Bias and Confounders Flashcards
random (non-differential) error
use of invalid outcome measure that equally misclassifies cases and controls
systematic (differential) erros
use of invalid measure that misclassifies cases in one direction and controls in another –> leads to inaccurate results
Bias is part of what error?
systematic/differential
Chance
cause of random error and leads to imprecise results
Selection Bias
study and control groups not representative of population and this factor has potential to affect results
Types of Selection Bias
sampling, volunteer, healthy worker, hospital, prevalence-incidence, surveillance, loss to follow up
healthy worker effect
those working generally healthier than rest of population and potentially less likely to participate in a study
hospital bias
occurs when hospital cases are compared to the rest of the population
Neyman’s Prevalence-Incidence Bias
prevalent cases used to study exposure-disease relationships
(think about the bathtub, how long they live)
Surveillance Bias
exposed individuals undergo a more thorough evaluation than non-exposed individuals
Ways to help control selection bias?
careful study design, reduce loss to follow up, “equalize” bias between groups
4 Types of Information Bias
- Exposure ID bias
- Outcome ID bias
- Temporal bias
- Lead Time bias
Exposure ID Bias
includes recall bias, Hawthorne (act differently if they’re in the study vs not), interviewer bias
Outcome ID Bias
overreporting or seeking exposures
Temporal Bias
when you can’t figure out what came first - risk factor or disease (so do a prospective study to avoid this)
Lead Time Bias
early diagnosis can happen since we’re looking so closely at our studies
Result of Information Bias?
misclassification of individuals
Ways to Prevent Misclassification
blinding, questionnaire (instead of interview), accuracy checks, CONSISTENCY
Publication Bias
bias in reporting study results (authors tend to only submit positive results)
confounding variable
causally associated with the outcome, could be related to the exposure, but not part of a causal pathway between exposure and outcome
Stratified Analysis
separate groups based on potential confounding variable and see if the statistical analysis holds up
3 Ways to Control for Confounders
- Restriction (keep potential confounders out)
- Random allocation (to even out confounders)
- Matching
- Model-fitting regression techniques in analysis
