Real-world validation of a Lynch syndrome screening model in a multi-ethnic Asian cohort with dMMR tumors.

169

Background: Current Lynch syndrome (LS) screening relies on mismatch repair (MMR) status. To optimize LS screening, we previously developed a 5-variable clinical nomogram (age, sex, personal/family history, and dMMR pattern). While originally trained on a Chinese dMMR colorectal cancer (CRC) cohort, its friendly design suggests immense potential for broader utility. This study aims to establish this nomogram as a universal pan-cancer screening tool by evaluating its real-world cross-ethnic generalizability and tumor-specific performance. Methods: We applied our nomogram to a real-world, multi-ethnic cohort (Chinese, Malay, Indian, others) of 283 patients with dMMR CRC and endometrial cancer (EC) undergoing germline testing at the National Cancer Centre Singapore. The model's inherent predictive performance was initially assessed using its predefined 0.435 cut-off. For the EC cohort, a high-sensitivity optimized threshold was explored to minimize missed diagnoses. Results: In the CRC cohort, our nomogram demonstrated robust discrimination, yielding an overall sensitivity of 78.5% and a specificity of 84.7% at the 0.435 cut-off, highly consistent with its original training performance (sensitivity 71.6%, specificity 88.9%). Notably, sensitivity increased to 84.2% in non-Chinese patients, translating to fewer missed diagnoses and highlighting the model's exceptional cross-ethnic generalizability. Conversely, applying this CRC-derived cut-off to the EC cohort resulted in a compromised sensitivity of 59.52% and specificity of 75.31%, reflecting distinct clinicopathological characteristics. Recognizing that screening tools must prioritize capturing true positives, we established an EC-specific optimized cut-off of 0.175. This pivotal adjustment successfully elevated the sensitivity to >90%, effectively mitigating the risk of missed LS diagnoses. Conclusions: Our 5-variable nomogram is a highly robust and user-friendly LS risk stratification tool for multi-ethnic CRC populations. To optimize the model for LS screening in endometrial cancer, we recommend adjusting the cut-off to 0.175 to pursue higher sensitivity, though further validation in larger cohorts is warranted.

Validation of the LS screening nomogram across tumor types and ethnicities.