Fundamentals of Epidemiology: Synthesis, Bias, Confounding, Effect Modification, and Causal Inference

Vocabulary practice flashcards covering systematic reviews, meta-analysis, measurement properties, information bias, selection bias, confounding, directed acyclic graphs (DAGs), effect modification, and causal inference frameworks from EPI 0850.

Systematic Review

A review of existing literature conducted using a pre-specified, reproducible, and documented methodology for identifying, selecting, appraising, and synthesizing all relevant evidence on a focused research question.

PICO/PECO Framework

A standardized framework for formulating an answerable epidemiological research question, defining Population, Intervention or Exposure, Comparison Group, and Outcome.

Meta-Analysis

A statistical technique for combining the quantitative results of multiple studies addressing the same or sufficiently similar question into a single pooled effect estimate with an associated measure of precision.

Forest Plot

A graphical display allowing visual assessment of each contributing study's point estimate, effect direction, and confidence interval, as well as the degree of consistency across studies and the pooled summary estimate.

Clinical Heterogeneity

Variability in effect estimates across studies arising from differences in participant populations, exposure or intervention definitions, or outcome definitions.

Methodological Heterogeneity

Variability in effect estimates across studies arising from differences in study design features or study-level risk of bias.

Statistical Heterogeneity

The observed variability in effect estimates across studies beyond what would be expected by chance alone, which can be formally quantified and stems from clinical or methodological differences.

Publication Bias

The systematic error in the evidence base caused by the tendency for studies with statistically significant, 'positive,' or novel findings to be published more frequently and rapidly than studies with null or unfavorable results.

Content Validity

The degree to which a measurement instrument, on its surface, appears to measure the target concept and adequately covers the entire scope and domain of that concept.

Criterion Validity

The degree to which a measurement instrument corresponds to an established reference standard, evaluated as concurrent validity (measured simultaneously) or predictive validity (measured at a later time).

Construct Validity

The degree to which a measurement instrument conforms to the theoretical concept it is intended to measure, demonstrated through convergent validity (correlation with related constructs) and divergent or discriminant validity (lack of correlation with unrelated constructs).

Test-Retest Reliability

The consistency of measurements obtained when the exact same instrument is administered to the same subjects on two or more separate occasions, assuming the underlying condition remains unchanged.

Kappa Statistic

A statistical measure of agreement beyond chance used to assess inter-rater reliability between different observers or intra-rater reliability for the same observer re-evaluating the same subjects.

Internal Consistency

The degree to which multiple items in a multi-item questionnaire or scale completed at the same time correlate with each other to reflect the same underlying construct, commonly quantified by Cronbach's alpha.

Sensitivity

The intrinsic ability of a test to correctly identify individuals who truly have the condition according to a reference standard: $\text{Sensitivity} = \frac{\text{True Positive}}{\text{True Positive} + \text{False Negative}}$.

Specificity

The intrinsic ability of a test to correctly identify individuals who do not have the condition according to a reference standard: $\text{Specificity} = \frac{\text{True Negative}}{\text{True Negative} + \text{False Positive}}$.

Positive Predictive Value (PPV)

The proportion of individuals testing positive who truly have the condition: $\text{Positive PV} = \frac{\text{True Positive}}{\text{True Positive} + \text{False Positive}}$, which increases as population prevalence increases.

Negative Predictive Value (NPV)

The proportion of individuals testing negative who truly do not have the condition: $\text{Negative PV} = \frac{\text{True Negative}}{\text{True Negative} + \text{False Negative}}$, which decreases as population prevalence increases.

Information Bias

Systematic error in the estimation of an epidemiological measure arising from how exposure, outcome, or covariate information is measured, collected, or classified.

Non-Differential Misclassification

Classification error occurring when study subjects are assigned to an incorrect category of exposure, outcome, or covariate status, but the probability of error is identical across comparison groups.

Differential Misclassification

Classification error occurring when study subjects are assigned to an incorrect category of exposure, outcome, or covariate status, and the probability of misclassification differs between comparison groups.

Recall Bias

A form of differential exposure misclassification in case-control studies where cases recall or report past exposures with different accuracy or sensitivity than controls because of awareness of their disease diagnosis.

Interviewer Bias

Differential measurement validity occurring when an assessor's questioning, probing, or recording technique differs systematically based on knowledge of a subject's exposure or outcome status.

Social Desirability Bias

A type of differential exposure misclassification occurring when participants systematically underreport stigmatized behaviors or overreport approved behaviors due to perceived stigma or social norms.

Detection Bias (Differential Detection)

A form of information bias in prospective studies where the frequency, intensity, or sensitivity of outcome surveillance differs systematically between exposure groups.

Digit Preference

Systematic measurement error caused by rounding or heaping continuous numerical readings at particular preferred digits (e.g., blood pressure values disproportionately ending in zero).

Instrument Drift

A progressive, gradual shift in a measurement instrument's calibration or baseline accuracy over the course of a long-term epidemiological investigation.

Diagnostic Suspicion

A bias that occurs when a healthcare provider's knowledge of a participant's exposure status leads them to examine that patient more intensively for suspected associated outcomes.

Selection Bias

Systematic error occurring when the association between exposure and outcome in the analyzed study population differs from that in the target population because eligibility, inclusion, or retention is jointly related to exposure and outcome.

Berkson's Bias

A hospital-based selection bias arising when hospitalized cases and controls have different probabilities of admission related to both their exposure status and their health condition.

Healthy Worker Effect

A form of selection bias in occupational epidemiology wherein employed workers exhibit lower morbidity and mortality than the general population because only relatively healthy individuals can enter and remain in the active workforce.

Non-Response Bias

Selection bias that arises when the decision to participate in a study differs systematically according to both exposure status and outcome status.

Volunteer Bias

Selection bias occurring when volunteers for a study systematically differ from non-volunteers in characteristics such as health consciousness and underlying risk in ways linked to both exposure and outcome.

Inverse Probability Weighting

An analytical method to adjust for selection bias or attrition where each analyzed individual is weighted by the inverse of their estimated probability of selection based on measured covariates.

Confounding

Distortion of an exposure-outcome association by a third variable that is associated with the exposure in the source population, is an independent risk factor for the outcome, and is not an intermediate on the causal pathway.

Directed Acyclic Graph (DAG)

A visual diagram of assumed causal structures consisting of variables (nodes) and directional causal arrows with no closed causal cycles, used to identify sources of confounding and guide covariate adjustment.

Fork (DAG Structure)

A causal structure where a single variable is a common cause of two others ($\text{Exposure} \leftarrow Z \rightarrow \text{Outcome}$), representing a confounder that creates a non-causal open path requiring adjustment.

Chain (DAG Structure)

A directed path where an exposure affects an intermediate variable which in turn affects the outcome ($\text{Exposure} \rightarrow Z \rightarrow \text{Outcome}$), representing a mediator on the causal pathway that should not be adjusted for when estimating total causal effect.

Collider (DAG Structure)

A variable on a path that receives incoming arrows from two or more variables ($\text{Exposure} \rightarrow Z \leftarrow \text{Outcome}$), which naturally blocks the path but opens an artificial non-causal association if conditioned upon.

Backdoor Path

Any non-causal path connecting exposure and outcome that begins with an arrow pointing into the exposure, which must be blocked to obtain an unbiased causal effect estimate.

Restriction

A design-stage technique to prevent confounding by limiting study eligibility to a single uniform level or narrow range of a potential confounder.

Mantel-Haenszel Method

A stratified analytical method used to calculate a single summary measure of association (such as an adjusted odds ratio or rate ratio) across strata of a confounding factor by weighting each stratum by its precision.

Propensity Score

A calculated scalar summarizing an individual's estimated probability of receiving an exposure or treatment conditional on a set of measured baseline confounders.

Residual Confounding

Confounding distortion that persists after analytical adjustment because a confounder was categorized too broadly, measured with error, or had poorly specified relationships.

E-Value

A quantitative sensitivity analysis metric defining the minimum association strength that an unmeasured confounder must have with both exposure and outcome to fully explain away an observed effect estimate.

Effect Modification (Statistical Interaction)

A phenomenon where the magnitude or direction of an exposure-disease association genuinely differs across subgroups defined by a third variable.

Additive Scale Interaction

Effect modification characterized by departure from additivity, in which the absolute excess risk (risk difference or rate difference) attributable to the exposure differs across strata of an effect modifier.

Multiplicative Scale Interaction

Effect modification characterized by departure from multiplicativity, in which relative measures of association (such as risk ratios, rate ratios, or odds ratios) differ across strata of an effect modifier.

Quantitative Interaction

A pattern of effect modification where the magnitude of the exposure-outcome association varies across subgroups, but the direction of effect remains consistent across all strata.

Qualitative Interaction

A pattern of effect modification where the direction of effect reverses across subgroups (e.g., the exposure is protective in one stratum but harmful in another).

Potential Outcomes Framework

A conceptual framework defining individual causal effects by contrasting the hypothetical outcome that would occur if an individual were exposed ($Y^1$) with the outcome that would occur if they were unexposed ($Y^0$).

Fundamental Problem of Causal Inference

The methodological limitation that individual causal effects are unobservable because only one potential outcome can ever be observed for any single person at a specific time.

Exchangeability

The condition under which the unexposed group's observed outcome experience accurately represents what the exposed group's outcome experience would have been had they been unexposed.

Average Causal Effect

The population-level effect comparing the average outcome if every individual in the population were exposed to the average outcome if every individual were unexposed.

Necessary Cause

A causal factor that must be present for a specific disease or outcome to occur.

Sufficient Cause

A complete causal mechanism (represented as a full causal pie) consisting of a set of minimal conditions and events that inevitably produces disease.

Causal Pie Model (Sufficient-Component Cause Model)

A conceptual model representing multi-causality where each whole pie is a sufficient cause mechanism and each individual slice is a component cause necessary to complete that mechanism.

Sir Austin Bradford Hill's Viewpoints

A set of nine viewpoints proposed in $1965$ (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy) to structure judgment when evaluating if an association is causal.

Target Trial Emulation

A methodology in which investigators first specify the protocol of a hypothetical randomized controlled trial that would answer their question, and then design and analyze an observational study to emulate that protocol as closely as possible.

Triangulation

An approach to causal inference that evaluates evidence across multiple distinct study designs or analytical methods, each possessing different, unrelated key sources of bias.