基于Logistic回归分析的胃癌前病变风险预测模型的建立及评价

发布时间：2018-06-06 02:16

本文选题：癌前病变 + 风险预测　；参考：《中国中医科学院》2017年硕士论文

【摘要】：研究背景:胃癌是世界范围内最为常见的恶性肿瘤之一。胃癌在全球范围内的发病率列居第五,死亡率位居第三。据世界卫生组织/国际癌症研究中心(IARC)公布的统计数据显示,2012年全球胃癌新发病例98.9万,约50%发生在东亚地区,尤以中国最多,占全球胃癌发病的46.8%;2012年全球胃癌的死亡病例为73.7万,中国死亡病例为35.2万,占全球胃癌死亡的47.8%。据统计,2015年我国胃癌发病率为22.7/10万,仅次于肺癌,位居第二;胃癌死亡率为17.9/10万,仅次于肺癌、肝癌,位居第三。胃癌的发生经历一系列演变过程,由Correa等提出了肠型胃腺癌的演化模式,即:正常胃黏膜—慢性浅表性胃炎—慢性萎缩性胃炎—肠上皮化生—异型增生—肠型胃癌。以胃黏膜萎缩为病变背景的肠上皮化生(Intestinal Meteplasia,IM)及异型增生(Dysplasia,DYS)是目前公认的胃癌前病变(precancerous lesions of gastric cancer,PLGC),也是当前胃癌防治领域的热点和难点。重度异型增生有明显癌变倾向,需积极进行内镜下切除治疗,中-重度肠化生及轻-中度异型增生者国内外一致建议定期进行内镜监测。电子胃镜下病理组织活检属于介入性操作,其创伤大、费用高,在我国现有医疗条件下作为普查手段进行大规模开展尚有一定困难。血清“ABC”法联合G-17检测助于提高萎缩、胃癌检出率,并且具有无创、简便、快速、低廉等优点,但这些指标是否对PLGC的发生、转归有预测意义尚不得而知。目前,国内外对PLGC的筛检工具开发研究较少,已有的模型建立多局限在环境因素方面,推广性欠佳,并且缺少相关中医证候学内容,在临床实际运用中存在一定局限性。研究目的:1.筛选PLGC相关的因素及中医证候要素,明确PLGC的重要影响因素。2.建立基于Logistic回归分析的PLGC风险预测模型,为PLGC筛检提供数据模型依据。3.评估PLGC风险预测模型准确性,为PLGC高危人群筛检提供科学依据。研究方法:1.根据慢性萎缩性胃炎、异型增生的诊断标准,设定纳入及排除标准。中医证型诊断的辩证依据分为两个方面:一是共识,即国家、协会等正式出版发行的教材、指南及专家意见等;二是两名从事慢性萎缩性胃炎中医诊治工作20余年以上的高年资中医师的经验性证型诊断。于中国中医科学院西苑医院脾胃科门诊、住院及胃镜室收集慢性萎缩性胃炎患者,并进行筛选。2.通过检测血清学相关指标及问卷调查的方法,采集PLGC临床资料:包括临床危险因素(一般信息、生活行为、饮食习惯、情志因素、家族史、胃镜及病理表现);血清学指标:PGI、PGII、HpIgG抗体、G-17及中医证候及其要素等;3.比较各因素组间差异,将影响因素纳入Logistic单因素及多因素回归分析,选择Forward前进法,α入=0.05,α出=0.1筛选出PLGC相关危险因素及中医证候要素;4.以极大似然值对自变量进行参数估计,建立基于Logistic回归的PLGC风险预测模型;5.运用Hosmer-Lemeshow拟合优度、PesudoR-square、AUC指标对PLGC风险预测模型拟合度及准确性进行评价,运用ROC曲线分析该模型富集高风险人群的能力。研究结果:1.PLGC相关因素①生活行为:吸烟在PLGC组与非PLGC组暴露分别为:31%vs24.8%(P0.05);饮酒在PLGC组与非PLGC组暴露分别为:40.5%vs27.7%(P0.05)。②饮食习惯:喜食蔬菜在PLGC组与非PLGC组暴露分别为:14.0%vs26.3%(P0.05),OR值为0.236(95%CI=0.058-0.952);豆制品在PLGC组与非PLGC组暴露分别为:34.9%vs17.5%,(P0.05);饮浓茶在PLGC组与非PLGC组暴露分别为:16.3%vs4.4%(P0.05),OR值为6.288(95%CI=1.465-26.989);进食腌制食品在PLGC组与非PLGC组暴露分别为:37.2%vs16.1%(P0.05),OR值为3.198(95%CI=1.289-7.930)。③Hp感染史:Hp感染史在PLGC组与非PLGC组暴露分别为:7.0%vs16.8%(P0.05);④社会心理因素:焦虑在PLGC组与非PLGC组暴露分别为:41.9%vs19.7%(P0.05),OR值为2.404(95%CI=1.058-5.511);⑤遗传因素:胃癌家族史在PLGC组与非PLGC组暴露分别为:14.0%vs10.2%(P0.05);消化道肿瘤家族史在PLGC组与非PLGC组暴露分别为:16.3%vs11.7%(P0.05)。⑥合并疾病:胃食管反流病在PLGC组与非PLGC组暴露分别为:39.8%vs23.3%(P0.05);十二指肠溃疡在PLGC组与非PLGC组暴露分别为:79.3%vs7.3%(P0.05);胆囊切除史在PLGC组与非PLGC组暴露分别为:97.0%vs1.5%(P0.05),OR值为7.351(95%CI=1.824-55.384)。⑦血清学检查:PLGC者血清PGI、PGII、血清G-17水平在PLGC组水平分别为:55.01 ±35.85ug/L,4.50±2.82 ug/L,4.21±7.69ng/L,在非PLGC组水平分别为:62.40士48.59 ug/L,5.40±5.33 ug/L,4.13±9.77 ng/L(P0.05);HpIgG抗体在PLGC组与非PLGC组暴露分别为:20.9%vs16.1%(P0.05)⑧中医证候:症状方面:症状总积分在PLGC组与非PLGC组分别为:14.19±11.21vs17.97±13.56(P0.05);主要症状积分在PLGC组与非PLGC组分别为:10.7±6.11vs11.31±6.21(P0.05);证候要素方面:血瘀在PLGC组与非PLGC组暴露分别为:44.2%vs28.5%(P0.05),OR值为2.420(95CI%=0.998-5.851);湿热在PLGC组与非PLGC组暴露分别为:41.9%vs27%(P0.05)。2.基于Logistic回归分析的PLGC危险因素筛选通过运用Logistic单因素及多因素回归分析,最终筛选出腌制、焦虑、胆囊切除史、血瘀、浓茶5个重要危险因素。其中,胆囊切除史与PLGC发生风险关系最密切,OR值为7.351(95%CI=1.824-55.384);其次是浓茶,OR值为5.351(95%CI=1.824-55.384);腌制、血瘀,OR值分别是3.198(95%CI=1.289-7.930),2.420(95%CI=0.998-5.851);焦虑再次之,OR值为2.404(95%CI=1.058-5.511)。3.基于Logistic回归分析的PLGC风险预测模型建立ln(p/1-p)=-2.507+2.153X1+1.995X2+1.162X3+0.884X4+0.877X51-P(X1=胆囊切除、X2=浓茶、X3=腌制、X4=血瘀、X5=焦虑)4.PLGC风险预测模型评价模型拟合优度:经 Hosmer-Lemeshow(H-L)检验,χ2值为 3.997,P=0.5500.05,不拒绝关于模型很好拟合数据的假设,即该模型拟合度较好。模型测准确性:关于 Pesudo R-square,Cox and Snell 值为 0.209,Nagelkerke值为 0.314;AUC=0.812(95%CI=0.738-0.885),P0.01,表明该模型预测准确性尚可。5.PLGC风险预测模型富集高风险能力分析根据模型ROC曲线,灵敏度设定为100%,对应最低1-特异度为66.7%,此条件下,高风险人群最低占比为87%,对应预测概率P值为0.00673502。对该高风险人群进行内镜筛检,4例高风险患者中可发现1例真正PLGC者需进行内镜监测,发现率之比为1.15。研究结论:1、腌制、胆囊切除、焦虑、浓茶为PLGC者的相关危险因素。2、血瘀证在PLGC演变过程具有重要意义。3、PLGC风险预测模型为:ln(p/1-p)=-2.507+2.153X1+1.995X2+1.162X3+0.884X4+0.877X5(X1=胆囊切除、X2=浓茶、X3=腌制、X4=血瘀、X5=焦虑)4、PLGC风险预测模型拟合度及预测准确性较好,模型进一步验证及校准尚需未来长时间随访。
[Abstract]:Background: gastric cancer is one of the most common malignant tumors in the world. The incidence of gastric cancer is fifth in the world and third in the world. According to the statistics published by the WHO / International Center for cancer research (IARC), 989 thousand of the new cases of global gastric cancer in 2012, about 50% are in East Asia, especially in the Middle East. The country is the largest, accounting for 46.8% of global gastric cancer. In 2012, the death cases of global gastric cancer were 737 thousand, the death cases in China were 352 thousand, accounting for 47.8%. according to the global gastric cancer death. In 2015, the incidence of gastric cancer in China was 22.7/10 million, second only to lung cancer, ranked second, the death rate of gastric cancer was 17.9/10 million, second only to lung cancer and liver cancer, the third. Gastric cancer. A series of evolution processes were developed, and the evolution patterns of intestinal gastric adenocarcinoma were proposed by Correa, such as normal gastric mucosa - chronic superficial gastritis - chronic atrophic gastritis - intestinal metaplasia - dysplasia - intestinal type of gastric cancer. Intestinal metaplasia (Intestinal Meteplasia, IM) and dysplasia (Dysp) with gastric mucosal atrophy as the background (Dysp Lasia, DYS) is currently recognized as the precancerous lesions of gastric cancer, PLGC. It is also a hot and difficult point in the field of prevention and control of gastric cancer. Severe dysplasia has a tendency of obvious canceration and needs to be actively treated by endoscopic resection. The moderate to severe intestinal metaplasia and mild to moderate dysplasia are consistently recommended at home and abroad. Pathological tissue biopsy under electronic gastroscopy belongs to interventional operation, which has great trauma and high cost. It is difficult to carry out large-scale development as a census method under the existing medical conditions in our country. The serum "ABC" method combined with G-17 detection helps to improve the atrophy, the detection rate of gastric cancer, and has the advantages of noninvasive, simple, rapid and low. It is not known whether these indicators have a predictive significance for the occurrence and outcome of PLGC. At present, there are few studies on the development of PLGC screening tools at home and abroad. The existing models are mostly limited to environmental factors, poor popularization, and lack of relevant TCM syndromes, and there are some limitations in the practical application of the hospital. The purpose of this study is 1. Screening PLGC related factors and TCM syndrome factors, and defining the important influence factors of PLGC,.2. set up the PLGC risk prediction model based on Logistic regression analysis, providing data model for PLGC screening based on.3. evaluation of the accuracy of PLGC risk prediction model, providing a scientific basis for screening PLGC high-risk population. Research methods: 1. according to chronic atrophy The diagnostic criteria of gastritis and dysplasia are set up and excluded. The dialectical basis of TCM syndrome diagnosis is divided into two aspects: one is the common understanding, that is, the textbook, guide and expert opinion of the state, the association and so on; two is the two senior Chinese medicine teacher who has been engaged in the diagnosis and treatment of chronic atrophic gastritis for more than 20 years. The patients with chronic atrophic gastritis were collected in the outpatient of the spleen and stomach department of Xiyuan Hospital of Chinese Academy of Chinese medicine, China Academy of science of traditional Chinese medicine (Xiyuan Hospital), and selected.2. to collect PLGC clinical data by screening the related indexes of serology and questionnaire survey, including clinical risk factors (general information, life behavior, dietary habits and emotional factors, Family history, gastroscopy and pathological manifestations); serological indexes: PGI, PGII, HpIgG antibody, G-17 and TCM syndrome and its factors, etc.; 3. compare the differences among the factors in various factors, take the influencing factors into Logistic single factor and multiple factor regression analysis, select Forward forward method, alpha into =0.05, and alpha out =0.1 to screen out PLGC related risk factors and TCM syndrome factors; 4. The maximum likelihood value is used to estimate the parameters of the independent variables, and the PLGC risk prediction model based on Logistic regression is established. 5. the fitting degree and accuracy of the PLGC risk prediction model are evaluated by using Hosmer-Lemeshow goodness of fit, PesudoR-square, AUC index, and the ability of the model type to enrich the high risk population by using the ROC curve. The result of the study: 1.PLGC Related factors: life behavior: smoking in group PLGC and non PLGC group were 31%vs24.8% (P0.05); drinking in group PLGC and non PLGC group were respectively 40.5%vs27.7% (P0.05). The exposure of non PLGC group was 34.9%vs17.5%, (P0.05), and the exposure of drinking thick tea in group PLGC and non PLGC group was 16.3%vs4.4% (P0.05), OR value was 6.288 (95%CI=1.465-26.989), and the food pickled food in PLGC group and non PLGC group were respectively: 37.2%vs16.1% (37.2%vs16.1%) and 3.198. The exposure of non PLGC group was 7.0%vs16.8% (P0.05); (4) social psychological factors: anxiety in group PLGC and non PLGC group were respectively: 41.9%vs19.7% (P0.05) and OR value 2.404 (95%CI=1.058-5.511); 5. The exposure of the group was 16.3%vs11.7% (P0.05). 6. The exposure to gastroesophageal reflux disease in group PLGC and non PLGC group was 39.8%vs23.3% (P0.05), and the exposure of duodenal ulcer in PLGC group and non PLGC group was 79.3%vs7.3% (P0.05), and the history of cholecystectomy in PLGC group and non PLGC group was 7.351. I=1.824-55.384). Serological examination: the level of serum PGI, PGII and serum G-17 in the group of PLGC were 55.01 + 35.85ug/L, 4.50 + 2.82 ug/L, 4.21 + 7.69ng/L respectively, in the non PLGC group, 62.40 and 48.59 ug/L, 5.40 + 5.33 ug/L, 4.13 + 9.77, respectively. 0.05) symptom: symptom: the total symptom score was 14.19 + 11.21vs17.97 + 13.56 (P0.05) in group PLGC and non PLGC group, and the main symptom integral was 10.7 + 6.11vs11.31 + 6.21 (P0.05) in PLGC group and non PLGC group, and syndrome factor: the blood stasis in PLGC group and non PLGC group were respectively 44.2%vs28.5% (P0.05) and 2.420 I%=0.998-5.851); heat and heat in group PLGC and non PLGC group exposure, respectively: 41.9%vs27% (P0.05).2. based on Logistic regression analysis of PLGC risk factors screening through the use of Logistic single factor and multiple factor regression analysis, and finally screened out the salting, anxiety, cholecystectomy history, blood stasis, thick tea 5 important risk factors. Among them, the history of cholecystectomy and PLGC hair. The most closely related risk relationship, OR value of 7.351 (95%CI=1.824-55.384), followed by thick tea, OR value of 5.351 (95%CI=1.824-55.384), pickled and blood stasis, OR value was 3.198 (95%CI=1.289-7.930), 2.420 (95%CI=0.998-5.851), OR value was 2.404 (95%CI=1.058-5.511).3. based on Logistic regression analysis of the PLGC risk prediction model. /1-p) =-2.507+2.153X1+1.995X2+1.162X3+0.884X4+0.877X51-P (X1= cholecystectomy, X2= thick tea, X3= pickled, X4= blood stasis, X5= anxiety) 4.PLGC risk prediction model evaluation model goodness of fit: through Hosmer-Lemeshow (H-L) test, the chi 2 value is 3.997, P=0.5500.05, do not reject the model is very good fitting data hypothesis, that is, the model fitting degree is better. Accuracy of model measurement: on Pesudo R-square, Cox and Snell value is 0.209, Nagelkerke value is 0.314; AUC=0.812 (95%CI=0.738-0.885), P0.01, indicating that the prediction accuracy of the model still can enrich the high risk capability analysis of the.5.PLGC risk prediction model based on the model ROC curve, the sensitivity is set to 100%, corresponding to the minimum 1- specificity of 66.7%, this The lowest percentage of the high risk population was 87%, and the corresponding predictive probability P was 0.00673502. for the high risk population. In 4 patients with high risk, 1 cases of real PLGC were found to be endoscopically monitored. The ratio of the rate of discovery was 1.15.: 1, pickled, cholecystectomy, anxiety, and thick tea as PLGC related risk factors.2, blood stasis. The PLGC evolution process has important significance.3. The PLGC risk prediction model is: ln (p/1-p) =-2.507+2.153X1+1.995X2+1.162X3+0.884X4+0.877X5 (X1= cholecystectomy, X2= thick tea, X3= pickling, X4= blood stasis, X5= anxiety), the PLGC risk prediction model is better and the prediction accuracy is better, the further verification and calibration of the model still needs long follow-up in the future.
【学位授予单位】：中国中医科学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R273

【参考文献】