Development and validation of machine learning models for predicting functional outcome after low-dose alteplase in the extended time window for acute ischemic stroke

 Predicting failure to recover never got anyone recovered! Get the hell out of stroke!

Development and validation of machine learning models for predicting functional outcome after low-dose alteplase in the extended time window for acute ischemic stroke

• H

  Huiru Chen 1†

• 

  Qian Gui 1†

• 
  

• 
  

• 
  

• 
  

• 

  Kangxiang Ji 2

• M

  Mengfan Ye 3

• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• 
  
• J

  Jieji Zhao 4

• Y

  Yan Kong 2*

• 
  

• 
  

• 
  

• 
  

• 
  

• 

  Guanhui Wu 1*

• 
  

• 
  

• 
  

• 
  

• 
  

• 

  Xin Tan 1*

• 1. Department of Neurology, The Affiliated Suzhou Hospital of Nanjing Medical University, Suzhou, Jiangsu, China

• 2. Department of Neurology, The First Affiliated Hospital of Soochow University, Suzhou, Jiangsu, China

Abstract

Background:

This study aims to develop machine learning (ML) models to predict 90-day functional outcomes for acute ischemic stroke (AIS) patients receiving thrombolysis with low-dose alteplase at 0.6 mg/kg between 4.5 and 9 h after symptom onset.

Methods:

We conducted a retrospective analysis of AIS patients receiving thrombolysis between August 1, 2019 and August 31, 2023. Eligible patients were randomly divided into training and validation sets in a 7:3 ratio. Good functional prognosis at 90 days were defined as modified Rankin scale score (mRS) ≤2. Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to select optimal features. Five ML algorithms were employed to construct prediction models. Model performance was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC) value, decision curve analysis (DCA), and calibration curves. SHapley Additive exPlanations (SHAP) plot was applied to interpret the model predictions.

Results:

A total of 202 patients were randomly divided into training (n = 142) and validation (n = 60) sets. The rate of poor functional prognosis at 90 days was 56.34% in the training set and 56.67% in the validation set. Random Forest (RF) model showed the best discriminative ability with the highest AUC of 0.854 in the validation set. Key predictive features included age, baseline systolic blood pressure, white blood cell count, baseline National Institutes of Health Stroke Scale (NIHSS) score, wake-up stroke, the absolute difference volume between the ischemic infarct and the penumbra, intracranial hemorrhage, hemorrhagic transformation classification, and occurrence of pneumonia.

Conclusion:

The RF-based ML model demonstrated clinical utility for post-intravenous thrombolysis risk stratification by identifying patients at higher risk of poor functional outcomes.