State Space Modeling of Event Count Time Series.

Area of Science:

• Biostatistics
• Time Series Analysis
• Computational Neuroscience

Background:

• Analyzing event count time series, particularly seizure counts in epilepsy, is complex due to multifactorial influences.
• Traditional methods struggle with the non-linear dynamics and external factors affecting seizure frequency.

Purpose of the Study:

• To develop and validate a novel state space modeling approach for analyzing event count time series.
• To objectively assess the impact of anti-epileptic drugs on seizure counts in drug-resistant epilepsy patients.

Main Methods:

• Utilized state space modeling with a nonlinear observation function and Gaussian linear dynamics.
• Employed an iterated extended Kalman filter for state estimation and a square-root filtering approach for covariance matrix stability.
• Incorporated exponential or "affinely distorted hyperbolic" observation functions to ensure non-negativity of count data.

Main Results:

• The developed algorithm successfully analyzed time series of daily seizure counts.
• External control inputs (anti-epileptic drug dosages) were integrated into the model.
• The analysis provided insights into whether specific drugs increase or decrease seizure frequency for individual patients.

Conclusions:

• State space modeling with Kalman filtering offers a robust method for analyzing complex event count time series like epilepsy seizures.
• This approach facilitates objective decision-making regarding the efficacy of anti-epileptic treatments.