- 學位論文
藥物經濟學中費用與效能比的不偏估計量之研究與探討
Bias and MSE of Incremental Cost-effectiveness Ratio
摘要
近年來，與民眾息息相關的健保局財務亮起紅燈，其相關人士積極研究如何減少健保支出，並且維持相關品質，甚至是提升醫療水準，是大家所關心的問題。而提高保費、或調整藥價基準等政策，皆會引起社會大眾的關心注意，甚至是批判，因此，減少可避免的醫療費用、適當分配有限的醫療資源，必能降低健保局財務方面的困難問題。在此篇文章當中，利用了Number Needed to Treat (NNT)的概念，來對藥物的治療效果做討論，並且加入Incremental Cost-effectiveness Ratio (ICER，增加成本效能比)的計算討論。 根據文獻研究，機率的倒數 之不偏估計量是找不到的，但可以找一較為不偏的估計量來作為估計，利用Truncated Power Series Estimators (TPSE)的方法對NNT與ICER作估計，進而與一般估計方法做MSE和Bias的比較。藉由R統計軟體產生隨機變數，計算NNT與ICER等相關數據，結果顯示出當樣本數越大，利用TPSE估計NNT的Bias較一般估計方法接近於0，且MSE會比一般估計方法估計NNT的MSE穩定並較小，也利用了TPSE的方法來做ICER與一般估計方法做ICER的估計比較，不同的相關係數與樣本數，Bias與MSE皆有所改變，其中當樣本數增加，TPSE之Bias與MSE皆比一般估計方法之Bias與MSE小。而根據模擬結果顯示，利用TPSE的方法來估計NNT與ICER的數值，MSE皆為較小，且MSE為Variance與Bias平方的和，因此我們可以說它是趨近於UMVUE。 文章的尾端，加入二個類風濕性關節炎之真實醫療數據，利用一般估計方法與TPSE方法對NNT與ICER作相關計算，並且求出其信賴區間。在Responders夠大，不論是估計NNT或是估計ICER，在相同信心水準之下，使用TPSE之估計所得到信賴區間，皆比利用一般估計方法之信賴區間較為窄短，因此，我們可認為利用TPSE之估計方法估計NNT與ICER較具為精確。
並列摘要
Recently, the finance of National Health Insurance Bureau is closely related to public has been alarmed, so everyone concerned is that stakeholders are researching that how to reduce the payment in National Health Insurance and keeping the quantity, even that trying to promote the level of medical treatment, it is a question for people. However, raising the premium or changing the price of drug which attracts to everyone even criticizes it. Therefore, to reduce avoidable medical costs, the appropriate allocation of limited medical resources, National Health Insurance Bureau will be able to reduce the financial difficulties. In this article, we discuss treatment effect of drug using NNT and ICER. According to the literature, the unbiased estimate of the reciprocal of the probability is not available, but can find a much unbiased as the estimative value is estimated, we can obtain the data of NNT and ICER to compare by TPSE and generally estimates. By R statistical software random variables, to calculate the NNT and the ICER and other related data. The results show that when the sample size increases, the Bias of NNT estimated by TPSE is closer to 0 than generally estimates, and the MSE of NNT estimated by TPSE will be much stable and smaller than by generally estimates. Compare the results of ICER estimated by TPSE and generally estimates, different correlation coefficients and sample size will change the average error and MSE. When the sample size increases, Bias and MSE of TPSE is smaller than of CE. According to the results, estimated MSE of NNT and ICER by TPSE is smaller than CE and MSE composes Variance and square of Bias. So it can be say that it is near UMVUE. Finally, adding two examples of medical data, in the same confidence level, the confidence interval of NNT and ICER by TPSE are narrower than generally estimates. The results shows that estimate the NNT and ICER by TPSE is more precisely and possess superior performance in terms of bias and MSE.