Bias in Epidemiological Studies

1
Bias in Epidemiological Studies
Dr. Sohinee Bhattacharya sohinee.bhattacharya_at_abdn
.ac.uk
2
Overview of Lecture

• Random vs Systematic Error
• What is Bias?
• Types of Bias
• Bias in Relation to Study Design
• Effect of Bias on Results
• Eliminating bias

3
News of the Week.
4
Error

• An error is by definition an act, an assertion,
  or a belief that deviates from what is right..but
  what is right?
• The true length of a metre is arbitrarily decided
  by agreeing a definition
• The difference between a "correct" metre stick
  and an erroneous one can be accurately measured
• For health and disease the truth is usually
  unknown and cannot be defined in the way we
  define metre
• Error should be considered as an inevitable and
  important part of human endeavor
• Popperian view is that science progresses by the
  rejection of hypotheses (by falsification) rather
  than the establishing of so called truths (by
  verification)

5
Figure 4.1
(b) Error is unequal in one of these groups
leading to a false interpretation of the pattern
of disease - here failure to detect differences
(a) Error is unequal in one of these groups
leading to a false interpretation of the pattern
of disease - falsely detecting differences
6
Error

• Due to Chance
• Random error type Ireject null in sample that
  is true in population (control with p value and
  CI)
• Due to Bias
• Systematic error (control in design, estimate
  effect size and direction in analysis phase)

7
RANDOM ERROR
8
Random Error

• Random deviation from the truth
• Incorrect assessment of exposure / outcome
• Continuous Incorrect measurement
• Binary / Categorical Incorrect categorisation
• May result from
• Poor instruments / tests
• Data-entry error
• Subject error

9
Random Error

• Also known as
• Random misclassification
• Random noise
• Decreases likelihood of observing an effect
• Bias findings towards the null
• Increases likelihood of Type 2 error (falsely
  accepting H0)
• Serves to underestimate any association

10
NON-RANDOM ERRORBIAS
11
Bias Websters Definitions

• A line diagonal to the grain of a fabric
• Highly personal and unreasoned distortion of
  judgment
• Systematic error introduced encouraging one
  outcome over others

12
Bias

• A preference or an inclination
• Bias may be intentional or unintentional
• In statistics a bias is an error caused by
  systematically favoring some outcomes over others
  
• Bias in epidemiology can be conceptualised as
  error which applies unequally to comparison
  groups.

13
Bias

• Sources of variation that distort the study
  findings in one direction
• Is a spurious association or effect on an outcome
  in a particular study
• Is a systematic error
• biased measurement vs biased results

14
A Biased Research Question

• Are girls cleverer than boys?
• The right question here would be that there are
  no gender differences in intelligence
• The underlying values of the researchers may be
  that girls are more intelligent than boys
• Likely to be revealed at the analysis and
  interpretation stage by biased interpretation
• It is problematic to describe difference without
  conveying a sense of superiority and inferiority

15
The research question

• Syphilis Study of the US Public Health Service
  followed up 600 African American men for some 40
  years
• The question does syphilis have different and,
  particularly, less serious outcomes in African
  Americans than European origin Americans?
• Investigators denied the study subjects treatment
  even when it was available and curative
  (penicillin)

16
Bias

• Systematic deviation from the truth
• Can increase or decrease an effect estimate
• Study design can (help to) eliminate
• Can be a problem, but may be useful
• Can be investigated
• Size of increase or decrease can be estimated

17
Types of Bias

• Selection bias
• Concerned about who is in your study
• Information bias
• Concerned about the information you elicit from
  your subjects

18
Selection bias

• Selection bias is inevitable, simply because
  investigators need to make choices
• Captive populations are popular-some may be
  representative, e.g. schoolchildren, others not
  at all, e.g. university students
• People are also missed either inadvertently or
  because they actively do not participate
• Selection bias matters much more in epidemiology
  than in biologically based medical sciences.
• Biological factors are usually generalisable
  between individuals and populations, so there is
  a prior presumption of generalisability
• If an anatomist describes the presence of a
  particular muscle, or cell type, based on one
  human being it is likely to be present in all
  human beings (and possibly all mammals)

19
Figure 4.2

• Ignoring populations
• Questions harming one population
• Measuring unequally
• Generalising
• from unrepresentative populations

Study population
Ignored population
Comparison population
20
Selection Bias

• Often occurs due to failure of intended sample to
  be representative of target population
• Hospitalized patients not representative of full
  range of patients with disease
• Randomized patients selected by response to
  therapy
• Protocol failures

21
Sources of Selection Bias

• Volunteer Bias
• Referral Bias
• Healthy worker effect
• Non-participation bias

22
Volunteer Bias High Grades
23
Referral Bias

• To examine mortality from cardiovascular disease
• Death reports from cardiovascular disease in 2
  tertiary referral hospitals

24
Healthy Worker Effect

• Typically, but not exclusively, in occupational
  environments
• Example
• Case-control study

25
Non-participation Bias

• Most frequent cause for concern in large-scale
  epidemiological surveys
• Non-participation
• Loss to follow-up
• Systematic differences between participants and
  non-participants
• With respect to the relationship under
  examination
• Typical non-participants
• Young / Male / Ethnic minorities

26
Types of Bias

• Selection bias
• Concerned about who is in your study
• Information bias
• Concerned about the information you elicit from
  your subjects

27
Information Bias

• What information are you getting from subjects?
• Concern
• Are there systematic differences in what is being
  collected, between study groups?
• Does each subject have an equal chance of
  providing the same information?
• Sources
• Observer bias
• Recall bias

28
Observer Bias

• Interviewer knowledge may influence structure of
  questions
• Preconceived expectations of study outcome
• Study methods may change over time
• Different investigators may examine different
  subjects
• Times / locations of interviews may vary

29
Attention Bias

• Hawthorne Effect
• An increase in worker productivity produced by
  the psychological stimulus of being singled out
  and made to feel important
• People may respond differently if they think they
  know what is being studied
• Potential effect
• ?? prevalence of disease
• ?? relationship under examination

30
Surveillance Bias

• Can arise if one group is over-researched, in
  comparison with the other
• Case-control study
• Tendency to examine more closely those with
  outcome of interest
• Association alcohol consumption vs oropharyngeal
  cancer
• Cohort study / Randomised trial
• Tendency to follow more closely (or for longer)
  those with exposure of interest
• Association CBT vs low back pain

31
Recall Bias

• Major concern where exposure data measured
  retrospectively
• Case-control studies (including case-control
  analysis of cross-sectional survey)
• Concern
• Differential recall between cases and controls

32
Recall Bias
33
(No Transcript)
34
Epidemiological Study Design
Assess exposure and outcome
35
Bias in Case-Control Studies

• Study design most susceptible to bias
• Separate sampling of cases and controls
• Retrospective measurement of predictors
• Selection
• Cases take all
• Controls similar in all respects to cases except
  in terms of disease occurrence
• Hospital vs community controls

36
Bias in Case-Control Studies

• Ascertainment
• Differential measurement in predictors
• Recall bias (use data recorded before outcome)
• Blinding

37
Bias in Cohort Studies

• Bias in cohort studies
• Loss to follow up or attrition bias
• Following up newborn infants (Victora, 1987)
• The loss to follow up differed systematically by
  socioeconomic status

38
Bias in RCTBias in RCT

• Selective failure to receive intervention
• - adhere to protocol
• - get follow-up
• Control with
• blinding is as important as randomization
• investigator, subjects and all others
• randomization controls for unknown
  confounders (also)
• analysis intention to treat
• design to prevent drop outs run in period

39
Other Types of Bias

• Investigator
• Conflicts of interest
• Publication
• Studies not representative
• Spectrum
• Relevant to research on medical tests
• Inappropriate range of tests results or disease
• Range of disease not representative of target
  population
• Difference of advanced disease vs normals

40
Spurious Associations from Bias

• Difference between research question and question
  actually asked
• For population and samples
• Target population all adults
• Intended sample patients in clinic
• For Phenomena of interest
• Cause coffee consumption
• Variable self reported coffee consumption

41
Spurious Associations from Bias

• In outcome
• Effect myocardial infarction
• Outcome diagnosis of infarction from claims
  data

42
Preventing Bias

• Design phase problem
• Dont introduce bias in analysis phase by
  attempting to correct for confounding or other
  problems
• In design phase assess direction (occasionally
  magnitude of effect) by use of outside
  information
• Eg positive effect measure found, bias toward
  null, minimal problem

43
Selection Bias What To Do

• Prevent
• Minimise
• Estimate effect
• What effect(s) might it have had on your study?
• How might this change the results?
• Does this change the conclusions?

44
Minimising Selection Bias

• Be aware
• Potential sources of selection bias
• Equal opportunity for participation and follow-up
• Cases / Controls
• Exposed / Unexposed groups
• Intervention / Control groups
• Tactics for high participation / follow-up rates
• Reminders / Postcards / Phone calls

45
Assessing Selection Bias

• Demographic approach / Alternative data
• What information is available on your
  non-responders?
• Where did you get sample from?
• Can you examine response by age / sex /
  occupation / etc?
• Examine reluctant responders

46
Estimating Effect of Selection Bias

• Study to estimate the prevalence of asthma in
  schoolchildren
• Subjects children aged 5 8 yrs
• Response rate 60
• What is the potential effect of non-response
  bias?
• Depends on characteristics of non-responders

47
Minimising Observer Bias

• Standardised techniques / instruments / etc
• Thorough training of data collection staff
• Test agreement between interviewers / instruments
• Use objective measurements where possible
• Where possible, researchers should be
• Randomly allocated to subjects
• Blind to study question
• Blind to case / control status

48
Minimising Attention Bias

• Mask true study question from participants
• Ethics
• Informed consent
• Health study
• Collect information about several outcomes
• Difficult in a case-control study
• Collect information about several exposures
• Ensure anonymity

49
Minimising Surveillance Bias

• Ensure identical methodological procedures for
  all study participants
• Where possible, blind researchers
• To study question
• To case / control status
• To exposure / non-exposure status
• To treatment / non-treatment group

50
Minimising Recall Bias

• Minimise period of recall (if possible)
• Measure exposure data objectively
• Medical notes
• Third-party verification of exposure information
• Triangulation of measurements
• Can you conduct a prospective study?

51
SUMMARY
52
Summary Bias

• Concerned with the internal validity of a study
• i.e. the extent to which, within the subjects
  studied, the results are true
• Deviation of results, or inferences, from the
  truth
• Or, processes leading to such deviation
• Results from some aspect(s) of study design or
  conduct

53
Bias Summary

• Increase / Decrease likelihood of observing an
  effect
• Bias findings towards / away from the null
• Increases likelihood of Type 1 error (falsely
  rejecting H0)
• Increases likelihood of Type 2 error (falsely
  accepting H0)
• Serves to over- / under-estimate any association

54
Bias Summary

• Can be prevented by design
• Can be estimated
• Cannot be overcome by analysis
• May be useful