BIOM-53. MENINGIOMA COPY NUMBER VARIANT THRESHOLDS REVEAL OPTIMAL SIZES FOR GAINS OR LOSSES THAT PREDICT LOCAL FREEDOM FROM RECURRENCE OR OVERALL SURVIVA

Abstract

Abstract INTRODUCTION Meningiomas demonstrate heterogeneous rates of postoperative local freedom from recurrence (LFFR) and overall survival (OS). Risk-stratification models incorporating copy number variants (CNVs) improve outcome prediction, but optimal size thresholds for defining CNVs in meningiomas are unknown, and previous models have used thresholds ranging from 5-80% that are applied uniformly across all chromosome arms. Here we show that optimizing chromosome-specific CNV size thresholds improves meningioma risk-stratification. METHODS 565 retrospective meningiomas with clinical data and DNA methylation profiles from 2 international institutions were included. CNVs were defined from DNA methylation profiles using iterative thresholds ranging from 5-95% for each chromosome arm. The area under the curve (AUC) for 5-year LFFR or OS was calculated using CNVs across iterative thresholds for each chromosome arm. LASSO and Elastic Net Cox regressions were used to generate multi-CNV models for LFFR or OS. RESULTS AUC for LFFR or OS peaked with CNV thresholds that were unique to each chromosome arm. Chromosome 1p deletion had optimal AUCs for LFFR and OS of 0.72 and 0.70 at thresholds of 40% and 30%, respectively. Chromosome 1q amplification had optimal AUCs for LFFR and OS of 0.58 and 0.60 at thresholds of 25% and 20%, respectively. Two previously published CNV-based risk-stratification models (integrated score and integrated grade, optimal LFFR thresholds of 5% and 20%, AUCs of 0.77 and 0.78, respectively) demonstrated heterogeneous sensitivity to CNV thresholds. LASSO and Elastic Net Cox regression identified size-dependent risk-stratification models incorporating loss of 1p, 3q, 4p, 5p/q, 6p/q, 9p, 10q, 11p, 12q, 13q, 14q, 17q, and 18p/q, and gain of 1q, 9q, 10p, and 17q. The optimal AUCs for LFFR and OS from multi-CNV models were 0.77 and 0.77 at thresholds of 25% and 35%, respectively. CONCLUSIONS Interrogating CNV thresholds reveals optimal chromosome-specific CNV sizes for predicting LFFR and OS.