intro to epidem W6-Randomized Trials Study Notes

Randomized Trials

• Focus on different forms of study designs, starting with randomized trials.

• Randomized trials are a form of experimental design.

• Researchers randomly allocate study participants into study groups:

  • Control group: placebo or standard treatment.

  • Experimental group: receives a new treatment or intervention.

Key Elements of Randomized Trials

• Independent variable: group membership (experimental vs. control).

• Three phases:

  • Pre-intervention assessment of the outcome.

  • Intervention period.

  • Post-intervention assessment of the outcome.

• Goal: to observe and assess the change in the outcome level from pre- to post-intervention.

Study Population and Recruitment

• Start with a sample population from a larger population.

• Define a population of interest that is already experiencing the outcome before the study begins (e.g., testing depression medication on individuals who already have depression).

• Create a detailed written recruitment plan to allow others to replicate the study.

• We'd like to generalize study participants to larger populations. This means that the study participants' demographics, characteristics, and outcome levels are similar to those of the larger population. Large sample sizes help ensure that results generalize to a larger population.

Randomization

• Random selection: bringing participants into the study (recruitment).

• Randomization: assigning participants to groups (experimental vs. control) after they are in the study.

• Ideally, participants should be equally likely to be in either group (true randomization).

  • Accomplished via coin flipping, drawing from a hat, or similar methods.

  • True randomization does not guarantee equal group sizes.

Benefits of Randomization

• Eliminates investigator bias in group assignment.

• Increases the likelihood that resulting groups will be similar regarding key characteristics (age, gender, known exposures, genetic predisposition, etc.).

• Good randomization increases the likelihood that the groups will be similar on underlying factors, but does not guarantee similarity; groups may differ due to chance.

Randomized Design Sequence

1. Start with a population specific to the outcome and relevant to the intervention.

2. Recruit a sample from that population.

3. Randomly allocate participants into experimental or control groups.

4. Pretest(pre-intervention assessment) outcome levels in both groups.

5. Implement the intervention.

6. Post-test(post-intervention assessment) outcome levels in both groups.

7. Assess the Level of outcome changed for some individuals in both the experimental group and the control group. Determine if the experimental group changed more than the control group.

Data Presentation and Interpretation

• Data includes pre-intervention and post-intervention outcome levels for both groups.

• Focus on the change in outcome within each group.

• With good randomization, any difference between experimental and control groups can be attributed to the intervention.

• Any change in the experimental group over and above the change in the control group is caused by the intervention.

Measures of Effect

• Efficacy: reduction in the rate of risk of an outcome in an intervention group compared to a control group.

  • Calculated as: \frac{rate\,in\,control\,group - rate\,in\,experimental\,group}{rate\,in\,control\,group}

• Number Needed to Treat (NNT): number of individuals who need to be vaccinated to prevent one outcome.

  • Calculated as: \frac{1}{rate\,of\,outcome\,in\,untreated\,group - rate\,of\,outcome\,in\,experimental\,group}

Design Options for Randomized Trials

• Blinding/Masking: hiding the study condition from participants and/or researchers.

  • Single-blind study: participants are unaware of their group assignment.

  • Double-blind study: both participants and assessors are unaware of group assignment (reduces bias).

• Stratified Randomization: controlling the distribution of key variables (e.g., gender, age) among experimental and control groups.

  • Ensures similar distribution of variables that influence the outcome.

Crossover Randomized Trial

• Study condition changes during the study; participants switch from experimental to control, and vice versa.

• Often planned in advance, controlling how long each participant is in each condition.

• Individuals serve as their own controls (change during intervention compared to change during control for each individual).

Factorial Design

• Used to compare two treatments when no treatment is also ethical.

• Examines effects of each treatment separately, in combination, or with no treatment.

Non-Randomized Trials

• Studies without control groups (e.g., case study, case series).

• Difficult to attribute change to the intervention without a control group.

• Historic controls: comparing current subjects to similar individuals from the past (using old data or medical records).

  • Problems: poor record keeping, inconsistent data, historical factors influencing the outcome.

Populations, Sample Size, and Interpretation Errors

• Larger sample size benefits:

  • Easier to generalize findings to the target population.

  • Easier to find statistically significant results.

• Challenges with large sample sizes: difficult with rare outcomes, can be expensive.

• Sample size calculators: statistical tools to determine adequate sample size.

Sample Size Effects.

• Sample size has a large impact on how representative our sample population will be in relation to the population that we are wishing to generalize to.

• Small Samples

  • With a small sample, it's often not possible to represent all the variation that's out there in the true population.

• Larger Samples

  • Might wind up with a sample that does not fully represent the population but as you increase you sample size the more likely your sample will represent that of the population.

Sampling and Reality

• Study data may not reflect what is true in the real population due to sampling problems or other study-related factors (e.g., participant noncompliance).

• Errors can occur when drawing conclusions from sample data.

Statistical Errors

• Type I Error (Alpha Error):

  • Study data indicates groups differ, but in reality, they do not.

    • Probability: p-value (researchers generally want a p value < 0.05 for a statistically significant difference).

    • Larger sample sizes generally equate to smaller p values (less chance of alpha error).

• Type II Error (Beta Error):

  • Study data indicates groups do not differ, but in reality, they do.

    • Probability: beta.

    • Related to the power of a study (the probability of correctly deciding if groups differ).

    • Power calculated as: 1 - \beta

    • Increased sample size increases power and the likelihood of seeing real differences.

• The larger your sample, the more likely it is that it will be truly representative of your population. If it is truly representative of your population, that also makes it less likely that you will make an error, such as a type one error or a beta error.

Overpowered Studies

• Samples that are too large can lead to statistically significant findings that are not meaningful in practice.

Validity of Study Designs

• External Validity: generalizability of findings to the target population.

  • Problems arise from how participants are brought into the study (non-representative samples).

• Internal Validity: problems with the study design itself.

  • Non-random assignment.

  • Investigator bias (lack of blinding).

  • Differential loss to follow-up between groups.

Reporting and Ethics

• Researchers should report both statistically significant and non-significant findings.

• Ethical considerations:

  • Is randomization ethical (assigning to placebo)?

  • Truly informed consent when participants don't know their condition?

  • Many research questions cannot be ethically studied with randomized trials.