A method to address community effect modification and avoid bias in evaluating mammogram interventions

BACKGROUND: Even though it is known that biologically identical community-based public health interventions may have significantly different outcomes because of contextual factors, Cochran reviewers note that community-based public health intervention research rarely accounts for effect modification by place.

METHODS: We first define a health disparity function D: I → R², where I is a set of communities identifiers and R² is the set of vectors of dimension. We then use a mathematical relation to create a classification scheme among communities based on the disparity function and prior experience. As an exemplar, we use data for regular screening mammography (≥ 1 per two years for women 65-74 years) following Medicare's 1991 decision to reimburse providers for screening mammography. Two county-based cohorts of 1,000,000 non-Hispanic women, each of whom were alive from 1992-95 (T1) or 2005-08 (T2), were obtained from all Medicare beneficiaries using complex probability sampling to enrich Black inclusion. Screening mammography was identified with an adapted Smith-Bondman algorithm.

RESULTS: Probability distributions are computed for each community set. Probability based stratified random sampling yields community sets across the spectrum of community-level outcomes which are significantly different from those which might be obtained through haphazard selection, selection based on co-location of an academic medical center, or selection based on socio-demographic characteristics.

CONCLUSIONS: The new methodology is theoretically advantageous as it makes no assumptions about community capability and achieves a greater likelihood of selecting communities that comprise the full spectrum of community contextual issues.