建立類神經網路模型預測住院跌倒之發生
中文摘要
近年來，世界各國愈來愈重視病人安全（ Patient Safety ）之相關議題，因此如何建立「病人安全管理體系」，儼已成為全球性醫療管理之新趨勢，在過去因不當醫療管理所 衍生之問題如：藥物不良事件（ Adverse Drug Event ）、手術相關（ Procedure Relate d ）、院內感染（ Nosocomial Infection ）、以及住院跌倒（ Inpatient Falls ）等意外事件 曾引發社會各界不同程度之關切，因此未來有關現代醫療管理等重要課題之研究將逐漸 成為各大醫院謀求服務品質提昇之重要目標。
據相關研究資料顯示，住院跌倒約佔醫院意外事件之 80 ％，而許多住院跌倒之案例更 將導致病人病情惡化、產生併發症及延長其住院日數，嚴重耗用醫療成本。為解決長存 於各醫療院所之此一困擾問題，本研究將以回溯性之方式收集台北市某醫學中心住院病 患電子病歷（ Electronic Medical Records ）資料（含三年跌倒組及非跌倒組計 3496 筆），希望藉由本研究所建立之類神經網路（ Artificial Neural Network ）模型，進一步地準 確預測住院跌倒之發生，以確保病人安全，降低醫療成本。
有關於住院跌倒之研究， O''connell 等人以「 Morse Fall Scale 跌倒評分 表」應用在澳洲某醫院兩個老人急性病房，結果顯示其敏感性為 83% ，特異性為 29%，陽性預測值為 18% 。 Hendrich A.L. 等人曾以「 Hendrich II 模組跌倒危險評估表」執 行相關之研究課題，而其所得之敏感性及特異性分別為 74.9% 及 73.9% 。此外， Oliver, D. 等人亦曾針對老人住院病患，應用「 STRATIFY 跌倒危險評估表」執行住院跌倒之 研究，並得出 93% 之敏感性及 88% 之特異性。而透過本論文之研究方法，電子病歷經 由本研究所採之類神經網路模型所得之敏感性與特異性則分別為 87.19 % 及 87.64% 。 研究者得到的結論是電子病歷在預測住院病人跌倒之發生上，可以做為預測因子很好的 資料來源。
Establishing an Artificial Neural Network Model for Predicting Inpatient Falls
英文摘要
Most of the countries have paid much attention to the issues of patient safety in recent years. The construction of a patient safe ty system has gradually turned into a global brand-new trend in the medical management. In the past, the medical accidents su ch as adverse drug event, procedure related, nosocomial infection, and inpatient falls etc., which normally resulted from the i mproper medical management, have caused different degrees of commotions in our society. Therefore, the study of a new ma nagement criterion for the medical issues will become the useful implement to enhance the service quality in a modern hospita l.
Some research data have shown that the chance of inpatient falls covers nearly 80% of the hospital’s contingency. Moreover, i n many cases inpatient falls along with serious injury will aggravate patients'' condition, produce complications and increase patient’s length of stay. Such situations always consume much amount of medical cost. In order to solve these aff lictive problems, the retrospective patients’ medical history data to the amount of 3496 from one medical center hospital in Ta ipei were employed in this research. These data include both faller and non-faller’s Electronic Medical Records （ EMR ） . The objective of this research is to establish an Artificial Neural Network （ ANN ） model combined with EMR for the pred iction of inpatient falls. Such method could be further used to provide a safe care for the patients and reduce the medical cost.
Regarding the research of inpatient falls, O''connell used the Morse Fall Scale as part of a fall prevention progra mme on two aged care wards in an acute care hospital setting in Australia. Findings revealed that the Morse Scale had a sensit ivity of 83%, a specificity of 29% and a positive predictive value of 18%. Hendrich A.L. used the Hendrich II fall risk model i n large concurrent case/control study of hospitalized patients, and herein obtained the significant values of sensitivity and spec ificity, which were 74.9% and 73.9% respectively. Meanwhile, applying the STRATIFY fall risk assessment tool, Oliver, D. g ot the 93% sensitivity and 88% specificity for the elderly inpatients study case. In this paper, an ANN model according with E MR have been used as an analysis tool; the sensitivity and specificity from the model analysis are 87.19% and 87.64% respect ively. The conclusion is that EMR can be employed as a useful source to accurately predict the occurrence of fall risk factor.
