Mining Temporal and Data Constraints Associated with Outcomes for Care Pathways.

A care/clinical pathway defines a standardized care process for a specific patient group, which consists of clinical goals, activities, data attributes, and constraints describing temporal dependencies and data preconditions of the activities. The constraints, which are the key elements to represent the best practices, are difficult to define due to the variations in different regions and populations. In this paper, we propose an approach to discover temporal and data constraints that are correlated with clinical outcomes for care pathways. For each activity of interest, we extract a set of associated event-condition-action (ECA) rules from electronic medical records (EMR) to represent the temporal and data preconditions of the activity, by using our modified association rule mining algorithm. Then the best ECA rule that is significantly more likely to lead to a positive outcome is translated into the constraint on the activity. The approach has been applied to real-world EMR, and discovered meaningful constraints for different groups of type 2 diabetes patients, which can be used to provide decision support during individual patient care.