臺灣博碩士論文加值系統

自民國84年3月全民健康保險開辦以來降低了民眾就醫時的財務障礙，也提高了民眾對醫療品質的期望，可視為我國近半世紀以來最重要的社會福利建設。實施初期主要是採論量計酬支付制度，在尚未有其他較有效的支付方式配合下，論量計酬制的確存在若干缺點。為了使醫療資源有效運用，並給予醫療提供者適當誘因，使其能自發性配合保險人自我節制，健保局於保險規劃初期，最先配合實施制度的係「論病例計酬制，Case payment」。「論病例計酬制」這種「定額給付」的支付制度對於臨床行為、醫療資源耗用及醫療品質等方面在其他文獻中都有相當的關聯。而在日益增加的健保醫療支出中，住院部份醫療費用就約占了32％，其中50%以上的住院醫療費用與住院日數長短有密切關係，因此各醫療院所嘗試利用控制病患的住院日數以降低醫療成本，也促使醫師提昇醫療效率。
本研究利用個案醫院論病例計酬案件(全髖關節置換手術及全膝關節置換手術住院病患)住院時的各項資料來分析影響住院日數因子，並以資料探勘中的C4.5決策樹及類神經網路技術建構一個預測模型，運用十摺交叉驗證法進行網路學習，用以判斷病患的住院日數是否超過健保局所提供的平均住院日數，以輔助臨床醫護人員儘早瞭解病患的情況，並因此達到住院日控管機制。本研究亦利用特徵選取程序以達到縮減屬性數量的目的，用以提升模式建構的效率。
當我們利用人工關節置換術資料所進行的結果顯示出適當的方法縮減描述資料屬性的數量後，可如預期提升C4.5決策樹所建構之模式的效能。而且，減少屬性的數量同樣也提升了類神經網路分類模式的效能。因此，對於其他的病例，本研究建議可以利用恰當的特徵選取程序挑選出重要屬性後，再以此類方法建構病患住院日數分類預測模式。未來更應實際應用在臨床上以評估其可行性。

Since 1995, The National Health Insurance Program (NHI) that is one of the most significant public policies in recent decades has implemented in Taiwan; this program not only changed the medical environment, but also improved health quality and patients’ right. However, this program is also facing the cost-increasing problem as all other countries. One of the strategies to control cost is to change the payment method. Taiwan government has tried to carry out the "Case Payment System" which is a prospective payment system resemblance to the Diagnosis-Related Groups to replace the current Fee For Service (FFS) payment method for inpatient services. Among the great deal of health insurance expenses, the ratio of hospitalization is 32%, and above 50% of hospitalization fee is related to days to hospital stay. In response to rapidly rising costs, governments and providers have become aggressive in searching for mechanisms to control their expenditures. And one of the most important reasons why budget of hospitalization is pruned off is often due to the long days to hospital stay. Many hospitals devoted themselves to manage patients’ stay length decreasing the costs and promoting health providers’ service to higher qualities.
This study carried out to the analysis of the payment data (patients with Total Knee Replacement and Total Hip Replacement) using individual cases of the hospital database and the data elements were explored as a mechanism useful in the prediction of patient length of stay (LOS). This study applied C4.5-based data mining with the past data and neural network technologies to establish the prediction model and to detect if the patients’ length of stay exceeded the average length of stay. This can help health providers to better understand patients’ recovery and have an early prediction profound effect on more efficient and effective medical resource deliveries.
This study adopted attribute selective procedures to reduce unnecessary properties. Making the model construction more efficient and to understand the simplified classification rules. The reduction of attribute numbers will also increase the efficiency of the neural network. This study recommends that using a proper attribute selective procedure to figure out some important attributes. And then applying this two data mining technologies to setup the predictive model will be a more efficient way to set up a model and to bring about the classification rules. Futures studies should assess the feasibility of implementing model in clinical practice.

誌謝　i
中文摘要　iii
Abstract　v
Table of Contents　vii
List of Tables　ix
List of Figures　x
List of Abbreviations　xi
Chapter I Introduction　12
1.1 Background　12
1.2 Case Payment　15
1.3 Introduction of Total Joint Replacement　17
1.4 Data Mining Technology　21
1.4.1 C4.5 Decision tree　23
1.4.2 Artificial Neural Networks　25
1.5 Specific Aims　28
1.6 Organization of This Thesis　28
Chapter II Literature Review　30
2.1 Impact of Inpatient Medical Utilization　30
2.1.1 Factors affecting Medical Expenses　31
2.1.2 Factors of Influence on Length of Stay　32
2.2 Prediction of Hospital Length of Stay　33
Chapter III Establishing Prediction Model　36
3.1 Reduce Model Attributes　36
3.1.1 Attribute Selection Techniques　36
3.1.2 Principal Component Analysis　39
3.2 Procedures of Establishing Model　42
Chapter IV Results　44
4.1 Research Datasets　44
4.1.1 Objects　44
4.1.2 Variables　45
4.1.3 Evaluation Measures　47
4.2 Establishing Classification Model　50
4.3 Attribute Selection　55
4.3.1 Information Gain　56
4.3.2 Principal Component Analysis　57
4.3.3 Evaluation　60
Chapter V Discussion and Conclusion　63
5.1 Discussion　63
5.2 Limitations of the Study　64
5.3 Conclusions　64
5.4 Recommendations for Future Research　65
Bibliography　67
Appendix　71
List of Tables
Table4.1 Description and Data attributes of Total Joint Replacement　46
Table4.2 Decision Confusion Matrix table　49
Table4.3 Establishing Classification model from original data set　51
Table4.4 ANNs network structure parameter setting　52
Table 4.5 Accuracy, TP Rate and Roc of different network type (I)　53
Table4.6 Accuracy, TP Rate and Roc of different network type (II)　53
Table4.8 Accuracy, TP Rate and Roc of different network type (IV)　54
Table4.9 Accuracy, TP Rate and Roc of different network type (V)　54
Table4.10 Results for Attribute Selection with Information Gain　56
Table4.11 Establishing Classification model from original data set　57
Table4.12 Total Variance Explained of Principal Component Analysis　58
Table4.13 Factor Analysis of each principal component　59
Table4.14 Comparison with Predict Models　60
List of Figures
Figure1.1 Figure of Hip implant　19
Figure1.2 Figure of Knee implant　20
Figure1.3 General Structure of an Artificial Neural Network　25
Figure1.4 Structure of a Node in an Artificial Neural Network　26
Figure3.1 Procedures of establishing model　42
Figure4.1 Schema of Total Joint Replacement　45
Figure4.2 C4.5 Decision Tree　51
Figure4.3 Scree Plot　58
Figure4.4 C4.5 Decision Tree　61

[1].M. R. Kraft, K.C. Desouza, and I. Androwich, “Length of Stay Predictions for Spinal Cord Injury Patients Using Neural Networks and Nursing Diagnoses Data: Methodological Issues, Results and Applicability,” Margaret Ross Kraft, Jul. 2002.
[2].M. Kraft, K.C. Desouza, and I. Androwich, “Knowledge Discovery in Clinical Databases: Issues and Process,” In M. Khosrow-Pour (Editor) Issues & Trends of Information Technology Management in Contemporary Organizations, Hershey, PA: Idea Group Publishing, Vol. I, pp.168-171, 2002.
[3].袁繼銓, “以類神經網路預測燒傷病患住院日之研究”, 國立中山大學資訊管理研究所碩士論文, 2003
[4].Canadian Joint Replacement Registry, 2005. http://secure.cihi.ca/cihiweb/dispPage.jsp?cw_page=services_cjrr_e
[5].Herng-Ching Lin, Yu-Chi Tung1, Chu-Chieh Chen, Chao-Hsiun Tang, “Relationships between Length of Stay and Hospital Characteristics under the Case-Payment System in Taiwan: Using Data for Vaginal Delivery Patients,” Chang Gung Med J, vol. 26, pp. 259-268, 2003.
[6].“Data from the National Hospital Discharge Survey: Detailed diagnoses and procedures, 1990.” Vital Health Stat, no. 113, 1992.
[7].中央健康保險局，”人工關節置換抗生素使用日數分析”, 全民健康保險專業檔案分析報告, 2005
[8].S. Dumais, J. Platt, D. Heckerman, and M. Sahami, “Inductive learning algorithms and representations for text categorization,” International Conference on Information and Knowledge Management, 1998.
[9].張佳琪, “論病例計酬支付制度對骨科住院醫療利用情形之影響分析-以全髖膝關節置換手術及全膝關節置換手術為例”, 陽明大學衛生福利研究所碩士論文, 1999.
[10].European Federation of National Associations of Orthopaedics and Traumatology. http://www.efort.org
[11].G.P. Forrest, J.M. Roque, S.T. Dawodu, “Decreasing length of stay after total joint arthroplasty: Effect on referrals to rehabilitation units,” Archives of Physical Medicine and Rehabilitation, vol. 80, pp. 192-194. Feb. 1999.
[12].Canadian Institute for Health Information,” Total Hip and Total Knee Replacements in Canada. Ottawa,” Canadian Joint Replacement Registry 2004 Report, 2004.
[13].Allan Gross, ”Report of the Total Hip and Knee Joint Replacement Expert Panel,” Nursing Leadership (CJNL), vol. 18, no. 4, Sep. 2005.
[14].Anderson Orthopaedic Research Institute “What is a Total Hip Replacement?,” http://www.aori.org/
[15].Scotter, Weingarten et al, “Total Knee Replacement A Guideline to Reduce Postoperative Length of Stay Los Angeles,” California Ann Rheum, vol. 63, pp.825–830, 2004.
[16].Data mining http://en.wikipedia.org/wiki/Data_mining#Definition.
[17].About Data Mining http://www.twocrows.com/about-dm.htm.
[18].R. Margaret Kraft, C. Kevin Desouza, Ida Androwich, “Data Mining in Healthcare Information Systems: Case Study of a Veterans? Administration Spinal Cord Injury Population,” IEEE Computer Society DL, 2003.
[19].Eui-Hong Han, Anurag Srivastava and Vipin Kumar, "Parallel Formulations of Inductive Classification Learning Algorithm," U of Mn CS Technical Report, TR 96-040, 1996.
[20].Ron Kohavi , Ross Quinlan , “Decision Tree Discovery,” Oct. 1999. http://tandem.bu.edu/classes/2005s/papers/trees.pdf
[21].E. B. Hunt, “Concept Learning: An Information Processing Problem”, Wiley. 1962.
[22].Building Classification Models: ID3 and C4.5 http://www.cis.temple.edu/~ingargio/cis587/readings/id3-c45.html
[23].R. P. Lippman, “An Introduction to Computing with Neural Nets,” IEEE ASSP Magazine, pp. 4-21, Apr. 1987.
[24].J. Cyr, and G.. Haley, “Use of Demographic and Clinical Characteristics in Predicting Length of Psychiatric Hospital Stay , A Final Evaluation. ” Journal of Consulting and Clinical Psychology, vol. 51, no. 4, pp. 637-640, 1983.
[25].E. J.MacKenzie, J. A. Morris, and S. L. Edelstein, ‘‘Effect of Pre-existing Disease on Length of Hospital Stay in Trauma Patients,’’ Journal of Trauma, vol. 29, no.6, pp. 757–765, 1989.
[26].J. H. Siegel, S. Shafi, S. Goodarzi, P. C. Dischinger “A Quantitative Method for Cost Reimbursement and Length of Stay Quality Assurance in Multiple Trauma Patients,” Journal of Trauma, vol. 37, no.6, pp.928–937, 1994.
[27].林千琳，’’腦中風復健病人住院時間長短之相關因子與預測模型”, 臺中健康暨管理學院健康管理研究所碩士論文，2004
[28].A. Mark Hall, Geoffrey Holmes, “Benchmarking Attribute Selection Techniques for Discrete Class Data Mining,” IEEE Transactions on Knowledge and Data Engineering, vol. 15, no. 6, pp. 1437-1447, 2003.
[29].João E. Steiner , “World University Rnkings A Principal Component Analysis,” May. 2006. http://www.iea.usp.br/iea/ensinosuperior/artigosteinerrankings.pdf
[30].Desouza, K.C, “Knowledge management with artificial intelligence,” Westport, CT: Quorum Books, 2002.
[31].I. L. Rosemont, “Staff AR: Primary total hip and total knee arthroplasty projections to 2030,” American Academy of Orthopaedic Surgeons, pp. 7, Aug. 1998. http://www.aaos.org/wordhtml/research/stats/stats_3.htm.
[32].余珮琦, “住院醫療服務市場之供給誘發需求現象探討,” 國立臺灣大學衛生政策與管理研究所碩士論文, 2003年
[33].D. Berry, W. Harmsen, M. Cabanela, B. Morrey, “ Twenty-five year survivorship of two thousand consecutive primary Charnley total hip replacements, Factors affecting survivorship of acetabular,” J Bone Joint Surg Am, vol.84, pp. 171-177, 2002.
[34].J.R. Lave, and R.G. Frank, “Effect of the structure of hospital payment on length of stay,” Health Serv Res, vol. 25, no. 2, pp. 327-347, Jun. 1990.
[35].J. Lindemann, G.. Fairweather, G. Stone, and R. Smith, “The Use of Demographic Characteristics in Predicting Length of Neuropsychiatric Hospital Stay,” Journal of Consulting Psychology, vol. 23, no.1, pp. 85-89, 1959.
[36].S. Mendenhall, “2005 Hip and Knee Implant Review,” In Orthopedic Network News, pp. 1-16, 2005.
[37].J. R. Quinlan, “Programs for Machine Learning,” Kluwer Academic Publishers, 1993.
[38].S. Walczak, and J. E.Scharf, “ReducingSurgical Patient Costs Through Use of an Artificial Neural Network to Predict Transfusion Requirements,” Decision Support Systems, vol. 30, no. 2, pp. 125-138, 2002.