Biases Flashcards
Selection Bias
Error due to systematic differences in characteristics between those selected:
1. INTO THE STUDY & NOT
2. INTO ONE ARM & NOT THE OTHER
This means these groups cannot be validly compared.
Types of Selection Bias (6)
- Sampling
- Allocation
- Spectrum
- Participation
- Attrition (loss to follow-up)
- Healthy Worker Effect
Sampling Bias
When individuals don’t have EQUAL chance to be selected into the STUDY because of characteristics (interchangeable with selection bias)
Allocation Bias
RCTs - selection for treatment arms affected by characteristics
(Avoided by randomisation & ITTA to sustain that randomisation)
Spectrum Bias
DTA - systematic error when specific groups are inappropriately excluded (‘difficult to diagnose’) making the test appear more accurate than it actually is in practice.
Participation Bias
See also:response/responder bias
Systematic error due to different characteristics btw those who CHOOSE whether to participate/in which arm they CHOOSE to participate in
Attrition Bias (LTF)
Systematic error due to differences in characteristics of those who were LTF vs those who COMPLETED the study.
Even though this may be representative of characteristics of non-compliance in practice, the ptpts must be followed up or representation of outcomes will be affected.
Healthy Worker Effect
When the study population is reaped from a working population (ie - nurses in a trust) this poses a defect in the representation of morbidity and mortality as workers are more likely to not be severely ill or chronically disabled, and hence exhibit lower overall death rates.
Measurement Bias
See also: information/observation bias
Systematic error in the way data is gathered between groups = differential quality/accuracy btw groups = difficult to compare them
NB: source either measured or measurer
Types of Measurement Bias (10)
- Interviewer
- Recall
- Recording
- Social Acceptability
- Detection
- Performance
- Reporting
- Verification
- Review
- Reflexive stance of researcher
Interviewer Bias
Error when the interviewer consciously/subconsciously gathers DTA differently between cases & controls.
Prevented by BLINDING.
Recall Bias
Systematic error when ptpts have different accuracy/completeness of recall of exposure/event depending on case/control status.
Recording Bias
Cases recorded in more detail than controls (medical notes; questionnaires)
Social Acceptability Bias
Selective suppression/revealing of relevant info so the most socially acceptable answer is given. (Compliance, smoking, illicit drug use etc)
To avoid: use anonymised questionnaire rather than interviewer
Detection Bias
Systematic differences btw groups wrt how outcomes are determined (method of assessment/diagnosis/verification)
Avoid by: BLINDING (outcome assessors - esp when subjective measures)
Performance Bias
(RCTs) - Systematic differences btw care provided/exposures outside of intervention (amount of attention, ancillary treatment, diagnostic investigations)
Avoid by: BLINDING participants & study personnel
Reporting Bias
(SR) - Systematic differences btw groups in what’s reported/unreported
(SR - significant differences more likely to be reported)
Verification Bias
DTA - systematic error when not all ptpts receive BOTH:
. INDEX TEST
. REFERENCE STANDARD
(most likely when index test = -ve)
Review Bias
DTA - interpretation of index test is NOT INDEPENDENT of the reference standard & vv
(could make index look more accurate than it actually is)
To avoid: BLINDING of study personnel
Reflexive stance of researcher
QUAL - relationships with ptpts &interpretation of data affected by researchers’ personal biases ( background, professional role, personality)
What can the consequences of confounding be?
- imply aberrant/false/spurious association
- exaggerate association
- mask association
What is confounding?
Additional factors that may give rise to the outcomes & be confused for a causality relationship btw exposure & outcome when not adjusted for
True confounders are related to BOTH Exposure & Outcome
How do you adjust for confounding?
Stratification/standardisation of data
Regression models
- linear
- logistic
- cox
- poisson
Rigour
The quality of being extremely thorough & careful (reducing the potential for bias at every opportunity)
