陳秀熙臺灣大學：預防醫學研究所許秀卿Hsu, Hsiu-ChingHsiu-ChingHsu2007-11-282018-06-292007-11-282018-06-292004http://ntur.lib.ntu.edu.tw//handle/246246/59187前言：最近部分研究以血清生化指標使用於胃癌篩檢，但是生化指標的切點在各個研究並無定論，尤其並沒有將年齡以及平衡偽陰性與偽陽性所決定的風險值列入考慮。 目的：以貝氏分析應用於胃癌及腸型化生之個別風險評估。 方法：將相關的生化指標以對數及平方根轉換使之符合常態分佈。透過貝氏方法進行單變項及雙變項分析。研究資料來自於馬祖地區所進行的胃癌及癌前病變社區篩檢，運用蒙地卡羅馬可夫鏈估計後驗風險比值及95%信賴區間。 結果：除了年齡因素，胃蛋白脢原I為診斷胃癌最重要的指標，其次為癌胚抗原，兩者具有邊緣統計上相關。在腸型化生以幽門螺旋桿菌感染分層，胃蛋白脢原I及胃蛋白脢原I/II比率為統計上顯著相關的因子。將上述因子進行單變項及雙變項分析，計算後驗風險比。再以預先訂定的風險值及效用比，求得適當的切點以及相對的敏感度及特異度。 結論：由方法學的角度而言，我們以貝氏分析模式進行個別胃癌及腸型化生的預測。由胃癌及癌前病變的篩檢角度而言，再以生化指標作為篩檢工具，對於切點的選擇本研究對於政策的擬定有很大幫助。Background: The recently proposed serum marker for gastric cancer screening has been criticized by lacking of appropriate cutoff point determined by age and predetermined risk level related to the trade-off between false negative cases and false positive cases. Objective: A Bayesian model was proposed to estimate individual risk for gastric cancer or intestinal metaplasia. Methods: Univariate and bivariate analysis using Bayesian approach were developed assuming normal distributions after log or square transformation of relevant serum markers. This model was applied to data from community-based screening for gastric cancer or its precursor in Matzu. Monte Carlo Markov Chain (MCMC) simulation was applied to estimate posterior odds ratios and 95% confidence interval. Results: In addition to age, PG I was selected the most important marker for ascertaining cancer and CEA was of borderline statistical significance. Similarly, PGI and PGI /PG II ratio were two significant factors for predicting IM after the stratification of the presence of HP infection. Posterior odds were all calculated for univariate and bivariate analysis. The selected cutoff points in relation to sensitivity and specificity given predetermined risk level and utility ratio were demonstrated. Conclusions: From the methodological viewpoint, we developed a Bayesian model for individual risk prediction for gastric cancer or intestinal metaplasia. From the aspect of screening for gastric cancer or precursor, this approach plays an important role in mass screening for gastric cancer or its precursor with serum marker.1 Introduction 1 1.1 Criticism on biological markers of screening for gastric cancer 1 1.2 Mass screening for gastric cancer and precursor with serum markers in Matzu 2 2 Literature Review 5 2.1 The role of H. pylori infection 5 2.2 Serum pepsinogen as a marker for gastric cancer and precancerous lesion 8 3. Material and Method 15 3.1 Study Subjects 15 3.2 Model Specification with Bayesian approach 16 3.2.1 Univariate analysis 16 3.2.2 Bivariate analysis 18 3.2.3 Marginal analysis 20 3.3 Optimal cutoff point by utility ratio 20 3.4 Bayesian inference with Markov Chain Monte Carlo (MCMC) simulation 22 3.4.1 Directed graphic model 23 3.4.2 Markov Chain Monte Carlo (MCMC) techniques 25 3.4.3 Gibbs Sampler 26 4. Results 29 4.1 Basic Findings 29 4.2 Risk prediction with PG I (Univariate analysis) 31 4.2.1 Gastric Cancer 31 4.2.2 Intestinal Metaplasia (IM) 32 4.3 Risk Prediction with PG I and CEA (Bivariate Analysis) 32 4.3.1 Gastric Cancer 32 4.3.2 Intestinal Metaplasia (IM) 33 4.4 The determination of optimal cutoff point 34 5. Discuss 37 Reference 41 Appendix 133 Figure List Figure 2.1 Multi-scale model for gastric carcinogenesis 45 Figure 3.1 Flow chart of two-stage screening for gastric cancer 46 Figure 4.1.1 The distribution of age of cancer and non-cancer 47 Figure 4.1.2 The distribution of PGI of cancer and non-cancer 47 Figure 4.1.3 The distribution of PGII of cancer and non-cancer 48 Figure 4.1.4 The distribution of PGI /PGII ratio of cancer and non-cancer 48 Figure 4.1.5 The distribution of CEA of cancer and non-cancer 49 Figure 4.2.1 The distribution of age of IM and non-IM 49 Figure 4.2.2 The distribution of PGI of IM and non-IM 50 Figure 4.2.3 The distribution of PGII of cancer and non-cancer 50 Figure 4.2.4 The distribution of PGI /PGII ratio of IM and non-IM 51 Figure 4.2.5 The distribution of CEA of IM and non-IM 51 Figure 5.1 The distribution of score of cancer and non-cancer 52 Figure 5.2 ROC curve for cancer prediction 52 Figure 5.3 ROC curve for IM prediction 53 Table List Table 2.1 Summary of epidemiological studies on the association between helicobacter pylori infection and the development of gastric cancer and precancerous lesions 55 Table 2.2 Summary of epidemiological studies of pepsinogen as a marker for gastric cancer or precancerous lesions 59 Table 4.1 The frequencies of demographic and biochemical variables by the presence of gastric cancer 68 Table 4.2 The frequencies of demographic and biochemical variables by the presence of intestinal metaplasia 69 Table 4.3 Frequency of demographic and biochemical variables by the presence of atrophic/superficial gastritis 70 Table 4.4.1 Posterior odds (1:n) of developing gastric cancer in Male given PGI level 71 Table 4.4.2 Posterior odds (1:n) of developing gastric cancer in females given PGI level 72 Table 4.4.3 Cumulative probability of developing gastric cancer in males less than certain PGI level 73 Table 4.4.4 Cumulative probability of developing gastric cancer in females less than certain PGI level 74 Table 4.5.1 Posterior odds of developing IM in HP seropositive 75 Table 4.5.2 Posterior odds of developing IM in HP seronegative 76 Table 4.5.3 Cumulative probability of developing IM in HP seropositive less than certain PGI level 77 Table 4.5.4 Cumulative probability of developing IM in HP seronegative less than certain PGI level 78 Table 4.6.1 Posterior odds of being gastric cancer given PGI and CEA for 30-34 y/o male 79 Table 4.6.2 Posterior odds of being gastric cancer given PGI and CEA for 35-39 y/o male 80 Table 4.6.3 Posterior odds of being gastric cancer given PGI and CEA for 40-44 y/o male 81 Table 4.6.4 Posterior odds of being gastric cancer given of PGI and CEA for 45-49 y/o male 82 Table 4.6.5 Posterior odds of being gastric cancer given of PGI and CEA for 50-54 y/o male 83 Table 4.6.6 Posterior odds of being gastric cancer given of PGI and CEA for 55-59 y/o male 84 Table 4.6.7 Posterior odds of being gastric cancer given of PGI and CEA for 60-64 y/o male 85 Table 4.6.8 Posterior odds of being gastric cancer given PGI and CEA for 65-69 y/o male 86 Table 4.6.9 Posterior odds of being gastric cancer given PGI and CEA for 70-74 y/o male 87 Table 4.6.10 Posterior odds of being gastric cancer given PGI and CEA for 75-79 y/o male 88 Table 4.6.11 Posterior odds of being gastric cancer given PGI and CEA for 80-84 y/o male 89 Table 4.6.12 Posterior odds of being gastric cancer given PGI and CEA for 85+ y/o male 90 Table 4.6.13 Posterior odds of being gastric cancer given PGI and CEA for 30-34 y/o female 91 Table 4.6.14 Posterior odds of being gastric cancer given PGI and CEA for 35-39 y/o female 92 Table 4.6.15 Posterior odds of being gastric cancer given PGI and CEA for 40-44 y/o female 93 Table 4.6.16 Posterior odds of being gastric cancer given PGI and CEA for 45-49 y/o female 94 Table 4.6.17 Posterior odds of being gastric cancer given of PGI and CEA for 50-54 y/o female 95 Table 4.6.18 Posterior odds of being gastric cancer given of PGI and CEA for 55-59 y/o female 96 Table 4.6.19 Posterior odds of being gastric cancer givenof PGI and CEA for 60-64 y/o female 97 Table 4.6.20 Posterior odds of being gastric cancer given PGI and CEA for 65-69 y/o female 98 Table 4.6.21 Posterior odds of being gastric cancer given PGI and CEA for 70-74 y/o female 99 Table 4.6.22 Posterior odds of being gastric cancer given PGI and CEA for 75-79 y/o female 100 Table 4.6.23 Posterior odds of being gastric cancer given PGI and CEA for 80-84 y/o female 101 Table 4.6.24 Posterior odds of being gastric cancer given PGI and CEA for 85+ y/o female 102 Table 4.7.1 Marginal posterior odds of developing cancer after adjustment for CEA in males 103 Table 4.7.2 Marginal posterior odds of developing cancer after adjustment for CEA in females 104 Table 4.8.1 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 30-34 years (HP seropositive) 105 Table 4.8.2 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 35-39 years (HP seropositive) 106 Table 4.8.3 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 40-44 years (HP seropositive) 107 Table 4.8.4 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 45-49 years (HP seropositive) 108 Table 4.8.5 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 50-54 years (HP seropositive) 109 Table 4.8.6 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 55-59 years (HP seropositive) 110 Table 4.8.7 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 60-64 years (HP seropositive) 111 Table 4.8.8 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 65-69 years (HP seropositive) 112 Table 4.8.9 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 70-74 years (HP seropositive) 113 Table 4.8.10 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 75-79 years (HP seropositive) 114 Table 4.8.11 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 80-84 years (HP seropositive) 115 Table 4.8.12 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 30-34 years (HP seronegative) 116 Table 4.8.13 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 35-39 years (HP seronegative) 117 Table 4.8.14 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 40-44 years (HP seronegative) 118 Table 4.8.15 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 45-49 years (HP seronegative) 119 Table 4.8.16 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 50-54 years (HP seronegative) 120 Table 4.8.17 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 55-59 years (HP seronegative) 121 Table 4.8.18 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 60-64 years (HP seronegative) 122 Table 4.8.19 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 65-69 years (HP seronegative) 123 Table 4.8.20 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 70-74 years (HP seronegative) 124 Table 4.8.21 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 75-79 years (HP seronegative) 125 Table 4.8.22 Posterior odds of developing IM with bivariate Bayesian analysis given PGI and PG ratio for age of 85+ years (HP seronegative) 126 Table 4.9.1 Marginal posterior odds of developing IM after adjustment for PG ratio (HP seropositive) 127 Table 4.9.2 Marginal posterior odds of developing IM after adjustment for PG ratio (HP seronegative) 128 Table 4.10.1 Determination of cut-off point for cancer screening given probability >1/1000 (Male) 129 Table 4.10.2 Determination of cut-off point for cancer screening given probability >1/1000 (Female) 129 Table 4.10.3 Determination of cut-off point for cancer screening given probability >1/2000 129 Table 4.10.4 Determination of cut-off point for cancer screening given probability >1/3000 130 Table 4.11.1 Determination of cut-off point for IM screening given probability >1/10 (HP seronegative) 130 Table 4.11.2 Determination of cut-off point for IM screening given probability >1/10 (HP seropositive) 130 Table 4.12.1 Determination of cut-off point for cancer screening given utility 131 Table 4.12.2 Determination of cut-off point for cancer IM screening given utility 131 Table 5.1 Logistic model prediction of cancer 131 Table 5.2 Comparison of the cancer predictive probability from two models 132699842 bytesapplication/pdfen-US蒙地卡羅馬可夫鏈貝氏分析生化指標胃癌biological measuresMonte Carlo Markov ChainBayesian modelgastric cancer[SDGs]SDG3thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/59187/1/ntu-93-R91846010-1.pdf