Predicting Recurrence for Patients With Ischemic Cerebrovascular Events Based on Process Discovery and Transfer...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Data Analysis

Background:

• Ischemic cerebrovascular events (ICE) recurrence leads to significant mortality and disability.
• Traditional machine learning models struggle with predicting ICE recurrence due to insufficient labeled hospital follow-up data.

Purpose of the Study:

• To develop a novel framework for predicting long-term ICE recurrence risk after hospital discharge.
• To identify high-risk patients for timely intervention by overcoming data limitations.

Main Methods:

• Utilized process mining on clinical guidelines to discover process models and extract control flow as patient characteristics.
• Employed transfer learning, specifically instance filter and weight-based methods, using in-hospital data (target domain) and national stroke screening data (source domain).
• Validated the framework on 205 tertiary hospital cases and 2954 screening cohort cases (2015-2017).

Main Results:

• The proposed framework demonstrated improved performance for three instance-based transfer learning algorithms.
• Successfully leveraged process mining and transfer learning to enhance the prediction of ICE recurrence risk.
• Addressed the limitation of insufficient labeled follow-up data in clinical settings.

Conclusions:

• The developed framework offers a comprehensive and efficient approach for predicting long-term ICE recurrence.
• Process mining and transfer learning integration can effectively mitigate data scarcity issues in predicting recurrent cerebrovascular events.
• This method facilitates proactive patient management and intervention for individuals at high risk of ICE recurrence.