P5-13-17: Multigene Signature Assays in Patients with Early-Stage Breast Cancer (ESBC) Receiving Neoadjuvant Chemotherapy: An NCI-Funded Systematic Review and Evidence Summary of Predictive Performance.

Abstract

Abstract Background: A comprehensive literature search and evidence synthesis of multigene signatures predictive of response to systemic chemotherapy in patients with breast cancer was initiated as a part of an NCI-funded program on Comparative Effectiveness Research. Methods: Validation studies were sought of multigene signatures for prediction of chemotherapy response (favorable vs unfavorable) in ESBC patient cohorts different from those used for signature development. Pooled estimates [±95% CI] of assay performance for predicting clinical outcome included sensitivity, specificity, likelihood ratio, predictive value (PV) and predictive odds ratio (POR) utilizing mixed effects models based on the method of Mantel-Haenszel. Exploratory metaregression analyses on log (POR) were also performed. Studies were classified by validation type including cell lines to patients, independent internal sample, random split sample, or external validation. Evidence for publication bias was assessed by Egger's regression intercept and Begg and Mazumdar's rank correction. Results: Dual-blind review of abstracts identified 33 studies of neoadjuvant chemotherapy response of which 29 stratified treatment response by signature classifier category. Classifier development was based on tumor response prediction in 20 studies, prognosis in 5, and molecular classification in 4. The Table shows assay performance measures overall and by study validation type. Assay performance based on the POR was positively associated with overall study quality (P=.015) and journal impact factor (P=.020). However, strong evidence for publication bias was observed based on both regression intercept (P<.001) and rank correlation (P=.005). No significant differences in assay performance were noted for assays originally developed for response prediction (POR=5.3), prognosis (POR=6.6) or molecular classification (P=6.9) (P=.770). Conclusions: While assay performance in predicting response to neoadjuvant chemotherapy based on multigene classifiers is encouraging, a compelling need exists for greater methodologic rigor and standardization of reporting. The predictive performance of multigene assay signatures varies with the type of validation sample utilized with external validation providing the most conservative estimates. No differences were seen for assays developed for prediction, prognosis or molecular classification. Considerable evidence for publication bias exists reflecting a paucity of smaller negative studies. The clinical validity of genomic response prediction assays should be evaluated in patient cohorts independent of those utilized for signature development. The clinical utility of these assays must then be further assessed in comparative effectiveness studies compared to commonly utilized clinical and laboratory measures. Funding: NCI: UC2CA14041-01 Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P5-13-17.