- 學位論文
利用全民健康保險資料庫建置門診類流感監測系統
Constructing the Influenza-like Illness ( ILI ) surveillance system by using the National Health Insurance Research Database in Taiwan
摘要
研究目的： 一、評估美國ESSENCE系統所歸類的類流感診斷與症候群組碼在台灣門診的可行性，並採各種不同的主、次診斷組合方式，找出具有本土化流感預測能力的症候群監測方式。二、嘗試建立台灣簡版的ESSENCE類流感診斷與症候群組。三、建立時間差(time lag)迴歸預測模式。 研究方法： 本研究是採用2005年5月22日至2006年12月30日的健保申報資料，以健保資料庫內的國際疾病分類號(ICD-9-CM)利用ESSENCE系統所歸類的類流感診斷症候群組碼進行資料的篩選，本研究以三種類流感病例定義「主診斷符合」、「主診斷且任一次診斷符合」、「任一個診斷符合」與醫院層級、就醫科別、地區交叉後共得二十四種模式，分別進行二十四種的類流感病例資料擷取，統計每一週符合ESSENCE類流感診斷與症候群組之病例數，進一步與我國疾病管制局之定點醫師監視系統中的通報病例數進行比對。且嘗試建立一個新的類流感診斷與症候群組碼，並以一週的時間差試圖了解所擷取的病例數,是否對於一定時間之後台灣類流感通報病例數具有預測能力。本研究採用pearson相關係數、簡單線性迴歸進行資料的分析與預測。 研究結果： 本研究的八十四週期間，經由Cook’s Distance影響力分析，找出極端值並去除後，其ESSENCE系統定義之類流感病例數與定點醫師監視系統的通報數之間，不論為何種模式，其相關係數均達0.80以上。最高的相關係數達0.875，二十四種模式當中，有十種模式的相關係數在0.80至0.845之間，更有十四種模式的相關係數高達0.85（含）以上，並且均達顯著性意義（p=0.000）。 三種的類流感病例定義，以「主診斷符合」下的八種模式其所有的相關係數皆達0.85以上（p=0.000），彼此之間差異不大，其中「全部醫院層級、全部科別、全台灣地區」此模式是最簡單方便的模式，與定醫通報病例數比較下，也可得到很高的相關係數為0.874（p=0.000），迴歸預測模式Y=795.346+3.257X，可解釋定醫的變異量為76.4%。 以「主診斷符合、全層、全科、全區」此一模式進行另一階段之分析，發現ESSENCE的ICD-9-CM碼中與定醫通報病例數相關係數達0.60以上的有：460、462、4640、46410、46420、4650、4658、4659、4660、4871共十個ICD-9-CM皆達統計上顯著性意義，並以這十個建立台灣簡版ESSENCE類流感診斷與症候群組，定點醫師通報的類流感病例人次數的相關係數達0.870（P=0.000），具有很強的線性相關，迴歸預測模式Y =1239.865＋3.385X，可解釋定醫變異量為75.7%。 一週的時間差(time lag)迴歸預測模式中，在預測定醫通報下一週類流感病例數的迴歸模式解釋能力上，以本研究所建立的台灣簡版的ESSENCE類流感診斷與症候群的解釋定點醫師通報數總變異量為最高（R2=74.1％），迴歸預測模式Y=1438.695+3.282X。 結論： 本研究發現ESSENCE類流感診斷與症候群組碼與台灣定醫類流感通報系統有高度相關，可適用於台灣門診的類流感監測。而在類流感的病例定義擷取時以使用「主診斷符合」定義病例、「全部層級醫院」、「全部科別」及「全台灣地區」此一組合模式下與定醫的通報達很高的線性相關，此也是所有模式中為最簡便的方式。所建立的台灣簡版的ESSENCE的類流感診斷與症候群組碼的相關性亦達有0.870，將來或可使用新的類流感診斷與症候群組碼進行資料的擷取，以提升類流感病例資料的特異性，而且發現ESSENCE類流感診斷與症候群篩檢法可預測一週後之流感趨勢。研究顯示利用現況健保資料系統下，ESSENCE類流感診斷與症候群組是可行的，若疾病管制局與中央健保局共同合作，可利用本研究之發現，發展出節省人力、具時效性之監測類流感系統，可作為國內流感防治監測上的另一個輔助方法。
並列摘要
Research Objectives：（1） Find out the appropriate surveillance system to predict influenza in Taiwan by assessing the feasibility of using the codes of the influenza-like illness (ILI) syndrome in U.S.A. ESSENCE system in the clinics in Taiwan and by adopting a variety of major diagnose and subdiagnose combination,（2）Try to set up a simplified influenza-like illness syndrome of ESSENCE system for influenza diseases in Taiwan. （3）Set up the time difference (time lag) mode of regression. Methods：This study adopted the influenza-like illness syndrome of ESSENCE system of ICD-9-CM to screen data from National Health Insurance Claim Data between May 22, 2005 to December 30, 2006..Three kinds of influenza definitions, [according to major diagnoses], was used. Data was cross-combined with hospital- level, department-level, and area code, and resulted in 24 models.. This research adopts Pearson correlation coefficient, simple linear return to conduct analyses and prediction of the data. Result： The study period was 84 weeks. We analyzed influence via Cook' s Distance, after finding out extreme value and removing, between influenza-like illness cases number of ESSENCE system definition with notification case number of sentinel surveillance system, no matter what kind of models, its coefficient correlation is more than 0.80. Supreme coefficient correlation up to 0.875 was highest, among the 24 models. Ten models had correlation coefficient between 0.80 and 0.845. The remaining 14 models had coefficient equal or greater than 0.85.and all reached significance variance (p =0.000). Of the three influenza-like illness cases definitions, the coefficient correlations of the eight models under [according with major diagnose] were all more than 0.85 (p =0.000), the differences among the models were not significant. Of the eight models, , the [level of all hospitals, all departments, the whole Taiwan area] is the simplest and most convenient model. When compared with sentinel notification case number, the coefficient correlation it was very high, 0.874 (p =0.000). The regression model Y =795.346+3.257X explained sentinel surveillance system as 76.4%. The data was analyzed by the models of 『according with major diagnose, all range, all department, all area』. The results showed that the group of 0.60 and above correlaton coefficients between ICD-9-CM of ESSENCE with sentinel notification case number included 460, 462, 4640, 46410, 46420, 4650, 4658, 4659, 4660, 4871. A total of ten ICD-9-CM reached statistical significance. The ten codes were used to establish a simplified Taiwanese influenza-like illness syndromic group of ESSENCE. The correlation coeffiecients between this group and the sentinel notification ILI case number reached 0.870 (P =0.000) and showed a strong linear correlation. The regression model Y =1239.865+ 3.385X explained sentinel surveillance system as 75.7%. A one-week time-lag to predict in the regression model, Predict sentinel notification in next week of ILI case number by regression model from simple way of Taiwan of ILI syndromic group of ESSENCE explain the sentinel notification total variance was most high(R2 =74.1% ), regression model Y=1438.695+3.282X. Conclusions：This study found the influenza-like illness syndromatic code of ESSENCE is highly relevant to sentinel surveillance system of Taiwan. It is appropriate for monitoring influenza-like illness in clinics in Taiwan.When using [according with major diagnose], [ all kind of level of hospital], [all department of hospital], and [whole area of Taiwan] to screen LI case, the results are linearly correlated with sentinel surveillance system. It was also the simplest method in all models. The simpilified edition of influenza-like illness syndromatic code of ESSENCE of Taiwan has a 0.870 relevance and may be used as a new screening method that would increase the specificity of data. In addition, we found ILI diagnosis of ESSENCE and syndromic screening can predict influenza trend in one week. The findings suggest that the utilization of the present health insurance data, together with influenza-like illness syndromes of ESSENCE to predict influenza is feasible. The collaboration between CDC authority and the Bureau of National Health Insurance can develop a more efficient and effective monitoring (ILI) system, and effectively monitor and prevent influenza in Taiwan.