address the imbalanced training data. The Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to filter the predictive features. Eight machine learning models, including Logistic Regression (LR), Support Vector Machine (SVM), Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Gradient Boosting Decision Tree (GBDT), Multi-layer Perceptron (MLP), and Random Forest (RF), were applied to identify the best prediction model. The optimal model was interpreted using the SHapley Additive exPlanations (SHAP).

Results: 

SAP incidence was 18.79%. LASSO identified 12 predictive features. The SVM demonstrated acceptable and stable predictive performance, achieving an accuracy of 0.773, sensitivity of 0.667, specificity of 0.798, F1 score of 0.524, Brier score of 0.156, and AUC of 0.794 (95% CI: 0.748–0.839) in the test set. SHAP analysis identified key factors influencing model predictions. An online platform was developed for clinical use.

Conclusion: 

This study demonstrates that an interpretable SVM-based machine learning model can effectively predict the risk of SAP in older patients with AIS using routinely available clinical and laboratory data. SHAP analysis further improved the model’s clinical interpretability by elucidating feature contributions. Our online prediction platform could serve as a promising tool for identifying high-risk older patients and facilitating the early prophylactic management of SAP.