利用時間序列廻歸分析糞便潛血檢驗值的時間趨勢和季節及環境溫度之變化

Abstract

中文摘要 背景 運用免疫法糞便潛血檢查之量性數值已在許多研究中證實對於大腸直腸腫瘤與病變之發生具有良好的預測能力，加上此檢查方法已被廣泛運用於族群大腸直腸癌篩檢，糞便血色素濃度檢測之穩定性相對形成感興趣的議題。運用不同的時間序列模式分析，可以探究溫度與季節變化等環境因素對於與血色素相關檢測結果之影響，這樣的研究議題在過去十分罕見。 研究目的 利用大規模族群篩檢所收集之糞便血色素濃度 以及血中血色素濃度之長期追蹤時間序列資料，本論文之研究目的如下： (1) 利用線性廻歸模式釐清時間相關因素如時間趨勢、溫度以及季節變化對於糞便血色素濃度數值量測所造成的影響， (2) 利用貝氏隨機效應自我廻歸模式(autoregression model)評估時間趨勢、溫度以及季節變異對於糞便血色素濃度數值量測之影響， (3) 利用貝氏隨機效應自我廻歸模式(autoregression model)評估時間趨勢、溫度以及季節變異對於血中血色素濃度數值量測之影響， (4) 利用傅立葉分析對於糞便血色素濃度之週期變化進行探討。

材料及方法 本研究利用2004年至2009年50-69歲共970,492位參與糞便潛血大腸癌篩檢，以及97,731位20歲以上2000年至2009年間參與社區整合式篩檢，具血色素檢測者之兩項時間數列資料進行相關分析。首先利用非貝式線性廻歸模型針對上述時間數列資料，針對年齡、性別、溫度、時間趨勢及季節變異對糞便血紅素之影響進行分析。再進一步利用貝氏隨機效益模式考量區域層級隨機效益後針對上述相關資料進行分析。 結果 分析結果顯示二階(second-order)自我廻歸對於後續糞便血色素濃度數值具有顯著的影響，因此後續在調整性別、年齡與疾病狀態後對於時間趨勢、季節變異與環境溫度分析模式主要基於此二階自我廻歸模式而進行。以週為時間單位之糞便血色素濃度不管有無調整性別、年齡與疾病狀態在時間趨勢上呈現顯著的增加效果 (2.67, 95%信賴區間: 1.02-2.46)。環境溫度對於糞便血色素濃度呈現負向之影響，其估計結果為-10.14 (95%信賴區間: -29.54-9.2)，但此一效果在調整疾病狀態後呈現相反之結果(15.84，95%信賴區間:6.17-25.8)。季節變異對於糞便血色素濃度之影響已有相類似的表現，與春季相較，冬季之糞便血色素濃度為最高(3.91，標準差:2.01，p=0.05)。季節對於糞便血色素濃度的影響在考二階自我廻歸的效應後即成為不顯著。 在對於血中血色素濃度的分析結果顯示，二階自我廻歸同樣對於血中血色素濃度亦有顯著影響。在調整性別、年齡、季節變異以及溫度後，季節變異對於血中血色素濃度即呈現與糞便血色素濃度相反之效果。在調整年齡、性別、二階自我廻歸血紅素、溫度、時間趨勢及季節變異等變項後，低血紅素較易增加糞便血紅素濃度(11.45, 95% CI: 0.14-22.73)。

結論 本研究以貝氏及非貝氏時間序列分析方法應用於糞便血紅素及血紅素相關時間數列包含時間趨勢、季節變異、溫度及自我廻歸等相關影響因子研究，主要研究發現包含由於氣溫或季節變化之效應，在夏天時產生較低糞便血紅素及較高血紅素數值，不論糞便血紅素或血紅素，均有明顯著之時間趨勢，以及糞便血紅素與血紅素間存有相關性。就方法學而言，這樣的季節循環也經由傅立葉轉換驗證。同時，藉由二階自我廻歸與受季節或氣溫影響之糞便潛血素降解關係中發現其中所隱含之生物性意義。

Background Due to the widely used fecal immunochemical test (FIT) in population-based cancer screening for colorectal cancer (CRC), elucidating time series data on fecal hemoglobin(f-Hb) is therefore of great interest on the ground of several reasons. As f-Hb has been demonstrated to be a good predictor for incident colorectal neoplasia it is therefore interesting to monitor the evolution of f-Hb so as to predict the time trend of CRC due to biological causes. This consideration together with seasonal variation and temperature both of which are central components of time series analysis model calls for the conduction of different types of time series analysis, which have never been addressed. Aims By using a longitudinal follow-up of time-series data on f-Hb and hemoglobin (H-b) from population-based colorectal cancer screening and community-based integrated screening data, my thesis aimed (1) to elucidate how time-series factors such as time trend, temperature, seasonal variation made contribution to f-Hb concentration using a linear regression model; (2) to evaluate the effect of time trend, temperature, and seasonal variation on fecal hemoglobin (f-Hb) concentration with Bayesian autoregression random-effect model; (3) to evaluate the effect of time trend, temperature, and seasonal variation on hemoglobin level of blood with Bayesian autoregression model; and (4) to use the Fourier analysis in modeling the periodical pattern of f-Hb concentration. Materials and Methods A total of 970,492 participates aged between 50 and 69 years who underwent CRC screening with FIT during the periods between 2004 and 2009 were enrolled and the totals of 97,731 subjects aged over 20 who attend the community-based integrated screening with hemoglobin information between 2000 and 2009 were enrolled in our analysis. The non-Bayesian linear regression models were applied to time-series data above to assess the effect of age, gender, temperature, time trend, and seasonal variation on f-Hb concentration. A Bayesian random-effect model with the incorporation of random effect at area level was further used to tackle the correlated property of data within the same region. Result As the second-order autocorrelation of f-Hb plays an important role in the subsequent outcomes of f-Hb, the results were mainly based on the second order autoregressive model for assessing the effects of time trend, seasonal variation or ambient temperature adjusting for sex, age, and possibly the disease status of CRC. There was a significantly increasing time trend (2.67, 95% CI: 1.02-4.26) on weekly f-Hb concentration regardless of whether seasonal variation or ambient temperature, age, gender, and disease status were adjusted. The inverse relationship of ambient temperature o f-Hb was found and estimated as -10.14 (95% CI: -29.54-9.2). However, the direction was reversed after adjusting for disease status (15.84, 95% CI: 6.17-25.8). Similar findings were noted for seasonal variation with f-Hb concentration mostly elevated in winter compared with spring (3.91, S.D.: 2.01, P=0.05). Such a relationship disappeared when second autoregressive order was incorporated. The results of H-b were similar to those of f-Hb, the second-order autocorrelation of H-b still plays the predominant role. The opposite findings compared with f-Hb were noted for seasonal variation of H-b when age, gender, seasonal variation, temperature, and the disease status were controlled. Considering the association between H-b and f-Hb, the weekly H-b was statistically associated with the corresponding f-Hb (11.45, 95% CI: 0.14-22.73) after controlling for age, gender, second autoregressive order, seasonal variation, and ambient temperature. Conclusion Time series analysis with Bayesian and non-Bayesian approach were applied to analyzing time trend, seasonal variation, ambient temperature, and autoregressive orders of the weekly fecal hemoglobin concentration (f-Hb) and hemoglobin (H-b). The main results included seasonal variation with lower f-Hb or higher H-b in summer or due to high ambient temperature, a significant time trend of f-Hb or H-b, and the positive association between concurrent f-Hb and H-b. From the aspect of methodology, such cycle of seasonal variation was confirmed by Fouier transformation. The role of second autoregressive order has a significant implication for the degradation of fecal sample attributed to seasonal or ambient temperature change when collected.