|
Abstract:
|
Preterm birth rates are increasing nationally. County rates show residual spatial correlation, even after adjustment for predictors such as race and poverty. The proposed model accounts for spatial dependence with an intrinsic autoregressive (IAR) prior on the risk of preterm birth, and then assumes that preterm births in a county follow a binomial distribution. Time is included to account for changes in risk factors and in the baseline risk surface. Known predictors are included in the model, and their coefficients are allowed to vary between time periods. An alternative model includes temporal autocorrelation. Methods for assessing changes in coefficients over time are considered. The models are fit to Arkansas preterm birth counts from 1994-2005. It is found that the IAR model is superior to one that does not include spatial correlation, and the apparent race effect changes over time.
|
