A Prognostic Model Study Of Intermediate And High Risk Gastrointestinal Stromal Tumors Based On Two Regression Methods

Objective:Based on a comparative study of COX proportional risk regression and LASSO regression methods,we constructed an ideal prognostic model for patients with moderate-to-high risk gastrointestinal stromal tumors and developed an effective column line graph and risk stratification to achieve precise medical treatment for patients with moderate-to-high risk gastrointestinal stromal tumors.Methods:The clinical data of 370 patients with gastrointestinal stromal tumors admitted to the First Affiliated Hospital of Nanchang University from January 2012 to December 2021 were retrospectively analyzed.All relevant variables were included in a one-way COX proportional risk regression model to initially screen for risk factors associated with survival and recurrence.Statistically significant variables were incorporated into multifactorial COX regression and LASSO regression methods to construct COX proportional risk models and COX-LASSO regression models,respectively,for comparative studies to screen for a more stable and reliable prognostic model for patients with moderate-to-high risk gastrointestinal stromal tumors,which was visualized by column line plots.ROC curves,calibration curves,clinical decision curves and risk stratification plots of the ideal prognostic model were applied to R software to assess the predictive power and degree of clinical benefit of the model.Validation sets were performed for internal validation of the prediction models.Results:Univariate COX proportional risk regression analysis yielded the variables associated with overall survival as age,mode of surgery,risk classification,Ki-67 index,nuclear division count,cell morphology,and targeted therapy;and those associated with recurrence-free survival as age,mode of surgery,primary site of tumor,risk classification,degree of tumor eradication,Ki-67 index,nuclear division count,cell morphology,and targeted therapy.In the multifactorial COX proportional risk regression model,the results showed that age,risk classification,cell morphology,and targeted therapy of GIST patients were independent prognostic factors affecting OS of GIST patients with moderate to high risk;age,risk classification,degree of surgical radicalization,Ki-67 index,cell morphology,and targeted therapy of GIST were independent prognostic factors affecting RFS of GIST patients with moderate to high risk.The results of the LASSO regression model showed that age,risk classification,surgical method,nuclear division count,and targeted therapy were independent prognostic factors for overall survival of GIST patients;age,risk classification,degree of surgical eradication,surgical method,nuclear division count,cell morphology,and targeted therapy were independent prognostic factors for recurrence-free survival of GIST patients with intermediate to high-risk GIST.In this study,the COX proportional risk regression model was defined as the old model and the COX-LASSO regression model was defined as the new model.Although both models performed well in terms of predictive performance,the COX-LASSO regression model had a greater C-index with NRI and IDI greater than 0 in overall survival and recurrence-free survival,and therefore,the COX-LASSO regression model was improved and more stable and reliable relative to the COX proportional risk regression model.The C-index for the training dataset was 0.809(0.727-0.891)for overall survival and 0.816(0.738-0.894)for relapse-free survival.The internal validation dataset had a C-index of 0.723(0.474-0.972)for OS and 0.654(0.474-0.834)for RRS,which had a good discriminatory power.In addition,the calibration curves demonstrated the agreement between the predicted outcomes of RFS and OS at 5 years and the observed actual survival outcomes.statistically significant differences between the "high risk" and "low risk" groups.Conclusion:This study developed a new COX-LASSO regression model nomogram for the overall survival and recurrence-free survival of intermediate and high-risk gastrointestinal stromal tumors.The nomogram is stable and reliable,which will help clinicians manage appropriate postoperative adjuvant therapy for GIST patients,thus aiding better clinical decision making and maximizing patient benefit.