Lynch syndrome testing in ethnically diverse patients: Mutation spectrum and performance of prediction models.

Abstract

1546 Background: Lynch syndrome (LS) is characterized by deficiency in DNA mismatch repair (MMR). Several models can predict the probability of MMR gene mutations, but all were derived from populations in Northern Europe and America. We aimed to characterize MMR mutations and to evaluate the performance of 4 models (Leiden, MMRpredict, MMRpro, and PREMM1,2,6) in an ethnically diverse high-risk population. Methods: Mutation and pedigree data from 387 distinct probands tested for germline MMR mutations at a single center between 2003-2012 were analyzed. Mutation testing was triggered by suggestive tumor microsatellite results and/or clinical/family history. Race/ethnicity was self-reported: 84 (22%) were ethnically-diverse (African American, 19; Hispanic, 29; Asian/Arabic, 34; and other, 2) and 303 (78%) were white. Area under the receiver operating curve (AUC) for all 4 models was analyzed by logistic regression. Results: Pathogenic mutations were detected in 152 patients (39%; MLH1 in 37, MSH2/EPCAM in 84, MSH6 in 21, and PMS2 in 10). The detection rate was significantly lower in ethnically-diverse patients when compared to white patients (30 vs. 42%; p=0.04). The mutation frequency of the major MMR genes MLH1 and MSH2 differed, although the overall mutation distribution did not reach statistical significance (MLH1, 40 vs. 22%; MSH2: 36 vs. 59%; MSH6: 20 vs. 13%; and PMS2: 4 vs. 7%; p=0.10 vs. white). The Leiden, MMRpredict, MMRpro, and PREMM1,2,6 models trended toward inferior discrimination for the ethnically diverse patients, with AUCs of 0.63, 0.64, 0.63, and 0.66 respectively (vs. AUCs of 0.71, 0.69, 0.72, and 0.76 for white patients), but the difference was not statistically significant(p=0.11, 0.17, 0.09, and 0.08, respectively). Adjusting for family size did not significantly alter results. Conclusions: In ethnically-diverse patients, the overall detection rate for pathogenic MMR mutations is significantly lower, with associated differences in mutation frequencies and in predictive model performance. We caution against relying solely on predictive scores, as high-risk ethnically-diverse patients without an identified mutation may deserve to be followed as clinical LS patients.