Background: Deficient mismatch repair (dMMR) and microsatellite stability (MSS) are key molecular subtypes in colorectal cancer (CRC). They have distinct clinical traits, prognoses, and immunotherapy responses. Accurate identification between dMMR and MSS, along with prognosis and immunotherapy efficacy prediction, is vital for personalized treatment planning.
Methods: This study aimed to construct a machine learning-based prediction model to assist in screening dMMR and MSS of CRC patients and predicting their prognosis and the efficacy of immunotherapy. 420 CRC patients’ data were retrospectively analyzed. The least absolute shrinkage and selection operator (LASSO) regression was used to screen out the features. Then, we adopted the random forest algorithm to construct the prediction model and used 10-fold cross-validation to evaluate the performance of the model.
Results: Among the 420 cases, 44 showed a high MSI. We successfully constructed a prediction model that can accurately distinguish between dMMR and MSS, with an accuracy rate of 92.5%, a sensitivity of 90.2%, and a specificity of 94.8%. In addition, we also constructed models that can predict the prognosis of patients and the efficacy of immunotherapy, with their C-indexes being 0.85 and 0.78 respectively. These models also demonstrated good predictive performance on the independent validation set.
Conclusion:The prediction models established can efficiently aid in the screening of CRC patients with dMMR and MSS, as well as accurately predict their prognoses and the effectiveness of immunotherapy. Furthermore, these models not only offer robust decision - making support but are also primed to drive the significant progress of precision medicine.