Epidemiology Exam 3 Flashcards

Validity and Bias

• Validity denotes the accuracy of a study, characterized by the absence of random error, bias, and confounding.
• Three critical alternative explanations to assess study validity:
  • Bias: Systematic error that skews the perceived relationship between exposure and disease.
  • Confounding: Influence of a third variable that distorts the true association between exposure and disease.
  • Random Error: Variability in study results that occurs by chance, lacking a specific cause.

Generalizability and Association

• Generalizability refers to extending study conclusions beyond the specific study population and setting.
• Bias can lead to both random and systematic errors simultaneously.
• When bias drives results toward the null hypothesis, the true association is underestimated; when it moves away, the true association is overestimated relative to the null.

Selection Bias

• Selection bias arises from differences in how study groups are chosen.
• More common in case-control and retrospective cohort studies; also a concern in prospective cohort studies due to differential follow-up.
• Strategies to mitigate selection bias: consistent case/control selection criteria, high participation rates, and effective tracking methods.
• The impact of selection bias can distort association metrics, shifting them either towards or away from the null.

Information Bias

• Information bias occurs from systematic discrepancies in how data on exposure and disease is gathered from participants.
• Types of information bias include:
  • Recall Bias: Differences in accuracy of reported information between groups.
  • Interviewer Bias: Systematic differences in data collection by interviewers.
  • Measurement Error: Nonsystematic errors that affect both exposure and outcome assessment.
• Preventive methods for information bias involve careful study design and data collection methods.

Confounding Variables

• Confounding is the interplay of an exposure, outcome, and a third variable that alters the perceived relationship.
• A valid confounder meets three criteria:
  • Associated with the exposure in the population being studied.
  • Acts independently as a predictor of the disease.
  • Is not part of the causal pathway between exposure and disease.

Misclassification

• Differential Misclassification: Errors are related to a specific axis (exposure or disease) and can bias outcomes in either direction.
• Non-Differential Misclassification: Errors on one axis unrelated to the other, generally biasing results towards the null hypothesis.
• Preventing misclassification focuses on enhancing data collection accuracy through various measurement techniques.

Case-Control Study Framework

• In case-control studies, determining how participants recall past exposure can affect odds ratio calculations.
• Bias introduced through inaccurate reporting can shift odds ratios toward the null value due to non-differential misclassification.
• Understanding how exposure and outcome relate in these studies is critical for evaluating true associations.

Statistical Concepts and Directions

• Adjusting for confounding can be achieved during both study design and analysis phases.
• The "crude measure of association" represents initial estimates without adjustment for confounding factors.
• Direction of bias can vary significantly depending on whether confounding leads to overestimation or underestimation of the true association.

Key Takeaways

• Recognizing and addressing different biases is essential to maintain the integrity of epidemiological research.
• Effective study design, critical evaluation of variables, and comprehensive data collection methodologies are key to minimizing errors.
• Continuous vigilance for biases at all study stages can enhance the reliability and applicability of research findings.### Bias in Epidemiological Studies
• The magnitude of the Odds Ratio (OR) is often less than the true value due to bias.
• Investigators commonly investigate types of bias such as recall bias and interviewer bias, both of which are forms of information bias.
• Selection bias is prevalent in case-control studies and retrospective cohort studies, leading to incorrect associations.

Confounding Control

• Confounding can be managed during the design phase using:
  • Randomization
  • Restriction
  • Matching
• In the analysis phase, confounding can be controlled through:
  • Standardization
  • Stratified analysis
  • Multivariate analysis
• Controlling confounding during the design phase is generally more effective than during the analysis phase.

Limitations in Confounding Control

• Matching is challenging due to its dependence on known confounders; may incur high costs and time.
• Restriction limits generalizability and sample size and may not fully control for confounders.
• Randomization is primarily applicable to experimental studies and is less effective with smaller samples.
• In the analysis phase, stratified analysis is complex with multiple confounders, and multivariate analysis requires stringent model assumptions.

Residual Confounding

• Residual confounding refers to unaddressed confounding variables that remain even after some confounding has been controlled.

Confounding Example

• If socioeconomic status (SES) shows different Relative Risks (RR) across stratifications, and an SES-adjusted RR differs significantly, SES is identified as a confounder.

Observed vs True Association

• Positive confounding results in the observed association exaggerating the true association, moving farther from the null value.
• Negative confounding causes the observed association to mask the true association, pulling it toward the null value.

Mediators vs Confounders

• A mediator is part of the causal pathway between an exposure and disease, while a confounder is not.
• It is possible to control for confounding in both design and analytic phases of research.

Random Error

• Random error encompasses unsystematic errors due to chance and can occur in all study types.
• It can create false associations or obscure true associations.
• Causes include measurement error and sampling variability, both contributing to random error in epidemiological studies.

Hypothesis Testing Outcomes

• Null hypothesis (H0) may be true and either not rejected (correct conclusion) or mistakenly rejected (Type I error).
• Alternative hypothesis (HA) may be true and either not confirmed (Type II error) or correctly supported (correct conclusion).

Statistical Practice

• The p-value indicates the probability of observing results due to chance if the null hypothesis is true.
• A confidence interval quantifies random error, indicating variability around a measure and providing a range in which the true effect lies.
• Wider confidence intervals signify greater random error and reduced precision, while narrower intervals indicate less random error and more precision.

Risk Interpretation

• A calculated Relative Risk (RR) of 4.3 indicates that sedentary computer programmers have 4.3 times higher risk of cardiovascular disease compared to non-computer programmers, based on a p-value of 0.02.