摘要:醫療費用成長過快,是全世界開發國家所面臨的問題,除了人口高齡化的影響, 新醫療科技不斷推陳出新,也是原因之一。傳統的醫療保險給付模式是依據提供的服 務計價,並沒有考慮新科技是否能提供較佳的治療結果,所以醫療費用無限制膨脹。在 此背景下,美國總統發起以病人預後為中心的醫療改革,其中,相對效應研究便在此 還環境下孕育而生自2010年逐漸發展的一項新型態研究,目的是以病患治療成果為中 心,進行治療選擇的一對一比較,提供臨床醫師、病患、醫療保險、新科技使用一個 客觀決策的數據。為此,美國也成立了 Patient centered outcome research institute,投入大量經費,促進相關領域發展與研究,並孕育利用相對效應研究成果 進行醫療決策的文化。相對效應研究的方法可以透過兩方面達成,第一、系統性回顧,並進行比較性統 合效用分析(comparative meta-analysis),現有的臨床研究為數相當龐大,但是受限 於研究族群的異質化,往往無法達成一致性結論,透過上述可以實現醫治性結論,或 指出知識鴻溝(Knowledge gap),引導未來研究。第二、利用原創性研究產生新的相對 效用證據,臨床試驗是決定相對效應研究最無偏差的方法之一,但受限於資源投入龐 大與耗時,同時也排除老年族群,臨床試驗的方法並不能普遍應用於相對效用研究。 觀察性研究目前是相對效用研究的主要方法,因為具有樣本大、觀察期長、符合臨床 情境等特點。台灣健保資料庫的建立近幾年來帶動了國内藥物流行病學、醫療經濟學、 以及其他臨床研究的蓬勃發展,也為相對效用研究提供一個沃土,然而健保資料庫的 研究本質是觀察性研究,並非隨機臨床試驗,在研究設計上有干擾誤差、選擇性誤差 的可能,不同的設計分析的結果差異很大,影響到結果的可信度。因此,建立嚴謹的 研究設計與分析方法,是健保資料庫研究的首要課題。另一方面,健保資料庫是海量 資料,傳統的流行病學設計與分析方法並不足以充分利用健保資料庫大資料的優勢, 新的高維度傾向計分法利用人工智慧電腦自動選取干擾因子,可以大幅提升干擾因子 的控制,但是對於許多相對少數的藥物暴露與預後,高維度傾向計分法卻無法有效控 制干擾。本計晝要發展的改良式高維度傾向計分,計晝利用人工或資料掘礦的技術縮 減資料維度,可以提升干擾因子控制的水平。相對效應研究相關的概念在台灣尚處於萌芽階段,醫師科學家或流行病學家投入相 關的領域為數尚少。本研究的目的在於推廣相對效應研究概念,研究内容有三部分, 第一部分,利用系統性文獻回顧與比較性薈萃分析,進行一系列研究,回答使用新生 物指標對於急性疾病的診斷是否優於傳統檢驗,第二部分,利用健保資料庫示範以病 患預後為中心的相對效應研究方法,研究疾病選取對健保資源耗用大年人最容易罹患 的社區肺炎為標的,第三部分,研究提升大型資料庫觀察性研究干擾控制的改良高維 度傾向計分方法。
Abstract: The rapid growth of health care cost is a severe problem faced by virtually allindustrialized countries. In addition to the aging population, the rapid emerging new medicaltechnology is another important cause for the sky-rocketing medical cost. Unfortunately,traditional payment system is based on the quantity of service, and not on the quality oroutcome of the treatment. Therefore the health care cost is rising without any control. Thegoal of “ObamaCare”, the healthcare reform proposed by the US president, is to rebuild aneffective and efficient health care for the US. Comparative effectiveness research (CER) is anew type of research that gradually developed out of “ObamaCare” in 2010. CER focused onpatient-centered outcome. CER promotes head-to-head comparison of different treatment,diagnosis, or health care delivery modalities. Results of CER would provide importantevidences for physicians and patients to select for the best clinical treatment. Large amount ofresearch funding are invested through the Patient Centered Outcome Research Institute(PCORI) in the US. This institute not only funds research, but also fosters a culture thatmedical decision should be informed by CER data. Unfortunately, this kind of institute islacking in Taiwan.CER could be done in two ways. First, CER can be done by a systemic reviewwith/without comparative meta-analysis. Clinical researches are often subjected to smallsample size and heterogeneous patient population. Systemic review and meta-analysis mayhelp partially address the problem, if not; it can at least point out a knowledge gap that guidesthe direction of future research. Second, CER can be also be done by conducting an originalresearch. Although randomized controlled trials are viewed as a gold standard for CER, it isoften limited by the high cost, long duration, and selected patient population, especially themulti-morbid elderly population. Observational studies using large database may be moreappropriate for CER in most circumstances. This is because, it can include a largehomogenous population with long term follow up that can reflect the real world situation. Therelease of national health insurance claims database (NHIRD) provides a good material forpharmacoepidemiology, pharmacoeconomics, and would also be suitable for CER. However,it has to be noted that study based on NHIRD is observational in nature. Non-randomizedobservational studies are usually subjected to selection and confounding bias. In addition, theNHIRD meets the big data definition. Traditional study design and analysis usuallyunder-utilize the data. The newly developed high dimension propensity score (hdPS) uses theartificial intelligence principle that has the statistical algorithm to mine “confounding factors”by itself. The introduction of hdPS has improved the confounding control in many cases, butin studies with rare exposure or rare outcome, hdPS may not be valid. Newer methods need tobe developed.In Taiwan, the concept of CER is still in its infancy. The goal of this project is topromote the use of CER to answer clinical and health policy decision questions. There arethree specific aims of this project. First, we will use systemic review and meta-analysis toanswer whether new biomarkers performs better than traditional laboratory tests in diagnosingacute critical illness. Second, we will demonstrate a patient-outcome based CER in NHIRD.Specifically, we will compare the treatment failure rate of guideline recommended antibioticsregimen, in the treatment of community acquired pneumonia. Third, we will modify the hdPSalgorithm by aggregating the medication and diagnosis codes into larger groupings. Theresults would lead to better confounding control in observational studies.