BASIC COURSE IN BIO-MEDICAL RESEARCH BASIC COURSE IN BIO-MEDICAL RESEARCH

nie.gov.in
nie.gov.in
11/28/2019

1
 11/28/2019

BASIC COURSE IN BIO-MEDICAL RESEARCH

To obtain an
estimate that is
generalizable to
relevant target
populations
To obtain an accurate
estimate of disease
frequency and effect
of exposure on
OBJECTIVES OF health outcomes in
EPIDEMIOLOGICAL study population

STUDIES
nie.gov.in

Accuracy = Precision and Validity

BASIC COURSE IN BIO-MEDICAL RESEARCH

• Precise • Precise
• Valid • Not valid

• Not precise • Not precise

• Valid • Not valid nie.gov.in

2
 11/28/2019

BASIC COURSE IN BIO-MEDICAL RESEARCH

Random errors Systematic errors
CHANCE BIAS
Variability in Errors in A process that
estimation due to estimation tends to produce
unknown/uncontrol
lable causes results that depart
systematically from
• Sampling error
true values (Murphy,
• Measurement 1976)
error

Association between exposure and outcome among the study nie.gov.in

participants differs from the causal association in the population

Threats to validity in epidemiologic

studies BASIC COURSE IN BIO-MEDICAL RESEARCH

Selection Information
Confounding
bias bias

nie.gov.in

3
 11/28/2019

Threats to validity in epidemiologic

BASIC COURSE IN BIO-MEDICAL RESEARCH

studies

Bias results from

Selection Do study participants
accurately represent
procedures used to select
bias individuals for inclusion in
target population?
the study and/or analyses

nie.gov.in

Selection bias in epidemiological studies

• Surveillance - Systematic
notification of exposed cases
• Screening / diagnosis -
Systematic case search among
exposed
• Admission to health care
facilities - Systematic
admission of:
• Case-patients exposed / unexposed
• Control-subjects exposed /
unexposed
BASIC COURSE IN BIO-MEDICAL RESEARCH
• Selective survival - Systematic
inclusion of cases who
survived and who may be
more or less exposed
• Non response / loss to follow
up - Systematic inclusion of
subjects more likely to
participate who may be:
• More or less exposed
• More or less at risk
nie.gov.in

4
 11/28/2019

Dealing with selection bias

BASIC COURSE IN BIO-MEDICAL RESEARCH

Designing stage of a study
• Use incident cases, not prevalent cases
• Case control studies
• Use population-based design
• Apply same eligibility criteria for selecting cases and
controls
• Both cases and controls undergo the same diagnostic
procedures and intensity of disease surveillance nie.gov.in

Dealing with selection bias

BASIC COURSE IN BIO-MEDICAL RESEARCH

Data collection stage of the study

• Minimize nonresponse, nonparticipation and
loss to follow-up (Cohort studies)
• Keep a record of all losses and collect baseline
data on them
• Make sure that diagnosis of disease is not
affected by exposure status (blinding) nie.gov.in

5
 11/28/2019

Dealing with selection bias

BASIC COURSE IN BIO-MEDICAL RESEARCH

Analysis stage of study
• Compare non-responders/dropouts with
responders/non-dropouts with respect to baseline
variables
• Large differences strongly suggest selection bias
• Small differences do not rule out selection bias
• Use study results and external information to deduce the
direction of biases and assess magnitude of biases
• Do sensitivity analysis nie.gov.in

Threats to validity in epidemiologic

studies BASIC COURSE IN BIO-MEDICAL RESEARCH

Do measurements Bias results from

Information procedures used to
accurately represent
measure “exposures”,
bias phenomena of “outcomes” and “other
interest? variables”

nie.gov.in

6
 11/28/2019

Information bias in epidemiological

BASIC COURSE IN BIO-MEDICAL RESEARCH

studies
• Case control study • Investigator - Systematic
• Collection of information collection of information
leaning towards specific
exposure status supporting expected
• Recall - Cases may recall conclusions
exposure more than controls (Unconsciously or
• Better exposure data available Consciously)
on cases compared to controls
• Cohort study • May be examined in the
analysis
• Collection of information
leaning towards specific • Prevarication -
outcome status Systematic distortion of
• Better outcome data available on
exposed compared to unexposed the truth by subjects
nie.gov.in

Dealing with information bias

BASIC COURSE IN BIO-MEDICAL RESEARCH

Precise operational definitions of variables

Detailed measurement protocols

Repeated measurements on key variables

Training, certification and re-certification

Data audits (of interviewers, of data centers)

Data cleaning – visual, computer

Re-running all analyses prior to publication nie.gov.in

7
 11/28/2019

Threats to validity in epidemiologic

BASIC COURSE IN BIO-MEDICAL RESEARCH

studies

The apparent effect of the

“confusion exposure of interest is distorted
Confounding of effects”
because the effect of extraneous
factors is mistaken for – or mixed
with – the actual exposure effect

nie.gov.in

Confounding in epidemiological studies

BASIC COURSE IN BIO-MEDICAL RESEARCH

May simulate an May hide an

association that does association that does
not exist exist

May increase or
May change the decrease the strength
direction of an effect of the association
nie.gov.in

8
 11/28/2019

Dealing with confounding

BASIC COURSE IN BIO-MEDICAL RESEARCH

Exposure Outcome

CONFOUNDER

• Design stage • Analysis stage

• Restriction • Stratification
• Matching • Multivariate analysis
• Randomization
• Experimental studies nie.gov.in

How to evaluate associations?

BASIC COURSE IN BIO-MEDICAL RESEARCH

Chance

Selection bias
Information bias

Confounding
nie.gov.in

9
 11/28/2019

Does coffee increase risk of heart attack?

BASIC COURSE IN BIO-MEDICAL RESEARCH

• Truth in Universe • Truth in Study
• Population: • Sample:
All Adults Selection bias Consenting adults
Low participation rate
Hospital patients
• Actual Coffee
intake Information • Reported Coffee intake
• Actual heart bias
• Reported/misdiagnosed
attack (MI) heart attack (MI) nie.gov.in

Does coffee increase risk of heart attack?

BASIC COURSE IN BIO-MEDICAL RESEARCH
• Was the association between coffee and MI due
to CONFOUNDING by smoking?
• “A confounder is associated with both the exposure
(coffee) and the outcome (MI).”
• Smoking in
• coffee (+) 86%
• coffee (-) 27% Coffee ? MI
• MI (+) 80%
Smoking
• MI (-) 40%
nie.gov.in

10
 11/28/2019

Bias can occur in Bias can occur

BASIC COURSE IN BIO-MEDICAL RESEARCH

all epidemiological during study design,
studies implementation and
analysis

Selection bias Biases threaten

Information bias internal and
Confounding external validity

Understand biases Prevent biases

you cannot avoid - you can avoid
nie.gov.in
limitations

BASIC COURSE IN BIO-MEDICAL RESEARCH

nie.gov.in

11