Study on survival prediction model of esophageal cancer based on random survival forests
Objective This study aimed to construct a survival prediction model for esophageal cancer using the random survival forests and traditional Cox regression to explore prognostic factors and evaluate the model's performance.Methods Clinical records of esophageal cancer patients diagnosed between January 1,2010,to December 1,2019,at the affiliated cancer hospital of Xinjiang Medical University were collected for a follow-up study,with the follow-up period being from January 1,2010,to December 31,2020,or patient death.A random survival forests model was established and compared with the traditional Cox regression model.The model predicted 1-year,3-year,5-year,and 10-year survival rates,and prediction accuracy was evaluated using error rates and Brier scores.To find the important factors influencing the prognosis of esophageal cancer,discover the interaction between factors and evaluate the model effect.R 3.6 and SPSS 21.0 software were used for data analysis.Data screening and model building and evaluation were performed using the program packages survival,rms,and randomForestSRC.Results Out of 3 018 esophageal cancer patients,623(20.64％)experienced an outcome event.Among censored data,1 309(54.66％)died from other causes,while 1 086(45.34％)remained alive at the end of the observation period.The median survival time was 1.41(1.30,1.52)years,with corresponding survival rates of 60.30％,33.34％,27.40％,and 19.08％at 1,3,5,and 10 years respectively.Clinical stage,age,distant metastasis,lymph node metastasis,and depth of infiltration were identified as important influencing factors in the random survival forests model.The prediction error rates for the model at 1,3,5,and 10 years were 0.384,0.385,0.347,and 0.320,which were lower than the Cox regression model and had higher accuracy.Conclusions The random survival forest model effectively predicted mortality rate for esophageal cancer patients outperforming the traditional model,which can provide a scientific basis for the improvement of prognosis and quality of life of esophageal cancer patients.
Esophageal cancerRandom survival forestsCox proportional hazards modelSurvival analysisPrediction model
NETL
NSTL
万方数据
