Modeling Depression Progression Dynamics from Electronic Health Record

Thesis (Master's)--University of Washington, 2016-06

To assess and monitor the progression dynamics of patients' depression severity conditions, Markov models are refined from other disease progression modeling methodologies to identify the characteristics and evolvement of disease severity state transitions among a cohort of patients. Explored in this thesis is an integrated evaluation approach on the Markov models with emphasis on their application of modeling depression progression dynamics. Using the Patient Health Questionnaire (PHQ) - 9 data from electronic health record, b-spline curving and k-means clustering methods convert individuals' irregular PHQ-9 measurements to complete longitudinal depression trajectories and group the entire dataset into custom subgroups. Multi-State Discrete Time Markov model (MSM), Hidden Markov model (HMM), Semi-Markov model (semi-M) and Hidden Semi-Markov model (HSMM) are applied to each of the five subgroups to model the depression progression dynamics and identify the characteristics of depression severity state transitions. Purpose served in this approach, including providing insights on long-term severity progression outcomes, can be demonstrated among measurements of stationary probability, expected first passage time, and the proportion of time in a depression state where appropriate computed from the subgroup-specific transition probability matrices based on four Markov models. The effectiveness and accuracy of the models are then further validated under bootstrap and cross validation.