Portal de Periódicos da CAPES

Causation, bias and confounding: a hitchhiker's guide to the epidemiological galaxy Part 2. Principles of causality in epidemiological research: confounding, effect modification and strength of association

2008; BMJ; Volume: 34; Issue: 3 Linguagem: Inglês

10.1783/147118908784734873

2045-2098

Samuel Shapiro,

Global Cancer Incidence and Screening

In Part 1 of this series the following principles of causation were considered: time order, specification of the study base, and specificity (bias due to random misclassification and bias due to systematic misclassification). 1 Part 2 will consider: 2a: Confounding 2b: Effect modification 2c: Strength of association.Part 3 concludes the consideration of causal principles and will discuss statistical stability, dose-and durationresponse effects, internal consistency, external consistency, analogy and biological plausibility. 22a: Confounding Confounding exists when a risk factor other than the exposure under study is associated, independently, both with the exposure and with the outcome.Confidence in causality is increased if an association can be judged to be reasonably unconfounded.When present, confounding is the same in a follow-up or case-control study carried out in the same study base.To the degree that confounding can be precisely measured, its effect can be removed or minimised.To illustrate, consider the effect of age on the association of oral contraceptive use with venous thromboembolism (VTE).With advancing age, oral contraceptive use declines, and the risk of VTE increases.If in any given study the compared groups are of different ages, that difference will confound the association.For example, in a case-control study, if the cases are older than the controls, and if age is not allowed for, the magnitude of the association will be underestimated.Age is an example of a risk factor that can be precisely measured, and its potentially confounding effect can therefore be eliminated ('controlled') by appropriate adjustment: for example, by setting age limits ('exclusion'); by comparing women of similar age ('comparisons within strata of age'); by making comparisons after applying percentages from strata of age to a single standard age distribution ('standardisation'); or by multivariate analysis (broadly, the more or less simultaneous statistical adjustment for the confounding effects of several factors, including age).In case-control studies age can also be allowed for by selecting controls matched to be of the same age as each individual case Causation, bias and confounding: a hitchhiker's guide to the epidemiological galaxy Part 2. Principles of causality in epidemiological research: confounding, effect modification and strength of association