Epi, Lec 6: Chance And Bias

Front 1

What may explain the differences for an observed association between an exposure and an outcome

Back 1

errors in design, conduct and analysis (chance and bias) – even if errors do not seem to be an obvious example for the observed effect, it is still necessary to assess likelihood that the association is causal

Front 2

Random error

Back 2

chance; sampling distribution; basis of statistics (e.g., confidence intervals); magnitude of random error will depend on: sample size (the larger the sample size the closer the estimate will be to the true population); Random error shown in CI, P-values of risk/rate/odds ratios and differences

Front 3

p-value

Back 3

tells us the probability of obtaining a relative risk (RR, measure) as high or higher than the value observed in the study assuming that the null hypothesis of no association is true

Front 4

Interpretation of p-values and CI

Back 4

avoid multiple testing; Cis are more useful than p-values since they give range of values that are consistent with the true unknown population value; they tell us nothing about bias or confounding

Front 5

Sampling distribution

Back 5

the concept that different random samples from the same population will give different results; larger the sample the closer the value will be to the true population

Front 6

Systematic error

Back 6

bias; non-random error; leads to incorrect estimate of the effect of an exposure on the development of a disease or outcome of interest > observed effect will be either above or below the true value; NOT dependent of sample size; Statistics CANNOT HELP us ‘solve’ the problem of bias

Front 7

Bias (2 major groups)

Back 7

deviation from the truth; 1. Selection bias; 2. Information bias

Front 8

Selection bias

Back 8

occurs when there is a SYSTEMATIC difference between the characteristics of the people who take part in a study and the characteristics and the characteristics of people who were eligible but did not take part, or who dropped out; problem in all epidemiological designs

Front 9

Selection bias (case-control)

Back 9

controls are recruited to provide an estimate of the exposure prevalence in the general population from which the cases come > sometimes it is not possible to define the population from which the cases/controls arise

Front 10

Selection bias (cohort)

Back 10

selection bias can arise from 1) non-response (e.g., missing data), 2) refusal to participate and 3) LTFU – missing data is of particular concern when related to the outcome or exposure; 4) choice of exposure group can be a source of selection bias (e.g., healthy worker effect)

Front 11

Selection bias (RCT)

Back 11

withdraws or LTFU, esp if related to exposure status or outcome status

Front 12

Information bias (or measurement bias)

Back 12

occurs when classifications or measurements of disease or exposure are inaccurate – can be introduced by the observer, participant or instrument aka misclassification > a) non-differential, b) differential (non-random)

Front 13

Non-differential misclassification

Back 13

occurs when an exposure of disease classification is incorrect for EQUAL proportions of subjects in the compared groups. Errors in the categorization of disease that are unrelated to the individual’s exposure status, or misclassification of exposure unrelated to the individual’s disease status > IT IS RANDOM > makes the two groups seem more alike and leads to UNDERESTIMATE the strength of the association (move it towards the association/estimate of effect toward the null hypothesis). ‘Dilute.’ Bias is a GREATER concern in interpreting studies that seem to indicate the absence of an effect.

Front 14

Differential misclassification

Back 14

1) error in classification are related to exposure; 2) errors in classification are related to disease status. This can bias estimates of association in EITHER direction. Types: responder, observer

Front 15

Responder bias

Back 15

occurs when the way in which study participants supply information about exposure differs according to outcome status, or the way in which study subjects supply information about outcomes differs according to exposure status. RECALL BIAS. Can be minimized: keeping study member unaware of the hypothesis and by ensuring they have similar incentives to provide accurate information.

Front 16

Recall bias

Back 16

may be both non-differential and differential bias depends on whether or not recall bias equally affects exposed and unexposed (non-differential) or just one or the other.

Front 17

Observer bias

Back 17

observers who know the exposure status of an individual may be consciously or unconsciously predisposed to assess the outcome variables according to the hypothesis under study; Minimize by: keep exposure status, or the disease status conceals from the observers, training, objective measure

Front 18

Due to the differences in groups?

Back 18

Confounding

Front 19

True rate ratio

Back 19

RR= rate among those exposed to X/rate among those not exposed to X