Construction of a prediction model for pain crisis in patients with advanced rectal cancer based on machine learning algorithms
ZHANG Lida
LIANG Sheng
NONG Shixiang
YUAN Jie
HUANG Lifang
Abstract：Objective:To construct a prediction model for pain crisis in patients with advanced rectal cancer based on machine learning algorithms and analyze the predictive performance of different models.Methods:A convenience sampling method was used to select 210 patients with advanced rectal cancer admitted to our hospital from September 2022 to September 2024 as the study subjects.Questionnaires were administered using the General Information Questionnaire,Social Support-Rating Scale,Connor-Davidson Resilience Scale,and Hospital Anxiety and Depression Scale.Based on whether patients experienced pain crisis,they were divided into a pain crisis group and a non-pain crisis group.Univariate and multivariate analyses were conducted to identify influencing factors of pain crisis.Prediction models were constructed using Logistic regression,random forest,and decision tree algorithms based on the univariate and multivariate analysis results.The receiver operating characteristic(ROC)curve and the area under the curve(AUC)were used to evaluate model efficacy and predictive value.Results:Among the 210 patients with advanced rectal cancer,64(30.48%)experienced pain crisis.Multivariate analysis showed that social support,psychological resilience,negative emotions,age,monthly household income per capita,and the number of radiotherapy/chemotherapy sessions were independent influencing factors for pain crisis in patients with advanced rectal cancer(all P<0.05).ROC curve analysis revealed that the AUC values for the Logistic regression model,decision tree model,and random forest prediction model were 0.902,0.901,and 0.933,respectively.The accuracy rates were 0.881,0.852,and 0.889;sensitivity rates were 0.750,0.734,and 0.824;specificity rates were 0.938,0.904,and 0.913;recall rates were 0.750,0.734,and 0.824;precision rates were 0.842,0.770,and 0.933;and F1 scores were 0.793,0.752,and 0.875,respectively.Except for specificity,the random forest model achieved the highest values in AUC,accuracy,sensitivity,recall,precision,and F1 score,demonstrating the best overall performance.Conclusions:The random forest model demonstrates superior predictive performance for pain crises in advanced rectal cancer patients compared to Logistic regression and decision tree models.Clinically,this model can help identify high-risk patients for early intervention with preventive measures,thereby reducing the incidence of pain crises.
Keywords：advanced rectal cancerpain crisismachine learning algorithmsprediction modelsocial supportpsychological resilience
Publication Date：2025-09-10
Online Publishing Date：2025-09-12(First online date of this platform, not the publication date of the document)
Pages：8( 2900-2907 )