Online Program

Return to main conference page

Wednesday, January 10
Wed, Jan 10, 5:30 PM - 7:00 PM
Crystal Ballroom CD & Prefunction
Welcome Reception & Poster Session I

Leveraging cluster level variation for patient level inference. (304270)

*Evan Carey, University of Colorado Denver 
Thomas J Glorioso, VA Eastern Colorado Healthcare System 
Gary K Grunwald, VA Eastern Colorado Healthcare System, University of Colorado School of Public Health 

Keywords: health outcomes, chronic pain, cluster exposure

Assessing exposure-outcome relationships with patient level modeling approaches using administrative data require the improbable assumption that there is no unmeasured confounding. However, variation in the same exposures across different hospitals with relatively homogenous patient populations often exists due to differences in medical provider behavior. There are multiple methods for defining cluster level variation in exposure patterns adjusted for patient mix, including fixed and random effects in frequentist models, and Bayesian inference. Using a cohort of 1,132,402 Veterans reporting incident chronic pain between 2010-2016, we investigate the differences in the stability of site level exposure variables over time, as well as the associations of exposures with outcomes using all three methods. The exposures variables are various treatment modalities relevant to chronic pain. The outcome is emergency department utilization. We also conduct simulations modeled after this dataset to explore the potential bias and statistical power of recovering patient level relationships via site level effects.