燒燙傷病患之住院死亡與醫療資源耗用評估研究

根據行政院衛生福利部統計，2012年事故傷害已經進入我國十大死因當中，且燒燙傷佔事故傷害中的前5名，而隨著近年來社會型態改變及科技進步，國民生活水準提高導致電器用品普遍化，加上國民對防火消防之忽視，各種原因引起之燒燙傷成了日常生活中常見且嚴重的意外傷害，即使在有完善醫療設備的今天，燒燙傷受傷後所引起的後遺症需要長時間的治療與復健，對於傷患的身心影響及所需的醫療費用，都是所有事故傷害中最為嚴重者。本研究以國內燒燙傷的病患資料庫為研究對象，運用粒子群最佳化演算法、基因邏輯斯迴歸演算法、交叉熵演算法相互結合倒傳遞類神經網路與支援向量機以及案例式推理系統，針對燒燙傷病患的死亡預測模型與醫療費用評估，研究結果顯示，死亡預測模型中以粒子群最佳化演算法結合倒傳遞類神經網路模型具有最好的預測效果，其平均測試準確度為91.05%，及其ROC值為0.893；在燒燙傷醫療費用評估系統方面，以基因邏輯斯迴歸演算法結合案例式推理系統之評估績效為最佳，其平均測試準確度達79.653%，及其ROC值為0.784，且得費用誤差於15%內的準確度達80.53%，故本研究可以做為國內燒燙傷疾病於相關醫療議題之參考，同時提供衛生福利部於未來擬定給付制度之參考依據。

關鍵字

燒燙傷；粒子群最佳化演算法；交叉熵演算法；基因邏輯斯迴歸演算法；倒傳遞類神經網路；支援向量機；案例式推理系統

並列摘要

According to statistics of Ministry of Health and Welfare, Executive Yuan.(MHA), accident injuries in 2012 are one of the ten leading causes of death in our country, especially burns that rank in the top five of accident injuries. With changes in social patterns and technological advancement in recent years, the improved standard of living has led to the increasing prevalence of electrical appliances. In addition, due to the public’ s negligence over fire prevention, burns that take place due to many reasons have become common and serious accident injuries in daily life. Despite the comprehensive medical equipment available today, burn-induced sequelae require long-term treatment and rehabilitation. The physical and psychological impact of burns and medical costs that arise are more devastating for patients than any other types of accident injuries. Burn patient databases in our country were adopted as research participants in this study. Targeting the burn patient mortality prediction model and medical cost assessment and through the use of particle warm optimization, genetic logistic regression algorithm, cross entropy algorithm, combined with back propagation network, support vector machine, and case-based reasoning system, findings show that the particle swarm optimization algorithm coupled with the back propagation neural network model produced the best predictive results, with the mean test accuracy of 91.05% and the ROC value of 0.893. As for the burn related medical expenditure assessment system, the genetic logistic regression algorithm combined with the case-based reasoning system produced the best assessment performance, with the mean test accuracy of 79.653% and the ROC value of 0.784. In addition, the accuracy of the expenditures within the 15% error range reached 80.53%. Hence, this study shall serve as a reference for domestic burn disease related medical issues and for the MHW to develop a payment system in the future.

並列關鍵字

Burn ； Particle Swarm Optimization ； Cross Entropy ； Genetic Logistic Regression Algorithms ； Back-Propagation Neural Network ； Support Vector Machines ； Case-Based Reasoning