1559 Background: Tumor mutational burden (TMB), has been recently granted FDA approval as a biomarker for ICI treatment in tumors with a high mutation load (≥ 10/MB). To leverage this biomarker in the clinical setting, it is necessary to evaluate its application in diverse patient populations. In particular, tumor-only sequencing may overestimate TMB in non-EUR populations for which reference panels are small or unavailable, and thus have poorer predictive performance. Herein, we investigate the effect of TMB overestimation in non-EURs on patient diagnosis and clinical outcomes in a real world patient cohort. Methods: TMB was computed using a tumor-only NGS platform (Oncopanel) for 8349 cancer patients (pts) of 7 cancer types (Table). Genetic ancestry was inferred directly from tumor sequencing data and confirmed for a subset of pts using germline SNP arrays. TMB was compared using Wilcoxon rank-sum test between European (EUR) and non-EUR pts. TMB percentile rank by ancestral group (East Asian, African, European) and tumor histology was computed. In non-EUR pts, TMB was calibrated by reassigning its value to the corresponding percentile rank in EUR pts having the same cancer (AH-TMB). Tumors with raw TMB ≥ 10/MB were assigned as TMB-high (TMBH; with the rest referred to as TMBL for TMB-low) and those with calibrated AH-TMB ≥ 10/MB as AH-TMBH. A subset of pts treated with ICIs at DFCI and of EUR ancestry were analyzed. TMB was intentionally mis-calibrated in EURs (MC-TMB), to mimic the TMB overestimation observed in non-EURs. Associations between TMBH status and overall survival (OS) was assessed using Cox regression. Results: Uncalibrated TMB was significantly higher in tumors from non-EUR pts overall (p <0.0001, Table) as anticipated, whereas this difference was not observed in tumors with matched germline filtering. We reassigned non-EUR pts into low/high groups using calibrated AH-TMB, which changed the assignment of 67/670 non-EURs, most of which (65/67) were downgraded from TMBH to AH-TMBL. In the ICI-treated cohort, there was a strong association between TMBH status and OS (p = 2e-11, HR = 0.6) whereas MC-TMBH (mimicking the miscalibration in non-EUR samples) had a weaker association with OS (p = 0.0013, HR = 0.8). Importantly, pts assigned to TMB-H due to miscalibration (90 TMB-L/MC-TMBH pts) had shorter OS compared to the true TMBH group (p = 3e-6, HR = 1.9). Conclusions: In this analysis, we showed that tumor-only sequencing platforms can overestimate TMB in non-EUR pts, impacting treatment decisions and outcomes. Replacing raw TMB counts with an ancestry-adjusted measure could optimize TMB application in the clinical setting when germline sequencing is not available. Accurate ancestry inference can be performed using tumor-only sequencing.[Table: see text]