Epidemiology Lecture 4: Study Designs and Bias

Lecture 4: Study Designs and Bias Sources in Epidemiology

• Epidemiology is the study of the distribution and determinants of health or disease in populations, and the application of this knowledge to control health problems.
• Epidemiological questions include what's causing a particular disease in a population, the average risk of disease, who is most at risk, and how risk can be reduced.
• A major goal is to explain patterns of disease occurrence and causation (etiology).
• Other questions include determining if there's an association between an exposure and outcome, if an exposure is the cause of the outcome/disease, and how much an exposure increases the risk.
• Epidemiological research aims to quantify exposure, outcome, and confounders.
• Different study designs are used to evaluate the association between an exposure and outcome.

Study Designs in Epidemiology

• Observational studies:
  • Ecological
  • Cross-sectional
  • Case-control
  • Cohort
• Intervention/Experimental studies:
  • Randomized controlled trials
  • Intervention/non-experimental
  • Difference-in-Difference, IV, ITS

Null Hypothesis

• The null hypothesis (H0) is a statement assuming no effect, association, or difference between groups/variables in statistical testing.
• It serves as the default assumption, and if evidence supports rejecting it, the alternative hypothesis (H1) is considered.

Type 1 and Type 2 Errors

• Type 1 Error (False Positive): Concluding an effect or association exists when none exists.
• Example: Concluding a new drug is effective when it's not.
• Significance Level (α): The probability of making a Type 1 error (commonly set at 0.05)
• Type 2 Error (False Negative): Failing to detect an effect or association that truly exists.
• Example: Concluding a drug doesn't lower blood pressure when it actually does.
• Power (1-β): Probability of avoiding a Type 2 error. Increased sample size and effect size increase power.

Bias

• Causal Effect: Random Error, Confounding
• Information Bias (misclassification)
• Selection bias
• Bias in inference
• Reporting bias
• Bias in knowledge use

Explanation of Biases

• Random Error: Arises from chance, affects precision, leads to variability in data, but doesn't favor one outcome over another.

• Example: Measuring blood pressure multiple times with different results.

• Mitigation: Increasing sample size, averaging multiple measurements.

• Confounding: An extraneous variable associated with both the exposure and outcome, distorting the relationship.

• Example: Studying alcohol consumption and lung cancer without accounting for smoking.

• Mitigation: Randomization, matching, stratification, or multivariable analysis.

• Information Bias: Errors in measuring exposure, outcome, or other variables leading to misclassification.

• Example: Recall bias in case-control studies.

• Mitigation: Using objective measures and blinding data collection.

• Selection Bias: Participants in a study aren't representative of the target population due to how they're selected.

• Example: Recruiting participants for a study from fitness centers only.

• Mitigation: Ensuring random sampling & minimizing loss to follow-up.

• Bias of Inference: Drawing incorrect conclusions from a study due to overgeneralization or ignoring limitations.

• Example: Concluding a drug is effective for all populations based on results from a small, homogenous sample.

• Mitigation: Carefully interpreting results and acknowledging limitations.

• Reporting Bias: Selection of results for reporting, typically favorable ones.

• Example: Reporting only the positive outcomes of a new drug.

• Mitigation: Preregistration of studies, requiring full reporting.

• Publication bias: Studies with positive results are more likely to be published than those with negative or null results.

• Example: Meta-analyses may overestimate treatment effects if only published studies are included.

• Mitigation: Encouraging publication of all results and use of trial registries to identify unpublished data.

Ecological Studies

• Compare overall population or group statistics of exposures and outcomes.
• Aggregated level; exposure prevalence arrow outcome prevalence
• Example: Average diet and life expectancy across multiple countries.
• Advantages: Cheap & quick, generate hypotheses
• Disadvantages: Difficult to draw conclusions at the individual level (ecological fallacy), populations can differ in other ways.

Cross-Sectional Studies

• Measure all factors (exposure, outcome, confounders) at one time point.
• Used to measure prevalence of health outcomes & describe features of a population.
• Don't follow individuals over time.
• Advantages: Quick
• Disadvantages: Time sequence of events difficult to ascertain, only measure prevalence, large surveys needed for rare diseases.

Case-Control Studies

• Identify people with a health outcome (cases) & select appropriate controls.
• Compare the proportion exposed between cases and controls.
• Example: Vaccine effectiveness studies.
• Advantages:Investigate multiple exposures (one outcome); relatively quick; good for rare outcomes (e.g., cancers)
• Disadvantages: limited to one outcome; selection bias if controls not comparable to cases; difficult to determine time sequence of events.

Cohort Studies

• Follow a group of people over time according to their exposure status to see who develops a health outcome.
• Example: Analyze association between talc powder use & ovarian cancer.
• Advantages: Measure incidence, time sequence clear, rare exposures can be studies; multiple exposures and outcomes
• Disadvantages: often time-consuming & expensive, losses to follow-up (selection bias); outcome may be influenced by knowledge of exposure.

Summary

• Study design is crucial for valid conclusions.
• Multiple studies over time often needed for evidence to influence policy.
• Study designs can vary, affecting conclusions.
• Factors such as the definition/measurement of exposure/outcome, biases in the selection of participants, how confounders are measured, and differences in the populations studied are important considerations.
• Important to interpret findings in relation to the study design.