應用決策樹演算法與邏輯式迴歸模式探討傷害就醫之相關因子

Application of Ddecision Ttree Algorithms and Llogistic Rregression Mmodels in the of Ffactors Aassociated with Mmedical Iinjuries

俞依良(I-Liang Yu)

指導教授 ： 詹前隆

元智大學/資訊學院/資訊管理學系/碩士(2011年)

https://doi.org/10.6838/YZU.2011.00279

摘要

傷害對個人、家庭和社會，都可能造成某種程度的負擔或損失。本研究旨在利用量化方法，了解因傷害而就醫的族群其人口學特質、個人健康狀態及健康相關行為等特性，並分析其相關危險因子，希望能對傷害防治策略提供有價值的資訊。
本研究是以民國91年1月16日戶籍資料在台灣地區之12歲以上個人為抽樣母群體之次級資料研究，並回溯接結2001年(民國90年)台灣國民健康問卷調查之家戶問卷及12歲以上個人問卷內容，加以分類整理後以其人口學特性、個人健康狀態、個人健康行為、自覺健康狀態、醫療資源利用等5大類內容為自變項，另以該抽樣族群當年(民國91年)之全民健康保險資料庫就醫記錄為依變項，藉由傳統生物醫學統計之多變項邏輯式迴歸模式及資料探勘之決策樹演算法加以分析，並比較兩種分析模式異同與優劣點。
使用「多變項邏輯式迴歸模式」顯示：學歷高低、身體質量指數、生理健康指數、過去一年神經肌肉功能、過去一年循環功能、過去一年醫療資源利用與否等六項變項為影響「是否因傷害而就醫」之相關因子。而利用決策樹演算法分析，所得到之規則顯示在過去一年中，其生理健康指數分數越低者、年齡低者、男性、身體質量指數越高者、學歷低者等，容易造成因傷害而就醫。重要性排名以醫療資源利用為最重要，其次為身體質量指數，再其次為年齡。綜合多變項邏輯式迴歸及決策樹演算法2種分析得知：在過去一年中，身體質量指數越高、生理健康指數分數越低、年齡輕及學歷低者、以及男性，在未來一年容易造成因傷害而就醫。政府衛政部門可應用本研究找出之易受傷害而就醫之危險因子，提供一般民眾適當的衛教建議，並擴大運用於職場之傷害防治策略。

關鍵字

台灣國民健康訪查調查 ； 全民健康保險資料庫 ； 傷害 ； 邏輯式迴歸 ； 決策樹演算法

並列摘要

Injury could be an important issue because that would result in some burden or loss to the individuals, their families and the society. This study aims to use the different quantitative methods to analyze its related risk factors, including the geographic characteristics of the injured population, their personal health status and their daily health behavior.
The present study was a retrospective case-control study based on the random sampling subjects aged 12 or more in 2002. These enrolled cases were traced back to their information of the household questionnaire and the individual questionnaire conducted by Taiwan National Health Interview Survey (NHIS) in 2001. At the same time, their medical records of Taiwan National Health Insurance (NHI) for the treatment of injuries in 2002 were evaluated to be the dependant variables in this study. Comparison of the multivariate logistic regression model and the decision tree analysis was the main statistic purpose.
The multivariate logistic regression model revealed that the education level, body mass index, physical health score, the past neuromuscular function over, the past circulative function, the past use of medical resources were the six major factors associated with the medical visits due to injuries. The decision tree analysis revealed that lower physical health score, younger age, male gender, higher body mass index and less educational attainment were the five major factors associated with the medical visits due to injuries. The ranking of importance of these associated factors were the past use of medical resources followed by body mass index and age. A final result combined the logistic regression model and the decision tree analysis showed that higher body mass index, lower physical health score, younger age, less educational attainment and male gender were the most significant factors to receive some medical treatment due to injuries. These findings could be useful for the healthy policymakers to promote the health education programs for the public and the injury prevention process in the workplace.

並列關鍵字

National Health Interview Survey ； Injury ； Logistic regression model ； Decision tree algorithm

參考文獻

4. 嚴玉華 、許碩芬、方世杰、龔志銘、牟聯瑞，「勞工職業病與職業傷害醫療資源耗用探討」，中華職業醫學雜誌，第十六卷第三期，第183-92頁，2009年。

5. 林莉茹、李蘭、莊莉菁、郭寶貞、白璐，「運用PRECEDE模式評估某社區之事故傷害問題」，台灣衛誌，第二十九卷第四期，第314-25頁，2010年。

6.鍾其祥、劉淳羽、簡戊鑑，「臺灣燒燙傷住院傷患之流行病學」，北市醫學雜誌，第七卷第一期，第 53-66頁，2010年。

8.李燕鳴，「台灣事故傷害之監控與防制」，台灣醫學，第八卷，第114-21頁，2004年。

10.賴建丞、白璐、蔡行瀚、簡戊鑑、張瑋庭、林佳欣、洪宇箴，「臺灣1986-2007年事故傷害死亡趨勢分析」，北市醫學雜誌，第6卷第3期，第174-84頁，2009年。