GASP: Graph-based Approximate Sequential Pattern Mining for Electronic Health Records

Area of Science:

• Health Informatics
• Data Mining
• Machine Learning

Background:

• Electronic health records (EHRs) contain valuable sequential patient data.
• Conventional sequential pattern mining is computationally intensive and sensitive to noise.
• Existing approximate methods struggle with multi-item event sequences common in healthcare.

Purpose of the Study:

• To propose GASP, a graph-based approximate sequential pattern mining approach.
• To address limitations of existing methods in handling multi-item event sequences.
• To improve the accuracy and computational efficiency of extracting frequent patterns from EHRs.

Main Methods:

• Developed GASP, a graph-based approximate sequential pattern mining algorithm.
• Compressed sequential EHR data into a concise graph structure.
• Evaluated GASP on two healthcare datasets.

Main Results:

• GASP effectively discovers frequent patterns in multi-item event sequences.
• The graph-based compression offers significant computational benefits.
• Empirical results show GASP outperforms existing approximate models in recoverability and predictive pattern extraction.

Conclusions:

• GASP provides an efficient and accurate solution for mining complex sequential patterns in EHRs.
• The method enhances the discovery of clinically relevant multi-item event sequences.
• GASP represents a significant advancement in approximate sequential pattern mining for healthcare applications.