The Application Of Semi-parametric Time-dependent AUC Method In The Prognosis Of DLBCL

Objective:The aims of this study were to introduce the time-dependent AUC semi-parametric method to assess the prognostic value of IRF4/MUM1 expression in diffuse large B cell lymphoma(DLBCL)patients.This study is divided into two sections,The first section is meta-analysis,comprehensively and quantitatively investigating the relationships between IRF4/MUM1 expression and pathological characteristics of DLBCL.The other is method application,using the time-dependent AUC semi-parametric method to evaluate whether the addition of one tumor marker(IRF4/MUM1 protein overexpression)improves the predictive accuracy of IPI with DLBCL.Methods:We performed a systematic electronic search in Pubmed,WanFang and CNKI for the studies that analyzed associations between IRF4/MUM1 and prognosis in DLBCL patients before December 2016.The hazard ratio(HR)estimates between elevated and normal groups of IRF4/MUM1 were derived or calculated.Fixed effects models were successively used to synthesize the HRs.The subgroup analysis and the sensitivity analysis were conducted to evaluate how the pooled effect would be changed by dividing into two groups or by eliminating the inferior quality of HRs,respectively.In this paper,we introduced the time-dependent AUC semi-parametric method to estimate AUC(t)as a function of time t using a flexible fractional polynomials model,without the middle step of modeling the time-dependent ROC.We develop a pseudo partial-likelihood procedure for parameter estimation and provide a test procedure to compare the predictive performance between biomarkers.Results:A total of 19 articles were included in the meta-analysis.The cumulative number of patients with DLBCL was 1699.Heterogeneity test showed that homogeneity between studies exists(2X =24.79,df =18,P =0.131,2I =27.4%).The pooled hazard ratio was 2.04(95%CI:1.75-2.3)and the overall effect were confirmed to be statistically significant(Z =9.24,P <0.001).Subgroup and sensitivity analysis also revealed that IRF4/MUM1 was an important hazard factor to DLBCL patients.A total of 315 DLBCL patients from Taiyuan,Shanxi,China,participated in a prospective cohort study between January 2014 and September 2016.Six risk factors were identified as tumor maker-adjusted IPI(T-IPI)at diagnosis,and a maximum of 6 points were assigned to each patient.The semi-parametric time-dependent AUC method to estimate AUC(t)curves and their 95% confidence bands for the IPI and T-IPI.The value of the AUC(t)of T-IPI is above IPI,implying that T-PI has the larger probability of a patient who has a progression of DLBCL on day t than IPI.To determine whether such advantage is statistically significant when accounting for the estimation variation,we tested the difference in the IAUCs between the IPI and T-IPI.The estimated difference in the IAUC is 0.138,with a standard error of 0.059,which leads to the conclusion that T-IPI has a significantly better prediction accuracy than that of IPI,with a p-value of 0.026.Conclusion:The findings of our method provide reliable evidence that IRF4/MUM1 expression may be strongly correlated with pathological characteristics of DLBCL.Combining IRF4/MUM1 with IPI may help to improve the prognostic accuracy of DLBCL.