Efficient Convex Optimization Methods for State-Space Heartbeat Dynamics Models with Gamma Generalized Linear Models

Abstract

Objective: This work proposes a dynamic state-space statistical model of heartbeat dynamics whose optimal state estimate can be solved efficiently with convex optimization and used to assess autonomic and cardiovascular health. Methods: This state-space model combines a Gauss-Markov process prior with gamma emissions using a log-link function. Parameters are estimated using a static version of the model, and alternating direction method of multipliers is used to efficiently find the optimal maximum a priori latent state estimate. Model selection is performed by finding the tradeoff between model size and model fit using the partial autocorrelation function. Results: We demonstrate in synthetically generated heartbeats that this model is able to closely recover the underlying latent weights. Then, we show using a tilt table dataset that this statistical model is able to better capture sharp transients in heart rate patterns than traditional local averaging methods. In addition, combining heart beat timings with information from other organs such as the respiratory phase, we show that our statistical modeling framework enables the extraction of information-theoretic measures to quantify inter-organ coupling such as respiratory sinus arrythmia. Conclusion: This statistical framework for modeling heartbeat dynamics provides an efficient framework that can provide physiologically meaningful insights. Significance: This work broadens potential tools that can be used to assess heartbeat dynamics.