Establishment And Validation Of Prognostic Model For Primary Gastrointestinal B-cell Lymphoma Based On SEER Database

Part I.Construction and internal validation of a predictive model for primary gastrointestinal B-cell lymphomaObjective: To construct a predictive model of primary gastrointestinal B-cell lymphoma(PGBL),followed by internal validation,to accurately predict overall survival OS and cancer-specific survival CSS in patients with PGBL.Methods: In this study,5740 patients with a diagnosis of PGBL confirmed by endoscopic or post-operative pathological biopsy were collected from the SEER database,randomly assigned using the verification package in the R language,and divided into modelling and internal validation groups in a ratio of 7:3.The Kaplan-Meier(KM)method was used to establish survival curves,log-rank sum tests(long-rank)to compare whether there were differences between the groups for the variables we collected.Independent predictors were obtained using the COX regression model to create a line graph(Nomogram)that could predict OS and CSS at1,3 and 5 years in PGBL patients.The predictive performance of the Nomogram was assessed using the Harrell consistency index(C-index),decision curve analysis(DCA),and calibration curve,validated internally,and then compared with the Ann Arbor staging system.Results: The Kaplan-Meier(KM)method was used to establish survival curves,and the log-rank test(long-rank)was used to compare whether there was a difference between the groups to obtain this result,where the senior age ≥75 years(P<0.001,HR=5.1299),Ann Arbor stage III(P<0.001,HR= 1.4843),and stage IV(P<0.001,HR=1.867)groups had a poorer prognosis.In contrast,women(P<0.001,HR=0.8121),with site of disease in the small intestine(P=0.003,HR=0.8351),undergoing radiotherapy(P=0.011,HR=0.842),undergoing chemotherapy(P<0.001,HR=0.7038),FL(P<0.001,HR=0.3303),MALT(P= 0.003,HR=0.4885),and married(P<0.001,HR=0.7186)patients had a better prognosis.Age size at diagnosis,race,gender,primary site of lymphoma,Ann Arbor stage,type of pathology,chemotherapy and marital status were obtained using multifactorial COX regression analysis and were associated with prognosis of OS.The risk factors independently associated with CSS prognosis included age,gender,race,clinical stage,type of pathology,chemotherapy and marital status,and the column line graphs were constructed using these factors,with C-index values of 0.735(95% CI = 0.719-0.751)and 0.761(95%CI = 0.739-0.783).In addition,the calibration curves showed that the model predicted values were similar to the actual values.The column line plots created in this study were compared with the C-index and DCA results of the Ann Arbor staging system and it was found that the DCA curves of our constructed model were further away from the two baselines than the Ann Arbor staging system.Conclusion: In this study,a prediction model for OS and CSS at 1,3 and 5 years for patients with PGBL was developed and can be used to determine prognosis.Also the above prediction model demonstrated satisfactory predictive performance and clinical applicability,and it can assist us in assessing the prognosis of patients with PGBL.Part II.External validation of a predictive model for primary gastrointestinal B-cell lymphomaObjective: To externally validate the predictive model for OS and CSS in primary gastrointestinal non-Hodgkin’s lymphoma(PGBL)that has been previously developed in this topic.To clarify whether this prognostic model prediction has value.Methods: Forty-one patients with pathologically confirmed PGBL at My Hospital between January 2013 and December 2021 were collected.Relevant case information such as patient gender,age size,type of pathology,primary site,chemotherapy,surgery or not,survival status,survival time and marital status were collected for statistical collation and then externally validated,and the predictive performance and clinical utility of the prediction model were assessed using AUC and DCA,while the predictive ability of the Ann Arbor staging system for the prognosis of patients with PGBL was compared.Results: The 1-,3-and 5-year AUC values for OS and CSS in the external validation group of the prognostic model were 0.8,0.857,0.857 and 0.8,0.865,0.865,respectively,whereas the 1-,3-and 5-year AUC values for OS and CSS in the external validation group of the Ann Arbor staging system were 0.7,0.772,0.7722 and 0.7,0.77,0.769,respectively.The DCA results show that the predictive model developed in this study is further away from the two baselines compared to the Ann Arbor staging system.Conclusion: The OS and CSS prognostic models for PGBL patients previously developed in this study have higher specificity and sensitivity than the Ann Arbor staging system,and have wider clinical applicability.