Methodological Exploration Of RWD-based Clinical Effectiveness Evaluation Of Anti-Lung Cancer Drugs

Objective: Malignant tumors have caused a severe disease burden to human society and patients’ families worldwide.Among them,lung cancer is the most common cause of death from malignant tumors in all regions of the country.However,there are a number of problems with current antineoplastic drug therapy and the need to evaluate the therapeutic efficacy of antineoplastic drugs in clinical treatment practice,and relevant guidelines and guidelines recommend the use of real-world data.As the standard statistical analysis methods for real-world studies are currently underdeveloped,this study explores the clinical effectiveness evaluation methods of antineoplastic drugs based on real-world data through descriptive analysis,regression models,and machine learning prediction models,using lung cancer as an example.Methods: In this study,medical data of patients hospitalized for lung cancer from July 2014 to September 2018 were collected from the Sichuan Provincial People’s Hospital.After the steps of data collection,integration,transformation,collation and selection of variables,and filling in of missing values,a dataset is obtained that can be used for subsequent analysis,including both an undifferentiated trade name cohort and a differentiated trade name cohort.Univariate Kaplan-Meier analyses,Cox proportional risk regression models,comparison of subgroup differences in antineoplastic drugs,univariate analyses,logistic regression models,and machine learning prediction models were performed for each of the two cohorts.Finally,the results of the data analysis after variable screening and the evaluation of model performance are discussed and evaluated.Results: After the variable selection,there were 1038 target cases data,including a total of 53 variables for the non-trade name cohort and 59 variables for the trade name cohort.The 1-year,2-year,and 3-year survival rates of lung cancer patients could be predicted by plotting the column line graphs of Cox proportional risk regression model;the 2-year risk of death of lung cancer patients could be predicted by the column line graphs plotted by logistic regression model;and the subgroup difference comparison could obtain the dominant population of different antineoplastic drugs in real-world lung cancer patients.Possible potential influencing factors related to treatment outcome in lung cancer patients are obtained by influence factor analysis.Machine learning prediction models can be used to predict the two-year survival of lung cancer patients and evaluate the clinical effectiveness of antineoplastic drugs.In the test set,the integrated learning algorithm model without screening,the Boruta-based integrated learning algorithm model,and the Boruta-based Bagging algorithm model ranked the top three in performance in anti-lung cancer drug clinical effectiveness prediction(AUC>0.615),with the integrated learning algorithm model without screening performing the best(AUC=0.6524).Conclusions: In this study,a methodological system combining survival analysis and machine learning prediction model was established.It can obtain the key influencing factors affecting drug treatment effects,obtain the dominant population of different antitumor drugs in real-world applications,and predict the therapeutic effects of antitumor drugs,which is beneficial for clinical drug use decision making.