Development of novel models for predicting mismatch repair protein deficiency and relevant disease-free survival in colorectal cancer patients.

ABSTRACT

PURPOSE: DNA mismatch repair (MMR) protein deficiency has attached more attention for its potential to be a biomarker of immunotherapy for colorectal cancer (CRC) patients. However, clinical models involving the expression status of MMR protein are rare. Herein, we sought to develop two clinical models (a diagnostic model for the prediction of MMR status and a prognostic model for the prediction of disease-free survival) for CRC patients. METHODS: A total of 582 CRC patients were finally included. There were 53 patients with deficient expression of MMR protein. The differences between the deficient MMR (dMMR) group and the proficient MMR (pMMR) group were analyzed. RESULTS: Compared to pMMR patients, those with dMMR status were younger and had better pathological features (depth of invasion, lymph node metastasis, distant metastasis, pathological stage, perineuronal invasion, and PLT level) and disease-free survival (DFS). The tumor location of the left colon, adenocarcinoma, and abnormal PLT level were identified as the independent predictors for pMMR. Based on these data, we developed the diagnostic model using Logistic regression analysis. It showed a satisfactory accuracy (AUC = 82.3% in the derivate set; AUC = 73.6% in the validation set). Furthermore, pMMR, poorer differentiation, perineuronal invasion, distant metastasis, lower hemoglobin level, and abnormal CEA level were established as the independent prognostic factors of poorer DFS. Based on them, a prognostic model with valuable performance (1-year AUC = 75.5%/3-year AUC = 76.9% in the derivate set; 1-year AUC = 72.3%/3-year AUC = 73.8% in the validation set) was developed. CONCLUSIONS: Our diagnostic and prognostic models could identify CRC patients at risk for pMMR protein expression and disease recurrence. It may contribute to improving the diagnosis and treatment of CRC patients at an individual level.