比較空間面疾病監控方法的新評估機制

Abstract

為了防止一些嚴重的傳染疾病發生範圍持續擴大，已有許多學者提供方法以期望能快速和準確地偵測到疾病族群(disease cluster)，稱之為疾病監控方法(disease surveillance method)。在許多種疾病監控方法中，需要一個好的機制評估該些方法的表現。所以本研究針對空間面的疾病監控(spatial disease surveillance)，提出了一套新的評估機制，並採用Tsui et al.(2012)所整理出的偵測統計量做為驗證本研究的範例。 疾病監控除了希望能偵測到疾病族群外，也注重判斷族群位置和範圍大小的精確程度，也可以將其視為區域的分類率。因此，醫學研究中所使用的False positive rate(FPR)和True positive rate(TPR)的概念，在評估空間面的監控表現上將是有用的工具。疾病監控統計量的表現受到許多因素影響，例如疾病族群位置、疾病族群範圍大小和患病比例改變的幅度(change of magnitude)是否已知，都會使得統計量的表現有所不同。本研究提供的評估機制，能方便且簡單地比較統計量在許多不同因素影響下整體的表現。 研究過程則是先以蒙地卡羅法模擬統計量的疾病監控結果，並採用圖形法觀察各個統計量的表現。由於圖形數據龐大，所以進一步利用FPR和TPR將繁雜的資訊統整為簡單明瞭的圖表，此作法不但不會漏失各種不同影響因素下的資訊，還能清楚看到各個統計量的整體表現。最後，本研究提供了評估指標，致使任兩種疾病監控方法之中，能有依據評斷誰的整體表現較好，並將實務中，FPR和TPR重要性不同的可能性列入考慮，當作影響指標的因素。

In order to prevent the outbreak of serious infectious diseases continues to expand, many disease surveillance have been developed with the attempt to detect the disease cluster as quickly and accurately as possible. In this study, we propose a new evaluation mechanism to compare the performance of various disease surveillance methods. The spatial disease surveillance statistics in Tsui et al. (2012) are used to demonstrate our evaluation mechanism. In addition to detecting the existence of the disease cluster, correct identification of the outbreak regions is important for disease surveillance. Thus, we also pay attention to the accuracy of the estimated cluster location and coverage size. The metrics of the false positive rate (FPR) and true positive rate (TPR), which have been used in medical research, could be useful criteria to evaluate the performance of disease surveillance methods. The performance of each disease surveillance method is under the influence of several factors, including the disease cluster location, cluster coverage size, and the magnitude of the outbreak. The evaluation mechanism provided in this study can easily and simply compare the overall performance of the surveillance statistics under several factors. We use Monte Carlo method to simulate incidence counts for the regions under various outbreak scenarios. Tsui et al. (2012) presented a useful graphical method to compare surveillance methods. However, it is quite tedious to compare so many pairs of graphs. Instead, we propose plotting TPR against 1-FPR for various values of the outbreak magnitude to organize complex information into a simple graph for comparing two methods. This approach not only keeps all important information under a variety of factors, but also allows us to clearly see the overall performance of each statistic. Because the comparison results may not be consistent with the two performance measures, we propose combining the two measures, FPR and TPR, into one composite metric so that a ranking can be made easily with an overall performance evaluation across various values of the outbreak magnitude. In practice, the importance of the FPR and TPR could be different and that should be taken into account as a factor in the composite metric.