BackgroundOur objective was to investigate why early studies regarding adoption of the 21-gene recurrence score (RS) assay did not show an initial reduction in the number of patients with breast cancer receiving real-world chemotherapy. Materials and MethodsWe addressed 2 sources of confounding suspected in previous studies: (1) the early time frame during the initial adoption phase of the RS assay, and (2) suspected selective, increased administration to patients more likely to have been chemotherapy candidates. To address selective use during initial adoption, we used updated SEER-Medicare data from 2004 and 2011. To address individual selection bias, we examined whether RS test utilization was negatively associated with rates of local chemotherapy use assessed at the hospital referral region level using conventional ordinary least squares and instrumental variable approaches to adjust for selection bias. ResultsA total of 26,009 patients met inclusion criteria. Assay use was associated with a decrease in absolute percentage use of chemotherapy of 4.5% (95% confidence interval [CI], 3.2%-5.7%), which was even more pronounced in sensitivity analyses limited to later study years (2008-2011), with a decrease of 6.8% (95% CI, 5.3%-8.3%). Instrumental variable models yielded similar point estimates but were insufficiently powered to draw conclusions. ConclusionReceipt of the 21-gene assay was associated with decreased utilization of chemotherapy by 2008.