Transcript The Road Map to a Successful Study Design

1
The Road Map to a
Successful Study Design
Lisa Kaltenbach, MS
Biostatistician II
February 6, 2007
[email protected]
Office: D2220 MCN
2
3
Can’t go back in time in research!
“To call in the statistician after the experiment is
done may be no more than asking him to
perform a postmortem examination: he may
be able to say what the experiment dies of.”
-R.A. Fisher, Indian Statistical Congress,
Sankhya, ca 1938
4
Outline: The Road to Success
• How to begin clinical research
• Important considerations when designing a
study
• Types of study designs
• Examples
5
Components of a Study Protocol:
•
•
The structure
of a research
project is set
out in its
protocol.
Protocols are
well known
as devices
for seeking
grant funds,
but they also
help the
investigator
organize
his/her
research in a
logical,
focused, and
efficient way.
Element
Purpose
Research questions
What questions will the study address?
Significance (background)
Why are these questions important?
Design
Time frame
Epidemiologic approach
How is the study structured?
Subjects
Selection criteria
Sampling design
Who are the subjects and how will the
be selected?
Variables
Predictor variables
Confounding variables
Outcome variables
What measurements will be
made/recorded?
Statistical issues
Hypotheses
Sample size
Analytic approach
How large is the study and how will it
be analyzed?
6
What do you wish to learn?
•
•
•
If only one question could be answered by the project, what would that question
be?
Often people would like to do describe how likely a theory or hypothesis is in light
of a particular set of data. This is not possible in the commonly used
classical/frequentist approach to statistics. Instead, statistics talks about the
probability of observing particular sets of data, assuming a theory holds. We are
not allowed to say, "Because I've seen these data, there is only a small probability
that this theory is true." Instead, we say, "The probability of seeing data like these
is very small if the theory is true."
In order to show an effect exists,
– statistics begins by assuming there is no effect.
– Prior to collecting data, rules are chosen to decide whether the data are consistent with
the assumption of no effect.
– If the data are found to be inconsistent with the assumption, the assumption must be
false and there is, in fact, an effect.
•
•
Classical statistics works by comparing study data to what is expected when there
is nothing.
If the data are not typical of what is seen when there is nothing, there must be
something. Usually "not typical" means that some summary of the data is so
extreme that it is seen less than 5% of the time when there is nothing.
7
Judging a project's feasibility
• Can everything that needs to be measured be measured?
• If the study involves some condition, can we define and recognize it?
– What is an unhealthy eating behavior?
– What's the difference between a cold and the flu?
– What do we mean by family income or improved nutritional status?
• How accurate and consistent are the measurements? How accurate do
they need to be? What causes them to be inaccurate or inconsistent?
– Calcium intake is easy to measure because there are only a few major
sources of calcium. Salt intake is hard because salt is everywhere.
– Will respondents reveal their income?
– Can others get the same value (inter-laboratory, inter-technician variability)?
• How do we choose among different measurement techniques?
– Is a mechanical blood pressure cuff better than using a stethoscope?
– Is there a gold standard? Is it worth paying for?
• Sometimes merely measuring something changes it in unexpected ways.
– Does asking people to keep records of dietary intake cause them to change
their intake?
• Resources (time and money)
8
Types of design
Analytic
Experimental
Descriptive
Non-experimental
Randomized
Clinical Trial
Cohort
Non-randomized
Clinical Trial
Cross-sectional
Case-control
Other
Community Survey
9
Considerations when choosing a Study Design
• No one approach is always better than the others.
• Each research question requires a judgment about
which design is the most efficient way to get a
satisfactory answer.
• A common sequence for studying a topic:
– Descriptive studies
• How common is estrogen treatment in women after menopause?
– Analytic studies to evaluate associations and discover
cause-and-effect relationships
• Is taking estrogen after menopause associated with lower risk of
CHD?
– Clinical trial to establish the effects of an intervention
• Does hormone treatment alter the incidence of CHD?
Examples of common clinical research designs used 10
to study whether hormone therapy after menopause
prevents coronary heart disease
Study Design
Key Feature
Example
Experimental Design
Randomized blinded trial
Two groups created by a
random process, and a
blinded intervention
The investigator randomly assigns women to receive hormone
or identical placebo, then follows both treatment groups for
several years to observe the incidence of heart attacks.
Cohort study
A group followed over time
The investigator examines a cohort of women yearly for
several years, observing the incidence of heart attacks in
hormone users and non-users.
Case-control study
Two groups, based on the
outcome
The investigator examines a group of women with heart
attacks (the “cases”) and compares them with a group of
healthy women (the controls) asking about hormone use.
Cross-sectional study
A group examined at one
point in time
The investigator examines the group of women once,
observing the prevalence of a history of heart attacks in
hormone users and non-users.
Observational Designs
11
Statistical Issues
3 Step Process:
1. Define specific aim
-Hypothesis: Women who receive estrogen treatment after
menopause will have fewer heart attacks than those who do not.
2. Calculate the sample size, the number of
subjects needed to observe the expected difference
in outcome between study groups with a reasonable
degree of probability, or power.
3. Select statistical methods needed to produce an
acceptable level of precision when confidence
intervals are calculated for the means, proportions,
or other descriptive statistics.
12
Randomized Clinical Trials
• In simplest implementation:
• Subject enrolls in study
• Randomly assigned to one of ≥ 2 treatments
• Followed up until end of study or outcome measure is obtained
• Outcome comparisons are made among treatment groups
• Treatment groups should be comparable on measured and
unmeasured covariates due to randomization
• Strongest design to establish causal relationships
• May be beneficial to blind subjects/investigators to treatment
groups
13
Randomized Clinical Trial Example
14
Cohort Studies
• Exposure not randomly assigned, but assessed
– Sample selection and analysis can minimize confounding
• Need sufficient # of subjects/events
• Prospective Cohort
– Outcomes are future events
• Retrospective Cohort
– Outcomes have already occurred
• Can study multiple outcomes
• No control over risk factors, or insufficient numbers
15
Cohort Study Example
Example with
dichotomous risk factor
and outcome
16
Cross-sectional Studies
• All variables are measured at same time
• Valuable for providing descriptive information
about prevalence
• But weaker evidence for causality as predictor
is not shown to precede outcome
17
Cross-Sectional Study Example
18
Case-Control
1. Subjects are identified as cases based on
outcome status
2. Identify comparable controls (challenging)
3. Retrospectively determine prior exposure
*Big challenge to account for all differences
between cases & controls that could explain
relationship between exposure & case
status
19
Case-Control Example
20
Summary of how research works
RESEARCH
QUESTION
STUDY
PLAN
design
Target
population
Women aged
50-69
Phenomena
of interest
The proportion
who take estrogen
TRUTH IN THE
UNIVERSE
Errors
Intended
sample
All women aged 5069 seen in UCSF
primary care clinic
in one year
implement
ACTUAL
STUDY
Actual Subjects
Errors
Errors
Actual
Measurements
Intended variables
Self reported
estrogen treatment
infer
TRUTH IN THE
STUDY
FINDINGS IN THE
STUDY
21
Sampling Errors: Threaten inferences from study subjects to
population of interest
•
Random error is a wrong result due to chance – unknown sources of variation
that are equally likely to distort the sample in either direction.
– If the true prevalence of estrogen treatment in 50-to-69-year-old women is 20%, a welldesigned sample of 100 patients from that population might contain exactly 20 patients
with this disease. More likely, however, the sample would contain a nearby number
such as 18, 19, 21, or 22. Occasionally, chance would produce a substantially different
number, such as 12 or 28.
– Reduce the influence of random error by increasing the sample size. The use of a
larger sample diminishes the likelihood of a wrong result by increasing the precision of
the estimate - the degree to which the observed prevalence approximates 20% each
time a sample is drawn.
•
Systematic error is a wrong result due to bias (sources of variation that distort
the study findings in one direction).
– Using patients who come to the primary care clinic, who might be more likely than
average to adopt medical treatments. Increasing the sample size has no effect on
systematic error. The only way to improve the accuracy of the estimate is to design the
study in a way that either reduces the size of the various biases or gives some
information about them. An example would be to draw a second sample of women from
a setting that may be less likely to bias the proportion of women treated with estrogen
(e.g., employees in a corporation), and to compare the observed prevalence in the two
samples.
22
Summary
• Plan ahead!
• We all want to do research that produces
valid results, is worthy of publication, and
meets with the approval of our peers. This
begins with a carefully crafted research
question and an appropriate study design.
23
References
• Dallal website
• Hulley, SB, et all. 2001, 2nd ed. Designing
Clinical Research, Lippincott Williams &
Williams; Philadelphia, PA.
• Wikipedia