Bias and Confounding

Non-Response Bias

-poor response in postal questionnaires -bias introduced if response is related to exposure and outcome

How to deal with confounding (2)

1) Design - randomization, restriction, matching 2) Analysis - stratified, multivariate regression

To help assess the role of chance, look at (2)

1) confidence interval around measure of effect 2) p-value

How does misclassification occur? (3)

1) error in instrument 2) subjects give incorrect info 3) researcher records info incorrectly

Steps to get the study right in order to fix bias (6)

1) select an appropriate study population 2) maximize response rates 3) minimize loss to follow-up 4) use objective rather than self-reported measures of exposure/outcome 5) follow standardized procedures in data collection 6) blind interviewer

Selection Bias is an issue particularly in what kind of study?

Case-control studies

What studies are Loss to Follow-up most relevant in?

Cohort studies

Solution to Chance

adequate sample size

Stratification

analyze data for each level (strata) of the confounder

Sampling Error

any finding could be a chance finding

What makes bias unlike confounding?

cannot correct for bias

Recall Bias

cases are more highly motivated to recall use of exposure than controls

Factors on the __________________ pathway should not be considered as confounders

causal

Selection Bias

criteria for inclusion of cases or controls is related to exposure of interest

Reverse causation

did the exposure cause the disease, or the disease cause the exposure?

Loss to Follow-up

drop-out related to exposure/outcome

Differential Misclassification

error in classification of disease status that is dependent upon exposure status or vice versa -can bias estimates of association in either direction

Non-differential (random) Misclassification

error in exposure that occurs equally for diseased and non-diseased -leads to measure of effect that is closer to unity (no effect)

What causes confounding?

estimate of exposure-outcome association is distorted by some other exposure (confounder) which is associated with the outcome and with exposure of interest

Matching

in case-control studies, can match a control to each case based on the potential cofounders -chose a control of the same age and gender of the case -must use special statistics to analyze data

The confounding factor must be associated with the disease how?

independently - not relying on any other factor

Observer/Interviewer Bias

interviewer may probe cases about their past exposure more than controls

What is confounding?

see an association between exposure and disease, but *not true association*

Types of bias (2)

selection bias information bias

Solution to Bias

study design

Solution to Reverse Causation

study design

Solution to Confounding

study design and/or analysis

Restriction

study one level of the confounding variable only -may make results less generalized/relevant to actual population

Information Bias

study subjects are misclassified either according to their disease status, their exposure status, or both

What is bias?

systematic error in an estimate arising from problems in the study design/execution

Reporting bias

those with disease are more likely to report exposure than healthy controls

Healthy worker effect

workers are generally healthier than general population -if comparison group is general population, may appear that occupation "protects" against disease