Oxygen Saturation Prediction Model Switching for Alert Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current oxygen saturation monitoring systems often inaccurately predict whether a patient's oxygen saturation levels will return above a desaturation threshold, leading to unnecessary alerts for trivial desaturation events and missed non-trivial events.
Innovation Solution
A computing system tracks the performance of oxygen saturation prediction models and performs corrective actions, such as switching to a different model or updating the existing one, to improve prediction accuracy by using ground truth data and machine learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-generated harmful factors
If an oxygen saturation prediction model is used to predict future oxygen saturation levels, then the number of alerts can be reduced by avoiding unnecessary notifications for trivial desaturation events, but the prediction accuracy may be insufficient leading to both false positives and false negatives
Solution Approach 1:
The system continuously monitors actual oxygen saturation levels and compares them against predicted values to generate performance metrics. This feedback loop enables the system to evaluate prediction accuracy and identify when model performance degrades, triggering retraining operations to maintain optimal performance and reduce both false positives and false negatives.
Solution Approach 2:
The system dynamically adjusts model parameters through automated retraining using ground truth data from actual patient monitoring. By changing model parameters based on performance metrics and retraining with updated data, the system maintains high prediction accuracy while effectively filtering out trivial desaturation events to reduce unnecessary alerts.
2Measurement precision
If the prediction model is frequently updated to improve accuracy, then prediction performance can be enhanced, but the system complexity and computational resources required increase
Solution Approach 1:
The system performs automated model retraining using ground truth data collected from actual patient monitoring without requiring manual intervention. The performance metric automatically triggers retraining when degradation is detected, and the system self-manages the entire process from data collection to model updates, reducing the need for complex manual management while maintaining high accuracy.
Solution Approach 2:
The system pre-processes and stores ground truth data during normal operation, preparing training datasets in advance before they are needed for model retraining. This preliminary action reduces the computational burden during actual retraining operations and simplifies the overall system architecture by organizing data management tasks beforehand.
3Measurement precision
If ground truth data is collected and used for model retraining, then prediction accuracy improves, but the time and computational resources required for data processing increase
Solution Approach 1:
The system continuously collects and stores ground truth data during normal patient monitoring operations, transforming idle data collection into useful training material. This continuous accumulation of data eliminates the need for separate data collection phases and allows for efficient batch processing during retraining operations, reducing overall processing time while maintaining high prediction accuracy.
Data Source
AI summary
In some examples, a computing system tracks, across a plurality of predictions, prediction performance of a first oxygen saturation prediction model used by one or more patient monitoring devices by comparing a respective prediction made by the first oxygen saturation prediction model to a corresponding ground truth. The computing system determines whether the prediction performance of the first oxygen saturation prediction model meets a performance metric, wherein the performance metric includes an accuracy level, a specificity level, a sensitivity level, or any combination thereof. The computing system may, in response to determining that the prediction performance of the first oxygen saturation prediction model does not meet the performance metric, cause the one or more patient monitoring devices to switch to a second oxygen saturation prediction model to predict future oxygen saturation levels of one or more patients.