Abstract
Many neighborhood socioeconomic index measures (nSES) that capture neighborhood deprivation exist but the impact of measure selection on liver cancer (LC) geographic disparities remains unclear. We introduce a Bayesian geoadditive modeling approach to identify clusters in Pennsylvania (PA) with higher than expected LC incidence rates, adjusted for individual-level factors (age, sex, race, diagnosis year) and compared them to models with 7 different nSES index measures to elucidate the impact of nSES and measure selection on LC geospatial variation. LC cases diagnosed from 2007-2014 were obtained from the PA Cancer Registry and linked to nSES measures from U.S. census at the Census Tract (CT) level. Relative Risks (RR) were estimated for each CT, adjusted for individual-level factors (baseline model). Each nSES measure was added to the baseline model and changes in model fit, geographic disparity and state-wide RR ranges were compared. All 7 nSES measures were strongly associated with high risk clusters. Tract-level RR ranges and geographic disparity from the baseline model were attenuated after adjustment for nSES measures. Depending on the nSES measure selected, up to 60% of the LC burden could be explained, suggesting methodologic evaluations of multiple nSES measures may be warranted in future studies to inform LC prevention efforts.