ENH522 Lecture 2: Randomized Control Trials (RCTs) PDF

Lecture 2: Introduction to Randomized Control Trials (RCTs) Design, Randomization, and Control Conditions ENH522 What is a variable? A variable is a characteristic measured on individuals drawn from a population under study. Variables differ from person to person, or measurement to mea...

Lecture 2: Introduction to Randomized Control Trials (RCTs) Design, Randomization, and Control Conditions ENH522 What is a variable? A variable is a characteristic measured on individuals drawn from a population under study. Variables differ from person to person, or measurement to measurement. Ex: height, weight, bp- things collected and are studied Data are measurements or values of one or more variables made on a collection of individuals Explanatory and response variables Research may try to to predict or Ex: studying certain drug on explain a response variable blood pressure. Explanatory: (often denoted y) from an drug b/c drug is affecting blood explanatory variable (often denoted x) pressure Changes in explanatory variable Other terminology for response lead to changes in response variable: when try to predict and explain dependent variable Outcome Other terminology for explanatory variable: what is affecting response variable independent variable exposure  Explanatory Response Variable Variable We can change …to observe the the value of this effect it has on variable… this variable Example 1: We change the dosage of the drug (explanatory variable) to observe the effect it has on blood pressure (response variable) Example 2: We change the type of diet (explanatory variable) to observe the effect it has on weight (response variable) What else can you think of? Smoking and lung cancer Smoking- explanatory Lung cancer- response Which of the following would best be categorized as an exposure/independent/explanatory variable for the study by Sana et al? A) Subject knowledge B) Test scores C) Laptop multitasking D) Students E) Distraction Which of the following would best be categorized as a dependent/response/outcome variable for the study by Sana et al? A) Subject knowledge B) Test scores C) Laptop multitasking D) Students E) Distraction Research questions First step to conducting research study When we have or create research questions, we have to make sure its clear and has explanatory and response variables A research question is a statement or query that summarizes the objectives of the research study and (typically) refers to the explanatory and response variables under study. Does exposure to asbestos (explanatory) cause lung Cancer (response)? Research questions are then framed as hypotheses, formal statements that can be tested with statistics and data. – Statistics usually require us to state a null hypothesis. What is the null hypothesis for the research question above? Null hypothesis- assumes no relationship between explanatory and response variable and it is what we want to test Null hypothesis: there is not relationship between expire to asbestos and lung cancer  For the following research questions, identify the explanatory and response variables and the null hypothesis. Does health and Does a restaurant safety training inspection system reduce injury at prevent foodborne work? infection? Will use of probiotic Is smoking a cause supplements of depression in improve symptoms adults? of chronic colitis?  For the following research questions, identify the explanatory and response variables and the null hypothesis. Does health and Does a restaurant safety training inspection system reduce injury at prevent foodborne work? infection? Will use of probiotic Is smoking a cause supplements of depression in improve symptoms adults? of chronic colitis?  For the following research questions, identify the explanatory and response variables and the null hypothesis. Does health and Does a restaurant safety training inspection system reduce injury at prevent foodborne work? infection? Will use of probiotic Is smoking a cause supplements of depression in improve symptoms adults? of chronic colitis?  For the following research questions, identify the explanatory and response variables and the null hypothesis. Does health and Does a restaurant safety training inspection system reduce injury at prevent foodborne work? infection? Will use of probiotic Is smoking a cause supplements of depression in improve symptoms adults? of chronic colitis?  For the following research questions, identify the explanatory and response variables and the null hypothesis. Does health and Does a restaurant safety training inspection system reduce injury at prevent foodborne work? infection? Will use of probiotic Is smoking a cause supplements of depression in improve symptoms adults? of chronic colitis? Research Question: Will use of probiotic supplements improve symptoms of chronic colitis? Which would be the best way to capture or measure the explanatory variable or exposure? a. Symptom checklist for chronic colitis b. Test for presence of healthy bacteria in the gut c. Whether or not the patient received a probiotic supplement d. Biopsy for definitive colitis diagnosis  We have a research question and null hypothesis, we know the explanatory and response variables of interest We have to TEST our hypothesis! A good test of a hypothesis would allow you to falsify the hypothesis Research studies are a great tool to test these hypothesis and if there is enough evident to reject null hypothesis the data should support or refute null hypothesis. Ex of statistical tests we can use pi squares, t-test, regression models  Not all studies are created equal Meta analyses is above all b/c it is summarizing all studies thar have happened to date and have been published- “are we finding the same thing across populations” A specific RCT will be conducted in one specific place- gold standard for experimental research b/c ability to minimize bias and look at cause and effect relationships- people are in lab in RCT Cohort study would have more bias than RCT b/c cohort study you are observing people but there’s factors that are out of experimental control- no control over variables  Experiments Want to change one thing at a time- ex: changing one variable while holding all In an experiment, the researcher can things constant manipulate (or change) the explanatory Ex: different diets on weight loss. To variable to observe its effects on the observe effect diet (expl) has on weight loss outcome (response) , have to control for all other Want to understand the effect on our factors, such as physical activity, portions explanatory to response they’re intaking, medication, health conditions Good experiments use a control condition to isolate the effects of the explanatory variable- Complete control over explanatory in an RCT and then observe effect has on outcome “All else being equal” Different manipulations to the explanatory variable are usually described as treatment conditions, even when these are not medical treatments What if instead of the complicated study Sana et al did, they just went into a typical university classroom and compared student performance by those who multitask during lecture vs. those who don’t? From last week  Occurs when a relationship between two variables can be explained by a third (hidden, unmeasured, Confounding confounding) variable. A hidden factor is explaining that observed association http://www.slate.com/blogs/crime/2 013/07/09/warm_weather_homicid e_rates_when_ice_cream_sales_rise _homicides_rise_coincidence.html Confounding Does eating ice cream cause crime? Consumption of ice cream Criminal behaviour Recall: Confounding Does eating ice cream cause crime? Hot weather: criminals are outside and not inside because it’s cold and not hot That is causing the association between ice cream and crime rates and we can now conclude with the confounder that it does cause higher crime rates Hot Weather Consumption of ice cream Criminal behaviour Recall: Confounding Does eating ice cream cause crime? Hot Weather Consumption of ice cream Criminal behaviour Confounding To be a confounder, a variable has to meet 3 criteria: (1) Must be related to both independent (exposure) and dependent variables (outcome).- arrows need to be pointing from the confounder into explanatory and response variable Hot Weather Consumption of ice cream Criminal behaviour Confounding To be a confounder, a variable: (1) Must be related to both independent (exposure) and dependent variables (outcome). (2) Must not be a common effect of both variables. Variables cannot produce the confounder Not confounding: Hot Weather Consumption of ice cream Criminal behaviour Confounding To be a confounder, a variable: (1) Must be related to both independent (exposure) and dependent variables (outcome). (2) Must not be a common effect of both variables. (3) Should not be in the “causal pathway” when we’re going to exposure and then it leads to the outcome Not confounding: Hot Weather Consumption of ice cream Criminal behaviour  Which of the following is the best suspect for a confounder in this study? A) Coffee consumption B) Bladder cancer C) Smoking D) Bladder surgery Confounding Smoking is third variable- smokers are more likely to drink coffee and smoking causes bladder cancer- it’s the third variable Smoking Coffee Consumption Bladder Cancer Confounding Smoking is: Related to both independent (exposure) and dependent variables (outcome). Not a common effect of both variables. Not on the “causal pathway”- it’s not that drinking coffee causes you to smoke which than causes bladder pathway So it is a confounder! Smoking Coffee Consumption Bladder Cancer Confounding in research design Any time we are interested in the relationship between an explanatory and response variable we must be concerned about confounding In epidemiology, age and sex are known as the “universal confounders” because they can influence a wide range of health relationship and always need to be accounted for Confounding especially difficult to manage when we don’t know what the confounding variables are (e.g. they are hidden, unmeasured)- might be a confounder we haven't measured so they can’t be accounted for- one of the biggest issues when there is a hidden variable and that is why RCT’S are so important Advantage: Experiments The key to experimental design is randomization: participants are assigned randomly to experimental condition (explanatory variable) – A way people account and control for cofounding On average, this means that unmeasured confounders are (statistically speaking) distributed evenly between the experimental treatments. – Process of randomization where unmeasured confounders are evenly distributed across the groups  By randomly assigning these people into the 2 groups, there is an equal spread out Experimental designs Randomize subjects/participants to one of two (or more than two) treatments. – Other words for treatment: factors, levels, arms One treatment should be a control. How we can directly compare is the treatment we’re applying on these groups working Randomized Control Trials People w/ intervention (RCTs) get or don’t get outcome and same thing with Intervention: people who do not get people who drug Gather sample receive drug Randomly assign people comparison- no drug  Types of Randomization in RCTs Simple Each participant has an equal chance of being assigned to any group. Randomization Example: Flipping a coin to assign participants to treatment or control. Ensures certain characteristics (e.g., age, gender) are evenly distributed Stratified across groups. Randomization Example: Stratifying participants by age group before randomizing within each stratum. Groups or clusters (e.g., schools, hospitals) are randomized instead of Cluster individuals. Randomization Example: Randomizing entire classrooms instead of students to treatment or control. Adjusts the probability of group assignment based on interim results. Adaptive Usually done for ethical purposes Randomization Example: Increasing the likelihood of assignment to a more effective treatment during the trial.  Cross-over Designs What is a Advantages of Disadvantages of When to Use Crossover Trial? Crossover Designs Crossover Designs Crossover Designs Participants Reduces Carryover Ideal for chronic receive both variability: Each effects: The conditions where treatments (e.g., participant effect of the first symptoms treatment and serves as their treatment may fluctuate over placebo) at own control. influence the time and different times. results of the treatments have Example: A second. short-lived participant Must utilize a effects. receives a new wash-out period Example: Studies migraine drug for Example: A on medications 2 weeks, migraine drug for conditions followed by a that has lingering like arthritis or placebo for effects might chronic pain. another 2 weeks. alter the response during the placebo period. Cross-over Design Selecting a control treatment or control condition Control conditions should be as similar to experimental treatment as possible, except for the specific explanatory variable of interest. Have no active component Selecting a control treatment or control condition Control conditions should be as similar to experimental treatment as possible, except for the specific explanatory variable of interest. In health, we often want a placebo control. From Porta (Dictionary of Epidemiology, 2014): Placebo: A medication or procedure that is inert (i.e., one having no pharmacological effect) but intended to give patients the perception that they are receiving treatment or assistance for their complaint. From Latin placebo “I shall please” Ex: giving treatment group a blue pill, but it has no medication in it” A placebo control helps address the placebo effect: “The beneficial effect resulting solely from the administration of a treatment, no matter whether strictly a placebo, an active drug, or another therapeutic procedure.” Porta, Dictionary of Epidemiology, 2014 Comes simply from administrating a treatment but it has no effect Placebo effect The expectation of improvement can be a potent effect on its own We don’t fully understand how this works or why, but its effects are widespread in both: – Subjective measures (mood, pain, energy) – Objective measures (blood pressure, movement flexibility) Example of a situation where you think the placebo effect is especially relevant? Chronic pain http://science.sciencemag.org/content/293/5532/1164  A study of the effectiveness of a training intervention on safe food handling practices in a restaurant. Control group What would get no training or a fake training would be A study of the effect of a knee surgery an on performance in athletes. No surgery appropriate but might make a fake incision to make control it look like surgery, put them to sleep condition A study of the benefits of a new blood for? pressure drug for people with hypertension. Give them a placebo drug When a part of an experiment or study relies on the researchers getting information from participants or watching those participants, the researchers can be considered observers. Observer-expectancy effect We can convey (as observers) and respond to (as the observed) subtle cues, often unconsciously- participants pick up on these cues and adjust their behaviours intentionally People act differently based on them being in a research study or watched  1958- person went to electric factory to conduct experiments on workers there, and he wanted to see if when he changed the lighting levels would change how the workers would be efficient in their work, and they fond during righting changes, productivity improved, but then suddenly, once the researcher left and lighting stayed the same, everybody went back to normal, productivity decreased- one workers realized they were being observed they started to work harder THE HAWTHORNE EFFECT Hawthorne Works, 1925  Knowledge of being observed can affect participant behavior Could affect response Effect of variables directly being observed Could influence the way the participants interact with researchers Could influence compliance with experimental protocol  Study blinding prevents observer Data analyst can be blind as well- they don’t expectancy effects by preventing know whether the observers and (often) the participant got the participants from knowing which control medication or condition they are assigned to. actual treatment Blinding Why Blinding is Important Reduces the likelihood Prevents observer bias of researchers and placebo effects. unconsciously influencing the results.  Blinding 01 02 03 Single Blinding Double Blinding Triple Blinding The participant does not Both the participants and Participants, researchers, know which group they the researchers are and the analysts of the are in. unaware of group data are all blinded. Example: A patient assignments. Example: Even the receiving a placebo pill Example: Neither doctors statisticians analyzing the but unaware whether it's nor patients know who is outcomes do not know the active drug or not. receiving the placebo vs. the group assignments. the actual treatment.  Ethical Considerations 1. Informed Consent – Participants must be informed about the study’s risks, benefits, and their rights. – Example: A clear explanation of possible side effects for drug trials. 2. Equipoise – Ethical requirement that there is genuine uncertainty about which treatment is better. – Example: Not conducting an RCT when one treatment is already known to be superior. 3. Stopping Rules – Trials may stop early if one treatment is found to be clearly beneficial or harmful. – Example: Halting a trial early when interim results show clear harm to participants. 4. Use of Placebos – Placebos should only be used when no proven effective treatment exists. – Example: A placebo-controlled trial for a new migraine treatment when no standard treatment is available.  Minimizes bias: Random allocation of participants reduces selection bias, ensuring that the treatment and control groups are comparable. Control Over Confounding Variables: Randomization helps control for both known and unknown confounding variables, reducing their impact on the results. RCT High Internal Validity: The rigorous design of RCTs provides strong evidence for the internal validity Advantages of the study findings. Gold Standard in Research: Considered the most reliable form of evidence for evaluating the efficacy of interventions, especially in clinical settings. Blinding Capabilities: Blinding (single, double, or triple) can reduce or eliminate observer, participant, and assessment biases.  RCT Disadvantages Limited Generalizability: The Ethical Constraints: controlled environment and Time-Consuming: RCTs often Randomization may not be require significant financial strict inclusion criteria can ethical in all situations, limit the external validity, and time investments, especially when withholding making them resource- making it difficult to treatment could harm generalize findings to a intensive. participants. broader population. Complexity in Not Always Feasible: In some Implementation: RCTs can be Loss to Follow-Up: High cases, it may be impractical complex to design, dropout rates or loss to or impossible to randomly implement, and monitor, follow-up can threaten the assign participants to requiring careful planning validity of the study results. different interventions, such and execution. as in observational studies.  EXPERIMENT DEMONSTRATION Experiment Demonstration We are going to do an in class activity to mimic an experiment. I will post the instructions on the next couple of slides. To ensure you understand the experiment, try the debrief questions on the following slide as homework. Collaborate with fellow students. Experiment Step 1: Write down your student number Look at the second last digit of your TMU student number – E.g. 100024601 Is this number even or odd? Logistics The groups are defined by odd or even tickets ODD NUMBERS: please open the link that I present in a few slides. EVEN NUMBERS: please sit quietly. Odd numbers To follow, I will present you will some simple addition problems via a link. Please do NOT use a calculator I will play some thinking music while we are completing the problems You will have 2 min to complete as many as you can  Odd numbers Please open the link but do not begin until I indicate https://forms.gle/VLPreGLfaQfnXPqUA Logistics The groups are defined by odd or even tickets ODD NUMBERS: please sit quietly. EVEN NUMBERS: please open the link that I present in a few slides. Even numbers To follow, I will present you will some simple addition problems using a link You will have 3 min to complete as many as you can. Even numbers Please open the link but do not begin until I indicate https://forms.gle/UzhQWv4aKR8uCdom8  I will determine the number of correct TIME’S UP! responses you had and discuss the results in 2 weeks. Activity Debrief: On your own and with classmates What do you think Who were the was the research What were the participants in the explanatory and question? Null experiment? hypothesis? response variables? Was assignment to What was the the experimental treatment What was the control condition? condition random? condition? Why or why not? Was the experiment What might you do What did the adequately differently in participants do? controlled? Why or conducting a similar why not? experiment? Next Class: Cohort Studies When RCTs are not feasible or ethical, we can use observational research designs, like cohort studies. Next class will be an introduction to cohort studies by Emma Buajitti