The association of preoperative hematologic parameters with short‐term clinical outcomes in rectal cancer: A feature importance analysis

Abstract

Abstract Background Various hematologic parameters have been proposed as prognostic factors in rectal cancer management, but data are conflicting and unclear. This study is designed to investigate the prognostic factor capability of preoperative hematologic parameters with postoperative morbidities and mortality in rectal cancer patients undergoing curative resection. Methods All 200 consecutive rectal cancer patients diagnosed at Ghaem University Hospital from 2017 to 2022 were retrospectively evaluated. The receiver operating characteristic (ROC) curves and machine learning (ML) algorithms of Random Forest, Recursive Feature Elimination, simulated annealing, Support Vector Machine, Decision Tree, and eXtreme Gradient Boosting were administered to investigate the role of preoperative hematologic parameters accompanied by baseline characteristics on three clinical outcomes including surgical infectious complications, recurrence, and death. Results The frequency of infectious complications was correlated with the surgical procedure, while tumor recurrence was significantly influenced by T stage and N stage. In terms of mortality, alongside T and N stage, the status of resection margin involvement was significantly correlated. Based on the ROC analysis, the NLR >2.69, MPV ≤9 fL, and PDW ≤10.5 fL were more classified patients to mortality status. Likewise, the PLT >220 109/L, MPV ≤9 fL, PDW ≤10.4 fL, and PLR >13.6 were correlated with recurrence. However, all factors examined in this study were not significant classifiers for the outcome of surgical infectious complications. The results of ML algorithms were also in line with ROC analysis. Conclusion According to the results of both ROC analysis and ML models, preoperative hematologic parameters are considerable prognostic factors of postoperative outcomes in rectal cancer patients, and are recommended to be monitored by clinicians to prevent unfavorable outcomes.

Keywords

• clinical outcomes
• hematologic parameters
• machine learning
• prognostic factor
• rectal cancer