Clinical Scorecard: External Assessment of Machine Learning Algorithms for Anticipating Prehospital Delays in Acute Ischemic Stroke: A Retrospective Cohort Analysis

At a Glance

Key Highlights

• Prehospital delay affected 80.9% and 83.2% of patients in the development and validation cohorts, respectively.
• Logistic regression model yielded an AUC of 0.774 without SMOTE.
• XGBoost showed the highest validation AUC of 0.771, but not significantly better than logistic regression.
• Complex algorithms did not consistently outperform simpler models.
• The study emphasizes the need for universal interventions to address prehospital delays.

Guideline-Based Recommendations

Diagnosis

• AIS diagnosis should follow World Health Organization criteria and be verified using imaging.

Management

• Timely reperfusion therapy is critical within narrow therapeutic windows.

Monitoring & Follow-up

• Prehospital delays should be monitored to improve stroke care delivery.

Risks

• Delays in treatment can lead to increased morbidity and mortality in AIS patients.

Patient & Prescribing Data

Consecutive adults with AIS admitted from January 1, 2019, through December 31, 2024.

Intravenous thrombolysis and thrombectomy were excluded as predictors to prevent reverse causality.

Clinical Best Practices

• Utilize logistic regression for initial modeling of prehospital delays.
• Incorporate a comprehensive set of candidate predictors for better model accuracy.
• Focus on quality improvement initiatives to reduce prehospital delays.

Related Resources & Content

• Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)