Understanding reversals of association between cancer screening and race/ethnicity

We used a composite variable composed of insurance status, income, and race/ethnicity to investigate access-enhancing programs as a possible reason for "reversals of association" and large percent changes (LPC), between race/ethnicity and cancer screening, when comparing the unadjusted and adjusted ORs.

Data were from women aged 40-64 years, using the combined 2008 and 2010 Behavioral Risk Factor Surveillance System surveys. Recent mammography was within the past 2 years, and recent Pap testing was within the past 3 years. Initial analyses using all variables singly were followed by analyses that used the composite variable with the remaining covariates.

Analyses with race/ethnicity singly indicated reversals of association for Hispanic women and higher estimated screening for black and Hispanic women than for white women. Analyses with the composite variable found no reversals of association, but there were several LPCs for Hispanic and black women, for lower income, and for uninsured women. White, uninsured, lower income women were among those with the lowest utilization.

Results were consistent with the possibility that access-enhancing programs for lower income, uninsured and often non-white women can lead to overestimates of screening, reversals of association, and LPCs in multivariable analyses. Attention should be given to identifying LPCs to unadjusted ORs. Lower income, uninsured, white women are also a group at risk of extremely low mammography and Pap test utilization.

Combining variables to create better-targeted population subgroups may help in the interpretation of analyses that produce reversals of association and LPCs for correlates of cancer screening utilization.