Presence Features for Sparse EHR Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electronic health record (EHR) data is highly sparse and lacks specific time stamps, making it challenging for neural networks to accurately predict adverse health events, as most features are not present at each time step and time stamps are often incomplete.
Innovation Solution
The system preprocesses EHR data by generating presence features to differentiate between absent and zero numerical values, using surrogate time steps for incomplete timestamps, and embedding features into a deep embedding neural network to model historical context, enabling the neural network to make predictions even with sparse data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional neural networks process raw EHR data directly, then the model structure remains simple, but prediction accuracy deteriorates due to data sparsity and missing timestamps
Solution Approach 1:
The patent applies preliminary action by preprocessing EHR data before neural network input. Specifically, it generates presence features that indicate whether a numerical feature was observed at each time step, and creates surrogate time steps for incomplete timestamps. This preprocessing transforms sparse, irregular EHR data into a dense, structured format suitable for neural network processing, thereby improving prediction accuracy without requiring changes to the underlying neural network architecture.
2Measurement precision
If presence features are generated to differentiate absent and zero values, then prediction accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent introduces presence features as an intermediary element between raw EHR data and neural network input. These binary indicators serve as mediators that explicitly encode the absence or presence of numerical measurements, allowing the neural network to distinguish between unobserved features and zero values. This intermediary representation resolves the ambiguity in sparse EHR data and improves prediction accuracy while maintaining a systematic approach to feature engineering.
3Stability of the object's composition
If surrogate time steps are created for incomplete timestamps, then temporal sequence completeness improves, but data processing time increases
Solution Approach 1:
The patent applies self-service by having the data processing system automatically generate surrogate time steps for incomplete timestamps without requiring manual intervention. When a time step is missing data, the system automatically creates a surrogate time step that carries forward the most recent available information, ensuring temporal sequence completeness. This automated approach maintains data integrity while minimizing manual processing time.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting future adverse health events using neural networks. One of the methods includes receiving electronic health record data for a patient; generating, from the electronic health record data, an input sequence comprising a respective feature representation at each of a plurality of time window time steps, comprising, for each time window time step: determining, for each of the possible numerical features, whether the numerical feature occurred during the time window; and generating, for each of the possible numerical features, one or more presence features that identify whether the numerical feature occurred during the time window; and processing the input sequence using a neural network to generate a neural network output that characterizes a predicted likelihood that an adverse health event will occur to the patient.