《流行病学试验》PPT课件.ppt

1,实验流行病学研究 Experimental Epidemiology,陈裕明 教授 医用统计与流行病学系 Tel 87330605,2,主要内容,回顾揭开古老的医术之谜 概述：基本特征、分类 设计 设计类型、 试验设计中的重要因素 实施 选择合适的实验现场 选择研究对象 样本量估计 随机化分组 设立对照 盲法（blinding or masking ) 的应用 资料分析 应注意的问题 优缺点,3,揭开古老的医术之谜,4,影响治疗后疾病转归的因素,5,对照研究及混杂问题,好转,不变或恶化,时间,Total population,Sample,Treatment group: 治疗+其他因素,Control group : 其他因素,好转,不变或恶化,6,其他因素的非特异作用,其他因素的非特异作用,临床转归,对照组,治疗组,观察到的效应,真实的治疗作用,7,其他因素的非特异作用,真实的治疗作用,其他因素的非特异作用,临床转归,对照组,治疗组,观察到的效应,8,其他因素的非特异作用,真实的治疗作用,临床转归,对照组,治疗组,其他因素的非特异作用,观察到的效应,9,概 述,10,第一节 概 述,一 、定义： 实验流行病学：是通过比较给予干预措施后的实验组人群与对照组人群的结局，从而判断干预措施效果的一种前瞻性研究方法。 二、基本原则 随机、 对照、盲法,11,Intervention 干预,Randomization随机,Control 对照,Blinding 盲法,二、基本原则,12,Classification of Experimental Studies,Healthy People,Patients,Experimental study,Study subject, Healthy or not?,Clinical trial,Field trial,Study unit,Individual trial,Community trial,社区试验,个体试验,临床试验,现场试验,If the trial meet basic requirements: Controlled, randomization,Quasi-experiment,类实验,NO,True-experiment,真实验,YES,Individual trial,13,第一节 概 述,三 主要类型 （main types),1 临床试验（clinical trial) 研究对象：病人 随机单位：个体 目的：评价药物/治疗手段的效果 举例：某新一代抗生素抗感染效果的评价,14,Phases of Clinical Trials,15,2. 现场试验 （field trial) 研究对象：正常人群或亚临床 干预单位：个体 目的：评价某种预防手段的效果 举例：乙肝疫苗效果的评价 3. 社区试验 （community trial) 研究对象：正常人群或亚临床 干预单位：社区/某一人群或亚人群 目的：评价某种预防手段的效果 举例：水加氟预防齿/食盐加碘预防地方性甲状腺肿,16,4 类试验 （quasi-experiment) or 半试验（semi-experiment) 为不严格的上述三种试验 缺少四个基本特征中的一个或几个特征（通常缺“3”和或“4”）。 无对照组：自身对照或历史对照 无随机对照组：试验对象进入对照组不是随机分配的,17,基本设计,18,流行病学试验设计 一、随机对照试验设计模式： A.Parallel Group Trials 平行设计 1. Simplest example - 2 groups 简单平行设计 2. Stratified design 分层平行设计 3. Matched pairs 配对设计 4. Factorial design 析因设计 B. Crossover Trials 交叉设计,19,Parallel Group Studies 平行设计,Eligible Patients /subjects,Drug A,Drug B,Informed consent,合格对象,知情同意,Outcome,Outcome,20,Parallel Group Studies Stratified Randomization 分层随机设计,Eligible Patients /subjects,Drug A,Drug B,Informed consent,合格对象,知情同意,Outcome,Outcome,Drug A,Drug B,Outcome,Outcome,Drug A,Drug B,Outcome,Outcome,第 一 层,第 二 层,第 三 层,21,“经皮冠状动脉腔内成形术后再狭窄的防治”设计中分层层次示意图,主层 性别 年龄 病情 次层数 （含2个次层） （含3个次层） （含3个次层） 总计,次 层,男,女,40-45,46-60,61-70,40-45,46-60,61-70,轻 中 重,轻 中 重,轻 中 重,轻 中 重,轻 中 重,轻 中 重,次层总数 2 3 3 = 18,3 3,3 3,22,3. 22 Factorial design 析因设计,Informed Consent,Yes, randomize,No,Eligible Patients/subjects,Drug A,Drug B,Drug None (无）,Drug A & B,23,22析因设计举例,A: high Protein 高蛋白 B: high fibre 高纤维,无A, 无B,无A, 有B,有A, 无B,有A,有B,Burke V, et al. Hypertension. 2001;38:821-826,Dietary Protein and Soluble Fiber Reduce Ambulatory Blood Pressure in Treated Hypertensives,24,Two-Period Crossover Trial 2期交叉设计,Drug B,Drug A,W A S H O U T,Phase 1,Period,Eligible Patients /subjects,Drug A,Drug B,Informed consent,合格对象,知情同意,Drug B,Drug A,Phase 2,洗脱期,25,Situations where crossover design is most applicable: 适于交叉设计的情况,1. Rapid response and response is transitory (反应快,持续短) Variability between patients is large compared to variability within patients 个体间差异大于个体内批间差异 3. Steady state physiological condition; disease or condition cannot be cured.病情稳定/不能治愈 4. No residual or carry-over effects of treatment expected 停药后与残余效应,26,Crossover Trial 交叉设计优缺点,Advantages: (comparison with parallel design) Within patient comparison more direct and precise (同一个体内比较) Gain in statistical power (增加统计效率) Fewer patients needed (较少样本量) Fewer ethical problems (较少医德问题),27,交叉设计优缺点,Disadvantages: Can only study short-term effects (只始于短期效应研究) Require chronic “stable” disease (需要慢性稳定的疾病) Order (period) effects a nuisance (阶段效应干扰) Carry-over effects and patient drop-outs ruin crossover design (携带效应与样本流失),28,实施步骤,29,实施步骤,Define a Clinical research Question Develop a Research Protocol Sample size planning Participant recruitment & Informed consent Measure baseline characteristics and describe sample Randomization / Binding / concealment /controlled Apply intervention Follow-up and adherence to protocol Measuring outcome,30,1. Define a Clinical research Question,The clinical question should incorporate four elements : Patient Intervention Comparison Outcomes,31,Example of A Research Question,Is diabetes preventable?,32,Example of a Better Research Question,Does treatment with metformin (二甲双胍) cause a decrease in the number of new cases of diabetes among people with impaired fasting glucose, compared to placebo?,Knowler et al., NEJM 2002;346:393-403.,33,Clear and Specific Statements,Which medicine is more effective in reducing mother-to-child transmission of HIVMedicine “A” or Medicine “B”?,34,2. Research Protocol,35,3. Sample size planning 样本量估计,Key questions defining the size Whats the primary outcome measure? (结果指标) What significance level is difference criterion? (显著性水准) With what degree of certainty (Power)? 把握度 How small a treatment difference needs detecting? 需要得到的最小差异大小 Event rates or the variance 发病率或方差.,36,样本量计算,N：为计算所得一个组的样本大小,对照组发生率,水平相应的标准正态差,试验组发生率,为1-水平相应的标准正态差,（p1+p2）/2,37,连续变量样本量,公式适用于N30时,估计的标准差,两组连续变量均值之差,水平相应的标准正态差,为1-水平相应的标准正态差,38,4. 选择实验现场考虑因素,人口相对稳定，流动性小，有足够数量 预期结局事件(如疾病)在该地区有较高而稳定的发生率 评价疫苗的免疫学效果，选择近期内未发生该疾病流行 的地区 较好的医疗卫生条件，诊断水平较好，卫生防疫保健机 构比较健全等 领导重视，群众愿意接受,39,5. 选择研究对象,对干预措施有效 研究对象的代表性好 预期结局事件发生率较高 容易随访的人群 干预措施对其有益或至少无害 依从性好、乐于接受并坚持试验,40,Study Participant Recruitment 对象招募,Identify eligible participants Explain study Provide informed consent Reassess eligibility Assign to one group,确定合格参与者 解释该研究 提供知情同意 再评估参与资格 安排接受干预,41,Participants 参与者的选取,Inclusion criteria (纳入标准) Define main characteristics of target population that are relevant to research question (定义与研究相关的目标人群的主要特征) Demographic characteristics (人口学特征) E.g., adults, aged 20-69 Clinical characteristics (临床特征) E.g., in good health Geographic characteristics (地区特征) E.g., living in Guangzhou Temporal characteristics (时间特征) E.g., inception period Jan 1, 2003 to Dec 31, 2003,42,Participants 参与者的选取,Other entry criteria (其它准入条件) Entry criteria should optimize (纳入标准必须适宜) Rate of primary outcome (主要结果的发生率) Expected effectiveness of active treatment (期望的效果) Generalizability of findings (结果外推) Ease of recruitment (容易招募) Likelihood of compliance with treatment, follow-up (依从性好，容易追踪),43,Reduction in the Incidence of Type 2 Diabetes with Lifestyle Intervention or Metformin,Inclusion criteria Eligibility criteria included an age of at least 25 years, a body-mass index (the weight in kilograms divided by the square of the height in meters) of 24 or higher (22 or higher in Asians), and a plasma glucose concentration of 95 to 125 mg per deciliter (5.3 to 6.9 mmol per liter) in the fasting state (lessthan/equal) 125 mg per deciliter in the American Indian clinics) and 140 to 199 mg per deciliter (7.8 to 11.0 mmol per liter) two hours after a 75-g oral glucose load. NEJM 2002 346(6): 393-403,44,Participants 参与者的选取,Exclusion criteria (排除标准) Subsets of people meeting inclusion criteria (potentially suitable for research question) except for characteristics that might interfere with quality of data etc. (排除对研究有干扰或不适于参加者) High likelihood of being lost to follow-up (容易流失) E.g., transients Inability to provide good data (不能提供好资料) E.g., language barrier or cognitive incapacitation High risk for side effects (对副作用敏感) E.g., pregnant, lactating Unethical to withhold study treatment (伦理上不宜) E.g., severe depression,45,Exclusion Criteria-example,Eligible persons were excluded if they were taking medicines known to alter glucose tolerance or if they had illnesses that could seriously reduce their life expectancy or their ability to participate in the trial.,46,Informed Consent,To make informed decisions, patients need to hear and understand specific information about the research.,47,2. Informed Consent： 知情同意保护参与者 A Part of Human Subject Protection,Objectives of Informed Consent 目的 To Ensure: Voluntariness 自愿 Comprehension 理解 Information 知情 To Demonstrate That: Person freely gave consent to participate 未压逼下同意 Consent given by a competent person 有决定权者同意 Person has been given all information 充分知情 Person knows this is research not treatment 知道是研究不是治疗,48,Informed Consent,Purpose Medicine to be studied Procedures and schedule,Risks Potential benefits Alternatives to participation Confidentiality,49,6. Baseline data collection,Need enough information to track subjects 联络信息 Contact persons; Alberta Health Registry (to track deaths, moves from province, etc.) Description of participants (社会人口学特征) Aid in assessing generalizability (e.g., gender, age, disease severity, etc.) Risk factors for outcome or to define subgroups 危险因素 E.g., smoking status, smoking status of spouse Measure of “outcome” variable 结局变量指标 E.g., if pain is “outcome”, need baseline pain,50,6. Baseline data collection,基线资料的用途 描述研究对象特征，用于结果外推性的判断 判断处理组间的可比性 在组间可比性差时，用于控制混杂因素 用于亚组分析或交互作用分析,51,7. Randomization,Note Usually, sampling is not random Group Assignment is random,52,Randomization,Investigational Group,Control Group,53,简单随机化法(simple randomization) 区组随机法 （blocked randomization) 分层随机法(stratified randomization) 整群随机法(cluster randomization),随机化分组方法,54,The aims of randomization 目的 To get a unbiased comparison of treatments 得到无偏倚的比较 To avoid any systematic difference in treatment groups 避免组间的系统误差 Two key principles: 关键原则 Each patient has an equal chance of each treatment option 分组机会同等 Neither patient nor investigator can predict in advance which treatment theyll get 受试者与研究者均不能预计将受到哪种干预,55,Advantages of Randomization 随机分组的好处,Bradford Hill: 1. Eliminates bias from treatment assignment (排除人为分组所致的偏倚) 2. Balances known and unknown differences between groups on average (平衡组间已知及未知的差异) 3. More credible study 更加可信 RA Fisher: Assures validity of statistical tests 确保统计的有效性,56,Simple Randomization 简单随机,The number and order of patients receiving treatments A and B is determined by chance. Example: Equal allocation Toss a coin(抛硬币): A = head, B = tails (正反) Random number table: (随机数字，奇数、偶数) A = odd, B = even (Pocock, page 74) Uniform random number generator: (随机数字大于或小于0.5) A if 0.5,57,Timing of Randomization Usual Sequence of Events 什么时候随机分组,1. Verify eligibility, informed consent, and completeness of baseline data. 2. Complete patient accession log. 3. Obtain assignment. 4. Record assignment on log and data forms. 5. Initiate treatment as soon as possible after randomization.,58,问 题,随机分组能消除哪些偏倚?,59,8. Blinding,Randomization Controls for confounding bias at baseline Does not control for confounding during follow-up E.g., differential attention to subjects in treatment arm (a type of co-intervention) Does not control for information bias Blinding Controls for Tampering with random assignment Information bias (e.g., observer bias, expectation bias),60,盲法（blinding or masking ),目的：避免主观因素所导致的偏倚。 如：来自实验对象霍桑效应和安慰剂效应。研究者本人多数希望得到阳性结果（实验组和对照具有差异性）而在设计资料收集或分析阶段采取的有利于出现阳性结果的主观行为。 分类：根据盲法的程度分为单盲、双盲和三盲,61,Blinding 盲法,Subject “blinded” to group assignment 受试者“盲” Care-giver (administering treatment) blinded to subjects group assignment 研究实施者“盲” Controls co-interventions, better care of treatment group, etc. Investigator (assessing outcome) blinded to subjects group assignment 研究者“盲” Controls biased assessment of outcome 控制结果测量偏倚 Especially important when outcome has some measure of subjectivity E.g., death difficult to misclassify even by the most biased assessor,62,盲法的应用,单盲（single blind） 研究对象不知分组情况 双盲（double blind） 研究对象、研究者不知分组情况 三盲（triple blind） 研究对、研究者、负责资料收集者不知分组情况,63,64,9. Allocation concealment 分组隐匿,将分组的每一个序号单独放入一个不透明的信封里，决定对象分组时才打开 一旦打开信封，如果研究对象不同意该分组即算退出或流失，该序号不能再分派给其它人。,65,10. 设立对照,原 因,不能预知的结局 霍桑效应（Hawthorne effect） 安慰剂效应（placebo effect） 潜在的未知因素的影响 向均数回归,方 式,安慰剂对照 自身对照 交叉对照,66,设立对照,1. 不能预知的结局(unpredicable outcome) 混杂因素的影响。如：性别，年龄，种族等；疾病的轻重；免疫状态；精神心理状态，遗传因素，等 2. 向均数回归(regression to the mean) 即一些极端的特征值，即使无任何干预，也可有趋向于均数的改变现象，向均数靠拢。（极大值缩小，极小值增大）。如血压特别高的5%的人，即使无治疗，过一段时间再次检测，也可能会降低一点。,67,设立对照,3. 霍桑效应(Hawthorne effect) 人们因为成了研究中特别感兴趣和受注意的目标而改变了其行为的一种倾向，与其接受的干预措施的特异性作用无关。因迷信或厌恶或不信任某个医疗单位或医生而产生的心理/生理效应，对干预产生正面或负面效应。 4. 安慰剂效应(placebo effect) 某些试验对象，由于依赖医药而表现的一种正向心理效应。因此，当以主观感觉的改善情况作为评价指标时，其效应可能包括安慰剂效应在内。 5. 潜在未知因素（potential unknown factors)影响,68,设立对照的类型,安慰剂对照(placebo control) 安慰剂是不具有真正预防、治疗或治病效应的制剂。通常用乳糖，淀粉，生理盐水等成分制成。不加任何有效成分，但外形，颜色，大小，味道等与试验药物或制剂极为接近。通常在评价药物预防效果，或在研究的疾病无有效的防治药物，或使用安慰剂对病情无重大影响时才使用。 通常用于盲法试验。 阳性对照：已知效果防治方法对照 出于医德方面考虑，对容易恶化（转坏）的疾病除非现行没有任何有效的病因甚至对症治疗措施，对对照组病人是不能给予安慰剂。应将试验疗法与现行标准或最好疗法作比较。可用于盲法试验。,69,设立对照的方式,1. 平行对照 2. 自身前后对照(self control): 自身试验前后对比。受试者在不同的试验阶段，分别接受不同的干预措施，最后比较各种干预措施的效果。 3. 随机同期对照+自身对照 交叉设计: 定义：在同一阶段试验中设立随机对照，在下一节段中 ，试验组与对照组交换，形成前后自身对照。 优缺点：见交叉设计,70,11. Follow-up and Adherence,Ideally All subjects follow adhere to treatment regimens; do not seek additional treatment; do not drop out, die, move or have to be withdrawn from study; attend follow-up sessions and provide outcome data 所有受试者跟随干预方案、不寻求额外治疗、不丢失、不退出、死亡、移民；参加追踪观察，提供结果资料。,71,Unplanned Crossovers 非计划交叉,When subjects choose the alternative treatment Ss in experimental group start using “control” treatment or vice versa 试验组用对照干预或相反。 Usually have selective cross-over (more Ss from one group cross over) （通常为选择性交叉） A large proportion of crossovers may invalidate study （大比例交叉会影响结果）,72,Other Pitfalls 其它缺陷,Drop-outs differential drop-out rate 退出 Lost-to-follow up 失访 Non-compliance differential non-compliance rate 不依从 E.g., if active drug has unpleasant side effects, subjects may skip doses Co-interventions 接受多种治疗 E.g., placebo group might seek out other treatments,73,Measuring Outcome 结果测量,选择评价指标的原则： 1. 定性+定量 2. 尽量用客观指标 3. 指标对干预的敏感性 4. 指标的可信性，可靠性，及可行性 5. 测量易为受试者所接受。,74,Measuring Outcome 结果测量,评价指标 健康效果 成本效益 健康效用：生存质量,75,资料的分析,统计描述 统计推断 临床和公共卫生意义分析,76,资料分析策略,1. 只分析合格完成者 (valid completers)或 称为遵循研究方案（Per-protocol）分析: 分析时剔除所有不合格者及退出者。 优点：提高检验效率 缺点：因剔除导致随机化分组效率下降，导致偏倚，增加假阳性率。 2. 按意向性(intention-to-treat, ITT)分析 优点：减少偏倚 缺点：降低实验检验效率。增加假阴性率，但不增加假阳性率 3. 接受干预措施分析 （per-treatment analysis) 按实际接受的干预进行分组分析,77,Intention to Treat意向治疗分析,Current practice in analyzing RCTs （现行方法） Analyze results as if all subjects stayed in study,complied, etc. Retains original group assignment and ignores deviations in actual treatment obtained Retains randomization (and therefore control of confounding) When results are negative, does not tell us whether treatment truly is ineffective or just not used When participants drop out or cross-over, intention to treat underestimates the effects of treatment,78,Explanatory Analysis,Is the experimental treatment better than the alternative? Analyze differences between those actually receiving experimental and alternative treatment But unless most patients receive the treatment randomized to, you no longer have random group assignment selective groups, confounding Can miss complications because those who discontinue therapy because of complications are not included in analysis,79,评价实验效果的主要指标,80,结局事件发生率指标,81,效应值指标,82,分析举例,治疗组,对照组组,随机分组人数,201,200,换组或失访人数,31,150,15,160,25,20,?,50,3,16,?,1,结局事件人数,ITT分析,P1= (50+1)/201 = 25.4%,P0= (16+3)/200 = 9.5%,接受干预分析,P1= (50+3)/(201-31-20+15) = 32.1%,P1= (16+1)/(200-25-15+20) = 9.4%,83,分析举例,治疗组,对照组组,随机分组人数,201,200,换组或失访人数,31,150,15,160,25,20,失访,50,3,16,失访,1,结局事件人数,ITT分析 发生率 效果指数 ARR NNT,P1= (50+1)/201 = 25.4%,P0= (16+3)/200 = 9.5%,= p1/p0 = 25.4/9.5 = 2.67,= p1- p0 = 25.4 - 9.5 = 24.45(%),= 1/ARR = 1/0.2445 = 4.1人,84,Example: Surgical vs Medical therapy in bilateral carotid stenosis 外科与内科治疗双侧颈动脉狭窄,Outcome in compliers (合格完成者的结果) Therapy Outcome Stroke or death Surgical 43/79 (54%) Medical 53/72 (74%) P=0.02 (a further 16 patients (15/16 due for surgery) had a stroke or died after randomization but before therapy started),85,Example: Surgical vs Medical therapy,2) Analysis by ITT Therapy Stro