RACIAL DISPARITIES IN CONTEXT: A MULTILEVEL ANALYSIS OF EARLY TREATMENT PATTERNS AMONG MEN WITH HIGH RISK PROSTATE CANCER

Purpose:

To quantify the impact of county-level characteristics on post-diagnosis early treatment patterns of African American (AA) and White-non Hispanic (W) older men with high risk prostate cancer (PCa)

Methods:

We analyzed AA and W men aged 66 years or older, diagnosed with high risk PCa (AJCC classification:T3/T4, PSA>20 ng ml^-1, Gleason ≥8) between 2000 and 2005 from the linked SEER-Medicare dataset. The data was enriched with US census (2000) and county business pattern datasets (2000-2005). Early treatment during the first 12 months of diagnosis was categorized as, 1) conservative management and, 2) any treatment receipt (hormone therapy/radical prostatectomy/radiation/chemotherapy).County-level measures included, percent below poverty-level and education-level as well as number of health care facilities and services available per-capita. The variation in early treatment patterns across US counties were examined using caterpillar plots. Variance partition coefficients (VPC) were estimated to quantify the variation in treatment patterns attributable to contextual differences across the counties. Two-level random intercept/slope logit models with cross-level interactions were used to quantify the extent to which area-level characteristics of the county modified race disparities in treatment receipt controlling for individual and county-level characteristics.

Results:

Application of the inclusion criteria resulted in 55,626 high risk PCa patients. Majority (89%) of the patients were W and 83% of the patients received treatment. The average age was 75 (SD:6.2) years. Ws were more likely to receive treatment than AAs (W:84%, AA:77%, p<0.01). The caterpillar plots illustrated the variation across counties in treatment receipt among AAs and Ws (Figures:1-2). VPCs showed that 7.1% (AA) and 3.5% (W) of the variation in treatment receipt was due to differences in contextual factors. The effect of race on treatment receipt varied across counties (p<0.01). An AA patient living in a richer county (<25% below poverty) was less likely to receive treatment compared to a W patient (OR: 0.55, 95% CI:0.49-0.62, p<0.01) living in the same county. However, the disparity was reduced when comparing AAs and Ws living in poorer counties (>75% below poverty) (OR:0.64, 95% CI:0.58- 0.71, p<0.01).

Conclusions:    

We observed geographic variation in treatment receipt among high risk PCa patients. The effect of area-level poverty is different among AAs and Ws, after adjusting for individual and area-level characteristics.