🦠Epidemiology Unit 4 – Bias, Confounding, and Effect Modification

Key Concepts and Definitions

• Bias refers to any systematic error in the design, conduct, or analysis of a study that results in an incorrect estimate of the association between exposure and disease
• Confounding occurs when a third variable is associated with both the exposure and the outcome, distorting the true relationship between them
• Effect modification happens when the magnitude of the association between an exposure and an outcome varies depending on the level of a third variable (effect modifier)
• Selection bias arises from the procedure used to select subjects for a study, leading to a distortion of the exposure-disease relationship
  • Can occur when the exposure and outcome are both related to participation in the study (healthy worker effect)
• Information bias results from systematic differences in the way data on exposure or outcome are obtained from the study groups
  • Includes recall bias, interviewer bias, and misclassification of exposure or outcome
• Validity is the degree to which a study measures what it intends to measure, and is affected by both random error and systematic error (bias)
• Precision refers to the degree of random error in a study's results, and is influenced by sample size and variability in the data

Types of Bias in Epidemiological Studies

• Selection bias occurs when the selection of study participants is related to both the exposure and outcome of interest
  • Examples include healthy worker effect, loss to follow-up, and self-selection bias
• Information bias arises from systematic differences in the way data on exposure or outcome are obtained from the study groups
  • Recall bias happens when participants' reporting of past exposures is influenced by their disease status
  • Interviewer bias occurs when interviewers gather information differently for exposed and unexposed groups
  • Misclassification bias results from inaccurate measurement or classification of exposure or outcome variables
• Confounding bias is caused by a third variable that is associated with both the exposure and the outcome, distorting their true relationship
• Lead time bias can occur in screening studies, where earlier detection of disease may appear to improve survival time without affecting the actual course of the disease
• Surveillance bias happens when one group is followed more closely than another, leading to increased detection of outcomes in that group
• Publication bias arises when studies with statistically significant results are more likely to be published than those with null findings
• Temporal bias can occur when the timing of exposure and outcome assessment affects the observed association between them

Understanding Confounding Variables

• Confounding variables are extraneous factors that are associated with both the exposure and the outcome, potentially distorting their true relationship
• For a variable to be a confounder, it must be associated with the exposure, associated with the outcome, and not be an intermediate step in the causal pathway between exposure and outcome
• Confounding can lead to an overestimation, underestimation, or even reversal of the true association between exposure and outcome
• Common confounders in epidemiological studies include age, sex, socioeconomic status, and lifestyle factors such as smoking and alcohol consumption
  • For example, if smoking is more common among individuals exposed to a certain occupational hazard, it may confound the relationship between the occupational exposure and a health outcome
• Confounding by indication can occur when a treatment or exposure is more likely to be given to individuals with a higher risk of the outcome, making the treatment appear less effective or even harmful
• Residual confounding refers to the distortion of the exposure-outcome relationship that remains after attempting to control for known confounders, often due to imperfect measurement or unknown confounding factors

Identifying and Controlling for Confounders

• To identify potential confounders, researchers should consider factors that are associated with both the exposure and the outcome based on prior knowledge and literature review
• Directed acyclic graphs (DAGs) can be used to visually represent the relationships between variables and help identify potential confounders
• Statistical methods for controlling confounding include stratification, matching, and multivariable regression analysis
  • Stratification involves dividing the study population into subgroups based on levels of the confounding variable and analyzing the exposure-outcome relationship within each stratum
  • Matching ensures that the distribution of potential confounders is similar between exposed and unexposed groups
  • Multivariable regression models simultaneously adjust for multiple confounders by including them as covariates in the analysis
• Randomization in experimental studies helps to distribute potential confounders evenly between study groups, minimizing their impact on the exposure-outcome relationship
• Sensitivity analyses can be conducted to assess the robustness of study findings to potential unmeasured confounders by simulating their effects on the observed association

Effect Modification: Concept and Examples

• Effect modification occurs when the magnitude or direction of the association between an exposure and an outcome varies depending on the level of a third variable (effect modifier)
  • For example, the association between alcohol consumption and cardiovascular disease may differ by sex, with a stronger protective effect observed in women compared to men
• Effect modification is also known as interaction, and can be synergistic (greater than additive effects) or antagonistic (less than additive effects)
• Identifying effect modifiers can help to target interventions to subgroups that may benefit most from them or to avoid potential harm in certain subpopulations
• Statistical methods for assessing effect modification include stratified analysis and the inclusion of interaction terms in regression models
  • Stratified analysis involves examining the exposure-outcome relationship within different levels of the potential effect modifier
  • Interaction terms in regression models test whether the association between exposure and outcome differs significantly across levels of the effect modifier
• Examples of common effect modifiers in epidemiological studies include age, sex, race/ethnicity, and genetic factors
  • The association between air pollution and respiratory health may be stronger among children and older adults compared to middle-aged individuals
  • The relationship between certain medications and adverse events may vary depending on an individual's genetic profile

Strategies for Addressing Bias and Confounding

• Careful study design is crucial for minimizing bias and confounding in epidemiological research
  • Randomization in experimental studies helps to distribute potential confounders evenly between study groups
  • Blinding of participants and researchers to exposure status can reduce information bias
  • Standardized data collection procedures and validated measurement tools can minimize misclassification bias
• Matching on potential confounders during the study design phase can help to ensure that exposed and unexposed groups are comparable
• Stratification and regression methods can be used to control for confounding during data analysis
  • Stratified analysis involves examining the exposure-outcome relationship within different levels of the confounding variable
  • Multivariable regression models simultaneously adjust for multiple confounders by including them as covariates
• Sensitivity analyses can assess the robustness of study findings to potential unmeasured confounders by simulating their effects on the observed association
• Triangulation of evidence from multiple studies with different designs and populations can help to establish the consistency and generalizability of findings
• Clear reporting of study methods, limitations, and potential sources of bias and confounding is essential for interpreting results and informing future research

Statistical Methods for Analysis

• Stratified analysis involves dividing the study population into subgroups based on levels of a potential confounder or effect modifier and examining the exposure-outcome relationship within each stratum
  • Mantel-Haenszel methods can be used to calculate summary measures of association across strata
• Multivariable regression models simultaneously adjust for multiple confounders by including them as covariates in the analysis
  • Logistic regression is commonly used for binary outcomes, while linear regression is used for continuous outcomes
  • Cox proportional hazards regression is used for time-to-event data in cohort studies
• Propensity score methods can be used to balance the distribution of potential confounders between exposed and unexposed groups
  • Propensity scores estimate the probability of exposure given a set of observed covariates
  • Matching, stratification, or weighting based on propensity scores can help to reduce confounding bias
• Interaction terms in regression models can be used to assess effect modification by testing whether the association between exposure and outcome differs significantly across levels of a potential effect modifier
• Mediation analysis can be used to examine the extent to which the association between an exposure and an outcome is explained by an intermediate variable (mediator)
• Sensitivity analyses can be conducted to assess the robustness of study findings to potential unmeasured confounders or sources of bias
  • E-value is a measure of the minimum strength of association that an unmeasured confounder would need to have with both the exposure and outcome to fully explain away the observed association

Real-world Applications and Case Studies

• The healthy worker effect is a classic example of selection bias in occupational epidemiology studies
  • Workers who are employed tend to be healthier than the general population, which can lead to an underestimation of the true association between occupational exposures and health outcomes
• Confounding by indication is a common issue in pharmacoepidemiological studies
  • Patients prescribed a certain medication may have a higher baseline risk of the outcome compared to those not prescribed the medication, making the drug appear less effective or even harmful
• The relationship between hormone replacement therapy (HRT) and cardiovascular disease risk in postmenopausal women is an example of effect modification by age and time since menopause
  • Observational studies suggested a protective effect of HRT on cardiovascular risk, but randomized trials found an increased risk, particularly among older women and those further from menopause onset
• The association between smoking and lung cancer was initially confounded by factors such as age, sex, and occupational exposures
  • Careful control for these confounders in epidemiological studies helped to establish the causal link between smoking and lung cancer risk
• The COVID-19 pandemic has highlighted the importance of addressing bias and confounding in epidemiological research
  • Studies examining risk factors for severe COVID-19 outcomes need to account for potential confounders such as age, comorbidities, and socioeconomic status
  • Differences in testing and surveillance across populations can lead to biased estimates of disease prevalence and risk factors