Prognostic correlation analysis of colorectal cancer patients based on monocyte to lymphocyte ratio and folate receptor-positive circulating tumor cells and construction of a machine learning survival prediction models.

PURPOSE

To evaluate the prognostic value of the monocyte to lymphocyte ratio (MLR) and folate receptor-positive circulating tumor cells (FR+CTCs) in patients with colorectal cancer (CRC) and to develop predictive model for post-treatment survival using machine learning (ML) algorithms.

METHODS

We retrospectively analyzed 67 CRC patients treated with radical surgery or chemoradiotherapy at The Central Hospital of Wuhan from January 2020 to December 2022. MLR, neutrophil to lymphocyte ratio (NLR), platelet to lymphocyte ratio (PLR) and FR+CTCs were categorized into high and low groups and clinicopathologic features were compared. Progression-Free Survival (PFS) and Overall Survival (OS) were analyzed using COX analysis and the Kaplan-Meier survival curve. Three ML algorithms, namely, random forest (RF), support vector machine (SVM), and logistic regression (LR), were utilized to construct the predictive models, and their performance metrics including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), precision, recall, F1 value, AUC, and calibration curve were compared.

RESULTS

MLR, FR+ CTCs, and T stage independently predicted PFS (P<0.05), both higher MLR and FR+CTCs levels indicating a significantly shorter PFS (P=0.004). The T stage was the only factor predictive of OS (P=0.043). NLR and PLR did not show significant prognostic effects on PFS or OS (P > 0.05). The RF model demonstrated superior performance with an accuracy of 0.63, sensitivity of 0.69, PPV of 0.75, a precision of 0.43, a recall of 0.5, and an F1 value of 0.43, outperforming the other models.

CONCLUSION

High MLR and high FR+CTCs are associated with a poorer PFS in CRC patients, suggesting their utility in prognostic assessment. NLR and PLR did not show significant prognostic value in this study. The RF algorithm-based model showed the best predictive performance for post-radical treatment outcomes in CRC.