建構老年病患非計劃性重返急診之預測模式：以全民健保資料庫為例

Construction the prediction model of Elderly Patients Revisit to the Emergency Department: A Case Study of National Health Insurance Database

林昱吟(Yu-Yin Lin)

指導教授 ： 胡雅涵

國立中正大學/管理學院/資訊管理學系/碩士(2013年)

摘要

隨著醫療技術進步與生活水準提昇，人口老化已經是明顯趨勢，由於急診的就醫流程較為快速，也導致老年急診就醫病患的比率日趨增加。老年人身體機能退化，疾病複雜性高，加上疾病臨床表現較不典型，造成老年急診病患獨特而複雜的臨床特性，在急診室緊迫的醫療流程與環境下，醫療服務提供者的確對老人之病程後續發展難以預測，然而，當老年病患急診就醫後，當發現到醫療需求並未獲得完全滿足時，即可能會導致重返急診以便解決醫療問題。 目標：資料探勘技術已經廣泛地被應用在各項醫療研究上，可從資料中所萃取關鍵資訊應用於醫療決策輔助的可行性。因此本研究目的為(一)探討全台灣老年急診病患三日內非計劃性重返急診之相關影響因素。(二)應用資料探勘技術建立老年急診病患三日內非計劃性重返急診之預測模式。 方法：本研究以2010年「全民健康保險研究資料庫」承保抽樣歸人檔，其案件分類代碼為【02】西醫急診且年齡65歲以上(含)者為研究對象，以決策樹與邏輯斯迴歸2種分類技術建構出可用以預測老年急診病患三日內非計劃性重返急診之分類預測模式，並經由效能指標進行分類模式之效能評估，從中選取最佳效能之分類預測模型。 實驗結果：以決策樹C4.5(J48)建構的預測模式，對於老人急診病患非計劃性重返急診的正確預測率平均為76.65%，對於未重返急診的正確預測率平均為76.95%，而決策樹模式整體預測準確率平均為76.80%。並由決策樹C4.5(J48)整理出老年病患非計劃性重返急診之預測規則共11項。因此本研究應能協助急診醫療人員建構適當的老年病患重返急診預測模型，而產生有效的篩選與評估決策建議。 結論：為能有效管控急診醫療資源並能同時確保醫療品質，深入瞭解老年族群急診醫療利用之情況，特別是老人重返急診的問題，將是現今急診醫療照護中非常重要的議題。資料探勘技術能協助處理龐大的資料，從中萃取出關鍵的資訊，並可建立更明確的篩選機制以供醫療臨床使用，可彌補一般統計分析之不足，若能應用資料探勘技術，有效率地篩選與掌握老人急診重返高危險群病患，不僅可減少急診就醫次數，減緩急診人員工作量，更重要的是讓老年病患於急診時獲得適當的醫療照護。

關鍵字

全民健保資料庫 ； 醫療品質 ； 老人急診 ； 重返急診 ； 資料探勘

並列摘要

With the advances in medical technology and living standards improve, the aging population trend is already evident, because the process is more rapid in emergency , emergency medical treatment also resulted in elderly patients with the increasing ratio. Elderly physical function degradation, disease complexity, coupled with a less typical clinical manifestations of disease, resulting in elderly patients with unique and complex emergency clinical characteristics, in the emergency room urgent medical processes and environment, medical services provided to the elderly who do difficult to predict the course of the subsequent development, however, when emergency medical treatment in elderly patients, when found to medical needs has not been fully met, that may lead back to the emergency in order to solve medical problems.
Objectives: Data mining technology has been widely applied in various medical research, can be extracted from the data in the key information used in medical decision support feasibility. Therefore, this study aims to (a) investigate emergency Taiwan elderly patients to return within three days of unplanned emergency relevant factors. (b) the application of data mining technology to establish emergency elderly patients to return within three days of unplanned emergency the prediction model.
Methods: In this study, in 2010 the "National Health Insurance Research Database," Insurance Beneficiaries who file their cases [02] classification code for Emergency Medicine and the age over 65 years for the study, decision tree and logistic Regression two kinds of classification techniques can be used to construct predictive elderly emergency patients to return within three days of unplanned emergency classification forecasting models and performance metrics are classified by mode of performance assessment, which selects the best classification performance prediction model.
Results: In the decision tree C4.5 (J48) constructed forecasting models for elderly patients with non-emergency emergency department planned to return to the average rate of correct predictions 76.65%, did not return to the emergency department for the correct prediction of the average rate of 76.95% , while the decision tree model overall prediction accuracy rate averaged 76.80%. By the decision tree C4.5 (J48) sorted out in elderly patients unplanned emergency return of the prediction rule 11. Therefore, this study should help build the appropriate emergency medical personnel to return to the emergency elderly patients forecasting model, and produce effective screening and evaluation of policy recommendations.
Conclusion: In order to effectively manage and control emergency medical resources and energy while ensuring quality of care, in-depth understanding of older populations emergency medical use of the situation, especially the elderly to return to the emergency of the problem, is now the emergency medical care is a very important issue. Data mining techniques can help deal with huge data, which extract critical information, and to make more explicit screening mechanism for medical clinical use, can make up for general statistical analysis of deficiencies, if application of data mining techniques to efficiently filter Emergency back and grasp elderly patients at high risk, not only can reduce the number of emergency room visits, reduce the workload of emergency personnel, more importantly, to allow older patients in the emergency department to obtain appropriate medical care.

並列關鍵字

National Health Insurance Database ； Medical Care Quality ； Elderly Emergency ； Revisit to the Emergency Department ； Data Mining

參考文獻

宋文娟、洪錦墩、陳文意（民97）。台灣老年人口醫療利用與多重慢性疾病之分析研究。臺灣老人保健學刊，4（2），75-87。.

周歆凱、黃興進、蔡明足、翁林仲、蘇喜、陳真吟（民98）。運用資料探勘之叢集分析技術探討急診72小時再返診病患特性。澄清醫護管理雜誌，5（3），13-20。

洪世昌、賴世偉、李亞欣、謝豐年、陳玉如、劉翎玲（民99）。急診近期返診之探討：以南投市某地區醫院為例。中華民國急救加護醫學會雜誌，21（S），21-27。

周歆凱、蘇喜、黃興進、蔡明足、翁林仲（民95）。運用決策樹技術探討急診病患醫療費用之消耗。臺灣公共衛生雜誌，25（6），430-439。

莊宗南、龔榮源、陳俊龍（民95）。以資料探勘技術建立病患就醫導引－以胃腸科病患為例。醫療資訊雜誌，15（1），17-34。