Académique Documents
Professionnel Documents
Culture Documents
Literature
Peter Flores-Quilala, RPh, MD
Faculty of Pharmacy
Pharmacy Informatics
The Age of
Information
Too much information
Evaluate critically
Abstract reading &
skimming not valid to
draw conclusions
(common)
Validity
Internal
Methodology
External
Can the sample be a
generalization for the
whole population?
Results
Syntax of PR
Submission of
manuscripts to
journals
Sent to experts
For comments and
opinions
Evaluation of reviews
Still mistakes may still
arise after publication
The ultimate responsibility in
interpreting the results correctly is left
to the reader.
If we choose to study
the effects of two
drugs in the treatment
of acute otitis media
then the sample is
those who are
actually studied, and
the population refers
to all patients with
acute otitis media.
Group A
Sample
Measure
Outcome
Compare
Results
Random Allocation of
Treatment
Group B
Apply
Intervention
B
Measure
Outcome
Draw
Conclusions
Association vs Causation
The ideal clinical study is
designed such that a
sample of patients with a
particular disease state is
selected at a random
from a population.
Patients are then
randomly assigned to two
or more treatment groups
Assessment are
conducted to determine if
the groups are similar to
each other at baseline
Then an intervention is
applied, and differences
in the outcome (if any)
are measured. If differences in
Therefore there is an
association between a
particular variable and the
outcome
Difficult to prove that
causation is the reason
for the difference
Example
Association
A retrospective cohort study to test whether Drug A
causes colon cancer. The results indicate that
patients who consume Drug A were twice as likely to
develop cancer as those patients not taking the drug.
However, we might not prove that Drug A causes
cancer since this is a retrospective study that control
of other variables would be necessary like other diet
and intake of other drugs within the study
Prospective
Retrospective
Observational
Parallel
Cross-Over
Longitudinal
Cross-sectional
Multicenter
Case-control
Cohort
Comparative Study
Defines what
conclusions may be
drawn
Draws Limitations
Methodology
Logistical
Financial
Ethical
One set of
independent variables
Thought to induce
changes in the outcome
of the dependent
variable
Which among the two Non-Steroidal anti-inflammatory drugs can lower core
body temperature?
Independent variablethe NSAiD
Dependent variable----- Core Body Temp
Study designs that seek to answer one question is desirable----few
samples
More questions-----require larger sample size
Prospective
For comparative
research
Planned in advance
Exerts maximum
amount of control
Minimizes biases and
confounders
Limitations
Expensive to conduct
Complex logistical
needs
Address ethical issues
for animal or human
subject use
Retrospective study
Not as powerful as
prospective
Low cost
Ease of data collection
Limitations
More biases and
confounders
Observational Study
Similar to retrospective
No attempt to control
certain variables
Patients are simply
observed
No outside
intervention is applied
beyond the normal
treatment of the
patient
Parallel Study
Designs
Enroll patients
Conducts the
evaluation
Collect data on the
same points in time for
each group being
studied
Utilized by the
comparative
prospective designs
Ensure that the
environmental
influence is the same
between groups
Cross-Over Study
Subjects are exposed
to more than one
intervention
Changes in the
dependent variable
are compared within
the same patient
Patient are allowed to
be their own controls
Minimizes inter-patient
variability
Requires a washout
out period between the
interventions
Can not be used if the
effects are permanent
(only transient effects)
Longitudinal Studies
Data are collected on an
extended period of time
Multicenter designs
Data collected from different
centers
More representative of the
population
The results also depend on
the site or environment
Use of Analysis of co variance
and Cochran Mantel-Haenszel
test controls the phenomenon
Cross-sectional
Comparison between two or
more groups are made at
specific time periods
Pharmacoepidemiologic
studies
Case-control
Compare groups of patients
with a disease to control
patients without the disease
and look for exposures to
certain factors like drugs
Cohort study
Identifies patients exposed to
some factor and compare
them to patient who have not
been exposed to determine if
there are any differences in a
particular clinical outcome
Phase 1
Assess pharmacology,
pharmacokinetics, and
safety of the drug
Conducted with small
number of healthy
human subjects
Efficacy is not studied
Phase 2
To evaluate the efficacy of
the drug
Tested in larger number of
patients
Patients with disease or to
prevent a disease
Strict inclusion criteria
Phase 3
Efficacy and safety of the
drug
Large scale studies
(hundred to thousands)
Uses randomization,
blinding and control groups
NDA
Drugs are allowed to be
marketed, prescribed and
sold
Phase 4
Post marketing
surveillance
Safety and efficacy
Case reports
Drug utilization evaluation
Less scientific designs
Long term data
Data may include new
indications for the drug
New dosage forms
New patient populations
Long term efficacy
ADR
Requires prolonged
exposure
Drug stability
Drug interactions
Outcomes research
Pharmacoeconomics
Should be evaluated carefully
Significant variation in their quality
Study designs may not be optimal
Statistical analyses may be
inadequate or inappropriate
Presence of bias and confounders
may lead to unsupported or invalid
conclusions
BIAS
? Favoritism
Systematic variation
Treatment groups under
study are treated and
measured differently in a
systematic consistent
fashion
Can mislead to conclude
erroneously
Most damaging is selection
bias may be intentional or
non intentional
Biases of Rhetoric
One-sided reference bias
Positive result bias
Hot stuff bias
Diagnostic access bias
Diagnostic suspicion bias
Wrong sample size bias
Admission rate bias
Confounding
A confounding variable is one that affects
the dependent variable the independent
variable or both
May be impossible to eliminate all
confounding variables
Solution: use the appropriate statistical
method
Control Groups
Used by comparative
studies
As a frame of
reference to use
when comparing
Without it, one can
not directly compare
the results of one
study with another
study
Blinding
To minimize bias
among the study
group and the
clinician
Open label studies
No blinding
Both the clinician and
patient are aware
Single blind
Blinding either the
clinician or the patient
Double blind
Both the clinician and
patient are unaware
Most desirable
Financially, logistically
and ethically
impossible
Randomization
To minimize bias in
the study
Random selection
All from a population
have an equal chance
of being chosen for the
study
Random Allocation
Assigned to a treatment
group by random
chance
Minimizes selection
bias
Hypothesis Testing
Null hypothesis (Ho)
Opposing hypothesis
Hypothesis of no
difference
No difference in the
outcomes measured
Data collected and
results are used to
either to accept the
Ho-no diff; or reject
Ho-a difference exists
Research hypothesis
Alternative hypothesis
States that there are
difference in between
groups
Random errors
Type I or alpha error
Ho incorrectly rejected
Thought that there is
difference but in fact there
is no difference
P value usually <0.05 level
of significance
Sample size parameters:
1. delta
2. reasonable statistical power
3. reasonable clinical difference
Example
The smaller the
difference one wishes
the larger sample size
is required
Difference of 20% ()
in the cure rates of
two antibiotics the
study needs 20
patients to achieve
80% power
To detect a difference
of 10% for the same
study 200 patients is
needed to achieve the
same power of 80%
Types of Analyses
One tailed
Two possibilities only
Ie Drug A is equal Drug B;
Drug A is better than Drug
B and is never worse than
Drug B
Used when one of the
group is a placebo group
that the active drug may be
similar to the placebo but
highly unlike that the active
drug is will be worse that
the placebo
Two tailed
Utilized in most studies
If you are not sure which
group of drugs is better
Ie Drug A is equal to Drug
B;
Drug A is better than Drug
B;
Drug Ais worse than drug B
Statistical Inference
Samples
Inclusion criteria
Exclusion criteria
Useful to maintain the
homogeneity of the sample
Assures patient safety and
welfare
Often excluded, allergy,
pregnant, breast feeding,
children
4 Types of Data
Nominal
Strictly categorical
Gender (male or female)
Survival outcome
Ordinal Data
Similar to nominal
Data are in groups or
categories but there are order
or magnitude
Pain scale (VAS 1-10)
Interval data
Integer values
Temperature scale
Points are arranged in
continuous integer values and
the difference between each
step is consistent
Ratio
Most sophisticated type of
data
There is an absolute zero
point or absence of a factor
Blood pressure; zero BP
means no BP unlike Temp 0F
or centigrade does not mean
an absence of temperature
Age
Ratio data
Ordinal scale
Evaluation of Variances
Mean Values
The SD square root of the
variance
If with common variances
homoscedastic but if with
different variances
heteroscedastic.
NON PARAMETRIC
Chi-Square Test
Comparison of
nominal data for
independent groups
(2x2)
McNemars Test
Comparison of
nominal data for two
matched or paired
groups
Contingency Table
Analysis (R x C)
Comparison of two
nominal data when
there are >2 groups or
>2 possible outcomes
NON PARAMETRIC
Cochran MantelHaenszel Test
Useful for comparing
nominal data for
multiple 2x2 tables
(when there are more
than one independent
variable being
considered)
NON PARAMETRIC
Wilcoxon Signed
Rank Test
Comparison of
continuous data taken
from 2 paired groups,
when the data are
nonparametric
Friedmans Test
Comparison of continuous data taken from >3
paired or matched samples, when the data are
nonparametric
PARAMETRIC TEST
Students t-test
Comparison of continuous
data taken from 2
independent groups
Paired t-test
Comparison of continuous
data taken from 2 paired or
matched groups
1 way ANOVA
Comparison of continuous
data taken from >3
independent groups
Repeated Measures
Comparison of continuous
data taken from >3 paired
or matched samples
2-way ANOVA
Similar to 1 way ANOVA
but comparisons can be
made for >2 factors
(independent variables
simultaneously)
Other Tests
Pearson Regression
Determines if there is
linear correlation
between 2 groups
when the data are
normally distributed
Spearman
Regression
Determines if there is
linear correlation
between 2 groups
when the data are
NOT normally
distributed
Multivariate
Regression
Determines
relationship of multiple
variables with a single
dependent continuous
variable
Logistic Regression
Determines
relationship of multiple
variables with a single
dependent
dichotomous variable
Analysis of Covariance
(ANCOVA)
Useful for controlling the
effects of a potentially
confounding variable on
the dependent variable
Survival Analysis
Evaluates the probability of
achieving or not achieving
a specified goal during a
specific period of time. A
common variable that is
often measured with this
method is survival,
however many other
variables with a
dichotomous endpoint can
be used as well. Other
statistical test, such as the
log-rank test, can be used
to compare the outcomes
between the groups
There is a 34%
chance of getting a
type I error even if
clinician deem this as
clinically significant
Introduction
Methodology
References
15 guidelines whew!!!!!!!
Am so exhausted na. Will just email the
test to you.
Good Nyt!!!!!!