Date of Award

Fall 12-2018

Degree Name

Doctor of Public Health (DrPH)

Advisor(s)

WENYAW CHAN, PHD

Second Advisor

LUNG-CHANG CHIEN, DRPH

Third Advisor

JOHN M. SWINT, PHD

Abstract

Count data are commonly used to report frequency statistics of diverse health outcomes. However, some data are marked intentionally to avoid leaking information that could be used to identify individuals when population sizes are small. The situation hinders the further use from those data in public health research. Thus, an accurate and efficient method for dealing with censored count data is needed.

We developed Integrated Nested Laplace Approximation algorithm to censored Poisson regression model to deal with censored count data and improve the computational efficiency. In addition, we applied three methods to deal with censored count data: 1) multiple imputation (MI); 2) small area estimation (SAE); 3) censored Poisson regression model (CPRM) and compared the accuracy and efficiency of these three methods.

A series of simulations results in that the censored Poisson regression method conducted the closest estimates to the true values (with the relative error = 0.21%), and MI had the worst results (with relative error=9.13%) under the censored proportion by 7.9 %. After comparing the results under the censored proportion by 33.61% and 54.1%, the censored Poisson regression method still showed a smaller relative error than the other two methods.

We also applied these three methods to assess the association between heat wave temperature and hospitalization due to cardiovascular diseases in Harris County, Texas, from 2006 to 2011. By comparing the relative errors and bar plots across different methods under different censored proportions, we concluded that by considering the balance of the estimation accuracy with computational time, the censored Poisson regression model is the best method for dealing with censored count datasets under different censored proportions, especially when the censored proportions were less than 30%.

Share

COinS