Abstract:
|
While opioid overdose surveillance commonly tracks population-level rates such as overdose rates or mortality rates, it is also important to consider the overdose case fatality rate (the proportion of total overdose events resulting in death) because increases in this rate may signal that overdose response resources should be increased or reallocated. Methodologically, the overdose case fatality rate is fundamentally different than related population-level rates because the denominator is itself a random process. This project compares methodologies for modeling the overdose case fatality rate over time including time series methods, logistic regression, frequentist beta regression, and Bayesian beta regression. We demonstrate the advantages of utilizing beta regression in a Bayesian framework to model the overdose case fatality rate, with the precision parameter specified to account for changes in the total number of overdoses. Bayesian beta regression is not a novel methodology, but its application in modeling public health surveillance data is currently scarce despite having several methodological benefits for modeling rates (such as the overdose case fatality rate) over time.
|