FMED-505 Epidemiology: Measures of Association and Sources of Error

Vocabulary flashcards covering measures of association, risk metrics, observational study analyses, random error, systematic biases, confounding, effect modification, and screening biases from FMED-505 Weeks 5 and 6.

Risk

A measure that generally refers to the incidence rate of disease, but can also be expressed as cumulative incidence, a mortality rate, an odds, or a hazard.

Relative Risk (RR)

A unitless measure of the strength of association between an exposure and an outcome, calculated as $\frac{\text{Risk in the exposed}}{\text{Risk in non-exposed}}$, which can be computed as a rate ratio, cumulative incidence ratio, odds ratio, or hazard ratio.

Attributable Risk (risk difference)

An absolute measure, also called risk difference, calculated as $\text{Risk in exposed} - \text{Risk in non-exposed}$, indicating how much disease in the exposed group could be prevented by eliminating the exposure while retaining the units of risk.

Attributable Fraction

The proportion of disease occurrence among exposed individuals that is directly attributable to the exposure, calculated as $\frac{\text{Risk in exposed} - \text{Risk in non-exposed}}{\text{Risk in exposed}}$.

Population Attributable Risk (PAR)

A public health measure indicating how much disease in the total population could be prevented by eliminating an exposure, calculated as $\text{Risk in Total Population} - \text{Risk in Unexposed}$.

Population Attributable Fraction (PAF)

The proportion of disease cases in the total population that is attributable to an exposure, calculated as $\frac{\text{Risk in total population} - \text{Risk in non-exposed}}{\text{Risk in total population}}$.

Relative Measures

Epidemiological metrics, such as the risk ratio, rate ratio, odds ratio, and hazard ratio, that describe the strength of an association between an exposure and an outcome.

Absolute Measures

Epidemiological metrics, such as risk difference, attributable risk, and population attributable risk, that describe the actual amount or burden of disease attributed to an exposure.

Odds Ratio (OR)

A comparative measure of association defined as $\frac{\text{Odds of outcome in exposure group}}{\text{Odds of outcome in control group}} = \frac{a/b}{c/d} = \frac{ad}{bc}$, used in case-control studies where baseline group risk cannot be computed and serving as an accurate estimate of relative risk when disease is rare.

Prevalence Difference

A measure in cross-sectional study analysis defined as the arithmetic difference in disease prevalence between exposed and unexposed groups: $(\frac{a}{a+b}) - (\frac{c}{c+d})$.

Prevalence Ratio

A cross-sectional measure calculated as the proportion of individuals with disease in the exposed group divided by the proportion with disease in the unexposed group: $(\frac{a}{a+b}) / (\frac{c}{c+d})$.

Prevalence Odds Ratio

A measure of the strength of association in cross-sectional analyses defined as the odds of having the disease among the exposed divided by the odds of having the disease among the unexposed.

Random Error

A form of study error that occurs when, purely by chance, the estimate obtained from a sample differs from the true population value.

Law of Large Numbers

A statistical principle stating that as sample size or the number of experiments increases, the observed sample estimate converges closer to the true population value, reducing random error and improving precision.

Confidence Interval (CI)

A calculated range of values that conveys the precision, direction, and magnitude of an association, where a $95\%$ CI indicates that approximately $95\%$ of such intervals from repeated samples will contain the true population parameter.

Type I Error

A statistical error in hypothesis testing occurring when researchers conclude that an association exists when in reality it does not, with the probability of this occurrence designated as $\alpha$.

P-value

The probability generated by a statistical test of obtaining a difference as large as or larger than that observed if there were truly no difference between the groups, corresponding to $\alpha$.

Type II Error

A statistical error in hypothesis testing occurring when researchers fail to detect an association that truly exists in reality, with the probability of this error designated as $\beta$.

Statistical Power

The probability of successfully detecting a specified association when that association is truly present, calculated as $1 - \beta$ and conventionally designed for $80\%$ or $90\%$.

Bias

A systematic error in the design, conduct, or analysis of a study that results in an incorrect estimate of the exposure-outcome association, which cannot be fixed or reduced by increasing sample size.

Selection Bias

A systematic distortion of an association that arises when the procedures used to recruit or retain participants during follow-up result in systematic differences between those included and excluded.

Information Bias

A systematic distortion, also known as misclassification, resulting from incorrect measurement, recording, or categorization of exposure, outcome, or covariate data.

Differential Misclassification

A type of information bias where the probability of erroneous classification of exposure or outcome differs systematically between study groups, which can bias the observed association in either direction.

Non-differential Misclassification

A type of information bias where the probability of misclassification is identical across comparison groups, which generally biases the observed association toward the null ($RR = 1.0$).

Recall Bias

A form of information bias occurring when individuals with a specific outcome recall past exposures differently or more intensely than controls without the outcome.

Surveillance Bias

Also called detection bias, an information bias occurring when patients receiving a specific exposure or medication are monitored more frequently, leading to higher detection of conditions regardless of true incidence.

Social Desirability Bias

A response bias characterized by the tendency of research participants to answer questions in a manner that will be viewed favorably or approved by others.

Acquiescence Bias

A response bias where survey respondents demonstrate a tendency to agree with statements whenever they are uncertain or in doubt.

Telescoping Bias

A cognitive recall bias in which individuals perceive distant past events as having happened more recently, or recent events as having occurred further in the past.

Confounding

A distortion in an exposure-outcome association caused by a third factor that is associated with the exposure, independently associated with the outcome, and not an intermediate on the causal pathway.

Stratification

An analytical method used to evaluate confounding by calculating the exposure-outcome association separately across distinct categories (strata) of a potential confounding variable.

Effect Modification

Also called interaction, an epidemiological phenomenon occurring when the true magnitude of an exposure-outcome association differs according to the level or category of another variable.

Precision

The degree to which repeated measurements or estimates show consistent results, reflecting the absence of random error and indicated by narrower confidence interval widths.

Internal Validity

The degree to which the results and inferences of an investigation accurately represent the true exposure-outcome relationship within the specific population being studied, free from bias and confounding.

External Validity

Also known as generalizability, the extent to which the empirical conclusions of a study can be reasonably applied to populations, settings, and patients outside the study group.

Lead Time

The duration of time between the early detection of a disease via screening and the time when the disease would have presented and been diagnosed clinically.

Lead Time Bias

An artificial inflation of survival time occurring when screening detects disease earlier without prolonging actual life, making survival from diagnosis appear longer simply because diagnosis occurred sooner.

Length Time Bias

A screening bias caused by the preferential detection of slowly progressive, indolent conditions over rapidly progressing, fatal ones because slower cases remain in the preclinical detectable phase for longer.

Receiver Operating Characteristic (ROC) Curve

A graphical representation plotting test sensitivity against $1 - \text{specificity}$ across all potential cut-off values for a continuous diagnostic test, where a larger area under the curve signifies superior accuracy.

Systematic Review

A scientific methodology that comprehensively locates, quality-evaluates, synthesizes, and interprets all relevant published evidence on a defined topic to generate an impartial, balanced summary.

Meta-analysis

A statistical procedure combining quantitative data from multiple independent studies sharing similar research designs to calculate a single summary effect estimate.

Efficacy

An evaluative measure of how well a therapeutic intervention or treatment performs under ideal, controlled clinical experimental settings.

Effectiveness

An evaluative measure assessing how successfully an intervention or clinical treatment works under routine, real-world conditions.

Efficiency

An evaluation of a treatment's cost-benefit ratio that considers economic resource expenditures in relation to health outcomes and intervention risks.

Intention to Treat (ITT) Analysis

An analytical approach in trials where participants are analyzed according to their assigned randomized groups regardless of compliance or drop-out, thereby preserving the benefits of randomization against confounding.

Per Protocol Analysis

An analytical approach evaluating only participants who fully complied with and completed their assigned trial regimen, which removes randomization protection and introduces selection bias.

Number Needed to Treat (NNT)

An absolute clinical measure estimating the average number of patients that must receive a specific therapeutic intervention to prevent one additional adverse outcome.