應用資料挖礦技術於全民健康保險研究資料庫-以骨質疏鬆症為例

Applying Data Mining Technology on National Health Insurance Research Database-For Example：Osteoporosis

陳曉芬(Hsiao-Fen Chen)

指導教授 ： 楊燕珠

大同大學/經營設計學院/資訊經營學系所/碩士(2006年)

摘要

隨著醫療水準提高，國人平均壽命逐年增加，一些慢性病也隨著壽命延長而影響到國人健康。骨質疏鬆症是慢性病的一種，通常要等到發生骨折或引起其它合併症後才明白罹患此病。而世界衛生組織曾宣稱骨質疏鬆症是僅次於心臟血管疾病的第二大難題，並特別提醒這是一個常受忽視、且診斷不足的疾病，各國都必須將該疾病認定是一個應該提高警覺的重大公共衛生問題。
由於健保資料庫累積了大量的門診就醫資料，而這些資料中可能隱藏許多的資訊是我們尚未發覺的知識，藉由資料挖礦的技術可將資料庫中有用的知識挖掘出來，以找出具參考意義的醫學知識。我們利用全民健康保險研究資料庫之1997~2000年的系統抽樣檔之門診處方及治療明細檔和基本資料檔之醫事機構基本資料檔為資料來源，擷取了相關的疾病分類代碼之骨質疏鬆症患者為研究對象，目的為分析骨質疏鬆症患者的人口學特性與就醫習慣、骨質疏鬆症與其他疾病的關連性、並分析各醫療院所申報費用之分類。
描述性統計分析結果顯示1997~2000年國人罹患骨質疏鬆症以女性居多，且年齡多集中在51~70歲，男性年齡則集中在61~70歲。罹患縣市別以金門縣居首、台中縣次之。國人就醫習慣以掛號骨科為主，若單就女性而言則是掛婦產科居多。又病患的就醫之醫事機構層級別則以地區醫院居多，其次是醫學中心。
關聯規則結果顯示出已知的疾病骨質疏鬆症與婦女停經狀況有關係，也發掘出未知的規則如骨質疏鬆症與女陰陰道炎所導致更年期徵候群之關係。而利用決策樹的技術分類結果，找出各醫療院所的申報費用皆為低費用群組較多，故此研究亦可提供給健保局參考。

關鍵字

決策樹 ； 關聯規則 ； 資料挖礦 ； 全民健康保險研究資料庫 ； 骨質疏鬆症

並列摘要

Along with the exaltation of medical treatment level, the local people’s life expectance has increased year by year; some chronic diseases have affected the health of local people gradually with the prolonged life. Osteoporosis, as one kind of the chronic diseases, has ever been claimed by WHO that this disease is the secondary serious problem only next to cardiovascular disease, and particularly emphasized that this is a common ignored disease under improper diagnosis procedure, therefore all countries in fact shall affirm such disease as a major public health problem which needs to be taken care with a heightened vigilance.
The present study utilized the claims data of the Bureau of National Health Insurance (BNHI) from hospitals throughout Taiwan during the period between 1996 and 2000 which contains “Prescriptions for Outpatients and Detailed Therapeutic File”, and “Registration file” as the data source, and retrieving the patients with osteoporosis among the related classification and code of disease as the research objects, so as to analyze both the demography characteristic and registration habits in seeking medical advice for the patients with osteoporosis, as well as the connectivity to other diseases, in addition to the analysis on the classification of medical expenditure claimed by each medical institution.
According to the results of Description Statistical Analysis, it showed that the majority of local patients with osteoporosis during 1997-2000 are female, between the ages of 51-70, and the most male patients are between 61-70 years old. The primary residence area for most patients is the Kinmen County, and the next one is the Taichung County. As to the registration habits in seeking medical advice for the local patients with osteoporosis, we found that most of them were registered for the treatment of orthopedics surgery, in which most female patients were registered for obstetrics & gynecology treatment. Where the local hospitals were the most popular medical institutions, and the next were the medicine centers.
In light of the results of Association Rule, it revealed that the known disease of osteoporosis has a closed relationship with postmenopausal women and also revealed some unknown rules, such as the relationship between menopausal syndrome with the osteoporosis and vulvovaginitis. Also, the classification results acquired by utilizing the technique of decision tree showed that medical expenditure claimed by most medical institutions were laid in the group of low expenditures, thus the results of this study can be provided for the National Health Insurance Bureau as an applicable reference.

並列關鍵字

Association Rule ； Data Mining ； Decision Tree ； National Health Insurance Research Database ； Osteoporosis

參考文獻

19. 曾淑芬，從醫院管理角度論全民健保資料庫，中華衛生公共雜誌，1999年，第18卷，第5期，頁363-372。

20. 葉啟昌，林宏達，男性骨質疏鬆症，中華民國內分泌暨糠尿病學會會訊，2001年，第14卷，第4期，頁34-37。

25. 鄭守夏，全民健保學術資料庫簡介，中華衛生公共雜誌，1999年，第18卷，第3期，頁235-236。

1. Agrawal, R. and R. Srikant, Fast Algorithms for Mining Association Rules in Large Databases, Proceedings of the 20th Conference VLDB, Chile: Santiago, 1994, pp. 487-499.

6. Fayyad, U., Piatetsky-Shapiro, G. and P. Smyth, The KDD Process for Extracting Useful Knowledge from Volumes of Data, Communication of the ACM, 1996, Vol. 39, No. 11, pp. 27–34.