Temporal Graphs for EHR Event Sequence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph-based methods for electronic phenotyping of electronic health records (EHRs) are static and fail to capture the temporality of events, which can overlook impending disease conditions.
Innovation Solution
Transforming EHRs into temporal graphs and learning temporal patterns from these graphs to identify sequences of events that indicate evolving health conditions, allowing for the prediction of potential diseases before they are diagnosed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If static graph-based methods are used for electronic phenotyping of EHRs, then the analysis is simpler and faster, but the temporality of events is lost and impending disease conditions cannot be identified
Solution Approach 1:
The patent transforms static graph-based methods into dynamic temporal graph methods that capture the evolving nature of health events over time. By introducing time-stamped nodes and temporal edges that represent event sequences, the system dynamically models patient journeys through health states, enabling detection of impending conditions while managing complexity through structured temporal representations.
Solution Approach 2:
The patent adds a temporal dimension to traditional static graph representations of EHR data. By organizing events along a time axis and creating temporal sequences of health events, the system transforms 2D static graphs into 3D temporal graphs, preserving the chronological order of events and enabling detection of patterns that unfold over time without excessive complexity.
2Measurement precision
If temporal patterns are learned from EHRs to predict impending disease conditions, then prediction accuracy improves, but computational requirements and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-processing EHR data into standardized temporal graph formats and pre-learning common temporal patterns from population data. By preparing temporal representations in advance and identifying recurring event sequences before clinical application, the system reduces real-time processing requirements while maintaining high prediction accuracy for individual patient assessment.
Solution Approach 2:
The patent applies partial action by focusing temporal pattern learning on specific disease domains or event types rather than analyzing all possible EHR events comprehensively. By selecting relevant temporal patterns and event sequences tailored to specific clinical questions, the system achieves high prediction accuracy for target conditions while reducing overall computational burden compared to exhaustive analysis.
Data Source
AI summary
In one embodiment, a computer-implemented method includes transforming a plurality of electronic health records into a plurality of temporal graphs indicating an order in which events observed in the plurality of electronic health records occur and learning a temporal pattern from the plurality of temporal graphs, wherein the temporal pattern indicates an order of events that is observed to occur repeatedly across the plurality of temporal graphs.