利用資料探勘技術探討超長天住院病人之相關屬性

ASSOCIATED FEATURES OF LONG-STAY PATIENTS USING DATA MINING TECHNIQUES

鄒淑英(Shu-Ying Chou)

指導教授 ： 詹前隆

元智大學/資訊學院/資訊管理學系/碩士(2006年)

https://doi.org/10.6838/YZU.2006.00199

摘要

中央健保局從民國91年總額制度實施以來，長久努力的目標：就是在不浪費醫療資源情況下，如何提升病人的醫療服務品質。從國外健保總額制度看來，似乎都面臨相同的課題。目前，該局針對各項醫療資源浪費的監測，均訂有明確的分析指標，並定期評估每項醫療服務品質指標的執行情形。從分析結果發現每季監測指標均有明顯的改善，抑或小幅改善，但其中有關「三十日以上超長住院率」，卻從未有明顯下降情形，明知長期住院病人對整體醫療資源是一項很龐大的支出，但卻無法有效地控制其成長率。本研究為能充份了解有關超長天住院病人之特質及就醫屬性，希望藉由「資料探勘」的技術，從超長天住院病人個案中，分析出超長住院所潛在之相關「屬性」及其「關聯性」，突顯出隱藏其間的問題。研究流程分為兩階段；第一階段為一般統計分析，期望找出有意義的變項；第二階段為資料探勘分析方法，將一般統計分析之結果與實際探勘之結果相互比較，以找出較具說服力之模式。一般統計分析架構中，自變項代表住院病患特質，包含病患屬性及就醫屬性。應變項代表住院病患住院醫療資源之耗用。控制變項代表病患住院期間住院日數分析包括小於等於120日與大於120日。本研究驗證了資料探勘技術與一般傳統統計分析結果相似，並發現資料探勘工具確實能夠挖掘出潛在之特性組合，對於異常資料之稽核也有極大的效用，分析超長天住院病患中潛在之醫療資源高耗用族群，不但可快速並視覺化出這些高危險群病患，更可由關聯規則中歸納出超長天住院病患之關聯性，可強化健保多元管理與審核的品質及時效。

關鍵字

中央健保局 ； 超長天住院 ； 資料探勘

並列摘要

With the measures of the global budget system in 2002, the Bureau of National Health Insurance (BNHI) permanently strives how to improve the medical service quality of the patients without squandering on medical resources. Considering from the global budget systems of overseas Health Insurance, the same issue came out a challenge to be dealt with. Currently, the BNHI has classified the explicit analysis indicators according to the monitor of all kinds of medical resource and periodically evaluates the measures condition of medical service quality. Based on the analysis, they can obviously find out somewhat improvement from the season monitor indicators while there was hardly decreasing of resident rate of the over-30-day for the long-stay-patient. The BNHI should know well the long-stay-patient would result in the enormous expenditure on the whole medical resources, but it was far from to effectively control its growing rate. In this thesis, we try to learn the characteristics and the medical-behavior properties of the long-stay-patient as well as the implicit properties and correlation of the patients from the long-stay cases with the techniques of data mining. The research procedure contains two stages. The first is the general statistic analysis in which the wanted meaningful variables would be set up. The second is the data mining analysis in which the results from the general statistic analysis would compare with that from data mining to search for proper operation-model. In the case of the general analysis, the independent variables represent the properties of patients resident in hospital, including patients and medical-behavior properties. The dependent variables represent the medical costs of patients resident in hospital. The control variables represent if the patients have combined diseases in the hospital around the day count less than or equal to 120 and more than 120. The experiments outcomes not only can verify the resemblance resulting from both the techniques of data mining and the general statistic analysis, but indeed can find out their implicit characteristic as well as provide great effectiveness on the check of abnormal data with the data mining tools. The analysis on the potential for high cost on medical resources from the long-stay-patient groups not only can fast visualize out those groups of high risk, but can induce out the correlation among the long-stay-patient from the correlation rules and accordingly enhance the multi-management, assessment quality and time effectiveness of the health insurance.

並列關鍵字

Bureau of National Health Insurance ； Long-stay patients ； Data mining

參考文獻

20. 吳肖琪、林麗嬋、藍忠孚、吳義勇，全民健保實施後急性病床住院病患超長住院情形之分析，中華衛誌，17(2)，pp. 139-147，1998。

1. Aslandogan, Y.A., Mahajani, G.A., and Taylor, S., “Evidence Combination in Medical Data Mining,” Proc. of the International Conference on Information Technology: Coding and Computing, Las Vegas, Nevada, USA, pp. 465-469, 2004.

5. Coenen, F., Goulbourne, G., and Leng, P., “Tree Structures for Mining Associaction Rule,” Data and Knowledge Discovery, 8, pp. 25-51, 2004.

6. Craven, M.W. and Shavlik, J.W., “Using Neural Networks for Data Mining,” Future Generation Computer Systems, 13, pp. 221-229, 1997.

7. Fayyad, U., Piatetsky-Shapiro, G., and Smyth, P., “The KDD Process for Extracting Useful Knowledge from Volumes of Data,” Communications of the ACM, 39, pp. 27-34, 1996.