SELECTION OF

SUBJECTS FOR STUDY

Sumber :
Critical Appraisal of Epidemiological Studies and Clinical Trials
Elwood M, 2007
 Description of this session:
• Deals with how subjects are selected for
inclusion in analytical & intervention study
• In observational cohort studies, select
subjects who are exposed and less
exposed / unexposed
• In case-control design, select subjects as
cases and controls
• In intervention studies, select subjects with
intervention and subjects with no
intervention
PART 1.
PRINCIPLES OF SUBJECT SELECTION
 LEVELS OF SUBJECT SELECTION (1)

Level of selection Main exclusions

Target pop.

Source pop. Not assessed, assessed and found not

eligible, and not classified because of
inadequate data.

Eligible subjects Death, inability to cooperate, admin issues,

confidentiality, voluntary non response...
(do not enter study)

Study entrants Failure to complete study requirements,

missing data, loss to follow up (do not
complete study)
Study Participants
 LEVELS OF SUBJECT SELECTION (2)
Target pop. The population to which the results can be applied
Direction of Direction of
selection of application of
subjects results

Source pop. The population defined in general terms and enumerated if

possible, from which eligible subjects are drawn

Selection Application

Eligible pop. The population of subjects eligible for inclusion in the study;
should be defined precisely

Selection Application

Study Individuals who enter the study; should be defined and

entrants counted. All non participants should be accounted for with
reasons for non-participations
Selection Application

Study Individuals who contribute data to the study; the results apply
participants directly only to these subjects
Ex: A clinical trial assessing different
treatments in the management of Acute
Myocardial Infarction, carried out in a major
teaching hospital.

• Study participants: patients who enter the

study and provide outcome data.
• Some subjects may be study entrants, but
their outcome information is not available,
so they are not fully participants (they do
not contribute to the key analysis)
Ex: (cont'd)
• Eligible population: all patients with an
appropriate diagnosis seen at the
participating hospital, within preset limits of
age and other factors, who do not have the
various clinical contraindications that will be
defined in the trial protocol
• Source population: patients admitted to the
teaching hospital, or to the particular clinical
unit, over a certain period of time.
• Target population: patients seen in other
geographical areas, or other countries,
include patients seen at future time.
The procedure for selection of
subjects who participate in the study
can affect not only how widely the
results can be applied, but also
whether the results are in fact valid
(internal and external validity)
 Internal validity
• A measure of how confident we can be that a
difference in outcome between two groups can
be attributed to the effects of the exposure or
intervention.
• Threat to internal validity: Bias (selection bias
and observation bias), Confounding, or Chance
variation
• In designing and interpreting studies, attention to
internal validity is very important.
 External validity
• Ability to apply the results of the study to a
wider population
• Due to application of results to eligible
population, source population, and target
population.
• Useful only if the internal validity of a study
is acceptable
 SELECTION BIAS
• Regarded as any potentially misleading
effect caused by the way subjects are
selected to participate in a study
• Have two types of effects on a study
1. Affect the internal validity
2. Affect the external validity
 Effects of Selection Bias on
Internal Validity
Occur only if the selection process has
different effects on the different groups being
compared
Ex: we compare the frequency of smoking in
men and women by sending a questionnaire
to all residents in a community. The response
rate for women > men, and probability being
smokers < non-smokers. The survey will
underestimate the prevalence of smoking, and
underestimation will be greater in women.
 Effects of Selection Bias on
Internal Validity
Occurs when lost to follow up has different
effects on the different groups being
compared
Ex: we compare the frequency of smoking in
men and women by sending a questionnaire
to all residents in a community. Lost to follow-
up rate for women > men, and probability
being smokers < non-smokers. The survey
will underestimate the prevalence of smoking,
and underestimation will be greater in women.
 Effects of Selection Bias on
External Validity
• Selection criteria control the nature of the
target population and limit the external
validity.
Ex: trial of two treatments for MI is restricted
to male patients, aged < 55, and have
particular pattern of infarction. Application to
population beyond these features needs to
be justified by other evidence.
• Influenced also by the participation rate, i.e
proportion of eligible subjects who participate
in the study
 METHODS OF REDUCING
SELECTION BIASES
1. Method to reduce effects of incomplete
participation:
Reporting, and optimizing participation
and response.
2. Method to reduce effects produced by
the selection criteria:
selection effects with different study
designs.
 Reporting, and optimizing,
participation and response
• Participation rates: number of study
participants divided by the number of eligible
subjects
• Response rates: number of study
participants divided by the number of eligible
subjects who are identified, contacted, and
asked to participate
• Ideally, all studies should report on the
participation and response rates.
 Reporting, and optimizing,
participation and response (cont'd)
• Response rate reduced when sensitive
questions were included, questionnaires
begins with the most general questions, or
participants were offered the opportunity to
drop out of the study
• Methods to increase response rates: monetary
and non-monetary incentives, a promotional
message on the envelope, a more interesting
topic, pre-notification of the subjects, follow up
contact, shorter questionnaires, an assurance
of confidentiality, etc
 Selection Levels in different designs
(a) (b) (c) (d)
Target Target Target Target

source
Source
Source Source Source
Exposed Unexposed
Eligible (or case) (or control)
Eligible

Participant
Eligible Eligible Eligible Eligible
Exposed Unexposed
Exposed Unexposed (or case) (or control)
Randomize (or case) (or control

Exposed Unexposed Participant Participant Participant Participant Participant Participant

participant participant

• (a) randomized trial design  higher internal validity, reduced external validity
• (b), (c), & (d) non-randomized intervention, cohort, or case-control design
• (d) higher external validity, reduced internal validity
Selection of subjects for comparative
subjects (exposed or case)

1. Should be truly “exposed” or a “case”

2. Should be newly exposed, or be a newly
incident case
3. Should be representative of a defined
eligible population
4. Should be available for study so that
necessary information can be collected
in the same way as it is for the
comparison subjects
 Selection of the comparison
subjects (non-exposed or
control)
1. Should be truly “non-exposed” or “non-
diseased”
2. Should be available for study so that necessary
information can be collected in the same way as
it is for the exposed or case subjects
3. Should be representative of a defined eligible
population, analogous to that for the “exposed”
or “case series”; this can be modified in that
comparison subjects may be chosen specifically
to be similar to the exposed or case series with
regard to particular factors (matched)
 Selection of the comparison
subjects (non-exposed or
control)
1. Controls must be representative to the group they are
designed to be:
1. In cohort  unexposed or minimally exposed
2. In case-control  free of outcome of interest
3. Misclassification will cause the association biased towards the
null value
2. Controls (comparison) should be chosen so that
relevant information can be collected in same manner
as exposed/case
3. Ideally, comparison subjects should be representative
of the unaffected (or all) members of eligible pop that
produced case or exposed group
 Issues in The Selection of Subjects
General questions Comparison of the groups
What is the definition of the eligible Are the definitions comparable?
population?
How do the participants relate to
the eligible population?
What is the participation rate? Compare the groups
What are the reasons for losses, Compare the groups.
and how frequent is each? Are differences likely to affect internal
validity?
Do the losses compromise external validity
What is the source population? Is it the same for each group?
How do the source and eligible Is the relationship similar for each group?
populations relate? Is the main result likely to apply to the
source population?
What is the target population Is the main result applicable to this target
population?
PART 2.
APPLICATION OF THE PRINCIPLES OF
SELECTION OF SUBJECTS TO EACH
TYPE OF STUDY, AND EXAMPLES
 SELECTION OF THE SUBJECTS
FOR A COHORT STUDY
• Groups: exposed group and comparison (non-
exposed group)
• Exposed group should truly be exposed.
• Misclassification by exposure status will bias the
result of the study towards the null value
• Misclassification often severe because an indirect
indicator of exposure is used.
Ex: to assess the health effects of exposure to
asbestos, an “exposed” group of subjects who
have worked in an environment where asbestos
was used may be chosen, even though many of
them may have had little or no exposure.
 SELECTION OF THE SUBJECTS
FOR A COHORT STUDY (cont’d)
• Exposure may change over time, introducing
further misclassification
Ex: in the study of asbestos, the initial cohorts
of exposed and unexposed workers may be
set, but as the study follow up proceeds some
exposed workers will cease exposure, and
some unexposed workers will begin
exposure.
 SELECTION OF THE SUBJECTS
FOR A COHORT STUDY (cont’d)
• Subjects should ideally be newly exposed to the
causative agent under study, taken along with the
definition of the exposure
Ex: Study of frequency of cancer in workers exposed
to particular chemical through their occupation, we
might define exposure as a minimum of 5 years
occupational exposure to chemical. Being newly
exposed means that this 5-year period has just been
completed, include the total follow up period
subsequent to that time for each subject, exclude
subjects who leave the workforce or change jobs
before they have 5 years exposure.
 Choices with regard to the unexposed
group in cohort studies
1. Comparison group can be chosen as a
representative sample of the unexposed
members of the eligible population from which the
exposed subjects are also drawn
2. Comparison group should be similar to the
exposed group in terms of other factors that
determine outcome
3. Comparison group from other population (external
group) will result in weaker design. The external
group (of unexposed) should relate to a common
target group (with exposed group)
 EXAMPLE OF PROSPECTIVE COHORT STUDY (1)
1400 volunteer GPs

Exposed Group Unexposed Group

First two women each month for whom Next woman in records with year of birth = that
an OC prescription (new or repeat) was of exposed women  3 years
issued. Excluded if:
Excluded if: • Not married, or living as married
• Not married, or living as married • Ever used OC by records or interview by
• Using OC for medical treatment rather doctor (or did not know)
than as a contraceptives. • Had left practice or could not be traced in
hospital when selected

Follow up by GP using form sent from central office every 6 months

Recording
all episodes of illness newly presenting
changes in contraception
pregnancies
Termination of follow up:
death
if patient leaves practice
if GP withdraws from study
if patient obtains OCs from another source
 EXAMPLE OF PROSPECTIVE COHORT STUDY (2)
17 Participating FPA clinics

Married, age 25-39, white, British subject,

willing to participate

Exposed Group Unexposed Group

Current user of OC for 5 Current user of diaphragm or
months intra-uterine device for 5 months

Follow up record of further clinic visits, or postal questionnaire;

telephone or visit to non-responders
Recording changes in contraception
pregnancies and their outcome
hospital admissions and outpatient
cervical smears
Termination of follow up death
emigration
withdrawal of cooperation
 Comparison of Two Prospective
Cohort Studies
• Both studies fail to fulfill one of criteria set
out in selection of the exposed group:
should be newly exposed.
The exposed group of both studies were
not defined from the time of first exposure,
but were identified as a prevalent sample
of OC users
• In both studies, external validity has been
sacrificed to facilitate follow-up and
achieve good internal validity
 Retrospective cohort studies
• If the essential information for a
particular study can be obtained from
records that already exist, the
advantages of a cohort study can be
exploited without the need for the
long length of time required for
prospective follow-up
• Study on occupational groups
 EXAMPLE OF A RETROSPECTIVE COHORT STUDY

All subscribers to cell phone services in Denmark from 1 Jan

1982 to 31 Dec 1995; n = 723,421
Excluded: 200,507 corporate customers
Non-corporate subscribers; n = 522,914
Failure of linkage: errors in name
49352, address 10183, address not
residential 11687, other 14
Linkage to central population register: successful linkage
n = 451,679 (86.4%)
Further exclusions: 10679 duplicates,
17921 after 1995, 2550 aged < 18,
394 residents outside Denmark
Linkage after exclusions; n = 420,135
Requested not to be included: 53
Participants; n = 420,095 (357 550 men, 62545 women)

Linkage to central population registry for cancers incident from

date of first subscription to 31 Dec 1996

Total person-years of follow up 1,128,493 (average 3.1 years)

Total cancers observed = 2876

Expected number based on national rates = 3327.6
Standarized Incidence ratio = 0.86
 SELECTION OF THE SUBJECTS
FOR A CASE CONTROL STUDY
• Groups: cases group and controls group
• The cases should truly be cases
• Inclusion of some individuals who do not
have outcome in question will tend to
dilute the case group and bias the result of
the study towards null value
• Definition of cases was stronger through
diagnostic tests, rather than a possible
diagnosis
 SELECTION OF THE SUBJECTS FOR
A CASE CONTROL STUDY (cont’d)
• Ideally, cases should be recently diagnosed.
• Frequent error in hospital-based case-control
studies is to study only those subjects who
are still under follow-up by hospital, rather
than all patients diagnosed with the disease.
• The patients not under follow-up include
those with particularly bad outcomes, who
may have died or been admitted elsewhere,
and sometimes those with particularly good
outcomes who need not return for further
care
SELECTION OF THE SUBJECTS FOR A
CASE CONTROL STUDY (cont’d)
• Major issue: choice between a case series
chosen from hospital (hospital-based), or one
derived from community (community-based).
• Hospital case series are acceptable if a very
high proportion of those developing the
disease will come into hospital for diagnosis
or treatment.
• On hospital-based, restriction may be
accepted, but considerable care is then
needed in the generalization of the results
Choice of the control group in case-
control studies
• Control group is a representative sample
of unaffected members from the same
source population (source population may
be a community or a health care source)
• For control group chosen from a health
care facility (hospital), patients with
diagnoses likely to be associated
(positively or negatively) with the exposure
factor under assessment should not be
eligible as controls.
• Examples of case-control studies:

1. Study on the association between cell phone

use and brain cancer (Muscat et al). A
hospital-based case-control study
2. Population-based case-control study of
breast cancer in New Zealand assessing
primarily OC use effect
 EXAMPLE OF A HOSPITAL-BASED CASE CONTROL STUDY

CASES CONTROLS
Patients in one of 5 Patients in same hospitals; excluding
hospitals with primary brain lymphoma, leukimia. Daily review of new
cancer between 1994 and admissions and first eligible patient
1998, aged 18-80, n = 668 approached, matched by hospital, aged
within 5 years, sex, race, month of
admissions; n = 469

Excluded: 42 not
english speaking, 55 Half of non-
too ill, 2 deaths, 25 participants
refusals, 75 consented refusals, half
but too ill or unsuccessful
uncommunicative interviews
Total 199 excluded

Interviewed n = 469 Respons rate given as 90%

of which 43 by proxy Interviewed: 422; 6 by proxy
 Hospital-based case-control on brain
cancer associated with cell-phone use

• Because there had been previous reports linking

radiofrequency exposures and lymphoma and
leukemia, patients with these conditions were
excluded from the control group
• Response rate case group= 469/571 = 82%
• Response rate control group= 422/469 = 90%
 EXAMPLE OF A COMMUNITY-BASED CASE CONTROL STUDY

THE CASE SERIES IN A CASE CONTROL STUDY

Source All New Zealand women aged 25-54 diagnosed with breast cancer
population between 1 July 1983 and 30 June 1985
Eligible Women aged 25-54, with histologically confirm breast cancer
population notified to the NZ National Cancer Registry or to the Auckland
Breast Cancer Study group between above dates (n=739); no
previous breast cancer; on current electoral roll, whose telephone
n=550 number was found; exclusions =189, eligible population =550

Participant Had to have permission from their physician (28 not given)
population Be identified in time to allow interview 4-8 months from diagnosis
(49 too late)
Still alive (8 had died)
Well enough (4 too ill)
Exclusions (14 other exclusions)
Agreed to participate (14 refused)
n=433 Participant population 433; exclusions 117
All participants gave usable information on OC use
Participants / eligible = 78.7%
Participants / source = 58.6%
EXAMPLE OF A COMMUNITY-BASED CASE CONTROL STUDY (cont’d)

THE CONTROL SERIES IN A CASE CONTROL STUDY

Source All New Zealand women aged 25-54 and without diagnosed
population breast cancer

Eligible Women on electoral register, aged 25-54, with a telephone

population number, with no history of breast cancer: selected by random
n=1110 sampling from register; estimate of number=1110

Participant Still alive (10 had died)

population Well enough (4 too ill)
In New Zealand (12 overseas)
No Language difficulty (13 excluded)
Agreed to participate (99 refused)
Traced (75 not traced, of whom some were likely to be outside the
age range)
n=897 Participant population 897; exclusions 213
Participants / eligible = 80.8%
Participants / source = unknown
 Pop-based case-control on breast Ca, New
Zealand
• Cases could be identified from cancer registry
• Controls were more difficult to obtain:
– Use electoral register  the most complete available
population records
– Need telephone to interview
• Cases had to be comparable to control 
– Cases had to be on the electoral register and have a
listed telephone number
 Comparison of Hospital-Based and
Community-Based Case Control Study
• Both studies have considerable difficulties that
could compromise the results. One cannot say
whether hospital design is inherently better or
worse than a community design
• Case series interviewed in the brain cancer
study represents a more complete series of
cases than the case series interviewed in the
breast cancer study
• Control series in the breast cancer study is more
likely to be representative of women in the
general community