INTERVENTIO

N STUDIES:
ALSO EXPERIMENTAL
STUDIES

Intervention Studies

Also called Clinical Trials.

Are designs that provide data of high quality

like that of controlled experiments by basic
researchers.

Individuals enrolled by exposure like cohort,

however, investigators allocate exposure.

Intervention

Studies cont..:

Public Clinical Practice is to prevent or delay

immediate death or disability and improve health

Trials assist in choice of effective agent or

procedure to diminish symptoms, prevent
recurrence or decrease death.

Achieves this by evaluating both effectiveness

and side effects of new interventions

Advantage of RCTs:

If well done on adequate sample, provides high

validity results unlike observational studies.

Measures small to moderate (10 20%)

difference in results which are difficult to do
reliably by observ. studies due to uncontrolled
confounding.

They are strongest and most direct evidence

for concluding observed associations.

Design and Conduct of RCTs:

They are a Gold Standard for epid.
Research if ethics, feasibility and costs are
well addressed.
To ensure valid results, must
Select study pop. well
Allocate treatment regimens well
Maintain and assess compliance well
Achieve high and uniform rates of
ascertainment

Typical Design of RCT.

Study
Pop.

Treatment
Treatment
Allocation
Allocation
(Randomiz
(Randomiz
))

Treatment
Grp

Non Treatment
Grp

Creating Comparative Grps:

Treatment procedures (or programs) can be

compared as follows;

With another dosage of same drug

With another therapy or program
With a standard medical practice
Placebo

Randomization Process: 1
Is non subjective (objective) criteria of
allocating grps to treatment grps.
Can use
- Table of random numbers
- Or computer generated random numbers
If outcome varies by sub grps of the study
pop. e.g. by sex or stage of disease,
efficiency can be increased by ensuring
grps are approx. equal with respect to such
characteristic.

Advantages of Randomization

Provides

subjects the same chance of

getting either of the possible
treatments.

It maximizes chance that grps

receiving the differing treatments will
be comparable

Advantages of Randomization
Best

in designing a trial intended to

remove predictability thus removes overt
or covert assignment subjective bias.

Brings

grps into comparability in terms of

sex, age, etc. Matching is difficult because
we can only match for variables what we
are aware of

Advantages of Randomization

Increases chance that grps will be

comparable even for variables we cant
measure but affect prognosis.

However randomization not guarantee to

comparability cause chance still plays a
role in process.

Advantages of Randomization:

There is no prior knowledge of assignment grps,

potential for allocation bias is removed.

On average, study grps will tend to be

comparable on all variables except for those
under study.

On average means that the larger the sample

size, the more successful the randomization
procedure in distributing factors equally among
the groups.

Advantages of Randomization :

All known confounders will be equally distributed

including those unsuspected by the investigator.

If exposure is assigned by method other than

randomization, burden of proof is on investigator
to show that all possible biases, known or
unknown did not account for observed results.

Randomization provides inherent confidence in

the results that can not be achieved with any
alternative allocation process.

Types of Randomization Procedures

Can use more complex form of

randomization called Block Randomization

Subjects are classified according to each

such variable (block) before randomization
and allocation.

In large samples, randomization virtually

guarantees comparability of grps while
blocking is best when study size is limited.

Ways of Allocating Groups Controls: 1

Controls:
Historical Controls: For comparison with new
treatment, look back into records of patients with same
disease treated before new treatment was available.
Problems:
Thus difference between drugs may be due to
differences in quality of the data and way it was
collected.
Changes

in calendar times including life styles, diets etc.

Ways of Allocating Groups Controls: 2

Simultaneous Non-Randomized Controls:

Assigning patients by day of Pt admission e.g. Odd
number day, one grp and even number another grp.
Problems:
People with certain characteristics may visit
in certain days than others

more

- So assignment is predictable i.e. researcher knows

in advance the grp pt will be assigned. Hence
selection bias.

Data Collection: 1

Those designed to take a given treatment

(A or B) must actually get it as specified.

Outcome must be measured equally in each

of the groups.

Blinding (Masking) of assessors reduce this

bias.

Data collection: 2
Masking (Blinding):
Subjects shouldnt know which group they
are assigned to (particularly if outcome is
subjective e.g. headache etc)

We may mask through placebo.

When subjects and data collectors are

masked, then this is double blinding.

Data arrangement: 1

Consider grps A and B receiving different

drugs to compare their effectiveness.

4 Possible treatment effectiveness

outcomes
- That they dont differ and we correctly
conclude that they dont differ

Data arrangement: 2

That they dont differ but we wrongly

conclude that they do.

That they differ and we correctly conclude

that they do

That they differ and we wrongly conclude

that they dont differ.

Data arrangement Cont: 3

Reality (Truth)
Decision
(Study)

Treatments Not Different Treatments are Different

(Null Hypothesis)
(Alternative Hypothesis)

Conclude Treatment
Not Different

CORRECT DECISION
(a)

Conclude treatment
are different

Type I Error
(Mistake)
(c)

Type II Error
(Mistake)
(b)
CORRECT DECISION
(d)

Data Arrangement Cont..:.

4

Ideally, all decisions should fall within the

correct ones (a and d) but it is rarely ever
so.

The two treatments may not be different

but because of the samples we have
selected, we wrongly conclude that they
are different (Type I Error)

Data Arrangement
Cont..:. 5

Or the treatments are different but we wrongly

conclude they are not different (Type II Error)
i.e. the treatments are different but we fail to
detect the difference in our study.

Probability of making a Type I Error is also

referred to as error

Probability of making a Type II Error is also

referred to as error.

Conclusions: 1
error is also called P value, i.e. the
chance that what is detected as different is
actually not different.

When P < 0.05, then it tells us that the

results are not different but we concluded
that they are different probability of this
occurring by chance (and that this
difference is not true) is only 0.05 (or 1 in
20).

Conclusions cont...: 2

Decision

Treatments are
Different

Conclude
are not
Different

Type II Error
(Probability = )
(b)

Conclude
are
Different

Correct Decision
(Probability = 1- )
(d)

Conclusion cont: 3
Probability

of making a Type II error

=
Total

of all probabilities = 1.
Probability of concluding grps are
different is (1-)

Conclusion cont: 4
Thus

1- is called the power of the

study.

Power

tells us how good study is at

correctly showing a difference
between the therapies if in reality
there is a difference (i.e. how likely is
our study not to miss a difference if it
exists).

Summary 1

= probability of making Type I error i.e.

probability of concluding that the treatments
differ while they actually dont.

= probability of making Type II Error i.e.

probability of concluding that the
treatments dont differ while they do.

Summary: 2

Power = 1- ( i.e. probability of making

type II Error)
= probability of correctly concluding that the
treatments differ (making correct
decisions).
= probability of detecting a difference
between the treatments if the treatments in
fact differ.

Importance of this is in Sample

Size
Must specify expected difference in response
rate to be detected

An estimate of the response rate in one of the

groups

Level of statistical significance ()

The value of the power desired (1- )

Whether the test should be one-sided or two

sided.

Importance of this is in Sample Size cont:

E.g. If existing therapy cures 40% of

patients and we are testing new therapy.

We then say new therapy cures 50% or 60%

of treated patients i.e. Will new treatment
be 10% or 20% better than current
therapy?.

Importance of this is in Sample Size cont:

So the size of difference between current

therapy and new therapy that we want to
detect with our study is what?

If we do not have the difference, then we

can guess e.g. 30% difference e.t.c.

Maintenance and Assessment of Compliance:

Extent of non compliance depends on

length of time of intervention as well as
complexity of study protocol.

To increase compliance, the use of pple who

are interested and reliable is suggested.

Frequent contact with participants by

telephone, home or clinic visits is also
useful in increasing compliance.

Maintenance and Assessment of Compliance:

Monitoring compliance is important because

non compliance decreases statistical power
of a trial to detect any true effect of the
study treatment.

All measures available to estimate

compliance have inherent limitations
- Simplest measure is self reporting
- Use of biochemical parameters (more
expensive and logistically difficult).

Ascertainment of Outcomes:

This is issue to consider in the design and

conduct of a clinical trial.

Ascertainment of Outcomes cont..:

However, procedures for unblinding must be

there on the patients physicians when
serious side effects e.t.c. occur.

HERE

Other Forms RCTs : 1

Factorial Randomized Studies:

These test two or more hypotheses
simultaneously.
This involves using same group to test two
different drugs whose anticipated outcome
are different and modes of action are
independent.

Other Forms RCTs : 2

This actually considers the cost and

feasibility of designing clinical trials and
applies a single technique of improving
efficiency.

Can utilize a 2X2 factorial design

Subjects first randomized to grps and to

address one hypothesis

Factorial Designs cont:

Then within each grp and further

randomization to grps A and B to evaluate a
second question is done.

Further in 2X2X2 factorial design, each of

these subgroups will be further randomized
into two additional intervention grps to
address a third hypothesis etc.

CROSS OVER TRIALS.

These are of 2 types

Planned
Unplanned

Cross Over Trials - Planned.

Subjects are randomized to therapy A or

therapy B and after being observed for a
certain period of time on one therapy they
are switched to the other therapy.
Each subject serves as own control, thus
holding constant the variation between
individuals in many characteristics that
could potentially affect a comparison of the
effectiveness of 2 agents.

Cross Over trials Unplanned:

Start with two groups. One for Coronary
bypass and the other medical treatment.
Then the medical get worse and are taken
for bypass. They are cross-overs to the
surgical group.