Prediction Of Progression-free Survival In Patients With Lung Cancer Based On RWD

Objectives:According to the ranking of cancer deaths,lung cancer ranks first in China.Relevant studies have shown that lung cancer is a malignant tumor that can metastasize in the early stage of onset.Most lung cancer patients have been in the relatively advanced stage when they are diagnosed.The progress of the disease is the main cause of death of lung cancer patients.The establishment of a localized prediction model for the prognosis of lung cancer patients will help doctors give accurate evaluation to patients,so as to formulate the corresponding individualized management plan.Due to the difficulty of obtaining image data and other reasons,at present,the prediction models for the prognosis of patients with lung cancer mostly focus on the prediction of the overall survival of lung cancer,while the prediction models related to the progression of lung cancer are relatively lacking.Therefore,this study attempts to obtain the real-world data（RWD）of local lung cancer patients,evaluate and sort out the image data,and explore the methodology of establishing the prediction model of lung cancer progression free survival through the method of machine learning.Methods: firstly,the relevant data of lung cancer inpatients from January 2015 to December 2018 were extracted from the hospital information system,and a standardized data set was established through a series of data preprocessing processes.Secondly,from the perspective of classification prediction model,the six-month disease progression model of lung cancer is established by using machine learning algorithm,and the performance of the model is evaluated.Then,from the perspective of survival analysis model,the prediction model of progression free survival in patients with lung cancer is established by using machine learning algorithm,and the performance of the model is evaluated.Finally,the main factors affecting the progression free survival of lung cancer are selected by machine learning algorithm and analyzed and discussed.Results: a total of 912 patients with lung cancer were included in this study.After data sorting,the relevant image data of patients were obtained,and 30 variables including demographic characteristics,oncological characteristics,comorbidity and routine blood tests were obtained.Then,the six-month disease progression model of lung cancer is established.After modeling the data through variable selection and machine learning algorithm,a total of 16 machine learning models are obtained.Among them,the best performance is the logistic regression model based on Lasso（AUC = 0.713）.Next,the prediction model of progression free survival of lung cancer is established by using the survival analysis model algorithm.The performance of the penalty COX model based on elastic network（C index = 0.664,average AUC= 0.725）is slightly better than that of the random survival forest model（C index= 0.649,average AUC = 0.700）.Through the survival analysis model,we screened the main factors affecting the progression free survival of lung cancer: TNM stage,platelet count,lymph node metastasis,lymphocyte rate and so on.Conclusion: it is feasible to use machine learning algorithm to predict the progression free survival of lung cancer patients,and the established prediction model has good performance.By establishing classification prediction model and survival analysis model respectively,this study systematically carried out methodological exploration related to progression free survival data analysis and prediction.The results will help to evaluate the prognosis of lung cancer population and provide reference for rational use of medicines and relevant intervention measures.