Integration of specific copy number imbalances and prediction models of early-stage breast cancer recurrence.

Abstract

1529 Background: DNA Copy number alterations (CNAs) are common in breast cancers. Their role as molecular predictors of disease recurrence risk is less clear. Methods: We conducted CN analyses using high- density molecular inversion probe (MIP) arrays in 971 tumors early stage breast cancers. A binning strategy and CoxBoost algorithm were used to reduce the CN data to chromosomal segments relevant to recurrence and applied to a multivariate model using a randomly selected training set (n = 728). A complete Cox proportional hazards risk prediction model was built including clinicopathological characteristics, intrinsic tumor subtype (approximated by immunohistochemistry methods) and CNA markers and validated independently in 243 cases. Validation in the independent sample set included comparisons to known clinical models of breast cancer recurrence that contained the intrinsic subtypes. Separate models for triple-negative and luminal breast cancers were also explored. Results: The dimensionality of the CNA data were reduced with circular binary segmentation and fusing of the segments across samples to 1593 common CNA segments across the genome. Following a step-wise model selection in training sample, age at diagnosis, tumor size, lymph node status, approximated tumor subtypes, and 19 CNA markers were retained in a full model for time to recurrence. The C-index for the full model (clinical + tumor subtype + 19 CNA segments) gave the strongest predictive estimate for recurrence (C-index = 0.71 ± 0.035) compared to the best clinical+ intrinsic subtype model (C-index = 0.62 ± 0.041). Models built for luminal and triple negative tumors incorporating CNAs significantly outperformed models with clinical characteristics alone. This was particularly true for triple negative tumors, where the combination of CNA markers + clinical co-variates yielded a strong predictive estimate for recurrence (C-Index = 0.79 ± 0.023) compared to the clinical model (C-Index = 0.64 ± 0.034). Conclusions: Measures of specific CNAs add significant additional predictive information for risk of recurrence and suggest additional opportunities to refine prognostication for early stage luminal and triple-negative breast cancers. No significant financial relationships to disclose.