防止病患跌倒事件一直都是護理照護品質指標之一,雖然,國內外機構致力於預防病人跌倒,但住院患者跌倒事件仍不斷的發生,主要因素在於跌倒危險因子複雜,且非單一因子所導致。為避免住院患者發生跌倒事件而造成病人及病人家屬額外的傷害以及醫療費用,及早掌握跌倒高危險因子以預防住院者跌倒,對患者及照護者而言,是非常重要的。 臨床上針對高危險跌倒患者評估方式大多只採用既有之Morse, STRATIFY, Hendrich跌倒評估量表於入院前進行評估,這些量表的變數不夠完整,也缺乏即時性,因此,本研究採用資料探勘之監督式學習分類技術來建立住院病患通用之跌倒預測模式,收集實際發生跌倒事件共197位病人資料及按科別比率抽取其他985位住院病患,做為本研究之實驗資料集。為避免實驗結果的偏誤,以隨機抽樣將未發生跌倒資料抽取30次和跌倒患者一致的數量後,採用十摺交叉驗證(10-fold Cross Validation)方式建構住院患者跌倒的預測模式。 實驗結果,以AdaboostM1 by決策樹(J48)分類技術所建構之分類器預測效能最佳。為了證明本研究之預測效能優於既有之跌倒量表。本研究使用相同資料集進行實證,實驗結果顯示,無論何種技術的預測精確度及敏感度都明顯優於三種跌倒評估量表。證實以資料探勘技術所建構的跌倒預測模式較能有效協助臨床照護人員找出跌倒的高危名單,進而能有效降低跌倒事件的發生率。
To prevent the patient's fall has been one of the indicators of care quality. The foreign and domestic healthcare institutions are devoted to prevent the patient's fall, but such events still continue to occur. The main reason is that the risk factors are complicated, and are not caused by a single factor. In order to avoid the inpatient's fall that leads to additional damages and medical expenses (costs) of patients and their family members, to early grasp the high-risk factors to prevent the inpatient's fall is very important for patients and nursing personnel. The assessment methods of the patients with high-risk falls mostly adopted by clinical personnel include the scales of Morse, STRATIFY and Hendrich, which are used before patients are hospitalized. The variables of these scales are not complete, and lack for immediateness. Therefore, this study adopts the supervised learning classification technique of data mining to establish the common fall prediction model of inpatients, and collects the actual fall events of 197 patients in total and extracts other 985 inpatients to be the experimental dataset of this study according to the rates of departments. In order to avoid the errors in the experimental result, we use random sampling to extract 30 times the fall data and the consistent amount of patients who fall, and adopts the 10-fold Cross Validation to construct the inpatient falls prediction model. As for the experimental result, the prediction performance of the classifier constructed by the classification technique of AdaboostM1 by decision tree (J48) is the best. In order to prove the prediction performance of this study is better than the one of existing falls scales, this study uses the same dataset of empirical data, and the result shows that the prediction accuracy and sensitivity are significantly better than the three fall assessment scales. Therefore, we prove that the falls prediction model constructed by data mining technique can better and more effectively assist clinical personnel to find out the name list of high-risk falls, and further effectively reduce the incidence of fall events.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。