Hierarchical Vectorisation for Healthcare Data Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The variability in coding systems used in electronic health and medical records systems poses a significant challenge for analyzing structured healthcare data, as different systems and data sources often lack compatible mappings, hindering the training of prediction models.
Innovation Solution
A computer-implemented method and system utilizing hierarchical vectorization to represent healthcare data, which involves mapping healthcare codes to a taxonomy, generating node embeddings, aggregating event embeddings, and producing patient embeddings for predicting future healthcare aspects through multi-task learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If multiple coding systems are used to encode healthcare data from different sources, then the quantity and diversity of healthcare data increases, but the compatibility and interoperability between different coding systems deteriorates
Solution Approach 1:
The patent introduces a mapping layer that acts as an intermediary between different coding systems (e.g., ICD-10, ICD-9, CPT, HCPCS) and a unified taxonomy. This mapping layer translates codes from various sources into a common representation, enabling interoperability without losing the ability to handle diverse data sources. The mapping relationships are stored and used to convert incoming codes to the standardized taxonomy structure.
Solution Approach 2:
The patent creates a universal taxonomy structure that can represent multiple coding systems within a single framework. The taxonomy is designed to be multi-functional, accommodating different code types (diagnoses, procedures, medications) and different versions (ICD-10, ICD-9) through a common hierarchical structure, allowing one system to handle diverse coding requirements.
2Adaptability or versatility
If coding systems are mapped to a unified taxonomy, then interoperability and analysis capability improve, but the complexity of the mapping process increases
Solution Approach 1:
The patent segments the complex mapping process into distinct components: code normalization (converting to standard formats), taxonomy mapping (assigning to hierarchical categories), and relationship establishment (defining parent-child and sibling relationships). This segmentation allows each component to be handled separately and simplifies the overall complexity by breaking down the monolithic mapping task into manageable segments.
Solution Approach 2:
The patent performs preliminary actions by pre-establishing mapping relationships between common coding systems and the taxonomy structure. Mapping templates and conversion rules are prepared in advance for frequently used coding systems, so that when data arrives, the mapping process can proceed efficiently using these pre-computed relationships rather than creating mappings from scratch each time.
3Measurement precision
If hierarchical vectorisation is applied to healthcare data, then prediction accuracy improves, but the computational processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing hierarchical embeddings for taxonomy nodes and code mappings. These pre-computed embeddings are cached and reused during prediction tasks, avoiding the need to perform complex hierarchical vectorization computations in real-time, thus reducing processing time while maintaining prediction accuracy.
Solution Approach 2:
The patent segments the vectorization process into hierarchical levels (taxonomy node embeddings, code embeddings, patient embedding). Each level is processed separately and cached independently. This segmentation allows the system to reuse computations at different hierarchical levels across multiple queries, reducing redundant calculations and processing time.
Data Source
AI summary
There are provided systems and methods for using a hierarchical vectoriser for representation of healthcare data. One such method includes: receiving the healthcare data; mapping the code type to a taxonomy and generating node embeddings using relationships in the taxonomy for each code type with a graph embedding model; generating an event embedding for each event including aggregating vectors associated with each parameter vector using a non-linear mapping to the node embeddings, the event embedding including the node embeddings related to said event; generating a patient embedding for each patient by encoding including the event embeddings related to said patient; and outputting the embedding for each patient.