Module 5.1- Bias

What is bias about?

• Ensuring trustworthiness

• We can only trust our measure of association (RR, OR, RD) once we are sure that we have ruled out alternative explanations

3 alternative explanations

1. Bias

2. Confounding

3. Random error

Bias

A systematic error in the design or conduct of a study that leads to an erroneous, “distorted” association between the exposure and outcome

Confounding 

The mixing of effects between the exposure and the outcome; and a third variable, which is called a confounder. Like bias, confounding distorts the relationship between exposure and outcome.

Random error

The probability that the observed result is attributable to the play of chance, which is an uncontrollable force that seems to have no assignable cause

Internal validity

The degree to which a study’s results are free of bias, confounding, and random error

External validity

The degree to which a study’s results can be applied to other populations than those in the study (aka generalizability)

Bias can…

1. Distorts the association measure by a small amount

2. Distorts the association measure by a large amount

3. Reverses the association direction

4. Removes an association when one is there

5. Creates an association when none is there

The null

RR = 1.0 or OR = 1.0 or RD = 0

Selection bias

• Error that results from procedures used to select participants into the study or analyses

• Gives a result among participants that is different from the result that would occur among all eligible individuals in source population

Selection bias in a case-control study

Selection of cases and controls is based on different criteria that are related to exposure status

How to control selection bias

Use the same selection criteria for cases and controls

Self-selection bias

• Refusal or non-response by participants that is related to both exposure and outcome

• Agreement to participate that is related to exposure and outcome

• If participation rates are not related to exposure or outcome, no bias

Solution for selection bias

• High participation rates among cases and controls

• Guideline: 80% participation of both

Differential Surveillance, Diagnosis, or Referral

• Detecting of cases is related to exposure

• Example: VTE and oral contraceptive users

  • Cases: women aged 20-44 hospitalized for VTE

  • Controls: women aged 20-44 hospitalized for something else

Selection Bias in prospective cohort / RCT

Differential loss to follow-up

• More exposed than unexposed drop out

• More “treatment” than control drop out

Loss to follow-up

Participants can no longer be reached or discontinue participation

Problem with loss to follow up

• Reduce study power

  • the chance that a study will detect a real effect if one truly exists

  • high power (>80%): less likely to miss something due to chance

• Bias the study results

  • If non-differential → observed disease incidence is too low

  • If differential → biases towards or away from the null

Selection bias in retrospective cohort

• Selection of exposed and unexposed individuals is related to developing the outcome of interest

• Follow-up of exposed and unexposed individuals is related to developing the outcome of interest

Healthy worker effect

A bias in occupational health studies where the working population appears healthier than the general population because severely ill or disabled individuals are typically excluded from employment

Information bias

A flaw in measuring exposure, covariate, or outcome that results in different quality (accuracy) of information between comparison groups

Information bias in a case-control study

Different interview techniques (Ca vs Co)

Information bias in a cohort study

Different procedure to assess outcome in exposed vs. unexposed

Recall bias

Differential level of accuracy of information provided by compared groups (e.g., cases vs. controls; exposed vs. unexposed)

Recall bias in a case-control study

• Cases may be more likely to recall exposure

• Can bias the associations toward or away from the null

Recall bias in a cohort study

Exposed participants more likely to report diseases

How to avoid recall bias in a case-control study

• Select a control group with an unrelated disease

• Structured questionnaire design

• Self-administered questionnaires for sensitive topics

• Biomarkers

Interviewer bias

Systematic difference in soliciting, recording, or interpreting information that occurs in studies using in-person interviews

Interviewer bias in case-control study

Interviewer is aware of disease status

Interviewer bias in cohort/RCT

When interviewers know treatment or exposure

Solutions to interviewer bias

• Mask interviewer to case status or exposure/treatment

• Standardize questionnaires (closed-ended)

  • Avoid ambiguous wording (“Have you ever smoked?”)

  • Avoid using 2 questions in 1 (“out of breath when you go up a hill”)

  • Avoid jargon (medical terms)

  • Avoid asking “unanswerable questions” (how many brands of TP used)

  • Avoid questions that are not self-explanatory (“through the vents”)

  • Avoid multiple choices (e.g., condoms, diaphragms, IUDs)

Misclassification

• An error in the classification of exposure or outcome

• Most common form of information bias

• Also called measurement error

Differential misclassification

• Errors on one axis (case/control) related to errors on the other (exposed/non-exposed)

• May bias towards or away from the null

Non-differential misclassification

• Errors on one axis (case/control) NOT related to errors on the other (exposed/non-exposed)

• Bias towards the null