Clustering of Longitudinal Trajectories with Multinomial EM Algorithm Based on State-Transition Templates

• Activity Number: 259 - SPEED: Missing Data and Causal Inference Methods, Part 2
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 3:05 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): John Rice* and Elizabeth Juarez-Colunga and James Feinstein
• Companies: Colorado School of Public Health and University of Colorado Denver and University of Colorado, Denver
• Keywords: polypharmacy; longitudinal data; clustering; multinomial

Abstract

Polypharmacy, the practice of prescribing multiple medications to a patient, is particularly prevalent among children with complex chronic conditions (CCC). However, there is limited understanding of how polypharmacy progresses over the life of the patient. It is hypothesized that there are a number of different phenotypes present in the population, including episodic conditions where additional drugs are prescribed as needed during flare-ups as well as consistent escalations over time. In order to characterize these subgroups statistically using billing data, we propose to summarize each patient’s data, a sequence of daily medication counts, into a frequency table defined by transitions between levels of polypharmacy. This comprises a multinomial distribution on the set of all possible transitions of a given order. Established applications of the EM algorithm may then be used to determine clusters of patients with similar patterns of transition in states of polypharmacy. This allows the clusters to account for diverse patterns, including patients who stay consistently at one level, those who escalate either gradually or rapidly, and those who episodically escalate and de-escalate.