Transcript Intervention Trials Principles of study

Intervention Trials
Principles of study design
James B. Spies M.D.MPH
Professor of Radiology
Georgetown University School of Medicine
Washington D. C.
Categories of Studies
Users Guide to Medical Literature

Studies answer one of 4 types of questions:

Therapy (efficacy)


Harm


Cohort or case control studies
Diagnosis


Randomized trials, other comparative studies
Diagnostic studies comparing new study to reference
standard
Prognosis

Cohort studies
Efficacy Trials

What type of study design?


Case reports
Case series (+/- historic controls)



Retrospective or prospective
Prospective comparative studies w/o
randomization.
Randomized comparative studies
Non-randomized design
Case Series- Weaknesses

Selection bias

The investigator controls recruitment and may
enroll those who are better candidates for the
intervention compared to the typical patient.

Does not allow for control of any known or
unknown confounders in subjects.

Often done retrospectively, in which incomplete
medical records are abstracted.

Does not allow for clear cut definition and recording of
baseline characteristics, technical aspects of therapy, or
outcome.
Non-randomized design
Case Series

No controls provides no means of assessing study
impact.

Before and after within subject control does not allow
assessment of placebo effect (if not blinded) and any
improvement that may occur without intervention.

Historical controls subject to even greater selection
bias

Research staff collecting data are typically not
blinded and significant assessment bias may be
introduced.
Non-randomized design
Case Series

Investigators not blinded

Researchers may have vested financial interest in
positive outcome.

Investigators may be on company speaker’s bureaus, be
consultants, or on medical advisory board.


This is why investigators required to report conflicts at
presentations and in publications. Red Flag.
May have a personal vested interest


The study is usually done with the intent and the hope
that the new therapy is an improvement over current
therapy.
New therapies become associated with the investigator,
potentially leading to a bias supporting positive
outcomes.
Non-randomized design
Non-randomized comparative prospective studies





Improved opportunity for parallel assessment
of subjects and contemporaneous controls.
Data definitions and collection can be more
complete
Can compare subjects and controls for
similarities.
Selection bias still key problem.
Asessors, patients usually not blinded.
Randomization

Controls selection bias


Best means of assigning the known and
unknown potential confounders in each arm.



Primary benefit
Factors that can impact outcome are randomly
distributed.
Even when unknown, confounders are controlled
for.
Allows for application of statistical models.
Randomization methods


Coin toss: subject to bias
Random number tables or computer program



Computer-generated less subject to bias.
Should be done by a third party
For procedure trials where physician not
blinded, assignment either done in central
call-in center or put in opaque envelopes and
revealed to investigator at last possible time.

Prevents investigator from gaming assignment.
Randomization schemes

Simple


Each assignment is independent of the previous
and the next.
Blocked

Blocks of assignment of varying lengths.


Blocks of 2, 4, 6, 8 mingled.
Provides for balanced assignment over a fixed number of
participants.


May be used to avoid very uneven numbers of patients in
each treatment group. Most important in small studies.
Length of blocks varied and secret- prevents guessing as
to which assignment is next.
Randomization schemes

Stratification

Dividing the assignment groups by some
important clinical parameter.




Race, age group, gender, study center, disease severity,
etc.
Assures equal numbers among important subgroups.
Usually only 2 groups or 3 groups stratified.
Can combine stratification and blocking

Will have different blocking groups for each subgroup.
Randomization schemes

May use unequal allocation of subjects



1:2 or 1:3
Helpful when one treatment inherently more
attractive to potential subjects
Applicable when a well-studied standard therapy
is used as control.

There may less need for data regarding safety on
standard therapy than the new therapy, so fewer
standard therapy patients may be needed.
Blinding

Extremely important to blind or mask as many
of the key participants as possible as to
assignment.



Physicians, assessors, nurses, patients.
All can contribute biases to the outcome
assessment.
Internal validity depends on removing
selection bias (randomization) and removing
assessment bias (blinding).
Therapeutic study designs

Parallel


Two groups randomized, treated and followed as
two groups.
Crossover


Two groups randomized, each starts one
treatment and then at a designated time point,
they crossover and receive the other treatment.
May need a washout period prior to starting drug
or therapy 2, to allow for effects of drug one to
wear off.
Choice of control

Placebo




Relatively easy to implement in drug trials.
Easy standard for approval or acceptance
of a new drug.
May not be useful if other effective similar
drugs are already approved.
Most recent standards for ethical conduct
of trials bar use of placebos
Choice of control

Active controls

Most ethical approach is a current standard of
care (if one is available).

Raises problems for analysis

Should the new treatment be better than the old?


May not be feasible if gold standard is highly effective.
It may be new therapy is less invasive, less expensive, or
otherwise advantageous, but is not better.
 May not be necessary to show superiority
Trial Design

Superiority, Equivalence, or Non-inferiority


Superiority ideal if a significant improvement over
conventional therapy anticipated.
Equivalence difficult to show


Need to have minimal difference between groups, which
requires large number of subjects.
Non-inferiority trials- the new therapy is not worse
than current.



Allows investigators to decide the margin that is
acceptable for non-inferiority.
Could be a margin of 10 or even 20 percent less
effective, but still non-inferior. Should be defined a priori
Allows for smaller studies.
Power Analysis

Depends on:


Superiority,equivalence, and non- inferiority
Difference between the two groups that will cause
the null-hypothesis to be discarded-i.e. what
difference do you want to detect.


What is the variability within each of the groups.


Usually estimate what is the inherent difference likely to
be between the two groups.
The standard deviation
Type of outcome measure

Continuous variable, dichotomous, survival time, etc.
Power Analysis
Also depends on:

Allocation ratio



1:1, 1:2 etc.
Defined acceptable power to detect a
difference (usually 80%).
Defined acceptable rate of detecting a
difference in error (usually 5%).
Study Design Considerations
Reporting results

Major journals have adopted standards
for reporting results of randomized
trials, the CONSORT Statement*


Any of these are applicable to any
therapeutic trial, even when not
randomized.
Our literature could be substantially
improved
*Maher D, Schulz KF, Altman D. The CONSORT Statement: Revised recommendations for
Improving the quality of reports of parallel-group randomized trials. JAMA 2001;285:1987-1991
Study Design Considerations
CONSORT Statement




Participants eligibility stated at the time
of study design, along with exclusion
criteria.
The interventions described in detail.
Primary and secondary outcome
measures defined.
Statistical analysis defined prior to start
of study
•Maher D, Schulz KF, Altman D. The CONSORT Statement: Revised recommendations for
Improving the quality of reports of parallel-group randomized trials. JAMA 2001;285:1987-1991
Reporting Results
CONSORT Statement

Flow of patients through study provided
in flow chart.



Number screened, qualified, randomized,
treated, and assessed at each stage of
follow-up.
Dates of recruitment provided
Baseline demographic and clinical
charateristics provided.
Study Design Considerations
CONSORT Statement

Intention to treat analysis




Based on how randomized, regardless of whether
actually treated as randomized.
A key component of internal validity: It is inherently
unbiased.
Provides an insight into the desirability of the two
treatments- non-adherence is not random.
If use “as treated” analysis, will be excluding
patients who may be different from those who
proceed.

Their analysis will be distorted
Study Design Considerations
CONSORT Statement


Adverse events reported in both groups.
Ancillary analyses take into account
multiplicity of analyses.


Less weight put on conclusions from secondary
analyses.
Discussion includes:



Interpretation of results and generalizability
Discussion of limitations, potential biases.
Discussion of the evidence in context of other
studies, current knowledge.
Summary

Key aspects of internal validity in a
therapeutic trial:





Population described, flow of subjects recorded.
Power analysis to determine proper number of
subjects.
Random allocation.
Blinded assessment of outcome.
Intention to treat analysis.
Summary



Randomized trials need to become the
standard by which we evaluate new
therapies.
Usually should be assessed compared to
current standard therapy.
CONSORT Statement guides reporting for
most journals, but is a useful tool to for
designing and reporting even nonrandomized studies.