EHRtemporalVariability

EHRtemporalVariability analyzes temporal variability in Electronic Health Records (EHR) datasets to identify shifts in data distributions over time and assess data quality and consistency across multisite biomedical repositories.

Key Features:

• Information-theoretic and geometric methods: Employs probabilistic methods rooted in information theory and geometry to assess distributional changes over time.
• Temporal Data Analysis: Estimates statistical distributions of EHR variables over time and projects them onto non-parametric statistical manifolds using Information Geometric Temporal (IGT) plots for visualization of latent temporal variability.
• Kinematic Modeling: Characterizes data probability distributions as continuous trajectories in IGT space and estimates velocity and acceleration to quantify gradual trends and abrupt shifts.
• Seasonal and Systematic Variability Detection: Applies the Auto-Parallelism of Velocity Vectors (APVV) method and APVVmap visualization to identify oriented seasonal patterns and systematic variability, including periodic components beyond traditional Fourier analysis.
• Exploratory Visual Analytics: Provides Data Temporal heatmaps and IGT plots to detect punctual anomalies, outlying clusters, and systematic data quality issues.

Scientific Applications:

• Data quality assessment: Identifies temporal and multisite variability that can compromise reuse and harmonization of large-scale EHR datasets.
• Multimodal and multivariate analysis: Supports analysis of multi-modal and multivariate biomedical data where distributional shifts affect downstream inference.
• Seasonality and coding-change detection: Detects seasonal effects and abrupt changes in coding practices that impact longitudinal studies and surveillance.
• Consistency evaluation across sites: Assesses consistency and systematic differences across multisite biomedical repositories to inform pooling and comparative analyses.

Methodology:

Partition data into temporal subgroups, estimate statistical distributions and project them onto non-parametric statistical manifolds, represent distributions as continuous trajectories on Information Geometric Temporal (IGT) plots, analyze trajectories with a kinematic model to obtain velocity and acceleration, apply APVV and APVVmap to detect oriented seasonal/systematic variability, and visualize shifts using Data Temporal heatmaps and IGT plots while correlating trajectory components with known temporal factors.