運用健保資料庫探討精神病患者的再自殺行為之研究

Analyzing Re-suice Behavior For Psychiatric Patients In Taiwan - A Population-based Study

童彙婷(Hui Ting Tone)

指導教授 ： 胡雅涵

國立中正大學/管理學院/資訊管理系醫療資訊管理研究所/碩士(2013年)

摘要

自從1995年起，自殺已連續11年列在台灣十大死因之中，並且自殺防治在2012與2013年度也列為台灣最新第十大的醫院醫療品質與病人安全年度工作目標，期望能建立更完善的評估與照護機制。但是效果仍不顯著，因為並無法事先掌握意圖自殺的個案，即不能有效預防。研究發現自殺身亡的個案，死因大多與個案身前的疾病相關，其中又以精神病最為顯著。並且指出有過自殺紀錄的精神病患者其再次自殺的機率高於一般人。因此，對於自殺防治單位而言，如何預測精神病患者再自殺是很重要的公共衛生議題。
而在過去文獻上，大多採用通用性的自殺評估量表來預測精神病患是否為高危險重覆自殺的族群，並且研究的對象大多以單一醫院之自殺個案。其研究結果可能因為樣本來源與量表原始設計研究對象的不同而有所偏誤。同時，防治單位之自殺通報紀錄也只能事後紀錄，不足以達到提早預測精神病患者未來再自殺行為的目的。基於以上理由，本研究以2010年百萬人承保抽樣歸人檔中曾開立自殺診斷的精神病人為基準，選擇與研究目的符合之再自殺患者的特徵值，以資料探勘技術中之決策樹與類神經網路技術建立重覆自殺傾向的預測模型。最後再以 Adaboost 技術提升分類預測模型的效能。
研究結果在單一分類器的平均預測正確率上，以原始資料集採用類神經網路進行再自殺的預測效能為最佳(92.54%)。加入Adaboost後，平均預測正確率以建築於只保留顯著性變項資料集決策樹分類預測模型為最高(96.55%)。因此，本研究之實驗結果證明採用決策樹加Adaboost來建立精神病患者再自殺之測模式建立了發生精神病患者發生再自殺的規則，將能夠準確以就醫行為來預測精神病患者的再自殺行為，本研究希望能透過本研究的結果，協助臨床人員能及早達到防治自殺行為之目的。

關鍵字

全民健保資料庫 ； 再自殺 ； 精神病患者 ； 資料探勘

並列摘要

Since 1995, suicide has been listed as the ten leading cause of death for 11 years, and suicide prevention is also further listed as the annual objective of medical quality and patient safety in 2012 and 2013. Health institutions hope to establish a more comprehensive assessment and care mechanism, but there is no significant effect. We can not grasp the cases who have intentions of committing suicide, so we can not also effectively prevent suicide behaviors. Past studies found that the suicide cases are mostly associated with their diseases, especially mental diseases. There are also studies suggest that the mental patients who have suicide records have higher probabilities of committing suicide again. Therefore, for the units of suicide prevention, how to predict the re-suicide of mental patients is a very important issue of public health.
Past relevant literature mostly take the suicide cases of a single hospital as the samples, and use versatile suicide assessment scales to predict whether a mental patient belongs to the group of having high risk of repeatedly committing suicide. The result has some errors due to we may not be able to exactly grasp the history of diseases of mental patients. Besides, the post-records of suicide reports can not achieve the purpose of early preventing the suicide behaviors of mental patients. For the above reason, this study takes LHID2010 of million people who have ever been diagnosed as mental patients as the basis, and selects the Eigen values of re-suicide patients which fit in with the research purpose, and use the decision tree of data mining technique and neural network technique to establish the repeated suicidal predictive model, and finally use Adaboost to build multiple classifier to enhance the predictive performance.
In terms of average accuracy rate of single classifier, we get the best effectiveness (92.54%) while adopting the original dataset and neural network to predict re-suicide. After adding Adaboost, the average accuracy rate of prediction reaches the highest standard (96.55%) while only retaining the decision tree classification model of dataset of significant variables. Therefore, the experimental result of this study proves that to adopt the decision tree and Adaboost to establish the prediction model of the re-suicide behaviors of mental patients can accurately predict the re-suicide behaviors of mental patients according to their medical behaviors. This study aims at assisting clinical personnel to early reach the purpose of preventing suicide behaviors.

並列關鍵字

National Health Insurance Research Database ； Re-suicide ； Psychiatric patient ； Data mining

參考文獻

劉介宇、洪永泰、莊義利、陳怡如、翁文舜、劉季鑫、梁賡義 (民95)。台灣地區鄉鎮市區發展類型應用於大型健康調查抽樣設計之研究。健康管理學刊，

劉嘉年、楊銘欽、楊志良(民90)。台灣成年民眾於死亡前三個月健保醫療費用支出之影響因素分析。臺灣公共衛生雜誌，20(6)，451-462。

周幸玲 (民96)。重鬱症個案自殺企圖與生活型態防衛機轉相關性之探討。(碩士論文)。取自http://libir.tmu.edu.tw/handle/987654321/5699

洪子棋(民97)。探討強迫症患者之焦慮、憂鬱症狀與自殺意念之相關。(碩士論文)。取自http://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi? o=dnclcdr&s=id=%220

王文華 (民97)。台灣地區自殺企圖者之重複自殺企圖次數統計模型探討。(碩士論文)。取自http://nccur.lib.nccu.edu.tw/handle/140.119/49596