Bias Flashcards
Is the study valid?
- -if the result of a study remains close to the truth then the study is said to have internal validity
- if the results of the study are applicable to the clinician seeking the evidence, then the study is said to have external validity
Validity
- may be compromised by three factors
1. random error (chance)
2. systematic error (bias)
3. confounding
Random error
- due to the play of chance
- all studies are prone to random error
- the probability of this is assessed using statistical measures such as p-values and confidence intervals
- random error can be reduced by repeating/replicating the study elsewhere
Systematic error
- bias
- error in the way we select our patients, measure our outcomes or analyse our data
- bias may result in inaccurate results
- this is predictable and repetition doesnt help
Negative bias
-bias can be negative and yield estimates closer to the null value ( no difference between two groups)
Positive bias
-higher magnitudes of estimates than the true ones
When can bias occur?
- Selection of participants- selection of bias
- measurement of variables- measurement bias
- analysis of data- analysis bias
Selection bias
- selected groups in a study differ in important factors other than the study variables
- adequate randomisation helps
- Berkson bias
- Neyman bias
- Response bias
- Unmasking bias
- lead-time bias
- referral bias
- diagnostic purity bias
- membership bias
Berkson bias
- admission rate bias
- results from a difference in the rates of admission of cases and controls due to influence of exposure
- e.g strength and ability to be well enough for trial
Neyman Bias
- incidence-prevalence bias
- one must look for an association between a risk factor and incidence not prevalence
Response bias
-persons who respond to an invitation to participate in a study differ systemically from those who do not respond
Unmasking bias
-said to exist when a risk factor unmasks rather than causes an event which it is suspected to cause
Lead-time bias
- lead time is defined as the difference in time between the date of diagnosis with screening and the date of diagnosis without screening
- if the lead time is not accounted for then the survival time should not be compared to an unscreened control group of patients
Referral bias
-refers to variation in concentration of rare exposures or diseases between patients in primary and secondary care
Diagnostic purity bias
-refers to the exclusion of comorbidities resulting in a non-representative sample
Membership bias
-refers to case identification using members of patient’s organisations leading to systematic differences and non-representativeness
e,g local self-help group
Measurement Bias
- when data are not collected in a uniform fashion
- e.g when cases are interviewed in person while controls are interviewed over the phone
- this type of bias is minimised by blinding
Recall bias
- type of measurement bias
- subjects often recall risk factors differently depending on their disease status
- case control studies are particularly vulnerable to this type of bias
Reporting bias
-results when a larger percentage of either case or control subjects are reluctant to report an exposure due to attitudes, perceptions or other concerns
Observer bias
-can occur whenever a researcher either knowingly or unknowingly evaluates a variable depending on the status of the individual under study
Surveillance bias
-disease may be better ascertained in a monitored population than in the general population
Work up bias
- aka verification bias
- during assessment of validity of a diagnostic test, the execution of the gold standard test may be influenced by the results of the assessed new instrument
- reference test may be less frequently performed when the test result is negative
Misclassification bias
- in extreme cases measurement bias may lead to misclassification
- cases may be misclassified as controls or ‘exposed group’ may be misclassified as ‘non-exposed’
- this misclassification amounts to bias only if it is differential ie. one sided
- errors in measurement instruments may lead to non-dfferential misclassification (both sides are affected equally)
Desirability bias
-patients may choose socially desirable answers to provide during data collection, distorting the true picture- reporting bias
Hawthorne effect
- refers to the observed respondents minimising perceived deviation form the norm
- occurs especially in cross-sectional surveys using questionnaires
Analysis bias
- said to occur when participants change group (contamination bias) or are lost to follow up (attrition bias) during the study
- this can be minimised by doing ‘intention to treat analysis’.
