Performance of PREMMplus in a racially and ethnically heterogenous population undergoing germline multigene panel testing.

Abstract

e22640 Background: PREMMplus is a risk prediction model that estimates the probability of carrying pathogenic germline variants (PGVs) across multiple hereditary cancer susceptibility genes. Prior development and validation cohorts demonstrated high performance but were composed predominantly of non-Hispanic White individuals. PREMMplus performance in more heterogeneous clinical populations has not been well characterized. Methods: We evaluated PREMMplus performance in the Hereditary Cancer Panel (HCP) cohort, a large prospective clinical population that met clinical testing criteria and undewent germline multigene panel testing. Participants were recruited from three clinical sites, including a large safety-net hospital serving the greater Los Angeles area, resulting in substantial racial, ethnic, and socioeconomic heterogeneity. The parent study was designed to evaluate the safety, diagnostic yield, and clinical utility of multiplex genetic testing with standardized genetic counseling.. PREMMplus scores were assessed at the ≥2.5% cutoff. Performance was evaluated for PGVs in 11 Category A (APC, BRCA1/2, CDH1, EPCAM, MLH1, MSH2, MSH6, biallelic MUTYH, PMS2, TP53) and 8 Category B genes (ATM, BRIP 1, CDKN2A, CHEK2, PALB2, PTEN, RAD51C, RAD51D), using sensitivity (SE), specificity (SP), positive predictive value (PPV), negative predictive value (NPV), number needed to test (NNT), and area under the receiver operating characteristic curve (AUC). Subgroup analyses examined performance by race, ethnicity, and personal/family cancer history. Results: Among 1,896 participants, 41.5% self-identified as Hispanic and 11.9% as Asian. Overall, performance results were comparable to prior validation studies (Table). Compared with non-Hispanic individuals, Hispanic participants demonstrated lower specificity and discrimination (AUC 0.58 vs 0.68 for Category A genes). Sensitivity and PPV remained comparable across racial and ethnic groups. Performance differences were most pronounced among participants reporting a personal history of cancer but no family history. Conclusions: In a large, multi-site, heterogenous clinical cohort, PREMMplus retained high sensitivity for PGV detection but demonstrated reduced specificity and discrimination in Hispanic and other non-White populations. These findings highlight the importance of accurate family history ascertainment and the impact of under-reporting on the interpretation of risk prediction models in real-world clinical settings. PREMMplus performance (≥2.5% cutoff) by gene category and ethnicity. Gene category Cohort SE % SP % PPV % NPV % NNT AUC A Entire 88.3 19.6 6.9 96.1 14.5 0.64 Non-Hispanic 91.4 24.0 6.2 98.1 16.1 0.68 Hispanic 85.5 13.1 7.8 91.3 12.9 0.58 A + B Entire 86.6 19.7 9.7 93.6 10.3 0.59 Non-Hispanic 86.0 24.0 8.7 95.3 11.5 0.60 Hispanic 87.2 13.3 11.0 89.4 9.1 0.56