Information bias Flashcards
Synonym for information bias?
Measurement error
What is information bias?
A systematic error that arises due to the collection of erroneous information from study participants
Regarding what types of variables is information bias relevant?
- Exposure
- Outcome
- Covariate
When is misclassification not differential?
When errors on the variable do not depend on the value of other variables
What is the effect of non-differential misclassification of a dichotomous exposure/outcome?
Bias towards the null
The degree of non-differential misclassification bias is a function of:
- Sensitivity
- Specificity
- Prevalence of exposure
About non-differential misclassification bias: when is specificity critical?
When exposure is rare
About non-differential misclassification bias: when is sensitivity critical?
When exposure is common
About non-differential misclassification bias: what is critical, at fixed values of specificity and sensitivity?
The prevalence of the exposure
What is the effect of non-differential misclassification of a multi-categorical exposure/outcome?
Bias in either direction (toward or away from the null)
When does non-differential misclassification NOT result in bias?
- When there is perfect specificity, for RR estimates (e.g., odds ratio)
When do dependent errors in the measurement of the treatment and outcome occur?
When the same source is used to gather information on the exposure and outcome
What is the result of dependent non-differential misclassification?
Bias in either direction
3 exceptions to “non-diff misclassification biases towards the null rule”
- Multi-categorical exposure (away or towards)
- Perfect specificity (no bias for RR estimates)
- Dependent non-differential misclassification (away or towards)
The effect of differential misclassification is a function of:
The sensitivity, the specificity and the prevalence of exposure between cases and control
3 types of information bias in the collection of the exposure information
- Recall bias
- Interviewer bias
- Changes in exposure status over time
2 types of information bias in the collection of outcome information
- Observer/detection/surveillance bias
2. Reporting bias
Difference between poor recall and recall bias?
Poor recall is non-differential
In what type of study design is recall bias especially an issue?
Case-control studies
What is recall bias?
Differential errors in recall of past exposures based on disease status
What is interviewer bias and when does it occur?
Bias in the ascertainment of the exposure due to the interviewer not being blind to the disease status (e.g., prompt or emphasize certain questions)
What is the observer bias?
The collection of different qualities of information about the outcome for exposed vs unexposed, affected by knowledge of the exposure
What is reporting bias?
When the information on the outcome is obtained by participant response, especially for events where it is hard to have objective information (may overreport if they know the study hypothesis or underreport if socially undesirable)
What is the link between information bias and confounding?
Information bias can cause residual confounding, such as when non-differential misclassification of a confounding results in imperfect adjustment
How can we prevent recall bias?
- Frame questions to aid accurate recall (design level)
- Use different control group (e.g., different disease)
- Objective/administrative information
- Nested case-control studies (provides longitudinal exposure information)
How can we prevent interviewer bias?
- Standardization
- Blinding interviewers to the outcome
- Validated instruments
- Quality control of interviews (must be systematic)
- Diagnose interviewer bias with “phantom” control
What is a phantom control?
Manipulated controls, e.g. they are mistaken for a case, so you can compare if they’re assessed like the other controls or like the other cases
How do we prevent observer bias?
- Blind to exposure status
- Multiple observers
How do we prevent reporting bias?
- Confirm self-report with administrative data
- Use of validated instruments and computerized surveys for sensitive information
