Understanding the Data: Study Design & Bias in Research

Lecture 4: Understanding Data in Research

Study Design & Bias in Research

• Course: KHPM324 Chronic Diseases of Modern Society

• Instructor: Hannah Oh, ScD

• Institution: Division of Health Policy & Management, College of Health Sciences, Korea University

Review Test

• Access test at www.socrative.com

  • Student log-in required

  • Enter name for participation points

  • Room name: [To be titled in class]

• A total of 3 questions on the test

Research Question Design Example

• Scenario: Testing whether a new drug (Drug A) reduces the risk of stroke.

• Study Design Question: How would you design a study to answer this question?

Randomized Experiment Design

• Components:

  • Drug A (new drug) vs. Placebo

  • Randomized assignment of subjects into groups

• Question: What would be the 1-year risk of stroke for both groups?

Epidemiologic Study Designs

• Experimental Studies:

  • Involves intervention by the investigator in treatment assignments.

  • Example: Randomized Controlled Trial (RCT)

• Observational Studies:

  • Involves observing outcomes based on natural treatment assignments.

  • Types:

  • Cross-Sectional Study

  • Cohort Study

  • Case-Control Study

Randomization in Studies

• Process:

  • Divide subjects into two groups: Group 1 & Group 2

  • Random assignment of treatment (e.g., drug A or placebo)

• Outcome of Randomization:

  • Ensures comparability between groups

  • 1-year risk in both groups remains the same regardless of treatment assignment

Study Types: Observational Study Designs

Cross-Sectional Study

• Focus: Disease frequency comparison (exposed vs. unexposed) at a specific time

• Limitations: Provides a snapshot, lack of temporal data

Cohort Study

• Structure:

  • Groups: Exposed group (at risk) vs. Unexposed group

• Outcome: Measure incidence of disease over time

Case-Control Study

• Definition:

  • Comparison of cases (with disease) and controls (without disease)

• Outcome: Frequency of exposure comparisons between cases and controls

Study Design Phases

Prospective Studies

• Definition: Exposures measured before outcome events occur

• Characteristics:

  • More costly and time-consuming

Retrospective Studies

• Definition: Exposures measured after outcome events have occurred

• Characteristics:

  • Prone to incorrect information gathering

Errors in Research

• Categories:

  • Random (chance errors)

  • Systematic (bias errors)

Implications of Errors

• Systematic Error:

  • Flaws in study design, data collection, or analysis can lead to incorrect conclusions

  • Broad types of bias include:

  • Confounding

  • Selection Bias

  • Information Bias (measurement error or misclassification)

Confounding Bias Examples

Example 1: Down Syndrome & Birth Order

• Question: Is birth order a cause of Down syndrome?

• Finding: Maternal age influences both factors (risk).

Example 2: Aspirin & Stroke

• Observation: People taking aspirin may show a higher stroke rate.

• Confounding Variable: The presence of atherosclerosis increases both stroke risk and the likelihood of aspirin use—thus misleading correlation.

  • Atherosclerosis is defined as the buildup of fats, cholesterol, and other substances in artery walls which can restrict blood flow.

Understanding Confounders

• Definition:

  • A confounding variable is both causally associated with the outcome and either causally or non-causally associated with the exposure without being an intermediate variable.

Reducing Confounding in Research

• Continuation:

  • Randomization provides similar risk factor distributions across groups.

  • Other methods:

  • Standardization (e.g., age-standardized measures)

  • Restriction and stratification (e.g., comparisons within the same age group)

  • Covariate adjustment in regression models

Selection Bias

• Description:

  • A systematic error occurring due to biased recruitment or retention of study participants.

  • Examples of analysis illustrating selection bias through disproportionate representation of smokers and non-smokers and their respective disease statuses are illustrated with ratios and percentages reflecting risks.

Recall Bias

• Definition:

  • A type of information bias resulting from inaccurate recall of past exposure.

• Example: In cases of birth defects, mothers may recall pregnancy exposures differently based on whether their child has a defect, leading to differential accuracy in recall.

Reducing Recall Bias

• Strategies:

  • Frame survey questions to improve recall accuracy.

  • Utilize a control group composed of mothers of babies born with defects unrelated to the study.

  • Gather historical data from medical records or biomarkers recorded pre-outcome.

Conclusion

• Recap and anticipation for the next lecture: See you next week!