Modified Needleman-Wunsch algorithm for clinical pathway clustering.

Area of Science:

• Health Informatics
• Data Mining
• Computational Biology

Background:

• Clinical pathways standardize care delivery to reduce variation in processes and patient outcomes.
• Patient pathways are often represented as strings for data mining and clustering.
• Existing distance metrics for string data lack contextual understanding, leading to arbitrary clustering results.

Purpose of the Study:

• To develop a data mining technique integrating expert information with patient pathway data.
• To create a novel string distance metric for process data, incorporating contextual information.
• To improve the meaningfulness of clustering results in clinical pathway analysis.

Main Methods:

• Focus on k-medoids clustering for string data representing patient pathways.
• Evaluation of eight common string distance metrics for their applicability and limitations.
• Development of the modified Needleman-Wunsch algorithm, incorporating expert-defined activity groupings and rankings for contextual similarity.

Main Results:

• Standard distance metrics provide arbitrary distances without context, yielding varied clustering outcomes.
• The modified Needleman-Wunsch algorithm successfully integrates expert knowledge into string similarity calculations.
• The new metric allows for more meaningful clustering by adding context to pathway strings.

Conclusions:

• Integrating expert knowledge via a novel string distance metric significantly enhances clinical pathway clustering.
• The developed algorithm, applied to a lung cancer pathway, offers a flexible approach applicable to any disease type.
• The Sim.Pro.Flow tool provides a publicly available decision support system for this enhanced pathway analysis.