Machine Learning Model for Hierarchical Medical Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for automating medical coding, such as adverse event information mapping to medical codes in dictionaries like MedDRA, face challenges due to the complexity and richness of healthcare data, often omitting significant portions of information and struggling with fine-grained semantic differences.
Innovation Solution
A machine learning model with multiple parallel output layers is trained to predict semantically and hierarchically related medical codes, using convolutional neural networks to assign both leaf node and parent node codes, thereby increasing prediction accuracy and utilizing training data more effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional coding methods are used, then the coding process is simple to implement, but coding completeness deteriorates significantly
Solution Approach 1:
The coding task is segmented into multiple hierarchical levels (system organ class, high level group term, high level term, preferred term, lowest level term). The model processes adverse event information through multiple sequential coding stages, assigning codes at each hierarchical level. This segmentation enables comprehensive coding coverage while maintaining implementation through modular processing steps.
2Productivity
If string matching techniques are used, then the coding process is fast and automated, but semantic precision deteriorates due to fine-grained differences
Solution Approach 1:
The patent replaces traditional mechanical string matching techniques with a deep learning-based neural network model. The model uses convolutional layers, pooling operations, and fully connected layers to process adverse event information and predict appropriate medical codes. This substitution enables both high-speed automated coding and high semantic precision by capturing complex semantic relationships through learned representations rather than simple pattern matching.
3Measurement precision
If multiple parallel output layers are used, then coding accuracy improves, but model complexity increases
Solution Approach 1:
The model architecture is segmented into functional modules: input embedding layer, convolutional layers, pooling layers, and multiple fully connected output layers corresponding to different hierarchical levels. Each output layer predicts codes at a specific hierarchical level, allowing the model to achieve high accuracy across multiple coding levels while maintaining manageable complexity through modular design.
4Loss of information
If hierarchical semantics are leveraged, then coding completeness improves, but computational requirements increase
Solution Approach 1:
The model performs preliminary processing by embedding medical terms into dense vector representations using word2vec or similar techniques before the main coding prediction. This preliminary action captures semantic relationships efficiently. During inference, the model processes adverse event information through convolutional and pooling layers to generate predictions across hierarchical levels, achieving comprehensive coding while managing computational resources through efficient layer design.
Data Source
AI summary
A method, a system, and a computer program product are provided. A machine learning model is generated to process adverse event information and produce multiple corresponding medical codes associated with the adverse event information, wherein the multiple medical codes are semantically and hierarchically related in a medical taxonomy. The machine learning model includes multiple parallel output layers, each of which is associated with a corresponding medical code. The machine learning model is trained with training data elements, each of which includes adverse event information mapped to respective multiple medical codes, wherein results from each of the output layers adjusts the machine learning model. After completing the training, information pertaining to an adverse event is applied to the machine learning model to determine the corresponding multiple medical codes within the medical taxonomy.