Can clinical audits be enhanced by pathway simulation and machine learning? An example from the acute stroke pathway

Area of Science:

• Medical Informatics
• Health Services Research
• Artificial Intelligence in Medicine

Background:

• Optimizing hyperacute stroke care is crucial for improving patient outcomes.
• Thrombolysis administration is time-sensitive and influenced by multiple clinical pathway factors.
• Understanding hospital-specific variations in thrombolysis use is essential for targeted improvements.

Purpose of the Study:

• To apply clinical pathway simulation and machine learning to clinical audit data.
• To identify key drivers for improving the use and speed of thrombolysis in stroke care.
• To evaluate the potential of these models for enhancing national clinical audits.

Main Methods:

• Utilized computer simulation modeling and machine learning techniques.
• Analyzed anonymized clinical audit data from 7864 patients across seven acute stroke units.
• Developed pathway simulation and machine learning models to predict the impact of pathway modifications.

Main Results:

• Identified three pivotal factors: known stroke onset time, pathway speed (arrival-to-scan, scan-to-thrombolysis times), and predisposition to use thrombolysis.
• Pathway simulation predicted benefits of optimizing individual stages, while machine learning predicted 'exportability' of decision-making.
• Combined models indicated a realistic thrombolysis use ceiling of 15%-25% and potential to double the number of patients with no significant disability.

Conclusions:

• A combination of pathway simulation and machine learning can enhance national clinical audits for hyperacute stroke pathways.
• These models identify key improvement levers, accommodating local patient population differences.
• Models based on audit data can be applied at scale, providing hospital-specific insights and realistic targets.