1954 Salk polio vaccine trials
► Biggest public health
► Polio epidemics hit
U.S. in 20th century
► Struck hardest at
► Responsible for 6% of
deaths among 5- to 9-
Number of polio cases in the U.S.
1930 to 1955
1930 1934 1938 1942 1946 1950 1954
1932 1936 1940 1944 1948 1952
Salk vaccine trial: Background
► Polio is rare but the virus itself is common
► Most adults experienced polio infection without
being aware of it.
► Children from higher-income families were more
vulnerable to polio!
► Children in less hygienic surroundings contract
mild polio early in childhood while still protected
from their mother’s antibodies. They develop
► Children from more hygienic surroundings don’t
develop such antibodies.
Salk trial: The need for testing
► By 1954, Salk’s research with a vaccine looked
► Government agencies were ready to try the
vaccine in the general population but some
scientists feared the vaccine was unsafe or
► There was enormous fear and desperation
throughout the country.
► Why not just distribute the vaccine to some and
see if it lowered the polio rate?
A yearly drop might mean the drug was
effective, or that that year was not an
► Vaccine could not be distributed without testing
Salk vaccine trial:
The need for controls
► An experiment requires controls.
► To test if the vaccine was effective the only variable that
should be considered is the vaccine itself
► This means that some children would get the vaccine and
some would not.
► This raises enormous ethical questions:
Is it ethical to not give children the vaccine?
Imagine yourself as a parent in these desperate times.
Would you participate in such an experiment.
Ultimately, does the benefit to society outweigh the risk
to those children who would not get the vaccine?
The need for massive trials
► Polio rate of occurrence is about 50 per100,000
► Suppose the vaccine was 50% effective and
10,000 subjects were recruited for each of the
control and treatment groups
You would expect 5 polio cases in control group and
2-3 in treatment group
Such a difference could be attributed to random
were needed on a massive scale
► Clinical trials
► The ultimate experiment involved over 1.6
million children, with over 600,000 children
Controversy over the
design of the experiment
► In order to isolate the vaccine as the only variable to be
considered, the treatment and control groups need to
be as similar as possible
► But how should subjects be recruited?
► Fact: volunteers tend to be better educated and more
well-to-do than those who don’t participate
► In the context of the polio disease, relying on volunteers
could potentially bias the results
Subjects would tend to have higher rates of polio
Subjects are not representative of the population
Results would be biased against the vaccine
► After much debate, the trials proceeded with two different
“Observed Control” approach
► Administer the experiment to 1st, 2nd, and 3rd graders
► Offer the vaccination to 2nd graders
This group would rely on volunteers (parental consent)
► Use 1st and 3rd graders as control group
These children would be observed for incidences of polio
► Supporters of this approach argued that there would not
be much variability between grades so treatment and
control groups would be similar
► And the control group would be “observed controls”
► But there were objections . . .
NFIP Observed Control study
► Volunteers would result in more children from higher
income families in treatment group
Treatment group is thus more vulnerable to disease than
Would expect more incidences of polio in the treatment
group than in the control group
Biases the experiment against the vaccine
► How would incidents of the disease be diagnosed?
Many forms of polio are hard to diagnose
In making the diagnosis physicians would naturally ask
whether a child was vaccinated or not
Diagnosis for borderline cases could be affected by
knowledge of what grade the child was in and whether the
child was vaccinated or not
Randomized control approach
► This experiment relied on volunteer subjects overall.
► But subjects were randomly assigned to treatment and
► Control group was given a placebo
► Placebo material was prepared to look
exactly like the vaccine so subjects didn’t
know what treatment they were getting
► Placebo-control group guards against the
► Many objected to the design on ethical
► Jonas Salk himself called it “A `beautiful’ experiment over
which the epidemiologist could become quite ecstatic but
which would make the humanitarian shudder.”
Randomized control approach
► Subjects were “blind”: they did not know to which
group they were assigned
► Also, those doing the evaluation
didn’t know which treatment
any subject received
► Each vial was identified by a code
number so no one involved in the
vaccination or the diagnostic
evaluation could know who got
► Experiment was double-blind:
neither subjects nor those doing
the evaluation knew which
treatment any subject received
Results of vaccine trials
The randomized, controlled experiment
Size Rate (per 100,000)
Treatment 200,000 28
Control 200,000 71
No consent 350,000 46
The Observed Control study
Size Rate (per 100,000)
Grade 2 (vaccine) 225,000 25
Grade 1, 3 (control) 725,000 54
Grade 2 (no consent) 125,000 44
Source: Thomas Francis, J r., “An evaluation of the 1954
Poliomyelitis vaccine trials---summary report,” American Journal
of Public Health vol 45 (1955) pp. 1-63.
Comparing the two studies
► Results show that the observed control study was biased
Treatment group got the vaccine but was more prone to higher
Control group didn’t get the vaccine but was more prone to lower
► It’s impossible to determine what’s the effect of the
vaccine and what’s the effect of socio-economic status
► This is called confounding—the inability to distinguish the
separate impacts of two or more variables on a single
► In a randomized controlled experiment, by making the
treatment and control groups as similar as possible (by
randomization), we are able to isolate the variable of
interest and eliminate confounding
Comparing the two studies:
are the results “significant”?
► In the “observed control” approach, chance
enters the study in an unplanned and
haphazard way based on what families will
► By contrast, for the randomized controlled
experiment chance enters the study in a
planned and simple way
Each child has 50-50 chance to be in the
treatment or control group
► Thisallows for the use of probability to
analyze the results
Are the results significant?
► Twocompeting positions—which side would you
Pro: “The vaccine is effective. There were less cases
of polio in the treatment group than in the control
group. We should undertake a massive vaccination
program throughout the general population.”
Con: “We are not convinced. The two groups were
randomly divided. There may have been fewer polio-
prone people in the treatment group. It was all done
by chance. We can’t be sure and we’re not willing to
commit millions of dollars of taxpayer’s money on a
vaccination program that might not be effective.”
Are the results significant?
► Assume the cons are right and that the
vaccine is worthless. What are the
chances of seeing such a large
difference in the two groups?
► Imagine a “polio” coin where the
chance of heads is equal to the
chance that a person gets polio.
Flip the coin in Room A for 200,000 times. Then flip it
in Room B for 200,000 times. What’s the chance that
we would get such a large difference as 28 heads in A
and 71 heads in B?
► They are over a billion to one against!
► In the face of such odds, we say that the outcome is
statistically significant. The effect is so large that it
would rarely occur by chance.
Salk vaccine trials aftermath
► The results, announced in 1955, showed good statistical
evidence that Jonas Salk's vaccine was 80-90% effective in
preventing paralytic poliomyelitis.
► Postscript: Polio was virtually eliminated from the
Americas in 1994, but still circulates in Asia and Africa,
paralyzing the world’s most vulnerable children.
► The Global Polio Eradication Initiative was begun in 1988.
That year, an estimated 350,000 children were paralyzed
with polio worldwide.
► In 2004, polio cases had fallen to just over 1,200 cases
The language of experimental design
In an experiment, we have at least one explanatory
variable, called a factor, to manipulate and at least
one response variable to measure
The specific values that the experimenter chooses
for a factor are called the levels of the factor.
A treatment is a combination of specific levels from
all the factors that an experimental unit receives.
The ability to manipulate factors, apply treatments,
and compare the responses is what differentiates an
experiment from an observational study
Nurses Health Study often in the news
– Over 100,000 registered nurses aged 30 to 55 have been
followed for more than 30 years
– Detailed questionnaires sent out every two years on a wide
variety of health and nutrition issues
– 90% response rate
– “One of the most significant studies ever conducted on the
health of women.” -- Donna Shalala, Former Secretary of
the U.S. Department of Health and Human Services
This is a prospective study. Subjects were identified
in advance and data collected as events unfolded.
Many observational studies are retrospective.
Subjects are selected and their previous conditions
or behaviors are determined.
Observational studies can suffer from
confounding and lurking variables
You’ll read about this over the weekend in
“Hormone Studies: What Went Wrong?”
The ability to control and manipulate
variables and compare groups allows for
eliminating confounding and the effect of
randomized comparative experiment:
The “gold standard” of statistics
Massive clinical trials industry
Complex ethical questions for experiments involving human
– Informed Consent, Institutional Review Board, Confidentiality
Placebo effect is a fascinating area of research
– In conditions such as pain, the percent of patients responding to
placebos has been shown to be 20% to 50%.
– Reflects the amount that the body can be coaxed/empowered to
heal itself, in the absence of other active agents.
Today, few clinical trials compare against placebo. Most new
drugs are improvements over existing therapies. If an existing
medicine exists it would be unethical to deny it to subjects
Other experimental design issues:
When groups of experimental units are similar, it’s
often a good idea to gather them together into
Blocking isolates the variability due to the differences
between the blocks so that we can see the
differences due to the treatments more clearly.
When randomization occurs only within the blocks,
we call the design a randomized block design
By contrast, a completely randomized design, all
subjects have an equal chance of receiving any
Diagram of a blocked experiment
Hypertension pharmacogenetics study
• Hypertension is most prevalent risk factor for diseases
of the heart, brain and kidneys, affecting 43 million in
• Complex disease affected by physical, physiological
and environmental factors
• State-of-the-art for treatment is trial-and-error
• Less than 40% of treated patients achieve blood
pressure control (systolic blood pressure < 140)
• Ultimate goal of this study is to identify unknown
genes that influence drug response with the potential of
tailoring antihypertensive therapy for individuals
GERA Clinical Trial
• Black and white patients react differently to blood pressure
• Blocked experimental design
• Mayo Clinic – Rochester, MN
– 300 white subjects with hypertension (150 women and 150 men, ages
30 to 60)
• Emory University – Atlanta, GA
– 300 Black subjects with hypertension (150 women and 150 men, ages
30 to 60)
• Subjects had previous medications discontinued for 4 weeks;
blood pressure rose and stabilized in hypertensive range
• Hydrochlorothiazide administered for 4 weeks
• Blood pressure measured at the beginning of therapy and after
• In each group, identify 100 “best” responders and 100 “worst”
responders by change in blood pressure
BP decrease BP increase
Yr N Race Drug Race N
1 100 B Hydrochlorothiazide B 100
2 100 W W 100
GERA clinical trial
• DNA collected for each patient
• Data consists of 100,000 genetic markers called Single-
Nucleotide Polymorphisms (SNPs)
• Goal: to find an association between blood pressure response
and genetic makeup
• Ultimate goal: to find those genes that affect blood pressure
• What makes this complicated is that we have only 400
observations (the patients) and over 100,000 variables (the
• Classically in statistics we had a “few” variables and “many”
observations. As datasets become larger and more complex,
this classic paradigm is shifting and the challenges are