EPID-25GERMLINE POLYMORPHISMS IN MGMT INCREASE ABILITY TO MODEL TEMOZOLOMIDE (TMZ)-RELATED MYELOTOXICITY RISK IN PATIENTS WITH GLIOBLASTOMA (GBM) TREATED ON NRG ONCOLOGY/RTOG 0825

BACKGROUND: Even though TMZ is well tolerated by most GBM patients, some experience severe myelotoxicity, resulting in treatment delay or cessation. Our ultimate goal is to develop robust risk prediction models for toxicity, which incorporate both clinical and genetic variables. Predicting risk of toxicity would allow for treatment optimization: lower dosing for at-risk patients; full dose for low-risk patients. METHODS: Logistic regression with backward variable selection (p ≤ 0.05) was used to select the best fitting clinical model. A genome-wide analysis of genetic variation associated with myelotoxicity in glioblastoma patients from RTOG 0825 was then conducted. The top hits were then incorporated into the risk modeling strategy to increase our ability to predict toxicity. Tagging variants were selected and examined individually and jointly (by summing the number of adverse genotypes). RESULTS: Clinical and genotype data from 367 cases were included. The final clinical model included the following variables: treatment arm, gender, Karnofsky performance status (KPS), anticonvulsant use, and body surface area (BSA). The area under the curve (AUC) for the clinical model was 0.69. The genome scan revealed ten polymorphisms in MGMT that were significantly associated with myelotoxicity (p < 10−8). The most significant variant in the genome scan resulted in the best model (AUC = 0.80) for an individual SNP. Individuals with the risk allele in this SNP exhibited a 5.3-fold increase in risk of myelotoxicity compared to those with the non-risk allele (p < 0.0001, 95%CI: 3.2-8.7) after adjusting for the clinical variables and independent of survival. The final model with genetic dose performed equally as well as the model with the single MGMT SNP. CONCLUSIONS: TMZ myelotoxicity is a significant clinical issue. The addition of genetic variants in MGMT greatly improved the performance of our clinical risk model. Future work will continue to improve prediction by examining other genetic variants and potential interactions between variants.