🦠Epidemiology Unit 5 – Causation and Causal Inference

Key Concepts and Definitions

• Causation establishes a direct relationship between an exposure and an outcome where the exposure is responsible for causing the outcome
• Causal inference involves using data and analytical methods to determine the causal effect of an exposure on an outcome
• Counterfactual outcomes consider what would have happened to an exposed group if they had not been exposed and compare that to the actual observed outcome
• Causal criteria (Hill's criteria) provide guidelines for evaluating evidence to determine if an observed association is causal
  • Includes strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy
• Causal effect estimates the difference in outcomes between an exposed and unexposed group while accounting for potential confounders
• Directed acyclic graphs (DAGs) visually represent the causal relationships between variables and help identify potential confounders and biases
• Potential outcomes framework compares the outcomes of the same individual under different exposure conditions

Types of Causal Relationships

• Direct causation occurs when an exposure directly affects an outcome without any intermediary variables (smoking directly causing lung cancer)
• Indirect causation involves the exposure affecting the outcome through one or more intermediary variables (obesity leading to diabetes which increases the risk of heart disease)
• Bidirectional causation happens when two variables cause each other (poverty and poor health outcomes)
• Reverse causation occurs when the outcome precedes and causes the exposure (cancer causing weight loss, not vice versa)
• Spurious association appears as a causal relationship between two variables but is actually due to a third variable causing both (ice cream sales and drowning rates both increase in summer due to hot weather)
• Effect modification happens when the effect of an exposure on an outcome varies depending on the level of a third variable (age modifying the effect of smoking on lung cancer risk)
• Sufficient cause is a cause that inevitably produces the outcome, while a necessary cause is a cause that must be present for the outcome to occur

Causal Models and Diagrams

• Structural causal models (SCMs) use mathematical equations to represent the causal relationships between variables
  • SCMs allow for the estimation of causal effects and the testing of causal assumptions
• Directed acyclic graphs (DAGs) are visual representations of the causal relationships between variables
  • DAGs consist of nodes representing variables and directed edges representing causal relationships
  • DAGs help identify potential confounders, mediators, and colliders
• Causal diagrams can be used to determine which variables need to be controlled for in order to estimate the causal effect of an exposure on an outcome
• Confounding can be represented in a DAG as a common cause of both the exposure and the outcome
• Mediation occurs when the effect of an exposure on an outcome is partially or fully explained by an intermediate variable
• Collider bias can occur when conditioning on a common effect of both the exposure and the outcome, creating a spurious association

Confounding and Bias

• Confounding occurs when a third variable is associated with both the exposure and the outcome, distorting the true causal relationship
  • Confounders can create spurious associations or mask true causal effects
• Selection bias happens when the study participants are not representative of the target population, leading to biased estimates of the causal effect
• Information bias arises when the exposure or outcome data are inaccurately measured or classified, leading to misclassification
• Recall bias occurs when participants' ability to accurately report past exposures or outcomes is influenced by their current health status or other factors
• Immortal time bias can occur in cohort studies when a period of follow-up time during which the outcome cannot occur is incorrectly attributed to the exposed group
• Confounding by indication happens when the indication for receiving the exposure is also associated with the outcome, creating a spurious association
• Strategies to address confounding include randomization, restriction, matching, stratification, and adjustment in statistical analyses

Study Designs for Causal Inference

• Randomized controlled trials (RCTs) randomly assign participants to exposure groups, ensuring that potential confounders are balanced between groups
  • RCTs are considered the gold standard for causal inference but may not always be feasible or ethical
• Cohort studies follow exposed and unexposed groups over time to compare the incidence of the outcome
  • Prospective cohort studies enroll participants before the outcome occurs, while retrospective cohort studies use existing data
• Case-control studies compare the exposure history of cases (those with the outcome) to controls (those without the outcome)
  • Case-control studies are useful for rare outcomes but are prone to selection and recall bias
• Cross-sectional studies measure the exposure and outcome at a single point in time, making it difficult to establish temporal relationships
• Natural experiments take advantage of naturally occurring variations in exposure to estimate causal effects (comparing outcomes before and after a policy change)
• Quasi-experimental designs, such as difference-in-differences and regression discontinuity, use non-random assignment to exposure groups to estimate causal effects

Statistical Methods for Causal Analysis

• Propensity score methods estimate the probability of receiving the exposure based on observed covariates and use this score to balance the exposure groups
  • Propensity score matching pairs exposed and unexposed individuals with similar propensity scores
  • Propensity score stratification groups individuals into strata based on their propensity scores and estimates the causal effect within each stratum
• Instrumental variable analysis uses a variable that is associated with the exposure but not directly with the outcome to estimate the causal effect
  • Instrumental variables help address unmeasured confounding but require strong assumptions
• Difference-in-differences estimators compare the change in outcomes over time between exposed and unexposed groups, assuming that any confounding factors remain constant over time
• Regression discontinuity designs estimate the causal effect by comparing outcomes just above and below a threshold for exposure assignment
• Mediation analysis decomposes the total effect of an exposure on an outcome into direct and indirect effects through a mediator variable
• Sensitivity analyses assess the robustness of causal effect estimates to potential unmeasured confounding or violations of assumptions

Challenges and Limitations

• Unmeasured confounding can bias causal effect estimates if important confounders are not accounted for in the analysis
• Measurement error in the exposure or outcome can lead to biased estimates of the causal effect
• Violations of the positivity assumption occur when there are no unexposed individuals for some levels of the confounders, making it difficult to estimate the causal effect
• Generalizability of causal effect estimates may be limited if the study population is not representative of the target population
• Causal inference methods rely on strong assumptions, such as exchangeability, consistency, and positivity, which may not always hold in practice
• Ethical considerations may limit the use of certain study designs or the ability to randomize exposure
• Complexity of real-world causal relationships, such as time-varying exposures and outcomes, can make causal inference challenging

Real-World Applications

• Estimating the causal effect of a new drug on patient outcomes in clinical trials
• Evaluating the impact of public health interventions, such as smoking cessation programs, on population health
• Assessing the causal relationship between environmental exposures and health outcomes in environmental epidemiology
• Investigating the causal effects of social determinants of health, such as education and income, on health disparities
• Analyzing the effectiveness of policy interventions, such as taxes on sugar-sweetened beverages, on health outcomes
• Studying the causal impact of lifestyle factors, such as diet and physical activity, on chronic disease risk
• Evaluating the causal effects of healthcare delivery systems and quality improvement initiatives on patient outcomes
• Examining the causal relationships between occupational exposures and workers' health in occupational epidemiology