What predicts an advanced-stage diagnosis of breast cancer? sorting out the influence of method of detection, access to care, and biologic factors

Abstract

Background: Multiple studies have yielded important findings regarding the determinants of an advanced-stage diagnosis of breast cancer. We seek to advance this line of inquiry through a broadened conceptual framework and accompanying statistical modeling strategy that recognize the dual importance of access-tocare and biologic factors on stage. Methods: The Centers for Disease Control and Prevention-sponsored Breast and Prostate Cancer Data Quality and Patterns of Care Study yielded a seven-state, cancer registry-derived population-based sample of 9,142 women diagnosed with a first primary in situ or invasive breast cancer in 2004. The likelihood of advanced-stage cancer (American Joint Committee on Cancer IIIB, IIIC, or IV) was investigated through multivariable regression modeling, with base-case analyses using the method of instrumental variables (IV) to detect and correct for possible selection bias. The robustness of base-case findings was examined through extensive sensitivity analyses. Results: Advanced-stage disease was negatively associated with detection by mammography (P < 0.001) and with age < 50 (P < 0.001), and positively related to black race (P = 0.07), not being privately insured [Medicaid (P = 0.01), Medicare (P = 0.04), uninsured (P = 0.07)], being single (P = 0.06), body mass index > 40 (P= 0.001), a HER2 type tumor (P < 0.001), and tumor grade not well differentiated (P < 0.001). This IV model detected and adjusted for significant selection effects associated with method of detection (P=0.02). Sensitivity analyses generally supported these base-case results. Conclusions: Through our comprehensive modeling strategy andsensitivity analyses,weprovide newestimates of themagnitude androbustness of the determinants ofadvanced-stage breast cancer. Impact: Statistical approaches frequently used to address observational data biases in treatment-outcome studies can be applied similarly in analyses of the determinants of stage at diagnosis.