Summary: | 碩士 === 高雄醫學大學 === 醫務管理暨醫療資訊學系碩士在職專班 === 102 === According to the National Health Department and the Taiwan Cancer Registry from 1998 to 2010 statistics, the standardized mortality rate of primary brain tumors is between 1.58 to 1.98% per 100,000 population, with annual average of 425 new cases. This study used Logistic regression model, Cox survival analysis and Artificial Neural Network to compare the forecasting performance of three models, and the weights of each predictor, and hope to assist clinicians in developing clinical decision-making of brain surgery, and assess the impact factors of postoperative mortality .
This study compared Logistic Regression、Cox Regression and ANN models based on initial clinical data from 1998~2010 for 7,740 brain tumor surgery patients, including age, gender, Charlson co-morbidity index (CCI) score, diseases of system , hospital level,hospital volume, surgeon volume 、time and outcome. A global sensitivity analysis was also performed to assess the relative significance of input parameters in the system model and to rank the importance of variables.
The Performance indices of predicted survival efficiency aspect in ANN 、LR and Cox Regression , the accuracy of after five years of surgery is ANN=83.59%、LR=69.05%、Cox Regression=62.95% . In AUROC aspect, ANN=0.87、LR=0.68、Cox Regression=0.70. In the aspect of important predicted factors, the top two factors of ANN 、 LR and Cox Regression are the same of age and time.
Based on the results of this study, the ANN model in the predicting accuracy is better than most of LR and Cox survival analysis . And artificial neural network applications can indeed improve the medical prediction in sensitivity, specificity, accuracy ,and the positive and negative predictive values . Therefore, data mining results can be considered as a clinical decision-making tool, in addition, age, time, DRS can be considered as important prognostic factors of brain surgery patients in clinical decision making.
|