Adherence Volatility Metric for Behavioral Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems lack effective methods for real-time detection of behavioral anomalies in treatment adherence patterns without prior model training, relying on static boundaries and not adapting to evolving patient behavior.
Innovation Solution
A method using adherence volatility metrics, calculated through binary or multi-state Markov chains, dynamically updates expectation bounds based on newly received data, allowing for real-time anomaly detection without prior training, using data from wearable sensors and mobile devices to monitor ingestion patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If static boundaries are used for anomaly detection, then the system is simple to implement, but it cannot adapt to evolving patient behavior
Solution Approach 1:
The patent implements dynamic boundaries that automatically adjust based on incoming adherence data. The system calculates rolling statistics (mean and standard deviation) from recent observations to define time-varying thresholds, allowing the anomaly detection system to adapt to changing patient behavior patterns without requiring manual reconfiguration or complex machine learning models.
Solution Approach 2:
The system performs self-adjustment by automatically computing its own boundaries from the data it collects. The adherence monitoring system calculates its own statistical parameters (mean adherence, standard deviation) from recent observations and uses these to dynamically set detection thresholds, eliminating the need for external training or manual calibration.
2Measurement precision
If prior model training is required for anomaly detection, then detection accuracy may be improved, but implementation time and complexity increase
Solution Approach 1:
The system performs preliminary calculations of statistical parameters (mean and standard deviation) from a small initial window of adherence observations, enabling immediate anomaly detection without requiring extensive prior training data. This allows the system to start detecting anomalies almost immediately after deployment while continuing to refine its parameters over time.
Solution Approach 2:
The system automatically computes its own detection parameters from the data it collects in real-time, eliminating the need for separate offline training phases. The adherence monitoring system continuously updates its statistical boundaries based on incoming observations, providing both rapid deployment and ongoing adaptation without manual intervention.
3Reliability
If dynamic boundary updates are implemented, then false notifications are reduced, but computational requirements increase
Solution Approach 1:
The system implements lightweight dynamic boundaries using simple rolling statistical calculations (mean and standard deviation) that can be efficiently computed with minimal computational resources. These dynamic thresholds adapt to changing patient behavior while maintaining low computational overhead compared to complex machine learning models.
Solution Approach 2:
The system changes its detection parameters (boundaries and thresholds) based on observed data characteristics. By computing statistical parameters from recent adherence observations and adjusting boundaries accordingly, the system adapts to individual patient patterns while using computationally efficient operations suitable for resource-constrained environments.
Data Source
AI summary
Methods, systems, apparatus, and computer programs, detecting behavioral anomalies in treatment adherence patterns. A method includes actions of obtaining data that represents whether an entity has complied with a therapeutic regimen or has not complied with a therapeutic regimen, determining a central tendency of an adherence volatility metric for the entity for at least n-time periods into the future, determining a plurality of boundaries around the central tendency, determining based on the data represented by the one or more data structures, an current observed adherence volatility metric, determining whether the current observed adherence volatility metric satisfies at least one of the plurality of boundaries around the central tendency, and based on a determination that the current observed adherence volatility metric satisfies at least one of the plurality of boundaries around the central tendency, generating a candidate anomaly data log record, the candidate anomaly data log record including data indicating that a candidate anomaly has been detected.