A varying-coefficient cox model for the effect of CA19-9 kinetics on overall survival in patients with advanced pancreatic cancer

PURPOSE To evaluate the relationship between serum CA19-9 and overall survival in patients with advanced pancreatic cancer. METHODS 109 advanced pancreatic cancer patients with gemcitabine based first-line chemotherapy were included. The effect of pretreatment CA19-9 level on overall survival was modeled by Cox proportional hazard regression. The effect of CA19-9 kinetics on overall survival was modeled by an extended Cox regression with a time varying coefficient and a time varying covariate. RESULTS Univariate analysis indicated that baseline CA19-9 correlated with OS (HR = 1.66, p < 0.01) and this association remained significant within multivariate analysis (HR = 1.56, P < 0.01). For the analysis of CA19-9 kinetics, the extended Cox model showed that the effect of CA19-9 on overall survival changed with time: increased in the first two months and reached the top at a HR of about 2, then decreased for the next two months to a HR of about 1.56 and finally tended to be stable. The combination of pretreatment CA19-9 and CA19-9 at 2 month may better evaluate the patients' prognosis compared to pretreatment CA19-9 alone. CONCLUSION Pretreatment CA19-9 and CA19-9 kinetics may serve as a useful serum biomarker in advanced pancreatic cancer.