HS 2801 Notes - Week 8 PDF

Week 8 Making groups as similar as possible to account for confounding - Like randomization in experimental studies (wait!) - Only difference: exposure (intervention) status; making exposure groups similar in other factors after randomization - Matching cases to contro...

Week 8 Making groups as similar as possible to account for confounding - Like randomization in experimental studies (wait!) - Only difference: exposure (intervention) status; making exposure groups similar in other factors after randomization - Matching cases to controls - Only difference: outcome status; making outcome groups similar in other factors Two broad categories of epidemiologic studies: - Observational Studies - researchers collect, record, and analyze the data on participants as they naturally divide themselves by potentially significant exposure and outcome variables. - Experimental Studies - involve some type of intervention and control of the exposure variable by the researchers to see how this control influences the outcome variable. Experimental Studies - key feature of is that the conditions of the study, particularly the exposure status, are directly controlled by the intervention of researchers. - Investigators assign study participants to two groupsL one group receives exposure of interest and the other group does not. THen the investigator follows the groups over time for incidence of disease - Defining feature is that the investigator assigns exposure to study participants - Sometimes called - Intervention studies - Clinical trials - Randomized controlled trials History - James lind - Scottish naval surgeon - Conducted experiment with 12 scurvy victims - Each pair received different treatment oranges and lemon recover - 1st recorded randomized clinical trial - Edward Jenner - Sarah Nelmes was infected with cowpox - James phipps is inoculated with cowpox pus from sarah nelmes - Phipps falls ill with mid disease of cow pox - Scabs are collected from smallpox patient - Phipps is inoculated with scabs of smallpox - Phipps is unaffected, protection is complete - This is experimental but neither randomized nor control group Experimental Studies Emulate Controlled Laboratory Experiments - In lab experiments, investigator regulates all important aspects of experimental conditions: genetically similar animals, same physical environment, same diet, same schedule, same route of administration of test chemical. Only difference is the dissimilar test chemical. - Experimental studies among free-living humans never achieve same degree of control, but many aspects of human experimental research emulate principles of lab and/or animal research. - Ethical issues are rampant - History - Monster study - Children were randomised to positive speech therapy and belittling to induce stuttering The process 1. Research questions for experimental studies - Experimental studies: investigator intervenes - Research question involves prevention or treatment - Feasible and ethical - Small population level effect expected - Observational studies: investigator watches - Research question involved prevention, treatment, or causal factor - Experimental not feasible or ethical - Moderate to large population level effect expected 2. Types of experimental studies - Prevention trial - Agents given to healthy or high risk individuals to prevent disease occurrence - Does removing lead contaminated soil prevent lead poisoning in high risk urban children - Does the drug tamoxifen lower the incidence of breast cancer in high risk women - Therapeutic (clinical) trial - Agent given to diseased individual to treat or cure disease - Does drug lower the risk of recurrence and improve survival among women diagnosed with breast cancer - Individual trial  - Treatment allocated to individuals - Study effective of treatment with two drugs vs three drugs among adults - Community (cluster) trial - Treatment allocated to entire community - Study impact of drinking water fluoridation on frequency of dental caries in one community vs no fluoridation in similar community 3. Study population - Choice of population depends on purpose of trial - Specific inclusions/exclusion criteria (eligibility criteria) based on scientific, safety, and practical considerations - People in population of interest, at risk for outcome, among whom the intervention could be effective - People with high likelihood of compliance with treatment, likely to be followed for total study period - May need to exclude certain people, such as those with conditions for which drug under study is contraindicated - generalizability : the extent to which the results from a study can be generalized to people who did not participate in the study - Recruitment of study population is important because it directly impacts to whom the results generalize once the study is over - If there are many inclusion/exclusion criteria, study loses generalizability - Reference population: group to whom results are applicable/generalizable - All human - All men and women - Urban children at risk of lead poisoning 4. Consent, enrollment, exposure assignment - All participants must give informed consent before enrolling in a trial - Informed consent includes clear explanation of research goals and methods, risks and benefits of participating, assurance of confidentiality, right to withdraw - Once consented and enrolled, treatment assignment occur - Random assignment vs non random assignment - Randomization: each study participant has the same probability of receiving treatment achieve by - Toss a coin, a or b - Computer generated randomization - Other assignment schemes have potential problems - Subject self selects - Day of week, order of visit - Randomization prevents biased group assignment - Randomization means that the assignment of participants into experimental and control groups is random - This creates similarity between groups for potential confounding variables  - Sometimes randomization doesn’t work, because not all groups are represented properly - Chance difference between groups due to sampling error are more likely to occur when the study sample is small, stratified randomization should be considered - Randomization balances confounder (a factor that distorts the true association between an exposure and outcome) - unequally distributed across groups - Randomization minimizes inaccurate results caused by confounding - Ultimate goal = treatment is the only different between study groups - Randomization creates 2 groups that are the same except one group has treatment and other does not - Same meaning: group characteristics are balanced - With randomization, this occurs for both known and unknown confounders - Only works when study if large enough and treatment assignment is not influenced by investigator 5. Minimizing bias - Blinding or masking = method of ensuring that participants and/or study investigators have no knowledge of whether a study participant has been assigned to the treatment or comparison group - Single blind: study participant does not know whether they are receiving treatment or no treatment - Double-blind: neither the (1) study participant nor the (2) study investigator administering the treatment knows who is receiving treatment or no treatment - Triple-blind: neither the (1) study participant nor the (2) study investigator administering the treatment nor the (3) study investigator monitoring the effects of the treatment knows who is receiving treatment or no treatment - Blinding is not always possible: certain interventions such as surgery and psychotherapy, exercise regimens, diet, with serious side effects - Placebo - Method of blinding - Makes exposed and unexposed groups’ experiences as comparable as possible (this goal harkens back to lab experiments) - Placebos cannot always be used. - Placebos are not always ethical; often use current standard of care instead of placebo  - “Placebo Effect” = Participants assigned to placebo groups improve because they are told that they will. This stems from the power of suggestion. - Compliance - Following the study protocol exactly as required throughout the course of the trial - Goal #1: Have groups be as alike as possible on other important characteristics (randomization) - Goal #2: Have groups be as different as possible on exposure (intervention) - No contamination - Deviations from protocol can occur for many reasons: side effects, illness, level of interest, and length of follow-up - Worst: : contamination. “Inadvertent application of the experimental procedure to the control group” or vise versa - Importance: - Noncompliance tends to make treatment and comparison groups more alike - This reduces the ability to detect a difference (in the outcome due to exposure) between groups. - We are comparing the difference between treatment and comparison groups. If they become more similar, we will not observe a difference, even if one exists. - Non compliance results to bias towards the null 6. Analysis - Descriptive analysis - Did randomization work - Is the distribution of potential confoudning factors the same in the two groups - Measures of event/association - Set up 2x2 table - Incidence rates in treatment and comparison groups - Relative comparing treatment and comparison group Cross over trial - Randomize so one group receives intervention then control and the other receives control then intervention - Each person acts as their own control Major strengths of RCT 1) Demonstrate causal relationship with high levels of confidence a) Higher validity 2) Researchers can control exposure levels Major weaknesses of RCT 1) Limited applicability due to ethical reasons and at times due to artificial setting of experiment 2) Potential for bias due to difference in compliance, withdrawal, and loss to follow-up in experimental and control groups 3) Usually costly and time consuming 4) May have limited external validity (low generalizability) 5) Risk of contamination is high even in blinded studies Cluster randomized trial 1. Same methodology 2. Lower risk of contamination 3. Some interventions only make sense at the group level 4. More convenient