Bias and error.final(1).ppt

error and bias in research

1. Error and Bias
PRESENTED BY
MN first year
Mary Pradhan, Saraswati Shrestha, Narayani
Lamichhane

2. Content
•Introduction of error
•Types of error
•Types of bias.

3. Error
•Falseormistakenresultobtainedina
studyorexperimentthattakesplacein
anystage/process.
•Affectontheaccuracy/validity/reliabilityof
thestudy
•Thetermerrorinepidemiologyreferstoa
phenemenoninwhichtheresultoffinding
ofthestudydoesnotreflectthetruthor
fact.

4. Good result = =
Erroneous result = =
Observed value Fact value
Observed value Fact + Distortion

5. •Itisdifficulttomakethestudyfreefrom
anytypeoferror.
•Therefore,theaimistomaximizefactand
minimizeerrorsothattheresearchwork
wouldrepresenttothepopulationthey
refer.

6. Basic Types of Error
•Random Error (Precision Problem)
•Systemic Error (Validity Problem)

7. Random error
•Bychanceerror
•Makesobservedvaluesdifferfromthetruevalue.
•Errorexisteverytimewhenwedrawarandom
sampleandmakeaconclusionregardingthe
respectivepopulation.
•Inepidemiologicstudies,randomerrorhasmany
componentsbutmajorcontributoristheprocessof
selectingthespecificstudysubjectsi.e.sampling
error

8. Sampling error
•Samplingerroristhedeviationoftheselectedsample
fromthetruecharacteristics,traits,behaviors,qualities
orfiguresoftheentirepopulation.
•Becauseofchance,differentsampleswillproduce
differentresultsandthereforemustbetakeninto
accountwhenusingasampletomakeinferences
aboutapopulation.Thisdifferenceisreferredtoas
thesamplingerroranditsvariabilityismeasuredby
thestandarderror.
•Theerrorwhicharisebecauseofstudyingonlyapart
ofthetotalpopulationarecalledsamplingerrors.

9. whenwetakeasample,itisonlya
subsetoftheentirepopulation;
therefore,theremaybeadifference
betweenthesampleandpopulation.

10. •Thesemayariseduetononrepresentativeness
ofthesamplesandtheinadequacyofsample
size.Whenseveralsamplesaredrawnfroma
population,theirresultswouldnotbeidentical.
•Samplesizeandsamplingerrorarethus
negativelycorrelated.
•Samplingerrorcanbereducedbyincreasingthe
samplingsizewithvalidscientificsample
selectioncriteria.

11. Asthesamplesizeincreases,itapproachesthe
sizeoftheentirepopulation,therefore,italso
approachesallthecharacteristicsofthe
population,thus,decreasingsamplingprocess
error.

12. Types of Random Error
1.Type I
2.Type II

13. Type I error (alpha error)
•Exists when an investigator or study
rejects a null hypothesis when it is
actually truein the population, so that the
test result is false positive.
•Example: a test that shows a patient to
have a disease when in fact the patient
does not have the disease.

14. Type II Error ( beta error)
•Exists when an investigator or a study
acceptsnull hypothesis when actually it
is falsein the population, so the test result
is false negative
•Example: a blood test failing to detect the
disease it was designed to detect, in a
patient who really has the disease

15. Systematic error or
Bias

16. Systematic error or bias
•It occurs when there is a difference between the
true value (in the population) and the observed
value (in the study) from any cause other than
random error.
•It is an error due to factors that exist in the study
design, data collection, analysis and interpretation
to yield results or conclusions that depart from the
truth.
•If there is misrepresentation of the effect, it is
called bias and If there is no misrepresentation, it
is called valid or no bias.
•Increasing of the sample size has no effect on it.

17. •In an epidemiological study, it is defined
as any systematic error that result an
incorrect estimate of the association
between exposure and risk of disease.
•Example -testing for antibodies will
consistently underestimate the prevalence
of HIV infection because individual who
have been infected for less than six
months will not yet have developed
antibodies.

18. Definition of bias
•Any trend in the collection, analysis,
interpretation, publication or review of
data that can lead to conclusions that
are systematically different from the
truth (Last, 2001)
•A process at any state of inference
tending to produce results that depart
systematically from the true values
(Fletcher et al, 1988)

19. Types of bias or systematic error
•Selection Bias
•Information Bias
•Confounding

20. Selection Bias
•Selection bias occurs when there is a
systematic difference between the
characteristics of the people selected for a
study and the characteristics of those who
are not and which distorts in the estimate of
effect (result).
•.sample obtained is not representative of
the population to be analyzed

21. Types
•Publicity bias
•Non-response bias
•Healthy worker effect
•Diagnostic bias
•Loss to follow-up bias

22. Publicity bias
•People referring themselves to the investigators
following publicity of the study.
•Publicity bias can also occur from news reports not
related to individuals.
•In a 1981 –1982 survey of individuals near two
hazardous waste disposal sites in Louisiana, people
were asked about various symptoms. Air and water
quality data showed little evidence of hazardous
concentrations of chemicals, but there had been
extensive media coverage at the time of the survey.
•Respondents living near the sites were two to three
times as likely to report symptoms as respondents in
an unexposed community because of the influence of
the publicity at that moment in time.

23. Non –response bias
•A type of bias when an individual chosen
for the sample cannot be contacted or
refuses to cooperate . When non response
bias occurs, there is an unrepresentative
sample.

24. Example
Consider a study that examines drug
abuse among adults. Many drug users
may be unwilling to talk about their views
toward drug abuse in light of their own
problems. Due to these participation
issues, the opinion of drug nonusers
should be overrepresented.

25. Healthy worker effect
•It is introduced when the disease or factor
under investigation itself make people
unavailable for study.
•Relatively healthy people become or
remain workers, whereas those who
remain unemployed, retired, disabled, or
otherwise out of the active worker
population are as a group less healthy.

26. Example…. ‘healthy worker
effect’
•Study : Association between
formaldehyde exposure and eye
irritation
•Subjects: factory workers exposed to
formaldehyde
•Bias: those who suffer most from eye
irritation are likely to leave the job at
their own request or on medical advice
•Result: remaining workers are less
affected; association effect is diluted

27. Diagnostic bias
•Diagnoses (case selection) may be
influenced by physician’s knowledge of
exposure
•E.g. Case control study –outcome is
pulmonary disease, exposure is smoking
•-Radiologist aware of patient’s smoking
status when reading x-ray –may look more
carefully for abnormalities on x-ray and
differentially select cases in exposed group
and less so in control group.

28. •Example: In a case-control study looking at
the relationship between DVT and oral
contraceptives.
•The GPs knew about the possible link
between OC and DVT so women with
suggestive symptoms and known use of OC
were more likely to be referred to the hospital
with “DVT”.
•This could lead to an over-estimation of the
effect of OC on DVT

29. Loss to follow-up bias
•lost to follow-uprefers to respondents who
are at one point in time were actively
participating in a clinical research trial, but
have become lost at the point of follow-up in
the trial.
•Design and implementation of the study
should try to minimize this and we should aim
to ensure that all groups are followed as
completely as possible and with equal rigor

30. Information bias
•Itisadistortionintheestimateofeffectdueto
measurementerrorormisclassificationof
subjectsononeormorevariables.

31. Contd…
•ItmayalsobecalledasMeasurementbias,
misclassificationbias.
•Majorsourcesofmeasurementbiasinclude
invalidmeasurement,incorrectdiagnostic
criteria,andomissionsandinadequaciesin
previouslyrecordeddata.

32. Common Types of Measurement Biases
•instrument bias,
•insensitive measure bias,
•expectation bias/observer bias
•recall or memory bias,
•attention bias, and
•verification or work-up bias.

33. Contd…
Instrumentbias:
•Instrumentbiasoccurswhencalibration
errorsleadtoinaccuratemeasurementsbeing
recorded,e.g.,anunbalancedweightscale.

34. Contd…
Insensitivemeasurebias:
•Insensitivemeasurebiasoccurswhenthe
measurementtool(s)usedarenotsensitive
(poorcalibration)enoughtodetectwhat
mightbeimportantdifferencesinthe
variableofinterest.

35. Expectation bias
•Expectation bias occurs in the absence of
masking or blinding, when observer may
have error in measuring data towards the
expected outcome.
•The observer-expectancy effect occurs
when a researcher's beliefs or
expectations unconsciously affect the
behavior of the observed subject(s)

36. Recall or memory bias.
•Systematic error due to differences in accuracy
or completeness of recall to memory of past
events or experience.
•Often a person recalls positive events more than
negative ones. Alternatively, certain subjects may
be questioned more vigorously than others,
thereby improving their recollections.

37. •Mothers of children with birth defects are likely to remember
drugs they took during pregnancy differently than mothers of
normal children.
•In this particular situation the bias is sometimes referred to
asmaternal recall bias.
•Mothers of the affected infants are likely to have thought
about their drug use and other exposures during pregnancy to
a much greater extent than the mothers of normal children.
The primary difference arises more from under reporting of
exposures in the control group rather than over reporting in
the case group. However, it is also possible for the mothers in
the case group to under report their past exposures
•For example, mothers of infants who died from SIDS may be
inclined to under report their use of alcohol or recreational
drugs during pregnancy.

38. Contd..
Attention bias:
•Attention bias occurs because people who
are part of a study are usually aware of their
involvement, and as a result of the attention
received may give more favorable responses
or perform better than people who are
unaware of the study’s intent.

39. Verification or workup or referral
bias
•It is a type of measurement bias in which
the results of a diagnostic test affect
whether the gold standard procedure is
used to verify the test result.
•It is mainly associated with test validation
studies.

40. Cont..
•In clinical practice, referral bias is more
likely to occur when a preliminary
diagnostic test is negative. Because many
gold standard tests can be invasive,
expensive, and carry a higher risk (eg:
angiography, biopsy, surgery) ,patients
and physicians may be more reluctant to
undergo further workup if a preliminary
test is negative.

41. Contd..
•In cohort studies, obtaining a gold
standard test on every patient may not be
ethical, practical, or cost effective. These
studies can thus be subjected to
verification bias.
•One method to limit verification bias in
clinical studies is to perform gold standard
testing in a random sample of study
participants.

42. CONFOUNDING

43. •The term ‘confounding’ refers to the effect of
an extraneous variable that entirely or
partially explains the apparent association
between the study exposure and the disease.
•Confounding is a distortion in the estimated
measure of effect due to the mixing of the
effect of the study factor with the effect of
other risk factors. Confounding effect may
distort the true association in either direction.

44. •In a study of the association between
exposure to a cause ( or risk factor )
and the occurrence of disease,
confounding can occur when another
exposure exists in the study
population and is associated both with
the disease and the exposure being
studied.

45. Eg-EXPOSURE DISEASE
(coffee drinking) ( heart
disease )
CONFOUNDING VARIABLE
(cigarette smoking)

46. Criteria for confounders
•It is the risk factor of the study disease
(but it is not the concequence)
•It is associated with exposure under study
•It is about of interest of current study(i.e.
an extraneous variables)
•In the absence of exposure it indendently
able to cause disease(outcome)

47. THE CONTROL OF CONFOUNDING
The method commonly used to control confounding
in the design of an epidemiological study is:
•Randomization
•Restriction
•Matching
•At the analysis stage , confounding can be
controlled by;
•Stratification
•Statistical modeling

48. Contd…
•It can be controlled either by research
design or during data analysis phase.
There are three methods that can be used
to control confounding during the design
phase of the study: randomization,
restriction and matching.

49. •RANDOMIZATION-is applicable only to
experimental studies, is the ideal method for
ensuring that potential confounding variables are
equally distributed among the groups being
compared .The sample size has to be sufficiently
large to avoid random misdistribution of such
variables . Randomization avoids the association
between potentially confounding variables and the
exposure that is being considered.

50. •RESTRICTIONcan be used to limit the
study to people who have particular
characteristics. For Eg. In a study on the
effects of coffee on coronary heart
disease, participation in the study could be
restricted to nonsmokers, thus removing
any potential effect of confounding by
cigarette smoking.

51. •MATCHING –If matching is used to control
confounding the study participants are selected
so as to ensure that potential confounding
variables are evenly distributed in the two
groups being compared .For eg in a case –
control study of exercise and coronary heart
disease , each patient with heart disease can be
matched with a control of the same age group
and sex to ensure that confounding by age and
sex does not occur.

52. •STRATIFCATION-Confounding can be
controlled by stratification in which
subject are split into groups or strata and
the association between exposure and
outcome of interest is then measure
separately in each stratum, analysis can be
done separately men and women to remove
confounding by sex, for different age
groups and so on. In practice, it is difficult
to remove all confounding .

53. •MODELLING-Although stratification is
conceptually simple and relatively easy to
carry out, it is often limited by the size of the
study and it cannot help to control many
factors simultaneously, if we want to control
by age, sex, and smoking by stratification,
each stratum will contain only few percent of
study population and it can be difficult to
obtain precise estimates of association in each.

54. Contd…
•Therefore, statistical modeling (multivariate) is
used to control for all potential confounders
while controlling for a number of confounding
variables simultaneously
•The most common multivariate approach for
unmatched case control study is multiple
logistic regression and for matched case
control study is conditional logistic regression
and common technique used for cohort is Cox
proportional hazards regression.

55. References
Beaglehole, R., Bonita, R. & Kjellstrom, T. (2006).
Basic epidemiology. Delhi : A.I.T.B.S.
Rao, B. S.(2009). Essentials of Epidemiology.
Delhi: A.I.T.B.S.
Adhikari.S. Foundation of epidemiology.Makalu
publications(2008) first edition.
•http://www.umdnj.edu/idsweb/shared/biases.htm
•http://www.nswphc.unsw.edu.au/pdf/ShortCourse
ResMetJul06/PPts/Introductionto_biasin_research
_DavidLyle.pdf
•Joshi Dr A, Banjara, M. (2007). Fundamental of
Epidemiology

56. THANK YOU!!!!