3. Bias Flashcards
Directed Acyclic Graphs (DAGs)
- Directed path
- Backdoor or undirected paths (are bias)
the presence of a common cause or backdoor path in a DAG identifies the presence of confounding.
In DAG terms, adjusting for confounding by means of restriction, stratification or multivariable analysis is called conditioning. (Block)
- A ‘collider’ is a common effect; a factor on which two arrowheads collide. A collider blocks a path.
A collider that has been conditioned on no longer blocks a path; conditioning on a collider could therefore introduce a form of selection bias and should be done with caution.
- Any path that contains non-colliders is open, unless a non-collider has been conditioned on, then it is blocked.
- A blocked path prevents the statistical association!
Key Facets of Exposure
1) Duration = length of exposure
2) Frequency = how often
3) Intensity = concentration
Study Validity (3 categories)
When a measure of association equals to the causal effect, then our estimate is considered valid.
Measurement validity
Internal study validity: refers to inferences made to the source population.
External study validity: refers to inferences made to people outside the population
What is measurement validity and three categories
•degree a measurement measure what it purports to measure
Content validity
Construct validity
Criterion validity
bias is the systematic error, random error is the smapling error. how to reduce.
–Increase sample size
–Improve sampling procedures
–Reduce measurement variability by using strict measurement protocols, better instrumentation or average of multiple measures
–Use more statistically efficient analytic methods
2 types of bias
–Selection bias
•Selection Bias occurs when individuals have a different probability of being included in the study sample according to the relevant study characteristics: the exposure and outcome of interest.
–Information bias
•Information Bias is being erroneously placed in either the wrong exposure or outcome category, leading to misclassification.
2 types of information bias
•Exposure Identification Bias
–An imperfect definition of the level of exposure or errors at the data collection phase may lead to bias. (exposure measurement error if continuous; exposure misclassification, if discrete)
•Outcome Identification Bias
–An imperfect definition of the outcome or errors at the data collection phase may lead to bias (outcome misclassification)
Reasons for Measurement Error (4)
•Errors in the design of the instrument.
e.g. didn’t cover all sources of exposure
•Poor execution of the study protocol
e.g. data collectors didn’t follow protocol in the same manner for all subjects
•Limitations due subject characteristics
e.g. recall error; over/under report
•Errors during data capture and analysis
e.g. data entry errors
2 types of misclassification
- Non-differential misclassification
Occurs when the degree of misclassification of exposure is independent of case non-case/control status (or vice versa)
The expected direction of bias is usually towards the null
- Differential misclassification
Occurs when the degree of misclassification differs between the groups being compared
The direction of the bias can be either towards or away from the null
2 sources of exposure misclassification and how to prevent
- Recall bias (esp in case-control)
Occurs when “recall” of past exposures is dependent or influenced by case or control status
Prevent by:
Verification of exposure information (hospital records, other)
Use diseased controls (case-control study) - rumination
Use objective markers of exposure, when possible
Use of nested case-control studies or cohort studies
- Observer Bias (Interviewer bias)
Can happen when case or control status is known to the interviewers. Can happen if interviewers deviate from the protocol
Prevent by:
- Strict protocols
- Validity studies using independent sources (e.g., medical charts)
- Blinding interviewer as to case/control status
- Phantom cases and controls
2 sources of disease misclassification and how to prevent
- Observer bias
Occurs when knowledge of the exposure may influence if the outcome is determined to be present
Prevent by:
- Strict protocols
- Blinding observer
- Respondent Bias
Information on outcome obtained by the participant response is biased
Prevent by:
- Confirmation (i.e., medical records confirmation)
- Standardized questionnaires
- Multiple questions to verify
When misclassification parameters (e.g., sensitivity and specificity) are NOT known…SENSITIVITY ANALYSES can be used to obtain a range of plausible “corrected” estimates under different assumptions on the levels of misclassification
Temporal Bias ( a type of information bias) and how to prevent
•When the inference about the temporal sequence of cause and effect is erroneous (reverse causality)\
Prevent by:
- Improve information on temporality from questionnaires (date of first exposure or date of disease onset)
- Attempt to assess if exposure a consequence of undiagnosed disease
Bias Related to the Evaluation of Screening Interventions
- Selection bias
- Incidence-Prevalence bias
- Length bias
- Lead time bias
Key element to selection bias
the relationship between the exposure and disease is different for those who participated from those who are theoretically eligible for the study, but did not participate
The probability of being included in the study should be different according to Both exposure and outcome
Conditioning on a common effect of disease and exposure will result in a selection bias. Interpret the OC and Cancer hypothesis (detection bias because exposure will lead to check)
The association between OC exposure and cancer may not be a causal relationship, but through the backdoor pathway.
With the introduction of “Vaginal bleeding”, there is a collider that introduced selection bias.
Loss to follow-up bias DAG
HIV therapy will be statistically associated with AIDS among those not lost to follow-up because both condition on “Loss to follow up”, one through a directed path, the other from the backdoor.
