HTLH 207 Week 13 - Information Bias

Information bias

Occurs when there is error in the measurement of the exposure or outcome that leads to a measure of association that is meaningfully different from what would have been obtained if all variables had been measured correctly

4 examples of potential sources of information bias

1. Faulty measurement devices

2. Poorly written questionnaires

3. Incorrect self-report

4. Data entry errors

2 terms related to information bias

1. Measurement error

2. Misclassification

Does misclassification occur in numeric or categorical variables?

Categorical

What does misclassification result in?

Individuals being misplaced in an RxC table

4 sources of information bias

1. Incorrect measurement devices or laboratory assays

2. Incorrect self-reported data

3. Investigator error

4. Data management error

2 reasons why incorrect self-reported data may occur

1. Problems with the data collection instrument

2. Participant barriers

Sources of information bias affect measurement of what factors?

It can affect measurement of the exposure, outcome, or both

Is information bias influenced by sample size?

No

3 reasons why medical devices or laboratory devices may yield incorrect results

1. Not properly calibrated

2. Tested on heterogenous populations

3. Disease in subclinical

2 examples of faulty medical devices

1. A scale that is not calibrated may yield incorrect measurements of weight

2. Pulse oximeters were not initially tested on darker skin and have been shown to yield incorrect measurements for black patients more frequently than white patients

False negative

Patient is incorrectly classified as not having the outcome

What 2 negative consequences can occur from false negative test results?

1. May contribute to worsening health outcomes if individuals do not receive the care and treatment they need due to receiving a negative test

2. A false negative result for an infectious disease can lead to uncontrolled transmission

3 problems with the data collection instrument that may lead to incorrect self-reported data

1. Survey or interview questions that use technical or medical jargon that is not understood by participants

2. Survey or interview questions that use ambiguous language

3. Poorly-constructed survey or interview questions that prevent participants from understanding the meaning of the question

3 participant barriers that may lead to incorrect self-reported data

1. Recall bias

2. Low health literacy, language barriers, and/or mental health concerns

3. Social desirability bias

Recall bias

Participants may not accurately remember certain behaviors or health information that occurred in the past

Social desirability bias

Intentional misreporting of stigmatized behaviors or health outcomes

When is social desirability bias more common?

When data are collected by interviewers rather than self-administered surveys

Observer bias

Investigators collecting data may make conclusions about participant exposure and/or outcome status in error

7 reasons why investigator error may occur

1. Fatigue

2. Lack of time

3. Limited knowledge

4. Preconceptions

5. Biases

6. Beliefs

7. Preferences

2 ways in which investigators might influence participants’ responses

1. Tone of voice

2. Use of leading phrases

Data management errors can lead to information bias in what 2 ways?

1. Incorrect entry

2. Coding of the exposure and/or outcome data

2 examples of data entry errors

1. Data entry personnel being unable to correctly interpret handwritten chart notes

2. Data entry personnel making incorrect keystrokes

When there is a misclassification of the exposure only, individuals are moving between which two cells on a 2×2 table?

Individuals are moving between cells A and B and between cells C and D in a 2x2 table

When there is a misclassification of the outcome only, individuals are moving between which two cells on a 2×2 table?

Individuals are moving between cells A and C and between cells B and D in a 2x2 table

When there is a misclassification of the both the exposure and outcome, individuals are moving between which two cells on a 2×2 table?

Individuals can move from their true cell to any other cell in a 2x2 table

6 ways to reduce information bias in the design stage

1. Masking

2. Validated, ideally gold standard, exposure and outcome classification methods

3. Strategies to support accurate self-report

4. Strategies to support accurate recall

5. Multiple, independent assessments

6. Planning and budgeting for validation studies

Masking

Prevents preconceived beliefs from influencing interactions with participants (making investigators), assessment of data (masking analysts), or participant reporting (masking participants)

What do validated, ideally gold standard, exposure and outcome classification methods allow for?

The most objective and accurate measurement of study variables

2 examples of strategies to support accurate self-report

1. Self-administered or computer-administered surveys instead of interviewer-administered surveys

2. Use of short, simple unambiguous, non-leading survey questions

3 examples of strategies to support accurate recall

1. Memory aids

2. Diaries

3. Framing questions to use short recall periods

What do multiple, independent assessments in the design stage allow for?

Adjudication of discrepant results

What does planning and budgeting for validation studies allow for?

Correction

2 things that misclassification tables allow us to explore

1. Degree of misclassification

2. Whether misclassification is differential or non-differential

How are misclassification tables oriented?

These tables are oriented such that the data that are not biased are presented in the columns and the data as classified are presented in the rows

Given the orientation of misclassification tables, what do cells A-D represent?

• Cell A: true positives

• Cell B: false positives

• Cell C: false negatives

• Cell D: true negatives

Using misclassification tables, information bias can be quantified using which 2 measures?

1. Specificity

2. Sensitivity

Specificity

Proportion of those without the attribute (i.e. exposure or outcome) who are correctly classified)

Sensitivity

Proportion of those with the attribute (i.e. exposure or outcome) who are correctly classified

Sensitivity (Se) calculation

True positives/total who truly have the attribute = true positives/true positives+false negatives = A/A+C

2 ways in which sensitivity can be reported

1. Decimal

2. Percentage

Specificity (Sp) calculation

True negatives/total who truly do not have the attribute = true negatives/false positives + true negatives = D/B+D

2 ways in which sensitivity can be reported

1. Decimal

2. Percentage

Proportion of false negatives calculation

1 - sensitivity

Proportion of false positive calclulation

1 - specificity

Pro and con of tests with high sensitivity and lower specificity

• Pro: good at identifying true cases

• Con: incorrectly identifies individuals as having the attribute when they truly do not

Pro and con of tests with high specificity have lower sensitivity

• Pro: probability of false positives is lower

• Con: test may miss individuals who truly have the attribute

2 classifications of misclassification

1. Differential

2. Non-differential

Differential misclassification

Occurs when the extent of misclassification of the exposure or outcome does depend on the status of the other variable

Non-differential misclassification

Occurs when the extent of misclassification of the exposure or outcome does not depend on the status of the other variable

Non-differential exposure misclassification

Occurs when the exposure is misclassified, independent of outcome status

In non-differential exposure misclassification, the extent of exposure misclassification is ___________ for those with and without the outcome

the same

Example of non-differential exposure misclassification

In a study of heart disease and previous physical activity, everyone had similar difficulty remembering their specific exercise frequency over many years, regardless of heart disease status

Non-differential outcome misclassification

Occurs when the outcome is misclassified, independent of exposure status

In non-differential outcome misclassification, the extent of outcome misclassification is _________ for the exposed and unexposed

the same

Example of non-differential outcome misclassification

In a cohort study of pesticide exposure and birth weight, there is a random amount of data entry error of birth weights on the birth certificate, irrespective of the mother’s pesticide exposure history

In general, non-differential misclassification of a binary exposure or outcome results in what kind of bias?

Bias towards the null

Differential exposure misclassification

Occurs when the exposure is misclassified to a different degree for those who have the outcome than it is for those who do not have the outcome

Example of differential exposure misclassification

In a study of birth defects and prescription drugs taken during pregnancy, mothers of children born with birth defects may remember the drugs they took during pregnancy better than mothers of children without birth defects

Differential outcome misclassification

Occurs when the outcome is misclassified to a different degree for the exposed than it is for the unexposed

Example of differential outcome misclassification

In a cohort study of health effects of diabetes, physician may monitor the health status of the diabetic patients more closely, and thus be more likely to detect diseases that are there compared to the group without diabetes

In general, differential misclassification of a binary exposure or outcome results in what kind of bias?

Bias in any dirrection

2 steps used to determine whether outcome misclassification is differential or non-differential

1. Create separate outcome misclassification tables for the exposed and unexposed groups

2. Calculate and compare the sensitivity and specificity of the outcome misclassification among the exposed and unexposed

Outcome misclassification is non-differential if…

The sensitivities and specifies are the same for the exposed and unexposed

Outcome misclassification is differential if…

The sensitivities and/or specificities are different for the exposed and unexposed

2 steps used to determine whether exposure misclassification is differential or non-differential

1. Create separate exposure misclassification tables for those with the outcome and those without the outcome

2. Calculate and compare the sensitivity and specificity of the exposure misclassification among those with the outcome and those without the outcome

Exposure misclassification is non-differential if…

The sensitivities and specifies are the same for those with the outcome and those without the outcome

Exposure misclassification is differential if…

The sensitivities and/or specificities are different for those with the outcome and those without the outcome

What does correcting information bias analytically require?

A measure of the sensitivity and specificity of variable classification

How do researchers estimate the sensitivity and specificity of a variable

They conduct validation studies

Example of validation study

A self-reported instrument may undergo validation by comparing the self-reported data to other more accurate sources of data, such as a lab assay

3 potential sources of validation data

1. Medical records

2. Pharmacy prescription records

3. Reports from family or friends

Is it typically necessary to perform validation on the entire sample? Why or why not?

No - saves both time and money

Sensitivity analyses

Conducted to understand how study findings may change under a set of varying, yet plausible, assumptions

What can be done if the sensitivity and specificity are unknown?

Multiple corrections using a range of plausible sensitivity and specificity estimates can be conducted to explore how different sets of assumptions affect the results

Screening tests are used in which individuals?

Individuals who appear healthy but may be at risk for a disease of interest

Function of screening tests

May detect disease before symptoms begin, which could lead to earlier treatment and reduced morbidity and mortality

What will individuals with a positive screening test often undergo?

Diagnostic testing

Diagnostic tests are used in which individuals?

Individuals who have signs or symptoms of a disease

4 ways in which the performance of screening and diagnostic tests can be quantified

1. Sensitivity

2. Specificity

3. Positive predictive value

4. Negative predictive value

Positive predictive value (PPV)

Proportion of those who test positive who do have the disease

PPV calculation

True positives/total who test positive = A/A+B

Negative predictive value (NPV)

Proportion of those who test negative who do not have the disease

NPV calculation

True negatives/total who test negative = C/C+D

Difference between sensitivity and specificity vs. PPV and NPV

• Sensitivity and specificity are not impacted by the prevalence of disease in a population

• PPV and NPV are impacted by the prevalence of disease in a population (they depend on sensitivity and specificity)

As the prevalence of disease increases, PPV _________ and NPV _________

increases, decreases

As the prevalence of disease decreases, PPV __________ and NPV _________

decreases, increases

Given that positive test results are more likely to be true when the prevalence of a disease is higher, what may occur?

Tests may be preferentially allocated to populations at greater risk to maximize screening affordability and feasibility

Is screening always beneficial?

No

4 considerations that support screning

1. The disease is serious and associated with severe morbidity and/or mortality

2. A high proportion of disease is detectable by screening before symptoms appear

3. Screening enables earlier disease detection and earlier treatment or disease management leads to better outcomes

4. Existing screening tests are accurate, reliable, affordable, straightforward to administer, and have few or acceptable side effects

5 considerations that do not support screening

1. Overdiagnosis and overtreatment if a screening tests leads to many asymptomatic cases that would not have required intervention

2. Undertreatment if a screening test results in a high proportion of false negatives

3. Physical and psychological side effects that do not outweigh benefits

4. High costs and unaffordability

5. Ethical concerns such as whether screening tests are widely available to the target population and whether adequate and timely diagnostic tests are available for individuals who screen positive