台灣癌症的空間分析

Abstract

癌症是台灣歷年主要死因之首，為能瞭解存在於環境中的可能影響因素，流行病學界常透過癌症地圖來探索，然而光靠視覺判定往往不夠精確。近年來空間分析方法發展迅速，使研究者得以利用空間統計方法來探索現象的空間特性。 本研究利用行政院衛生署2003年出版之「中華民國癌症發生率及死亡率分佈地圖集」內女、男性標準化比率的電子檔資料，使用Moran’s I和LISA(Local Moran)指標來探討各種癌症的分布是否呈現聚集的型態，並瞭解癌症發生率、不同時期死亡率的空間變異。研究結果發現歷年女、男性五大主要癌症高發生率和死亡率類別均具有空間聚集的特性。癌症發生率具有空間聚集特徵的類型者，其死亡率亦有空間聚集的特徵；在空間上高發生率的聚集區通常也是高死亡率的分布位置。若以性別來分，胃癌是不分性別發生率與死亡率聚集區都位在東部；結腸直腸癌女、男性發生率聚集區位置相近、死亡率相近；口腔癌則不同性別聚集區位置不同等情形。 本研究選擇另以女性結腸直腸癌發生率和男性胃癌死亡率為例，透過傳統迴歸模型(ordinary least squares model)和空間誤差模型(spatial error regression model)分析，除了獲得地區社經差異的確影響空間聚集特徵的形成，本研究獲得尚有社經因子以外，具有空間聚集特徵之影響因子的存在，提供後續研究的基礎。另外，利用空間誤差模型，將影響癌症的因子具有空間相依性並非獨立存在的特性納入考慮，使參數估計結果較傳統迴歸模式適合且有效，值得相關學界使用。

Cancer is the leading causes of death in Taiwan. In epidemiology, disease maps are often used to explore the causes of diseases. Visual interpretation of these maps, however, usually leads to concerns of imprecision. This research seeks to take advantage of recent developments in spatial statistics to improve researchers’ exploration of the spatial characteristics and patterns revealed by data. The case in discussion is whether the distribution of female and male cancers exhibit spatial clustering by Moran’s I and Local Indicator of Spatial Association (LISA, local Moran) and to further understand the spatial variations. The dataset used in this study is from the statistical data of electric atlas of cancer mortality and incidence published by Taiwan’s Department of Health in 2003, which included age-standardized incidence and mortality rates. As shown in the results, the five main incidence and mortality rates of cancers over the past years demonstrated a spatial clustering characteristic. Spatially, areas of high incidence rates are usually of high mortality rate. The incidence and mortality rate of both male and female with stomach cancer (ICD151) are clustered in the east, the incidence rate cluster and mortality rate of male and female with colon, rectum, rectosigmoid junction & anus cancer (ICD153-154) tend to be relatively close, while the incidence and mortality rate of the population with lip, oral cavity (ICD140,141,143-146,148-149) will vary according to sex. The characteristic of spatial cluster can reveal whether the pattern correlates with socioeconomic differences in the area. Using the incidence of female with colorectal cancers and mortality of male with stomach cancer as examples, these cases show a clear characteristic of spatial clustering relevant to the socioeconomic factors of the area. Utilizing ordinary least squares model and spatial error regression model, this research found that other than socioeconomic factors, integrating spatial dependency is also important in tracing the causes of spatial clusters. The spatial error regression model also took into consideration of the fact that factors effecting cancer distribution are spatially dependent. The result has proven the model to be much more robust and effective in parameter estimations.