# Analysis of prognostic risk factors and risk management measures for patients with ischemic stroke and bloodstream infection based on machine learning

## TL;DR

This paper uses machine learning to identify key risk factors for patients with ischemic stroke and bloodstream infections, creating a predictive model to guide clinical risk management.

## Contribution

The novel contribution is the application of LASSO regression to build a precise predictive model for prognosis and risk management in this patient group.

## Key findings

- The LASSO model outperformed traditional regression in predicting survival rates at 7, 14, and 28 days.
- Key risk factors include pneumonia, heart failure, and elevated C-reactive protein-lymphocyte ratio.
- The model showed high predictive accuracy across training and validation datasets.

## Abstract

Through the machine learning Least absolute shrinkage and selection operator (LASSO) algorithm, the system screened core prognostic risk factors for patients with ischaemic stroke and bloodstream infection, constructed and validated a predictive model, provides a basis for formulating precise risk management strategies in clinical practice.

Clinical data from patients with bloodstream infections secondary to ischaemic stroke were retrospectively included. The dataset was randomly allocated into training and validation sets at a 7:3 ratio. Within the training set, Model 1 was constructed using traditional univariate and multivariate Cox regression methods, while Model 2 employed the machine learning LASSO regression algorithm. Models were compared using metrics including R-squared, C-index. The superior model was selected for validation on both training and validation sets using Area Under the Curve (AUC) of the receiver operating characteristic curve, calibration curves, and Decision-Making Curves (DCA).

Model 2 was adopted as the final model, with a nomogram generated for the training set. As demonstrated by the nomogram, an increase in the total score was observed in patients with concomitant pneumonia, heart failure (HF), or coronary atherosclerotic heart disease (CHD), in cases where mechanical ventilation (MV) was utilised, and in instances of elevated alanine aminotransferase (ALT) and C-reactive protein-lymphocyte ratio (CLR) values, and reduced albumin levels. In the training set, the AUC values for predicting 7-day, 14-day, and 28-day survival rates were 0.875, 0.886, and 0.861, respectively. The AUC value for the 28-day prognosis on the internal test set was 0.844, while that on the external validation dataset was 0.860. The model demonstrated high concordance between predicted and actual probabilities across three distinct cohorts. The clinical decision curve indicates that the model provides good net benefit within the 5%-25% range at all three time points (7,14,and 28 days) across the three datasets.

Comorbidities such as pneumonia and hypoalbuminaemia constitute prognostic risk factors affecting patients with ischaemic stroke and bloodstream infections. LASSO regression enables precise identification of these risk factors, yielding a prognostic prediction model with outstanding predictive efficacy and clinical utility. Clinicians may utilise model variables to implement targeted risk management strategies.