近年來，非緊急病患急診次數大幅成長，醫學中心的急診服務量也同時增長。為探討最適政策，落實分級醫療的可能性，本計劃首先研究民眾面臨醫療院所選擇時，在部分負擔費率，交通成本，以及各層級醫院在醫療品質上的差異取捨的判斷準則。本研究從顯示性偏好的角度以最大分數估計法（Maximum Score Estimation）對2005至2015年間輕症急診病患醫療機構之選擇進行估計。這樣的估計策略有以下幾點優勢。首先，由於民眾的醫療機構選擇集合係依隨市場而定義，因此在估計時，龐大的候選集合對於最大概似估計所造成的計算負擔隨市場規模而指數增長，而使logit與probit在實務上難以執行。其次，在分析健保資料時，基於個資保護的原則，我們僅能觀察到病患的最終決策，因而無法控制影響病患選擇的重要變數。顯示性偏好分析的特性恰好能夠適當地處理未觀測到的異質性（Unobserved Heterogeneity）所導致的偏誤。本計劃將利用估計參數進行反事實分析（Counterfactual Analysis），模擬在各種不同自負費率與交通成本下輕症病患越級就診的變化。計畫執行初期估計範圍將以臺北都會區為主，並視進度延伸至全臺六個主要一級醫療行政區。 ;Non-urgent emergency department visits have been increasing in recent years in Taiwan. In order to explore the optimal policy design for the implementation of the referral system, this project first studies the decision rules patients hinge upon when confronting tradeoffs among co-payment rates, transportation costs, and quality of healthcare from hospitals of different size.Using the 2005-2015 NIH dataset, this project estimates non-urgent patients’ discrete choices on the emergency services, via the maximum score estimation from the perspective of revealed preference analysis. Our estimation strategy has the following two advantages. First, the agents’ choice sets are constructed by the way markets are defined. Therefore, complexity of computation grows exponentially with the size of markets during estimation, rendering multinomial logit and probit models practically infeasible.Furthermore, the NHI data is highly sensitive, and hence only hospital choices made by agents are observed, whereas many important characteristics of agents in the markets might not be disclosed. In this regard, revealed preference analysis serves as a good solution to address the bias arising from the unobserved heterogeneity. In this project, the counterfactual analysis will be conducted based on the estimated coefficients, where changes in the non-urgent use of emergency room at medical centers will be examined under various counterfactual co-payment rates and transportation costs. The project will first focus on observations in the Taipei metropolitan, and data from all six metropolitans in Taiwan will then be incorporated into analyses in the future.
