透過您的圖書館登入
IP:174.129.190.10
  • 學位論文

呼吸器使用病人存活預測模式準確率之探討

Accuracy of Survival Models of Patients Who Use Mechanical Ventilators

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


研究目的 根據統計資料,我們發現呼吸器使用患者的醫療耗用、住院天數甚至死亡人數都是令健保當局手足無措。有鑑於此,如果臨床醫、護能與家屬的想法與觀點進行討論,評價臨床生存預測和其結果,這對於呼吸器使用患者及其家屬是重要的。因此本研究希望透過類神經網路(ANN)與複迴歸(MLR)分析找出影響呼吸器使用患者住院期間死亡之重要因子。以利醫療團隊於醫學的應用領域,故本研究目的如下: 目的一:探討呼吸器使用病人的病人特性及醫院特性之趨勢分析 目的二:利用效益指標(Performance Indices)比較兩種預測模式(Forecasting)之準確性(Accuracy)。 目的三:探討呼吸器使用病人的病人特性及醫院特性之趨勢分析利用全域敏感度分析 (Global Sensitivity Analysis)評估各個重要預測因子之權重。 研究方法 本研究採回溯性之研究設計,研究時間為 2004年至2009年,資料使用來源為全民健康保險學術資料庫之醫事機構基本資料檔(HOSB)進行回溯性次級資料之研究分析。研究對象為年齡>18歲且連續使用呼吸器≧96小時的病人,利用ICD-9-CM 518.81-518.84、518.89之診斷碼,以及處置用的ICD-9-CM 96.72,截取出研究個案,並排除<18歲及使用呼吸器<96小時之個案,總樣本數共213,945人。使用ANN模式及MLR模式,探討呼吸器使用病人存活模式之準確率,希望透過研究分析出各項重要影響因子,對於患者的臨床存活之預測能幫助家屬及醫師有效利用醫療資源,提供最好的醫療照護品質。 研究結果 研究目的一:呼吸器使用病人隨著時間的推移有年齡、傳染與寄生蟲疾病、教學醫院、醫院服務量、醫師服務量及住院天數、CCI分數、循環系統疾病、泌尿系統疾病、區域醫院及地區醫院等是隨著時間的增加有顯著相關。 研究目的二:使用18個重要變項中以類神經網路預測模式優於邏輯斯迴歸預測模式。呼吸器使用病人在住院期間存活,類神經網路與邏輯斯迴歸於正確率(Accuracy)分別為89.9%及81.7%,敏感性(Sensitivity)及1-特異性(1-Specificity)分別為86.9%、3.4%及84.8 %、25.4 %,接受者操作特徵曲線(Receiver Operating Characteristic Curve,ROC)分別為95 %及80 %。呼吸器使用病人出院後一年之存活,類神經網路與邏輯斯迴歸於正確率(Accuracy)分別為86.7%及85.5 %,敏感性(Sensitivity)及1-特異性(1-Specificity)分別為64.1%、1.2 %及61.4%、1.5 %,接受者操作特徵曲線(ROC)分別為91%%及80%。 研究目的三:利用全域敏感度分析(Global Sensitivity Analysis)在住院期間及出院後一年影響存活顯著因子之預測,類神經網路模式與羅吉斯迴歸模式存活預測重要因子皆為住院天數。 結論與建議 從預測結果得知,類神經網路預測模式之內、外部驗證皆優於邏輯斯迴歸預測模式。但邏輯斯迴歸較能清楚知道變項的方向性及重要性,且不同的預測模式皆有其優缺點,故建議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,以利各類疾病進行系統性之相關探討。

並列摘要


Research Purposes This study purposes to identify the impact factors of hospitalized survival rate and 1-year survival rate after discharge among patients with respirator use by using artificial neural network (ANN) and multiple logistic regression (MLR) analysis. For this purpose, this study purposes: 1. To evaluate the changing trends of patient characteristics and hospital characteristics during the study period; 2. To compare the performance indices between these two prediction modes; 3. To conduct the global sensitivity analysis in order to weight these significant predictors Research Methods This nation-wide population-based study retrospectively included 213,945 patients with respirator use from 2004 to 2009. Included criteria were the subjects aged larger than 18 years old and continuous use of respirator larger than 96 hours; the patients with ICD-9-CM diagnostic codes of 518. 81-518.84, 518.89, and disposal of used ICD-9-CM 96.72 were excluded. The ANN model and the MLR model were employed to compare the performance indices on hospitalized survival rate and 1-year survival rate after discharge. The global sensitivity analysis was also used to weight these significant predictors. Results According to the trend analysis, advanced age, infectious and parasitic diseases, teaching hospitals, hospital volume, physician volume and hospital days, CCI score, diseases of the circulatory system, urinary system diseases, regional hospitals and district hospitals are significantly associated with time. Moreover, according to the prediction of hospitalized survival, the ANN model showed the better performance indices than the MLR model on correct rate (89.9% vs. 81.7%), sensitivity (86.9% vs. 84.8%), 1-specificity (3.4% vs.25.4%), area under the accept operating characteristic curve (AUROC) (95.0% vs. 80.0%). For the 1-year survival rate, the ANN model also showed the better performance indices than the MLR model on correct rate (86.7% vs. 85.5%), sensitivity (64.1% vs. 61.4%), 1-specificity (1.2% vs.1.5%), AUROC (91.0% vs. 80.0%). Additionally, the global sensitivity analysis showed the most significant factor of both hospitalization survival rate and 1-year survival rates was the lengths of stay (LOS). Conclusions and Suggestions In conclusion, compared with the conventional MLR model, the ANN model in this study was more accurate in predicting both hospitalization survival and 1-year survival rates and had higher overall performance indices. The global sensitivity analysis also showed that LOS was the best predictor of both hospitalization survival and 1-year survival rates among patients with respirator use. The predictors analyzed in this study could be addressed by healthcare professionals during preoperative and postoperative health care consultations with candidates for patients with respirator use to educate them in the expected course of recovery and health outcomes. Further studies of this model may consider the effect of a more detailed database that includes complications and clinical examination findings as well as more detailed outcome data. Hopefully, the model will evolve into an effective adjunctive clinical decision making tool.

參考文獻


參考文獻
英文文獻
1. Aboussouan, L. S., Lattin, C. D., & Kline, J. L. (2008). Determinants of long-term mortality after prolonged mechanical ventilation. Lung, 186(5), 299-306.
2. Ata Murat Kaynar, M. C. E. M. R. P., MD, CM, FCCP, FC CM. http://emedicine.medscape.com/article/167981-overview
3. Azoulay, E., Pochard, F., Chevret, S., Lemaire, F., Mokhtari, M., LE GALL, J. R., Schlemmer, B. (2001). Meeting the needs of intensive care unit patient families. American journal of respiratory and critical care medicine, 163(1), 135-139.

延伸閱讀