1. Primary Research Question
and Definition of Endpoints
Mario Chen
The Fundamentals of Biostatistics in
Clinical Research Workshop
India, March 2007
2. Research Questions
Uncertainty that the investigator wants to
resolve.
β Interesting, Novel, and Relevant:
Literature
Colleagues
Stakeholders
β Feasible: Concrete, researchable issue
3. Primary and Secondary
Questions/Objectives
Common error β Sinking ship: Avoid overloading
the study with too many objectives and too much
data collection
A single primary question around which to focus
the development of the protocol and sample size
estimates
Secondary research questions: can be related to
the primary question or to other hypotheses
4. Main study variables
Longitudinal Studies (cohort, RCT):
β Endpoint
β Outcome Measure
β Response Variable
Case-Control Study:
β Exposure Variable
β Case Definitions
Other Study Designs:
β Main Analysis Variables
5. Example
Primary Objective: To assess the
effectiveness of a new malaria vaccine
Possible endpoints:
β Occurrence of a malaria episode
β Mortality
β Occurrence of malaria related anemia
Secondary objectives and endpoints
β Acceptability
6. Framework
Objective
Objective Population
Population
Endpoint
Endpoint Individual
Individual
Summary measure*
Summary measure* Sample
Sample
Parameter
Parameter Population
Population
* Depends of primary analysis method
7. Example
Objective
Objective Vaccine effectiveness
Vaccine effectiveness
Endpoint
Endpoint Occurrence of disease
Occurrence of disease
Summary measure
Summary measure VE=1-RR (Preventable Risk)
VE=1-RR (Preventable Risk)
Parameter
Parameter 95% CI for VE
95% CI for VE
8. Desired (Required)
Characteristics
Clinical relevance (clearly reflects research
question, mechanism of action, impact on
well-being of individuals).
9. Desired (Required)
Characteristics
Single primary endpoint:
β If more than one primary endpoint is used, the
probability of getting a nominally significant
result by chance alone is increased (Type I
Error).
β If the analysis, based on several endpoints,
gives conflicting results, interpretation becomes
difficult.
10. Desired (Required)
Characteristics
Consistency. Primary endpoint must be
capable of being assessed in all subjects
consistently:
β Avoid having different endpoints for different
subjects for the same primary objective.
β Avoid having different instruments or
techniques applied for the measurement of the
endpoint.
11. Desired (Required)
Characteristics
Validity of the comparisons. Unbiased
ascertainment of endpoints across
comparison groups:
β The issue of blinding. Objective endpoints
β The Misclassification problem:
Non-differential.
Differential.
12. Misclassification (Example)
Scenario: P.falciparum malaria in children
living in holo-endemic area (EIR>100)
~60% infected asymptomatically
MOI ~ 5 strains/child on average
Clinical malaria diagnosis?
low high
Specificity scale
Fever fever+ fever+ Severe
parasitaemia high anaemia Cerebral
parasitaemia
Malaria
13. Misclassification
(Non Differential)
Issue: Low specificity (Non-differential) in
clinical outcomes
=
gross underestimation of true efficacy of
intervention
14. Misclassification
(Differential)
Issue: Low specificity (Differential) in
clinical outcomes
=
underestimation or overestimation
of true efficacy of intervention
15. Misclassification
RCT:
β Misclassification can occur both before and
after intervention. Usually non-differential due
to randomization and blinding
Cohort:
β Misclassification can occur in the classification
of exposure or disease
Case-Control:
β Misclassification can occur in the classification
of disease or exposure
16. Desired (Required)
Characteristics
Reliability. The extent to which
measurement obtained is reproducible in
repeated administrations. Lack of random
measurement error
17. Desired (Required)
Characteristics
Completeness. Ascertainment of endpoints
should be as complete as possible:
β Consequences of Missing Data:
Sample size β Loss of power
Bias β Loss of validity
β Data collection procedures and instruments
β Follow up procedures. Participant retention
18. Desired (Required)
Characteristics
Statistical Significance. Selected endpoint
should be such that it has the potential to
show clinical significance statistically:
β Clinical meaningful difference worth detecting
(Effect size)
19. Example
RCT to study the efficacy of prophylactic
doxycycline at IUD insertion1
Primary Endpoint: Pelvic Inflammatory
Disease (PID)
Secondary Endpoint: Unscheduled Visit for
an IUD-related Problem, e.g., bleeding,
pain, or discharge
1. Sinei SKA, Schulz KF, Lamptey PR, et al. Preventing IUCD-related pelvic infection:
the efficacy of prophylactic doxycycline at insertion. Br J Obstet Gynaecol 1990;97(5):412-19.
20. Estimating the Incidence of PID
for Sample Size Calculations
Government officials estimated 40%
Ob/GYN from Med School estimated 12%
We conservatively set initially at 6%
Readjusted to 4% based on pilot trial
And after all that, found 1.9% in placebo
group
Effect on power?
22. Alternative Endpoint
Definitions
Surrogate Endpoints:
β Indicator of effect in lieu of the one of
substantive interest,
e.g CD4 counts for AIDS mortality
β Rationale: Measuring effect sooner and/or for
less cost
β Highly correlated to the clinical outcome of
interest:
Biological plausibility
Trial measuring both true and surrogate endpoint
and studying their correlation
23. Alternative Endpoint
Definitions
Surrogate Endpoints (Example)
β Cardiac Arrhythmia Suppression Trial (CAST)
compared encainide and flecainide to placebo
β Trial established that the drugs were extremely
beneficial in suppressing arrhythmia
β Surprisingly to cardiologists, CAST showed that
the drugs tripled the death rate [Senn S. Statistical
Issues in Drug Development. John Wiley, 1997.]
24. Alternative Endpoint
Definitions
Composite endpoints:
β Combines multiple measurements into a single
composite endpoint using a pre-specified
algorithm
β Any one event occurs too infrequently
Sample size
Length of follow-up
β Meaningful interpretation
β Possibility of conflicting results
25. Alternative Endpoint
Definitions
Composite endpoints (example):
β Primary endpoint. Occurrence of one or more
of the following critical events associated with
severe disease:
Death
Cardiac index less than or equal to 2.2
Ventricular tachycardia or fibrillation
Pulseless electrical activity
26. Final Remarks
Choose your study endpoints (especially the
primary endpoint) carefully by considering the
desired characteristics discussed. Involve
colleagues and recent research
Define endpoints in the protocol. Rationale and
measurement procedures should be specified a
priori.
β βRedefinition of primary endpoints after unblinding
will almost always be unacceptable,β ICH 9