Establishment Of A Prognostic Model For Patients With Ovarian Cancer

Objective: Ovarian malignant tumors are one of the most common malignant tumors in female genital organs,which seriously threaten the health of women.Because ovaries are deep in the pelvic cavity,small in size and lack of typical symptoms,most of them have spread to pelvic and abdominal organs at the time of discovery,and the prognosis is poor.This study aimed at collecting data of patients with ovarian cancer and constructing a 5-year survival model to provide services for evaluation of surgical effect and prediction of prognosis of patients.Methods: Records of ovarian cancer patients who were followed up for more than five years from 2004 to 2010 were downloaded by software SEER*Stat 8.3.5.Referring to NCCN clinical guidelines,AJCC clinical guidelines,CS cancer information collection system and clinical experts' opinions,totally 20 prognostic variables were preliminarily selected for the study,including race,residential area,site of onset,histological grade,affected side,type of in situ surgery,degree of invasion,degree of lymphatic involvement,marital status,stage of cancer,distant metastasis,number of tumors,primary malignant tumors,lymph node removal,histological type,size of tumors,number of lymph nodes,number of positive lymph nodes,age of diagnosis and survival time.The baseline time was 5 years after operation,and the survival period of?5 years(60 months)was recorded as "survival" and <5 years as "death".Survival and death were taken as the outcome variables of this study.The data were cleaned up according to the United States collaborative staging data collection system CS;continuous data were discretized by interval method;training sets and test sets were randomly generated in total samples according to about 70% and 30%.Single factor analysis and logistic regression were used to preliminarily screen variables for training set data.Then Bayesian network model was constructed by Tabu search algorithm,and the network structure was adjusted according to the opinions of clinical experts to complete the final modeling.The model was tested with test set and compared with support vector machine,artificial neural network and decision tree algorithm.Results: A total of 7827 patients were selected.The 5-year survival-to-death ratio was about 5?3,and the proportion of strain categories was relatively average.After screening,the prognostic variables were: tissue grade,lymphatic involvement,infiltration,lymph node removal,distant metastasis,age of diagnosis,and marital status.The prediction accuracy of the model based on Bayesian network was 76.3%,which was better than that of support vector machine(70.6%),artificial neural network(70.8%)and decision tree(73.1%).Conclusion: The predictive accuracy of the survival prognosis model for patients with ovarian cancer is 76.3%.By constructing Bayesian network to further explore the relationship between prognostic variables,compared with simple logistic regression analysis,it can better explain the relationship between prognostic factors and achieve the combination of methods and experience.