Using Multi-failure Models To Evaluating The Effects Of Time-varying Covariates On The Prognosis Of Patients With Gastric Cancer

Background:Currently, analyses of prognostic factors for patients with gastric cancer are usually utilizing the baseline information such as tumor biomarkers, inflammatory cytokines and albumin measured at the entry of the subjects. However, most of these reflect the current status of the tumor and are closely related to the baseline clinical stage of the patients. What’s more, many factors that have impact on the prognosis of cancer patients would vary as the time passing by, even with no intervention. If the baseline exposure level of a time-varying factor is related to the prognosis of the cancer patients, we suppose that the exposure levels in different follow-up periods are also associated with the prognosis. If the exposure level of the factors can be changed by certain kinds of intervention, we want to know about the relation between prognosis and the exposure levels in the follow-up period.Objective:We will investigate the statistical characteristics of multiple failure survival models when dealing with the time-varying covariates. We will use this model to evaluate the factors which may have impact on the prognosis of the patients with gastric cancer and compare the results with these of other generally used models. Meanwhile, we will compare the survival status and related factors between those progressing after surgery and those diagnosed with advanced GC at beginning. We will also investigate the relation between the survival time before and after disease progression.Methods:Taking the existing data into consideration, we generate data to simulate oncology trials under different situation to evaluate the the predictive effects of the multi-level survival model with time-varying covariates and utilize this model to investigate impact of systemic inflammatory cytokines (eg. CRP), albumin, NLR, PLR as well as tumor markers in different follow-up periods on the PFS and OS time, and compare with that of the baseline information. Besides, use the general Cox model to compare the survival status and related factors between those progressing after surgery and those diagnosed with advanced GC at beginning. We will also investigate the relation between the survival time before and after disease progression by redefining a new survival time. In addition, we use Pearson correlation to investigate the relations between different types of laboratory test results (eg. CD19+, CEA, CA199) in different follow-up periods.Main results & Conclusion:The coverage rates of 95%CI of the multi-failure survival model are around 95%, and the coverage rates of 95%CI of the multi-failure survival model are not greatly affected by the sample size.Afer controlling for the factors including age and gender, the higher level of CRP leads to the greater hazard for disease progression. In patients who have received radical surgery, HR=1.062 (95% CI:1.011,1.115), while in patients who have not received radical surgery, HR= 1.038 (95% CI:1.025,1.052). The higher level of albumin leads to the smaller hazard for disease progression. In patients who have received radical surgery, HR= 0.908 (95% CI:0.825,0.998), while in patients who have not received radical surgery, HR= 0.948 (95% CI:0.914,0.982).Patients with longer PFS time have significantly smaller hazard for disease progression. One month extension of PFS would lead to 11% reduction of hazard for death after disease progression, and this would not be significantly affected by the application of different chemotherapies before.PLR, NLR, AST, cyfra21-1, albumin and total protein show statistical significance and similar HR in both multi-failure survival models and general Cox model with single outcome, which indicates that not only the baseline but also the follow-up information of these indicators is closely related with the prognosis of patients with gastric cancer. The multi-failure survival model with different HRs for a covariate in different visits shows that the tumor status in the early stage and the overall body condition in the later stage may have close relation with the prognosis of GC patients.