Trajectory Analysis And Dynamic Prediction Study Based On Perioperative Serum Tumor Markers Of Colorectal Cancer

BACKGROUND Colorectal cancer(CRC)is a malignant tumor of the digestive tract,ranking third in incidence and second in mortality worldwide.Currently,locally advanced CRC is mainly treated by radical resection.Some patients can be cured by surgery,however,30-50%of CRC patients experience recurrence after treatment,with poor prognosis.Early identification of patients with poor prognosis is the main goal of postoperative surveillance for CRC.And how to use perioperative tumor markers to achieve more individualized survival prediction than tumor-node-metastasis staging is a key clinical concern.Current guidelines for CRC recommend perioperative routine measurements of carcinoembryonic antigen(CEA)to monitor recurrence.However,the accuracy of the prognosis prediction using only CEA is limited,and it is necessary to combine other markers to improve the accuracy of prediction.The association of carbohydrate antigen 19-9(CA19-9)and carbohydrate antigen 125(CA125)with CRC prognosis has been extensively studied.However,whether CA19-9 and CA125 should be added for prognostic surveilliance of CRC remains controversial.In this study,based on a retrospective cohort,we collected the repeated measurements of perioperative CEA,CA19-9 and CA125 of CRC patients,to fit the perioperative trajectories of tumor markers,evaluate the prognostic value of the simultaneous measurements of CEA,CA19-9 and CA125,and construct prediction models to achieve dynamic prediction of CRC prognosis.METHODS CRC patients undergoing curative resection for stage I to III colorectal adenocarcinoma between January 2011 and February 2019 from three hospitals(Yunnan Cancer Hospital,Guangdong Provincial People’s Hospital,and the Sixth Affiliated Hospital of Sun Yat-Sen University)were retrospectively identified.Perioperative measurements of serum tumor markers and prognostic information of all patients were collected.The follow-up of this corhort ended on June 30,2020,with the primary end point being death and the secondary end point being recurrence.The latent class growth mixed model(LCGMM)was used to fit perioperative CEA trajectories and identify latent trajectory groups.Cox proportional hazards models were used to evaluate the associations of trajectory group memberships with overall survival(OS)or recurrence-free survival(RFS).The perioperative trajectories of CA19-9 and CA125 were further identified based on LCGMMs.Jointly,patients were divided into preoperative and trajectory joint groups to evaluate the prognostic value of CA19-9 and CA125 in addition to CEA,according to preoperative levels and dynamic trajectories of CEA,CA19-9 and CA125.A functional principal component analysis(FPCA)was used to extract trajectory patterns of perioperative CEA,CA19-9,and CA125.In consideration of the correlation between CEA,CA19-9,and CA125,multivariate principal components analysis(MFPCA)was applied to characterize the changing patterns of the multivariate longitudinal processes.Prediction models to predict CRC overall survival were constructed with demographic and clinicopathological variables,by incorporating preoperative CEA,CA19-9,and CA125 as well as their perioperative longitudinal measurements.The prediction performance of the prediction model was measured by area under the receiver operating characteristic curve(AUC),brier score(BS),net reclassification improvement(NRI)and integrated discrimination improvement(IDI)over time.Predictive models were constructed and validated with 10-fold crossvalidation using data from Yunnan Province.Data from Guangdong Province was used for external validation.RESULTS 2160 CRC patients with preoperative serum CEA and at least three postoperative serum CEA measurements taken within three years after surgery were included in the CEA trajectory analysis.Three distinct trajectory groups of perioperative CEA were indentified,labeled as low-stable(n=1680,77.78%),earlyrising(n=291,13.47%)and later-rising(n=189,8.75%).In the low-stable group,the CEA persisted at a low level during 36 months after surgery(0-5.0 ng/ml).In the earlyrising group,CEA declined rapidly from elevated preoperative level(>5.0 ng/ml)towards the normal range(0-5.0 ng/ml)within three months of surgery,then increased rapidly at 9 months after surgery.In the later-rising group,CEA declined rapidly towards the normal range(0-5.0 ng/ml)within three months of surgery,and kept stable up to 18 months after surgery,then increased slowly.Compared with the low-stable group,the hazard ratio(HR)and 95%confidence interval(CI)of death were 1.68(1.192.36)and 2.46(1.75-3.47)for the early-rising and later-rising group,and the corresponding HR(95%CI)of recurrence were 1.54(1.21-1.96)and 1.68(1.27-2.21),respectively.Serum CA19-9,and CA125 also had low-stable,early-rising,and later-rising trajectory groups,and the rising groups had higher risks of death and recurrence than the stable group.A total of 1974 patients were included for joint analyses of CEA,CA19-9 and CA125.Jointly,patients were classified into six preoperative(trajectory)groups:three-low(three-stable),CEA-high(CEA-rising),CA19-9-high(CA19-9rising),CA125-high(CA125-rising),two-high(two-rising)and three-high(threerising).Compared with the three-low group,the adjusted HRs(95%Cls)associated with death were 1.87(1.29-2.70),3.82(2.37-6.17),1.87(0.97-3.61),2.81(1.93-4.11),and 4.99(2.80-8.86)for the CEA-high,CA19-9-high,CA125-high,two-high,and three-high group,respectively.The trajectory joint group had prognostic significance after adjusting for preoperative levels.And compared with the three-stable trajectory group,the corresponding HRs(95%CIs)were 1.60(1.09-2.35),1.52(0.76-3.05),6.13(3.75-10.00),3.99(2.63-6.05),and 12.00(5.37-26.60)for the five rising trajectory groups,respectively.Similar associations between joint groups and RFS were observed.The model with preoperative CEA,CA19-9,and CA125 outperformed the model including CEA only,with the better AUCs(0.774 vs.0.716),BSs(0.057 vs.0.058),NRI of 33.5%(95%CI:12.3-54.8%),and IDI of 0.017(95%CI:0.005-0.029)at 36 months after surgery.Furthermore,by incorporating longitudinal measurements of CEA,CA19-9,and CA125 within 12 months after surgery,the prediction accuracy of prediction models improved,with higher AUC(0.849)and lower BS(0.049),NRI of 40.8%(95%CI:19.6-62.1%)and IDI of 0.071(95%CI:0.045-0.098)at 36 months after surgery.External validation shows similar results to internal validation.The AUCs at 60 months after surgery for the basic model,preoperative CEA model,preoperative CEA&CA19-9&CA125 model,longitudinal CEA model,and longitudinal CEA&CA19-9&CA125 model were 0.581,0.597,0.620,0.696,and 0.736.The relative importance of variables showed that the first and second principal components of the three tumor markers’ perioperative measurements were important prognostic predictors,second only to the pathological stage.The proposed longitudinal prediction model can provide individualized dynamic predictions for a new patient,with estimated survival probability updated when new measurements of the patinent obtained.CONCLUSIONS Perioperative CEA trajectories can provide important prognostic information independent of preoperative CEA levels,and close attention should be paid to patients with an upward trend of postoperative CEA.Simultaneous measurement of CEA,CA19-9 and CA125 will contribute to the identification of CRC patients with poor prognosis,and the predictive performance of CRC prognostic models improved with the incorporation of preoperative CEA,CA19-9,and CA125,as well as their perioperative longitudinal measurements.In addition to CEA,the dynamic measurements of CA19-9 and CA125 are recommended to monitor the prognosis of CRC patients.Based on the proposed longitudinal prediction model,dynamic prediction can be achieved using repeated measurements of tumor markers of a patient.