Machine learning using multimodal and autonomic nervous system parameters predicts clinically apparent...

Area of Science:

• Neurology
• Infectious Disease
• Biomedical Informatics

Background:

• Stroke-associated pneumonia (SAP) significantly worsens patient outcomes.
• Predicting SAP is crucial for timely interventions.
• Machine learning (ML) offers potential for accurate SAP prediction using clinical data.

Purpose of the Study:

• To develop and validate an automated ML model for predicting clinically apparent SAP.
• To identify key clinical and physiological parameters for SAP prediction.
• To compare the ML model's performance against existing clinical scores.

Main Methods:

• Utilized a logistic regression classifier trained on clinical, laboratory, heart rate (HR), heart rate variability (HRV), and blood pressure (BP) data from the first 48 hours post-stroke admission.
• Employed nested cross-validation (nCV) for internal validation.
• Benchmarked the ML model against the A2DS2 score.

Main Results:

• The ML model achieved an AUC of 0.91, outperforming the A2DS2 score (AUC 0.84).
• Key predictors included CRP, mRS, leukocyte count, HRV high-frequency power, stroke severity, sex, and diastolic BP.
• The ML model demonstrated superior sensitivity (0.87) and specificity (0.82) compared to A2DS2.

Conclusions:

• Automated ML prediction of SAP in stroke units is feasible and effective.
• Incorporating vital signs significantly enhances prediction accuracy.
• This model can aid in identifying high-risk patients for prophylactic pneumonia management.