透過您的圖書館登入
IP:3.82.3.33
  • 學位論文

住院跌倒病人跌倒傷害預測模式之探討

The Prediction Models of Falls Injury among Hospitalized Patients

指導教授 : 許弘毅
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


中文摘要 研究目的 跌倒事件在醫院為高危險性且經常發生的問題,佔世界各國家醫療機構中所有異常事件的24%~84%。跌倒後果可能導致病人生理的損傷,影響心理感受、降低個人及家庭生活品質,影響療程及延長住院天數,增加醫療成本支出及耗損健保費用,故有效降低跌倒發生率及傷害程度已是刻不容緩的議題。過去國內外對於跌倒的相關研究大都為針對國內單一醫療機構及某一特定時間進行探討,對於對跨機構和長期追蹤的研究付之闕如,故本研究希望透過運用類神經網路(ANN)分析模式及邏輯斯迴歸(MLR)分析模式找出造成跌倒傷害的重要影響因子,提供給醫護人更有效的篩選出需特別注意照護的病人,降低跌倒傷害發生率。本研究的目的如下: 目的一:探討住院跌倒病人跌倒傷害的病人特性、生理特性、藥物特性、環境特性及照護特性相關影響因子及其長期趨勢。 目的二:利用利用效益指標(Performance Indices)比較二種預測模式(Forecasting)在住院跌倒病人跌倒傷害之準確性。 目的三:利用全域敏感度分析(Global Sensitivity Analysis)評估影響住院病人跌倒傷害各個預測因子之權重。 研究方法 本研究為回溯性之研究設計,研究時間為2010年至2012年,資料來源為是南區某醫療體系3家醫院的病人異常事件通報資料庫及醫囑處方資料檔,進行回溯性次級資料之研究分析。研究樣本為住院發生跌倒的病人,排除年齡≦18歲、跌倒地點不在院內、跌倒因素為其他及異常事件通報資料不完整,總樣本數共1,151人次,針對病人特性、生理特性、藥物特性、環境特性、照護特性等5個構面,使用ANN分析模式及MLR分析模式進行不同預測模式之準確性比較,再以全域敏感度分析(Global Sensitivity Analysis)找出造成影響跌倒傷害重要的預測因子。 研究結果 目的一:病人特性(男性、年長)、生理特性(步態與平衡障礙、肌肉軟弱無力)、藥物特性(服用安眠鎮靜劑、抗高血壓劑、降血糖劑、抗組織胺藥及瀉劑)為影響住院病人發生跌倒傷害的重要預測的因子。藥物使用是隨著時間的增加有顯著相關。 目的二:使用12個重要變項預測住院跌倒病人跌倒傷害的準確性,以ANN分析模式優於MLR分析模式。住院病人跌倒傷害,ANN與MLR之準確率(Accuracy)分別為73.0%及68.7%;敏感度(Sensitivity)分別為81.5%及83.1%;特異度(1-Specificity)分別為62.0%及50.0%;操作者特徵曲線檢定(ROC curve)分別為74%及67%。 目的三:利用全域敏感度分析(Global Sensitivity Analysis)評估影響住院病人跌倒傷害各個預測因子之權重要預測因子方面,最重要的因子為步態與平衡障礙,次之為服用嗎啡類複方止痛藥及高年齡者。 結論與建議 由本研究結果顯示,影響住院病人跌倒傷害的因素是多面向的,特別是步態與平衡障礙、服用嗎啡類複方止痛藥及降血壓藥、高年齡及肌肉軟弱無力等為住院病人跌倒傷害的重要預測因子。本研究結果可作為醫院管理者及臨床醫護人員制訂跌倒高危險評估量表之參考,及早確認高度危險傾向之病人,以期採取有效的防範措施或計畫,提升住院病人照護安全。 從研究結果證明,類神經網路預測模式之內、外部驗證優於邏輯斯迴歸預測模式,但不同的預測模式皆有其優缺點,臨床醫護人員於臨床運用面臨評估及決策時,可以使用類神經網路預測模式並擴大預測變項(例如日常生活活動功能(ADLs)分數、血壓變化情形、血色素等變數),並能利用此預測模式推廣到其他疾病之相關研究。 關鍵詞:跌倒傷害, 類神經網路, 邏輯式迴歸, 全域敏感度分析

並列摘要


Abstract Research Purposes Falls are a high risk and most common problem for hospitalized patients and accounted with 24%~84% hospitalized patients around the world. Such falls can result in serious physical or emotional injury, decrease the quality of life for many patients, and even prolong length of a hospital stay, admission to a long-term care facility, increased medical expense and healthcare insurance cost. For falls of related research mostly for aim at single medical institutions and certain a particular time carried on the study in the past, for to across the multi-centers and longitudinal follow up research reports for very rarely. The study purposed to identify the risk factors of falls by using artificial neural network(ANN) and multiple logistic regression (MLR) analysis, thus provided doctors and nurses to more effectively screen to need to particularly notice a caring patient, reduce to falls injury incidence rate. For this purpose, we listed the purposes of the study as follow: 1. To evaluate risk factors of falls injury in particular the demographic, physiological, medication, environment and care factors and the changing trends. 2. To compare the performance indices between these two prediction modes. 3. To conduct the global sensitivity analysis in order to weight theses significant predictors. Research Methods This study was a retrospective review of the web-based incident reporting system form three hospitals in southern Taiwan. Data on a total of 1,151 patient falls during the year 2010 to2012. Excluded criteria were age ≦18 years old, fall down don't in the hospital, factors of falls with others and not complete. The ANN model and the the MLR model were employed to compare the performance indices on falls injury versus non- injury. The global sensitivity analysis was used to weight these significant predictors. Results 1. In the 1,151 reported case of falls, the patients’ demographic (male gender, age), physiological (gait and balance disturbance, muscle weakness), medication use (hypnotics, antihypertensive, hypoglycemia, antihistamine, miscellaneous analgesic, cathartics) were significantly positively related injury. According to trend analysis, medication use is significant associated with temporal trend. 2. According to the prediction of falls injury, the ANN model showed the better performance indices than the MLR model on accuracy rate (73.0% vs. 68.7%),sensitivity (81.5% vs. 83.1%),1-Specificity (62.0% vs. 50.0%), area under the operating characteristic curve (AUROC) (74% vs. 67%). 3. Additionally, the global sensitivity analysis showed the important predicted factors, the top three factors was gait and balance disturbance, take for miscellaneous analgesic, and old age. Conclusions and Suggestions It is concluded that falls injury among hospitalized patient were a combination, gait and balance disturbance, take for miscellaneous analgesic, old age, use to antihypertensive and muscle weakness. It could be used for hospital managers and clinical nurses to set up fall-risk assessments or instruments, to identify in advance the most in danger of falling and to promote the safety of patient care. In comparison with the conventional MLR model, the ANN model in this study was more accuracy rate, 1-Specificity and AUROC in predicting fall injury among hospitalized patient. Moreover, the ANN model had higher overall performance indices. But the different predicted mode all has its strength and weakness. Further studies of this model may consider more objective factors (such as ADLs, blood pressure change, hemoglobin). Hopefully, the model will evolve into effective adjunctive clinical decision making tool. Key words: falls injury, artificial neural network, logistic regression, the global sensitivity analysis

參考文獻


參考文獻
中文文獻
方靜玉、黃錫培、柯宣妤、姜秀滿(2007).預防透析病人跌倒的照護措施.腎臟與透析,19(2),102-106。
台灣病人安全通報系統網站(2011)。統計報表,2007年2010年年報表資料。2013年2月15日取自http://www.tpr.org.tw/index03.php
何德威(2009).三種資料探勘法及邏輯斯迴預測效能和預測因子之比較.高雄醫學大學:醫務管理研究所碩士論文.

延伸閱讀