應用資料探勘於醫療資源之輔助規劃研究-以子宮肌瘤手術為例

A Study of Applying Data Mining to Medical Resource Planning — Using Myomectomy as an Example

鄭勝志(Sheng-Chih Cheng)

指導教授 ： 張俊郎

國立虎尾科技大學/管理學院/工業工程與管理研究所/碩士(2010年)

摘要

全民健保自 86 年3 月1日開辦以來，醫療費用與成本逐年增加，隨著全民健保之經營日益艱困下，其財政收支課題也備受各界所關心與爭議。在龐大的健保醫療支出中，住院費用就佔了33%，其中50%以上的住院醫療費用與住院日數有密切關係。根據臨床統計約四分之一女性發現有子宮肌瘤，子宮肌瘤是女性生殖器的一種常見的腫瘤，也是婦產科最為常見手術之一。通常在生育年齡的婦女中，發生率約為20%，隨著年紀愈大比例愈高。在國內每年健保支付醫療成本更不容小覷，故子宮肌瘤手術及住院費用一直是醫療研究重要的議題之一。
本研究應用粒子群最佳化演算法結合資料探勘中的決策樹與倒傳遞類神經網路作為探討子宮肌瘤手術住院日預測評估之工具。研究結果顯示，在腹腔鏡手術上粒子群最佳化演算法結合倒傳遞類神經網路正確分類率為89.47%；在開腹式手術上粒子群最佳化演算法結合倒傳遞類神經網路正確分類率為90.91%；在子宮鏡手術上粒子群最佳化演算法結合倒傳遞類神經網路正確分類率為87.88%；總結而言，本研究之結果已顯示系統的實際可行性，藉由評估住院日，提供醫師診斷時參考並提昇臨床醫療品質與醫療資源分配，使病患能獲得更好的醫療照護與服務品質。

關鍵字

子宮肌瘤 ； 住院日 ； 資料探勘 ； 粒子群演算法 ； 倒傳遞類神經 ； 決策樹 ； 腹腔鏡 ； 開腹式 ； 子宮鏡

並列摘要

Since the National Health Insurance launched in March 1st, 1997, the medical cost and expenditure is increasing yearly in Taiwan. With the hardships of managing National Health Insurance, the financial status, especially the balance, has been put under the spot light for public examination. Among the huge amount of health insurance expenditures, the hospital stay fees account for 33%, in which over 50% has high correlation with the length of stay. According to the clinical statistics, approximately a quarter of females are diagnosed with uterine fibroids, which are common benign tumors in female reproductive organs, and myomectomy is a common procedure in the department of obstetrician and gynecology (OB/GYN) to remove uterine fibroids. About 20% of the women during their reproductive years can have uterine fibroids and the likelihood increases with age. In Taiwan, myomectomies and the hospital stay fees associated with the procedure have accounted for a great proportion of the total costs in the National Health Insurance and thus remain top concern in the health care research.

In this study, particle swarm optimization (PSO) algorithm was used along with decision tree and back propagation neural (BPN) network of data mining as the tool to review the predictive assessment of the length of hospital stays after hysteroscopic myometomies. The results indicate that the correct classification rate for the laparoscopic myomectomy using PSO algorithm and BPN Network is 89.47%; that for the laparotomic myomectomy is 90.91%, and 87.88% for the hysteroscopic myomectomy. In sum, such results prove the practicality and feasibility of the system. With the evaluation on length of hospital stays, the system provides physicians a reference while making diagnosis as well as increases clinical medical care and allocations of medical resources so that patients can receive better health care and service quality.

並列關鍵字

Uterine Fibroid ； Length of Stay ； Data-Mining ； Particle Swarm Optimization (PSO) ； Back-Propagation Neural Network (BPN) ； Decision Tree ； Laparoscopy ； Hysteroscopic

參考文獻

1.中山醫學大學婦產部，http://w2.csh.org.tw/Web/A23010/?p=35，中華民國九十九年四月十二日。

2.中央健保局（即時統計資訊），http://www.nhi.gov.tw/index.asp，中華民國九十九年四月十二日。

3.王人澍、張寶源、熊雅意（2008），「應用決策樹理論於中醫辨證－以慢性咳嗽為例 」，中西整合醫學雜誌，10卷，2期，頁25-33。

4.王派洲 (2008)，資料探勘蓋念與方法，頁386-387，滄海書局，台中。

5.王偉驎、林文燦、賴政皓、陳慧敏（2008），「應用資料探勘技術提升急診醫學檢傷分類之一致性－以台灣某醫學中心急診醫學部為例」，品質學報，15卷，4期，頁283-291。