1

Epidemiologic Study Design

- Cohort and RCT

Pekka Nuorti MD, PhD

Professor of Epidemiology
Tampere University, Finland

Hung Vuong Hospital, Ho Chi Minh City, Vietnam

December 7, 2023
STUDY FORMAT
Non-experimental
• “Observational” studies
• Case-control, cohort

Experimental (intervention)
• Randomised trials
• Random allocation
• Intervention as generated exposure
• Type of intervention
• Preventive
• Diagnostic
• Therapeutic
NON-EXPERIMENTAL STUDIES
‘Natural experiment’
• “Quasi-experimental” try to emulate an experimental study
• Utilize circumstances
• Exposure distribution ‘given’
• Can be selected!

No random allocation
• Comparability crucial issue for maximal validity, minimal bias
• Balance of extraneous factors (confounding)
• Comparability of populations
• Comparability of information
AIMS
Descriptive
• Simple quantification
• No hypothesis, no comparison
• Particularistic: Bound by time and place
• Not generalisable
• For administrative, planning purposes

Analytical
• Comparison: Exposed vs non-exposed
• Index vs. referent
• Exposure contrast: None vs any, different amount…
Grimes&Schulz
Lancet 2005
 Classification of study designs

Study Descriptive/ Retrospective/ Experimental/

design analytical Prospective Non-experimental

Cross- Descriptive and Single time Non-experimental

sectional analytical point

Case- Analytical Retrospective Non-experimental

control

Cohort Analytical (or Retrospective/ Non-experimental

Descriptive) Prospective

Intervention Analytical Prospective Experimental

Bhopal, 2002
POPULATION CAN BE
DIVIDED IN TWO WAYS

Assessment of disease Assessment of exposure

Disease No disease Exposure No exposure

 2X2 TABLE

Exposure No exposure

Disease a b a+b

No disease c d c+d

a+c b+d Disease No disease

a+b c+d

Exposure No exposure
a+c b+d
 COHORT STUDY
1. Obtain the column totals

2. Break the total down to

components

Exposure No exposure

Disease a b a+b

No disease c d c+d

a+c b+d

1. Define study populations: Who should be included?

Which groups are compared?
Exposed (a+c) Non-exposed (b+d)
2. Obtain information on outcome of interest in the two groups
For a+c: a=? c=? For b+d: b=? d=?
CASE-CONTROL STUDY
1. Obtain the row totals

2. Break the row total down to

components

Exposure No exposure

Disease a b a+b Cases

No disease c d c+d
Controls
a+c b+d

1. Define study populations:

Who should be included? Which groups are compared?
Cases (a+b) Controls (c+d)
2. Obtain information on exposure to the agent of interest in the two groups

For a+b: a=? b=? For c+d: c=? d=?

POPULATION

Grey: Non-exposed
Red: Exposed

Circle: No disease
Square: Disease

60 grey circles
8 red circles
12 grey squares
8 red squares
 WHO WOULD BE INCLUDED
IN A COHORT STUDY?

Blue: Non-exposed
Red: Exposed

Circle: No disease
Square: Disease

60 grey circles
8 red circles
10 grey squares
8 red squares
1. Exposed group: All 16 red shapes

2. Non-exposed reference: Sample grey shapes

3. Follow-up for disease occurrence
WHO WOULD BE INCLUDED IN A
CASE-CONTROL STUDY?

Grey: Non-exposed
Red: Exposed

Circle: No disease
Square: Disease

60 grey circles
8 red circles
10 grey squares
8 red squares

1. Cases: All 18 squares

2. Controls: Sample circles
3. Assess exposure among cases and controls (how many red, grey)
 COHORT STUDY

Start of study

Exposed Outcome +
Study population subjects Outcome -
(cohort)
Non-exposed Outcome +
subjects Outcome -

Study direction
Time
COHORT STUDY: EXAMPLE MOBILE
PHONE USE AND BRAIN TUMORS

Start of study

Exposed: Brain tumor

Cohort Mobile phone No brain tumor
of children
Non-exposed: Brain tumor
No mobile phone No brain tumor

Study direction
Time
COHORT STUDY

 Exposure-based study
 Primary task: Evaluate exposure
 Form groups on the basis of exposure
 Divide population exposure strata
 Measure disease frequency among
exposed and non-exposed
RATIONALE
Hypothesis, study question
• Clear and detailed definition of exposure and outcome
Full information extracted from study base
• All subjects fulfilling eligibility criteria
• Often larger sample size than case-control studies: More
subjects without the outcome
Start with exposure assessment
• Minimum: Exposure at baseline (entry)
• Preferable: Exposure history up to baseline
• Optimal: Over time, also during follow-up
Then assessment of occurrence of end-point
• Over time: Follow-up
COHORT STUDIES
Comparison of outcome across exposure
groups
Selection based on exposure status
• Two or more groups
• Qualitative and quantitative difference
Follow-up to assess occurrence
• Possibility to evaluate several end-points
Closed population as a paradigm
Synonyms: Follow-up/longitudinal study
• Incidence/mortality study
BASELINE

Define population (setting)

Identify subjects (selection, sampling)
Assess eligibility: At risk for outcome
Obtain consent
T0, start of the study
• Start of follow-up, baseline, entry
Assess exposure and other determinants
Ascertain availability of follow-up information
EXPOSURE ASSESSMENT
Classification of population-time
Both needed for estimation
Classification of events of outcome occurrence

A subject may contribute to several categories at

different times
Switch from non-exposed to exposed (rarely also
vice versa)
COMPARISON
Internal comparison
• Within cohort: Recruitment of both groups for the study
• Similar method (follow-up): Comparability of information
External comparison
• Outside cohort
• General population: Standardised mortality/incidence ratio
(SMR, SIR) – observed vs expected numbers of cases
• Source of information: comparable?
Comparability of populations? Selection into study population
Comparability of information? Similar coverage/completeness in
the data sources
COMPARABILITY OF POPULATIONS

Reference group represents index group in the hypothetical

absence of exposure
Index group would have the same outcome if there had been
no exposure
Counterfactual model: Not actual, counter to the fact
Exchangeability: If the members of the groups (exposed
vs non-exposed) were switched, results would remain unchanged
Exposure is the only difference between the groups
Comparability of populations: Lack of selection bias
COMPARABILITY OF INFORMATION

Similar coverage and quality of information for both groups

Same sources and procedures
Outcomes equally likely to be captured for both groups
Unaffected by exposure
Sensitivity and specificity
Comparability of information: Lack of information bias
FOLLOW-UP
Observation of events
• Obtain information
• Classify
Measurement of occurrence
• Relate events to population size or time (follow-up time, person-
years))
• Incidence proportion, incidence rate
Estimate outcome
• Relative effect: Occurrence ratio (risk/rate ratio, RR)
• Absolute effect: Occurrence difference (risk/rate difference, RD)
FOLLOW-UP (2)
Information collected on
• Exposure(s) of interest
• Extraneous determinants (potential confounders)
• Follow-up status: Death, emigration
• End-point(s): Several outcomes can be studied
FOLLOW-UP (3)
Active follow-up: Generated for the study
• Actual measurements
• Self-report
• Frequent contacts with the subject needed, e.g. annually

Routine records: Existing data

• Registers or databases, hospital records, death certificates
• Environmental measures
• Feasible also retrospectively

Exit criteria
• End of follow-up time
• Loss to follow-up (emigration, death, etc.)
• Outcome event
TEMPORAL DIRECTION
Prospective Concurrent: Record events as they occur
• Real-time measurements
• Possibility for repeated exposure assessment

Retrospective Historical: Obtain information on past events

• Records, registers
• Sources outside the study, not produced specifically for it
PROSPECTIVE VS. RETROSPECTIVE
COHORT STUDY

Now

Prospective
Exposure Outcome

Retrospective
Exposure Outcome

Time
EFFECT MEASURES

Relative effect
• Incidence proportion ratio (risk ratio): RR=R1/R0
• R1 risk among the exposed, R0 risk among the non-exposed
• Incidence rate ratio, mortality rate ratio: RR=R1/R0
• R1 rate among the exposed, R0 rate among the non-exposed
• Hazard ratio
• Standardised incidence ratio (SIR), standardised mortality ratio (SMR)
• Observed number of cases vs expected number (assuming rates similar to
the base population)

Absolute effect
• Risk difference: RD=R1 - R0
• Rate difference: RD=R1 - R0
FOLLOW-UP: CALENDAR TIME

2010 2011 2012 2013 2014 2015 2016

FOLLOW-UP: AGE

30 40 50 60 70 80 90
FOLLOW-UP: STUDY TIME

0 1 2 3 4 5
FU start Dg Death

Person-time for incidence study

Person-time for mortality study

Person-time for survival study

 Cohort study
STRENGTHS
+Temporal relation between
exposure and outcome clear
+Direct observation of disease
occurrence
+Direct calculation of relative and
absolute effects measures,
impact measures,
+Several outcomes can be
studied simultaneously
+Suitable for rare exposures
+Little selection bias if
prospective
WEAKNESSES
−Requires large sample size
for rare diseases
−Prospective studies slow and
expensive
−Loss to follow-up may lower
power and induce selection
bias
−Control of confounding
crucial
−Information bias a concern
 The Nurses’ Health Study

OC use Cases Person- Rate per Rate ratio

of years 100 000 (95% confidence
breast ca. pyrs interval)

Never 240 128 528 187 1.00

(reference group)
Past use
< 48 months 106 54 080 196 1.05 (0.84-1.32)
> 48 months 86 36 039 239 1.28 (1.00-1.64)

Example from dos Santos Silva 1999

BRITISH DOCTORS’ STUDY
Subjects identified from British medical register in 1951
Questionnaire about smoking habits
34,439 men and 6,194 women
Renewed 1957, 1966, 1972, 1978, 1990
2/3 deceased by 1990
Causes of death
 BRITISH DOCTORS
MORTALITY

Cause of Non-smokers Current

death smokers
Lung cancer 14 209
Ischemic heart 572 892
disease
Rate ratio Rate difference
Lung cancer 209/14=15 209-14=195
Heart disease 892/572=1.6 892-572=320
SMOKING AND SURVIVAL
100
90
80
70
60
50 7.5 yrs
40
30
20
10
0

Never Current Former (Q<35)

 SMOKING AND MORTALITY
Respi- Vas- Neo- All
ratory cular plastic

Never 107 1037 305 1706

Former 192 1221 384 2113

Current 313 1643 656 3038

Doll et al. BMJ 1994

EVALUATION OF A COHORT STUDY
•What was the source population of the study?
–Was a comprehensive population roster available?
–What were exclusion criteria?
•What was the exposure of interest?
–How was it assessed?
•Sources of information
•Timing of assessment
–Comment on validity
•Sensitivity, specificity; quality control
•Availability of information?
–How was the exposure contrast formed?
•Comparability of populations?
•Exposure distribution
EVALUATION OF A COHORT STUDY

What was the outcome of interest?

–Were cases prevalent at baseline excluded?
–How was the outcome evaluated?
•Sources of information, timing
•Completeness of follow-up, validity
•Comparability of information
–What was its occurrence?
•Comment on power
•Comparable to other populations?
EVALUATION OF COHORT STUDY

•What was the main result?

–Effect measure
–Dose-response?
–Consistent with crude results?
–Consistent across sub-groups?
–Residual confounding?
–Direction and magnitude of selection bias?
Information bias? Misclassification?
INTERVENTION
STUDIES
RANDOMISED TRIAL
A study where people are allocated randomly to receiving a
particular intervention or not
• Two different interventions, or an intervention vs. no
intervention (possibly placebo)
WHY RCT?
Random allocation ensures comparability of groups
Eliminates bias from treatment assignment
• Removes selection bias (ensures comparable case mix)
Balances known and unknown differences between groups
on average
• Minimises confounding
Any observed difference attributable to intervention

BUT this is not automatic - can be achieved only if a trial is

carried out with vigorous adherence to key principle
PICO(T): SUMMARY OF STUDY
QUESTION
Patients/Population: Who?
Intervention: What was the experimental procedure?
Comparator/Control: What did the control group receive?
Outcome: Which end-point was used?
Time: How long follow-up, when was the outcome evaluated?
 RECRUITMENT AND
ALLOCATION

First stage
A sample is selected from a defined population and
consent obtained.

Second stage
This sample is randomly assigned to intervention
and comparison groups.
 Eligible
Population

Recruitment

Study
population

Randomized
assignment
Intervention Comparison
arm arm
RCT SETUP
Population

Inclusion
Criteria

Sample

Consent

Baseline Follow-up
Assessment assessments
INTERVENTION TRIAL
- ETHICAL PREREQUISITES
Study should give useful information
• Genuinely open issue addressed (equipoise)
• Either one of the interventions can be better than the other
• Effectiveness, complications, costs
Primum est non nocere
• Do no harm!
• Safety
• Risks should be known in advance
Comparability of interventions
• The control arm should receive standard care (which may be no
intervention) what would have been offered outside the trial
Informed consent
It may also be unethical not to perform the study!
THE LAW OF LARGE NUMBERS

When a random sample is drawn from a large

population, and
when the sample size increases,
the average of any characteristic in the sample
will tend to become closer to the expected
value.
When the number of units in the sample
increases, on average the sample will
increasingly resemble the original population.
LAW OF LARGE NUMBERS
FOR RANDOMISATION
Only substantial numbers of repeated randomisation will
results in comparable distributions
Cluster randomization: Randomise groups rather than
individuals
Note: Number of randomized units important, not the
numbers of subjects within such units
BASIC TRIAL DESIGN

Intervention Outcome

Enrolment Randomization Blinding

Control Outcome
RANDOMISATION
A random process should be used to generate treatment
allocations or assignments
Treatment allocations should be concealed until the time of
randomization – “allocation concealment” is critical to
prevent selection bias
• Neither investigator nor subject knows the intervention to be
assigned at the time the patient is registered
• Prevent tampering, manipulation
Allocation procedures: 1:1 sequence
• Toss a coin: heads vs. tail
• How to document, prevent possibility of multiple tries?
• Random number table: Odds vs. even
• Random number generator: <0.5 vs >0.5
ENSURING ALLOCATION
CONCEALMENT

Computer-generated allocation: (pseudo)random numbers

• Centrally upon recruitment, after registering
• Best, least prone to exposing the allocation
Envelopes
• Generated in advance, delivered to recruitment centres
• Could be manipulated or opened in advance
Allocation based on fixed features
• Sequence: Every other subject
• Unit A uses x and Unit B uses z
• Date of birth, dg, start of treatment…
• Other characteristic
• Not true random allocation!
KEY REQUIREMENTS OF A
RANDOMIZATION SCHEDULE

1. Unpredictability (allocation concealment)

2. Approximate balance (desired allocation

ratio) within strata

3. Well documented; reproducible

Note: Does not maintain comparability AFTER

randomisation, only at entry (baseline)
BLINDING
Keeping allocation secret, unknown
Unaware of which arm a subject belongs to
Intervention or control/placebo
Single blind - participants are not aware of assignment/arm
Double blind - both participants and investigators unaware
Triple blind - various meanings
• Persons in charge of intervention
• Outcome adjudicators
• Data analysts
• Safety monitoring group
WHY BLINDING?
To avoid information bias due to differential assessment of
outcome
Study staff who decide if a change or outcome has occurred
may
• classify similar events differently in treatment groups
Problematic with “soft” outcomes
• Investigator judgment
• Participant reported symptoms, scales
Should report WHO was blinded and
HOW it was done
CHALLENGES FOR BLINDING
May be impossible
• Surgery, diet, exercise, education etc
May be dangerous, cumbersome
Identical placebo difficult to prepare
Drug may smell, taste, feel different
Drug may cause side effects
Test results may unblind (reveal assignment, trial arm)
Participants may test drug

Always useful to assess degree of unblinding

• ask participants to guess treatment
• ask study staff to guess treatment
COMPLIANCE
o Not everyone assigned to the intervention will
actually receive the treatment (non-compliance,
non-adherence)
o Refusal
o Some of those assigned to control arm may still
receive the intervention (non-compliance)
o Contamination (spill-over)
INTENTION TO TREAT PRINCIPLE

Principle:
• Once a patient is randomized, s/he should be analyzed in the
group randomized to - even if they discontinue, never receive
treatment, or crossover.
Grouping in the analysis follows random allocation
• Maintaining the original allocation
Exception: If patient is found on BLIND
reassessment to be ineligible based on pre-
randomization criteria
 INTERVENTION STUDY,
ANALYSIS

Per-protocol Intention-to-
analysis treat analysis

Non-participants Excluded Included

Contamination Excluded Included
Not complying with protocol Excluded Included
HIGH QUALITY RANDOMIZED TRIALS

Tamper-proof randomization
Blinding of participants, study staff, lab staff,
outcome ascertainment and adjudication
Adherence to study intervention
Complete follow-up
Adequate power
CONSORT DIAGRAM: FLOW OF RCT
 VALIDITY

QUESTION: Recruitment
Participants
Allocation
concealment?

Intervention Group (I)

& Comparison Group (C)
I C Maintenance of
comparable groups:
Treated equally?
Compliant?
+ -
Outcome + A B Measurements:
blind? Subjective or

-
objective?
C D
LIMITATIONS OF RCTS
Ethical issues
Can only address a narrowly defined, rigid research question
Expensive
Time consuming
Generalizability: How restrictive inclusion/exclusion criteria?
Efficiency vs. effectiveness: Maximal effect in specialised
centres vs. average effect across a wide-scale of practices
Real-life effectiveness