利用貝氏網路評估急診室之急性闌尾炎病患
Applications of Bayesian Network in Evaluating Acute Appendicitis in the Emergency Department
中文摘要
背景: 急性闌尾炎是世界上最常見的外科急症。但研究顯示急性闌尾炎的臨床診 斷率卻只有 76 到 92%。所以改善急性闌尾炎的診斷率以避免不必要的手術，一 直都是臨床上被熱烈討論的話題。過去數年來已經有許多種臨床決策（電腦資訊）
系統被利用於臨床診斷上，但根據統計結果顯示臨床上急性闌尾炎的診斷率在過 去二十幾年來並無明顯改善。所以在本研究裡，我們嘗試利用貝氏理論，藉由探 討數個有關急性闌尾炎的重要臨床問題，以評估並建構急性闌尾炎的完整臨床推 論過程。
目的: 由於延遲或錯誤地診斷急性闌尾炎將會導致病情及死亡率攀升，故本研究 目的在於利用貝氏網路來評估並建構完整的臨床疾病推論過程。另一方面，我們 也希望藉由這個研究來評估貝氏網路於預測急性闌尾炎之相關臨床因子的可行 性及適切性。
方法: 我們收集了 30 個月(2005 年 1 月 1 日至 2007 年 6 月 30 日)急診室病患出 院診斷碼為急性闌尾炎，且最後有接受手術治療的個案。接著我們又收集了 14 項在急診室或住院期間較容易取得的臨床參數，包括:性別、年齡、體溫、轉移 痛、厭食、拉肚子、噁心、白血球指數、發炎指數、病理報告，電腦斷層攝影與 否、症狀發作至接受開刀時間、開刀術式、住院天數等。接著再利用這些參數建 構貝氏推理網路。最後，我們藉由指定貝氏網路內的變因為觀察值的方式來探討 前述有關急性闌尾炎的疾病因子並釐清其間的相關性。
結果: 本研究結果顯示，貝氏網路推論模組在臨床上所顯示出來的意義與我們所 熟知的傳統統計理論所呈現出來的結果並無明顯的差異。例如：較長的症狀發作 至接受手術的時間將導致較長的住院天數；而較嚴重的病理報告結果將意味著較 長的住院天數及較多的傳統開刀術式。
結論: 由於在急診室利用傳統臨床方法診斷急性闌尾炎的診斷率只有不到
50%，而本研究結果顯示利用貝氏網路來快速分析有關急性闌尾炎的重要臨床因 子是相當合適的。故本實驗證明了貝氏網路可以運用在急診室評估急性闌尾炎的 病患上。
英文摘要
Background: Appendicitis is the most common surgical emergency in the Worldwide.
But it has been estimated that the accuracy of the clinical diagnosis of acute
appendicitis is only between 76 percent and 92 percent. So, improving the diagnosis of acute appendicitis in order to prevent unneeded surgery is a critical topic that has been debated often. Over the years various clinical scoring systems (some computer assisted) have been used, but it concluded that the rate of misdiagnosis of acute appendicitis still has not changed over the last twenty years. So we selected some key clinical problems of acute appendicitis and try to evaluate and construct inference process overall using Bayesian Network in this study.
Objectives: Because of delay or mistake in diagnosis and inference will leads to increased rates of morbidity and mortality, the purpose of this study was to evaluate and construct inference process overall using Bayesian Network. On the other hand, we also hope to evaluate the feasibility and suitability of Bayesian networks for predicting variant variables for the patients with acute appendicitis in this study.
Methods: We included patients presenting to the ED during a 30 -months period ( January 1, 2005 – June 30, 2007 ) and were assigned a coded final ED visit
diagnosis as acute appendicitis, and received operation. Then we collect 14 variables that are commonly available during ED and hospitalization period as following: sex, age, temperature, shifting pain, anorexia, diarrhea, nausea, WBC level, hsCRP level, pathologic reports, undergo CT, symptom signs to operation period, operation type, length of hospitalization, and then modeling a Bayesian Network using these variables.
Finally, for the purpose for realizing the outline and inference process overall of above key clinical problem, we try to observe some definite variables in the network and can understand the relationslip of them.
Results: The result of this study revealed that no specific clinical difference compared with Bayesian Network inference model and traditional statistical outcome as we know. For example: longer symptoms onset to operation period will lead to longer length of hospitalization period; more severe pathologic state will mean longer length of hospitalization period and increasing rate of open appendectomy.
Conclusions: The accuracy of diagnosis of acute appendicitis in the emergency
department using the traditional approach has been shown to be less than fifty percent, and the result of this study revealed that Bayesian analysis was seen a suitable
approach to the important clinical problems of analysis of acute appendicitis in the Emergency Department. So this study demonstrated that a BN can be applied to
evaluate acute appendicitis using routinely available electronic data in the emergency.
