牙醫門診風險計價模式之建立

Development of Risk Assessment Models for Dental Care

張明芳(Ming-Fang Chang)

指導教授 ： 張睿詒

國立臺灣大學/公共衛生學院/醫療機構管理研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.10430

摘要

國際間許多國家之健康保險制度皆面臨醫療費用高漲的問題，採前瞻性支付制度以為因應。前瞻式預算近年來多採用校正論人方式計算預算額度。台灣全民健保支付制度亦引入總額預算之方式，擬有效抑制費用成長，然其總額預算額度之設定並非藉由論人計酬方式，本研究首先發展牙醫風險校正論人計酬之計價模式，期提供未來前瞻性預算額度設定之參考。

本研究利用全民健康保險研究資料庫中2001年與2002年保險對象相關資料進行研究，隨機抽取樣本共計2,131,067人，以迴歸(Weighted Least Square)方式及分類與迴歸樹(Classification and Regression Trees, CART)方式分別建構三種不同之風險計價模式。人口因子模式選用年齡與性別為自變項，先前利用模式採用年齡、性別與前一年牙醫門診費用，診斷資料模式則利用年齡、性別以及9種與牙醫醫療利用相關之疾病，而6個模式的依變項皆為保險對象2002年年化後牙醫門診費用。

在PR2方面，利用CART方法之風險計價模式其預測力較迴歸方法略高，其中診斷資料模式預測力皆優於人口因子模式且不低於先前利用模式。在特定群體預測比方面，不論迴歸或CART方法，診斷資料模式與先前利用模式相當，此兩模式皆優於人口因子模式。

人口因子模式資料容易取得但預測力偏低，先前利用模式相較人口因子模式其PR2的提升有限，推論應與牙醫醫療特性及國內民眾就醫習慣有關，而診斷資料模式中3項牙科疾病可掌握樣本大部分的牙醫醫療利用，但其前後年個人醫療費用相關係數僅0.125，因此預測力有所侷限。未來可思考不同牙科疾病其第二年牙醫門診照護需要的差異，且是否有更細緻的資訊可提供區分同一疾病者對於次年需要之差異，藉此建構較為細緻的診斷群組，應可對未來牙醫門診醫療需要提供更為精確的預測。

關鍵字

牙醫 ； 風險校正 ； 分類與迴歸樹

並列摘要

Many countries have experienced the escalation of health care expenditure and have adopted prospective payment system. Under prospective payment system, risk-adjusted capitation has been considered and implemented to set budgets in recent years. In order to contain the growth of expenditure, Taiwan’s National Health Insurance program has adopted a global budget payment system, and budgets were not set through capitation. This study intends to develop risk assessment capitation models for dental care. The results may provide alternative approaches of allocating prospective budgets in Taiwan’s NHI program.

By using NHI’s research data bank of 2001 and 2002, 2,131,067 enrollees was randomly selected as the study sample. Three risk assessment models were developed through weight least square estimation and Classification and Regression Trees (CART) respectively. The independent variables of demographic model are age and sex. In addition to age and sex, while the prior utilization model includes individual’s 2001 outpatient expenditure of dental care as an independent variable, and the diagnostic model includes nine diseases as the independent variables. The dependent variable of the six models is enrollee’s 2002 annualized dental care expenditure.

The PR2 of risk assessment model through CART are slightly higher than those through regression. The PR2 of diagnostic model is better than demographic model and at least as good as the prior utilization model. The PR values of diagnostic model are similar to prior utilization model, and both of them are better than demographic model.

Demographic data are easy to gain but the PR2 is low. Although the PR2 of prior utilization model is higher than that of the demographic model, the improvement is somewhat limited. The reason may be related to the characteristics of dental care and the utilization behavior. The three oral diseases included in the diagnostic model account for most utilization of outpatient dental care; however, the correlation of individual expenditures between 2001 and 2002 is relatively low, only 0.125. The predictability of the diagnostic model is therefore restricted. It is encouraged to investigate the difference of the need for outpatient dental care in subsequent year for individual oral diseases. This subtle diagnostic information can provide a basis of developing refined diagnostic groups which may predict the need of outpatient dental care more accurately.

並列關鍵字

Dental Care ； Risk Adjustment ； Classification and Regression Trees

參考文獻

張睿詒、賴秋伶：風險校正因子：論人計酬醫療費用預測之基礎。台灣衛誌，23(2)，2004。

張舒婷：建構所有診斷資訊群組及其風險預測模式。台北：台灣大學公共衛生學院醫療機構管理研究所碩士論文，2005。

謝孟甫，張睿詒：處方資訊用於台灣風險校正模型之初探，台灣公共衛生雜誌2006：25(3)，189-200。

Chang RE, Lai CL: Using Diagnosis-Based Risk Adjustment Models to Predict Individual Healthcare Expenditure under the National Health Insurance in Taiwan. J Formos Med Assoc 2006. (Forthcoming).

Lin WD, Chang RE, Hsieh CJ, Yaung CL, Chiang TL: The development of a risk-adjusted capitation model based on principal inpatient diagnoses in Taiwan. Journal of Formosan Medical Association 2003;102:637-43.