統計模式應用於胃幽門螺旋桿菌及胃癌前病變之家戶聚集性研究

Statistical Models for Family Aggregation of Helicobacter Pylori and Gastric Neoplasm

范僑芸(Chiao-Yun Fan)

指導教授 ： 陳秀熙

國立臺灣大學/公共衛生學院/流行病學與預防醫學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902662

摘要

背景
已有許多研究證實家戶內的胃幽門螺旋桿菌感染，但尚未有研究利用基因流行病學研究去量化家族聚集的影響。此外，盡管胃幽門螺旋桿菌的家族聚集已被推廣至胃部癌前病變如：萎縮性胃炎、胃黏膜腸上皮化生，卻忽略了家族聚集對疾病進程的動態影響。

目的
本論文的目的為利用Correa模式與先證者(Proband)的病例對照研究設計及其推展，配合不同統計模式去推估家族聚集對於胃幽門螺旋桿菌傳染及胃部癌前病變(萎縮性胃炎、胃黏膜腸上皮化生)進程的動態影響。

材料與方法
在此論文中，我們利用兩個資料集來研究家族聚集對胃幽門螺旋桿菌與和胃幽門螺旋桿菌相關之胃部癌前病變的影響。第一個資料集採用35個日本被感染胃幽門螺旋桿菌兒童的家族研究，其中包含兩種基因分型結果：多位點序列分型(MLST)及隨機擴增多態性DNA標記(RAPD)，用於檢測胃幽門螺旋桿菌的家族聚集與傳播途徑。第二個資料集是利用Correa模式與馬祖的胃部癌前病變的社區預防資料，包含三個時間段：1996-2003為篩檢介入之前，2004-2007為篩檢介入之後，2008為全面性投藥之化學預防。
我們利用兩階段先證者的病例對照研究去推估在相同MLST、RAPD家族聚集的影響，與貝式有向無環圖(DAG)計算在給定觀察資料與未知資料之下的條件機率，利用隨機效應模式計算家庭內部成員胃幽門螺旋桿菌感染的家族聚集影響。
我們利用多元邏輯斯迴歸分析、離散狀態及時間之馬可夫鍊模式與連續時間下馬可夫過程三種模型估計在Correa模式之下家族聚集對胃幽門螺旋桿菌感染與胃部癌前病變動態轉移的影響。
結果
第一部份：日本基因序列與胃幽門螺旋桿菌感染的家族聚集研究
當使用第一個孩子作為指標個案時，貝式DAG模型的DNA指紋序列的估計結果顯示匹配序列的家族聚集在MLST為56倍（95％CI：3.99-1878.07），及RAPD為68.10倍（95％CI：4.85-2171.12）。
第二部分：家族聚集對胃幽門螺旋桿菌感染及胃部癌前病變動態轉移平衡的影響
應用馬祖資料與多元邏輯斯迴歸模式，在調整抽菸與飲食習慣之下，家族聚集的影響在胃幽門螺旋桿菌感染、萎縮性胃炎與胃黏膜腸上皮化生皆有統計上的顯著，其估計值自1.39 (95% CI: 1.09-1.77) 至 2.02 (95% CI: 1.26-3.55)。
符合遍歷理論的馬可夫鍊的估計結果表明，在Correa模式之下於1996年存在強烈的家族聚集趨勢，其上三角形與下三角形的比為2.10。然而在2004與2008的相應數值為0.91與0.87。
連續時間馬可夫過程的估計結果證明家族聚集與Correa模式的動態轉移平衡存在高度相關。主要貢獻為從正常狀態到胃幽門螺旋桿菌感染，指標個案疾病狀態為胃幽門螺旋桿菌感染、萎縮性胃炎、及胃黏膜腸上皮化生其倍數分別為：1.30 (95% CI: 1.18-1.44), 1.21 (95% CI: 1.07-1.36), 和1.35 (95% CI: 1.14-1.59)。而在第二階段自胃幽門螺旋桿菌感染轉移至萎縮性胃炎，其倍數根據不同指標個案狀態為0.74 (95% CI: 0.62-0.88), 1.33 (95% CI: 1.10-1.60), 和1.06 (95% CI: 0.82-1.37)。在第三階段萎縮性胃炎轉移至胃黏膜腸上皮化生結果未達統計上顯著，倍數分別為1.28 (95% CI: 0.94-1.75), 0.95 (95% CI: 0.68-1.34), 和 1.14 (95% CI: 0.72-1.81)。
結論
本論文利用統計模式去評估胃幽門螺旋桿菌感染與Correa模式下胃部癌前病變動態轉移的影響。所提出的模式應用於一個具有基因分型結果的資料集，與一基於社區的胃部癌前病變資料在預防措施介入前與介入後。

關鍵字

胃幽門螺旋桿菌 ； 胃癌 ； 家戶聚集

並列摘要

Introduction
Although intra-familial aggregation of helicobacter pylori infection has been well studied before, the effect size of family aggregation has been scarcely elucidated by using a well-designed genetic epidemiological study. While the familial aggregation of HP has been extended to include pre-cancerous lesions like atrophy gastritis (AG) and intestine metaplasia (IM), the effect size of familial aggregation accounting for the dynamic transition study has been even neglected.
Aims
The objective of this thesis is to employ a case-control proband study design and its variants in conjunction with different statistical models to estimate the effect size of familial aggregation of genetic typing associated with HP infection and the dynamic transitions between HP, AG, and IM under the context of the Correa model.
Material and Methods
Two data sources were used for studying familial aggregation of HP and HP-related pre-cancerous lesions. The first dataset was derived from a 35 index Japanese pediatric patients familial study with the available information on two genetic typing procedures, Multilocus Sequence Typing (MLST) and random amplified polymorphic DNA (RAPD) fingerprinting used for detecting transmission route of familial aggregation. The second dataset was descended from the Matsu community-based prevention of gastric neoplasm under the context of Correa model with three periods, 1996-2003 before intervention, 2004-2007 after screening, and 2008 after chemoprevention. The case-control proband study design with two-state was designed to estimate the effect size of familial aggregation associated with the same sequence of MLST and also RAPD. Bayesian directed acyclic graphic (DAG) model was built up to develop fully conditional distribution given the observed data and unknown quantity to estimate the effect size of family aggregation of HP infection making allowance for the correlation of HP infection across the same family members with the random effect model.
To model how family aggregation affects the dynamic transition of HP and gastric pre-cancerous neoplasm under the context of the Correa multistate model, three statistical approaches were used, including the multi-nominal logistic regression model, the discrete-state and discrete-time Markov chain model, and continuous-time Markov process.
Results
Part I Japanese familial aggregation study on genotyping associated with HP infection
When using first child as index case the estimated results on the DNA fingerprint sequence with Bayesian DAG model show the tendency of familial aggregation for the matched sequence of RAPD was 56-fold (95% CI: 3.99-1878.07) for RAPD and 68.10-fold (95% CI: 4.85-2171.12) for MLST compared with the unmatched sequence.
Part II Family aggregation for the dynamic transition of HP and gastric neoplasm
With the application of multi-nominal logistic regression model to the Matsu data, the effect sizes of familial aggregation of HP infection, AG, and IM were statistically significant with the range from 1.39 (95% CI: 1.09-1.77) to 2.02 (95% CI: 1.26-3.55) after adjustment for smoking and dietary factors.
The estimated results of Markov chain with the ergodicity theory show there was a strong tendency of familial aggregation for reaching equilibrium with the property of the Correa model in 1996 but not in the two periods after interventions as the ratio of the summation of transition probability in the upper triangle to that of in the lower triangle was around 2.24 whether the corresponding values in 2004 and 2008 were 0.91 and 0.87.
The estimated results of continuous-time Markov process demonstrate family aggregation was highly associated with the dynamic transition of the Correa model mainly in the transition from normal to HP infection with the order of 1.30 (95% CI: 1.18-1.44), 1.21 (95% CI: 1.07-1.36), and 1.35 (95% CI: 1.14-1.59) for the index case of HP infection , AG, and IM. The influence of family aggregation dwindled from HP infection to AG with the order of 0.74 (95% CI: 0.62-0.88), 1.33 (95% CI: 1.10-1.60), and 1.06 (95% CI: 0.82-1.37) to AG to IM with insignificant findings, 1.28 (95% CI: 0.94-1.75), 0.95 (95% CI: 0.68-1.34), and 1.14 (95% CI: 0.72-1.81)
Conclusions
Statistical models are proposed here to model familial aggregation of HP infection and the dynamic transition of HP-related gastric precancerous lesion under the context of the Correa model. The proposed models were applied to one data with genotyping HP infection and the community-based data before and after prevention of gastric neoplasm.

並列關鍵字

Helicobacter Pylori ； Gastric Neoplasm ； Family aggregation

參考文獻

Aziz RK, Khalifa MM, Sharaf RR. Contaminated water as a source of Helicobacter pylori infection: A review. J Adv Res. 2015;6(4):539-47.

Bamford KB, Bickley J, Collins JS, Johnston BT, Potts S, Boston V, Owen RJ, Sloan JM. Helicobacter pylori: comparison of DNA fingerprints provides evidence for intrafamilial infection. Gut 1993;34:1348–50.

Blaser MJ. The bacteria behind ulcers. Sci Am 1996;274:104–7.

Chiu YH, Lin WY, Wang PE, Chen YD, Wang TT, Warwick J, Chen TH. Population-based family case-control proband study on familial aggregation of metabolic syndrome: finding from Taiwanese people involved in Keelung community-based integrated screening (KCIS no. 5). Diabetes Res Clin Pract. 2007;75(3):348-56. Epub 2006 Aug 28.

Chiu SY, Chen LS, Yen AM, Chen HH. Population-based proband-oriented pedigree information system: application to hypertension with population-based screening data (KCIS No. 25). J Am Med Inform Assoc. 2012;19(1):102-10.