Integer linear programming for Optimal Clinical Classification Schemes: New Scoring system for Predicting Target Lesion Revascularization after Paclitaxel-Coated Balloon

• Main Authors: Min-I Su, 蘇珉一
• Other Authors: Gwo-Liang Liao
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/yk789s

碩士 === 國立臺東大學 === 資訊管理學系碩士班 === 106 === Background: Medical scoring systems are linear classification models widely using for clinical decision-making. Linear programming is a method to achieve the best outcome in a linearly mathematical model. We demonstrate a new model using integer linear programming (ILP) to create optimal data-driven scoring systems. The clinical efficacy of paclitaxel-coated balloon (PCB) have been well proven in the treatment of instent restenosis (ISR), but the failure prediction models of PCB are not developed. The aim of this study was to use ILP to create a new target lesion revascularization (TLR) prediction models of PCB. Methods: We used ILP in medical scoring systems and the AUC of ROC curve was utilized for evaluating optimal solution. Results: The variables such as DES-ISR and statin using are superior to other variables and D2-S score was formed by assigning 2 points for the presence of DES-ISR and by assigning -1 points for statin using had the optimal predicting performance. The area under the receiver operating characteristic curve of new model (D2-S score) is 0.75. Conclusion: We developed new scoring systems for predicting TLR of PCB which we refer to as D2-S Score, and on top of that, ILP is a new method for creating optimal data-driven medical scoring systems and it can be utilized with data that is routinely available in electronic medical records.