Incidence | Occurrence of the disease outcome over a specified time period. Incidence is generally assessed as a risk/proportion over a fixed time period (e.g., risk for 1-year mortality) or as a rate defined by persons and time (e.g., mortality rate per person-year). Incidence is often defined as first occurrence of the outcome of interest, a definition that requires prior absence of the outcome. | Etiologic studies are based on incidence of the outcome of interest rather than prevalence, because prevalence is a function of disease incidence and duration of disease. |

Prevalence | Proportion of persons with the exposure/outcome at a specific point in time. | Because prevalence is a function of the incidence and the mean duration of the disease, incidence is generally used to study etiology. |

Measures of association | Measures needed to compare outcomes across treatment groups. The main epidemiologic measures of association are ratio measures (risk ratio, incidence rate ratio, odds ratio, hazard ratio) and difference measures (risk difference, incidence rate difference). | Difference measures have some very specific advantages over ratio measures, including the possibility of calculating numbers needed to treat (or harm) and the fact that they provide a biologically more meaningful scale to assess heterogeneity.^{5} Ratio measures nevertheless abound in medical research. All measures of association should be accompanied by a measure of precision, e.g., a confidence interval. |

Confounding | Mixing of effects. The effect of the treatments is mixed with the effect of the underlying risk for the outcome being different in the treatment groups compared. | Confounding leads to biased treatment effect estimates unless controlled for by design (randomization, matching, restriction) or analysis (stratification, multivariable models). |

Selection bias | Distortion of treatment effect estimate as a result of procedures used to select subjects, and distortion of factors that influence study participation. | While procedures to select subjects usually lead to confounding that can be controlled for, factors affecting study participation cannot be controlled for. Factors affecting study participation are referred to as selection bias throughout this chapter to differentiate selection bias from confounding. |

Information bias | Distortion of treatment effect estimate as a result of measurement error in any variable used in a study; i.e., exposure, confounder, outcome. | Often measurement error is used for continuous variables, and misclassification for categorical variables. It is important to separate nondifferential from differential measurement error. Nondifferential measurement error in exposures and outcomes tends to bias treatment effect estimates towards the null (no effect); nondifferential measurement error in confounders leads to residual confounding (in any direction); differential measurement error leads to bias in any direction. |