Module 8: Validity, Bias, and Confounding

Module Overview

  • Title: Validity, Bias, and Confounding

  • Course: BSPH 421

  • Professor: Kimberly Tseng

  • Dates: October 5 - October 7

Learning Objectives

  • Define validity and reliability

  • Describe the difference between internal and external validity

  • Define sources of bias in epidemiologic study designs

  • Define confounding and methods to control for it

Assignments

  • Due: 8.1 First submission for final projects (project planning stage)

    • Due Date: Monday, 10/20 at 11:59am

  • Reading for this week: Friis' chapter 7, pages 145-147

Validity

Definition of Validity

  • Validity: The degree to which a study’s findings accurately reflect the truth.

    • Questions to consider: "Is our study a true representation of what actually occurs in the population?"

    • Example: From an observational study: "All dogs have four legs… All tables have four legs… So all tables are dogs!"

Validity and Causality

  • Exploration of Validity in Relation to Causality:

    • Studies aim to answer: "Does the exposure cause the outcome?"

    • Validity assesses whether the study accurately reflects the real world concerning causation.

Internal vs. External Validity

Internal Validity

  • Question: "Are the results true for the study population?"

  • Encompasses concerns about:

    • Control: Minimizing confounding factors that may impact observed effects.

    • Generalizability: Can results be applied beyond the study?

External Validity

  • Question: "Would the conclusions hold true in other populations, times, or places?"

  • Influenced by various factors:

    • Strict inclusion/exclusion criteria (e.g., only healthy participants)

    • Standardized interventions and follow-ups

    • Controlled environments such as labs or hospitals

    • Diversity in participant enrollment

    • Actual real-world testing conditions

    • Implementation flexibility

Reliability

Definition of Reliability

  • Reliability: The consistency, reproducibility, or repeatability of results or measures.

    • Example Scenarios:

    • A radiologist interprets imaging similarly regardless of personal circumstances.

    • Multiple lab readers produce similar results.

    • A basketball player's shooting form remains consistent, but success may vary.

Relationship Between Reliability and Validity

  • Variations:

    • Poor validity can exist with poor reliability.

    • Good reliability does not necessarily imply good validity.

Sources of Error in Epidemiologic Research

Types of Error

  1. Random Error: Variability due to chance fluctuations.

  2. Systematic Error: Includes biases such as:

    • Selection Bias

    • Information Bias

    • Confounding

Bias in Epidemiologic Studies

Definition of Bias

  • Bias: Systematic error in the design or conduct of a study, leading to erroneous association between exposure and disease.

  • Can lead to either overestimation or underestimation of the true effect.

  • Affects internal and possibly external validity.

Impact of Bias on Validity

  • Types of Bias:

    • Selection Bias: Affects participant selection and follow-up.

    • Information Bias: Distorts data collection on exposure or outcomes.

  • Examples of Bias Types:[Indented List]

    • Recall Bias: Discrepancies in memory for past exposures between cases and controls.

    • Selection Bias: Participants chosen do not represent the larger population.

    • Attrition Bias: Participants who drop out differ from those who stay.

Selection Bias

Definition and Types

  • Selection Bias: Systematic error during participant selection or retention affecting internal validity.

  • Types of Selection Bias:

    • Healthy Worker Effect: Working populations are typically healthier than the general population.

    • Impact: Can lead to underestimation of risk in workplace studies due to healthier compared groups.

    • Self-selection/Vacancy Effect: Individuals who voluntarily participate may differ from non-participants.

    • Non-response Bias: Differences in response rates can skew data results.

    • Attrition Bias: Dropouts from studies differ based on outcomes, which can distort findings.

Information Bias

Definition

  • Information Bias: Systematic inaccuracies in measuring or classifying disease or exposure.

    • Discrepancies occur post-enrollment (study participation).

    • Types of Information Bias:

    • Recall Bias: Cases may remember past exposures better than non-cases.

    • Reporting Bias: Participants may misreport due to stigma or fear.

    • Hawthorne Effect: Participants change behavior when they know they are being observed.

  • Each form introduces risk to internal validity.

Confounding

Definition and Conditions

  • Confounding: The distortion of an association due to another variable influencing both the exposure and outcome (the 'third variable' problem).

Conditions for Confounding:

  1. Confounder is a risk factor for the disease, independent of exposure.

  2. Confounder is associated with exposure.

  3. Confounder is not in the causal pathway between exposure and disease.

Examples of Confounding:

  • Coffee Consumption and Pancreatic Cancer: Cigarette smoking mediates the relationship.

  • Parental Myopia and Nighttime Light Exposure: Uncontrolled genetic factors influence findings.

Controlling for Confounding

Study Design Methods:

  • Individual or group matching: Match confounder variables for exposed/unexposed subjects.

  • Restriction: Limit participant selection based on confounders (e.g., recruit only non-smokers).

  • Randomization: Randomly assign exposure, distributing variables evenly across groups.

Statistical Adjustment Methods:

  • Stratification: Analyze outcomes within specific confounder levels (e.g., maternal age stratified analyses).

  • Multivariate Regression Models: Adjust for multiple confounders simultaneously.

Conclusion

  • Study validity encompasses the accuracy of results obtained, with reliability playing a critical role.

  • Bias poses significant threats to internal and external validity and must be meticulously managed through design and analysis.

  • Confounding can alter the perception of causal relationships among variables, and thus needs thorough evaluation and control in study methodologies.