Online Program

Calibration of Alzheimer’s Disease Microsimulation with Approximate Bayesian Computation (304199)

Stavroula A. Chrysanthopoulou, Brown School of Public Health 
Rowan Iskandar, Brown School of Public Health 
Eric Jutkowitz, Brown School of Public Health 
*Peter Tadashi Shewmaker, Brown School of Public Health 

Keywords: calibration, microsimulation, approximate bayesian computation

We will share findings from our calibration of a dementia health policy microsimulation that predicts people living with dementia’s cost and time in the community. The monthly probability of transitioning from the community to nursing home is defined by a Weibull regression model that includes terms for the person with dementia’s age, gender, race, cognition, physical function, and behaviors. The regression was estimated using data from the Uniform Data Set of the National Alzheimer’s Coordinating Center. We compared model predicted time in the nursing home stratified by gender, race, and life expectancy following diagnosis to external estimates from Medicare data. We used a rejection based Approximate Bayesian Computation (ABC) algorithm and ABC algorithm combined with Markov Chain Monte Carlo to calibrate the Weibull model coefficients for gender, and race, towards outcomes in Medicare data. We graphically compared the approximate posterior and posterior predictive distributions. ABC is currently being applied in a wide variety of fields. Our poster will illustrate the findings of our applied analysis and will aid practitioners working on parameter estimation.