Transcript Case control studies: - Kashan University of Medical Sciences

Medical Statistics
Dr. Gholamreza Khalili
Department of Epidemiology & Biostatistics
School of public health
Tehran University of Medical Sciences
Why does good evidence from
research fail to get into practice??
- 75% cannot understand the statistics
- 70% cannot critically appraise a research
paper
Using Research for Practice: A UK Experience of the barriers scale
Dunn V, Crichton C, Williams K, Roe B, Seers K
Why is statistics necessary?




58% of the population had GERD
Mean age of the respondents was 25+8
25% of women and 50% of men lied about their
age (4 in each group!?)
Doctors live longer than normal people
Why is statistics necessary?

Descriptive statistics



58% of the population had GERD
Mean age of the respondents was 25+8
Inferential statistics


25% of women and 50% of men lied about their
age (4 in each group!?)
Doctors live longer than normal people.
Descriptive statistics


Point estimates: Mean, median, mode, relative
frequency
Distribution: Standard deviation
Inferential statistics: exploring
associations and differences
Differences

Continuous variables (blood pressure, age):
109+11 vs. 140+10

Categorical variables (proportion of blind
people): 10% vs. 2%
Measures of Association
Relative Risk (RR)
Odds Ratio (OR)
Absolute risk reduction
Number needed to treat
Treatment
Medical
CABG
dead
404
alive
921
Total
1325
350
974
1324
Number Needed to Treat (NNT)


only defined for a specific intervention!
only defined for a specific outcome!


eg. Pravastatin™ 40 mg nocte x10 years, in a 65 year old
male, ex-smoker with high BP and Diabetes, to reduce MI or
Death.
NNT is the inverse of Absolute Risk Reduction: i.e.
NNT = 1/ARR
Measures of Association

Linear Correlation



Regression





Conditions
r and p, CI
Univariate
Multiple Regression
Logistic Regression
Cox Proportional Hazard Model
Do they mean causation?
Associations may be due to

Chance (random error)



Bias (Systematic error)


must be considered in the design of the study
Confounding


statistics are used to reduce it by appropriate design of
the study
statistics are used to estimate the probability that the
observed results are due to chance
can be dealt with during both the design and the analysis
of the study
True association
Dealing with chance error

During design of study



Sample size
Power
During analysis (Statistical measures of
chance)


Test of statistical significance (P value)
Confidence intervals
Statistical measures of chance I
(Test of statistical significance)
Association in Reality
Yes
Type I
error
Yes
Observed
association
No
No
Type II
error
The p-value in a nutshell
Could the result have occurred by chance?
The result is
unlikely to be due
to chance
The result is
likely to be due
to chance
0
1
p < 0.05
a statistically
significant result
p > 0.05
not a statistically
significant result
p = 0.05
p = 0.5
1
20
1
2
or 1 in 20
result fairly
unlikely to be due
to chance
or 1 in 2
result quite likely
to be due to
chance
Significantitis

Significantitis is the plague of our time.


A. Etemadi, 21st century epidemiologist
The drug reduced blood pressure by 1mmHg
(p<0.0000000000000001)
Confidence Interval (CI)
Is the range within which the true size
of effect (never exactly known) lies,
with a given degree of assurance
(usually 95%)

The ACE inhibitor group had a 5%
(95% CI: 1-9) higher survival.
Associations may be due to

Chance (random error)



Bias (Systematic error)


must be considered in the design of the study
Confounding


statistics are used to reduce it by appropriate design of
the study
statistics are used to estimate the probability that the
observed results are due to chance
can be dealt with during both the design and the analysis
of the study
True association
Types of Bias


Selection bias – identification of individual subjects
for inclusion in study on the basis of either exposure
or disease status depends in some way on the other
axis of interest
Observation (information) bias – results from
systematic differences in the way data on exposure
or outcome are obtained from the various study
groups
Associations may be due to

Chance (random error)



Bias (Systematic error)


must be considered in the design of the study
Confounding


statistics are used to reduce it by appropriate design of
the study
statistics are used to estimate the probability that the
observed results are due to chance
can be dealt with during both the design and the analysis
of the study
True association
Confounding
coffee
smoking
Pancreatic
cancer
Confounding
coffee
smoking
Pancreatic
cancer
Confounding
Possible
cause
confounder
effect
Associations may be due to

Chance (random error)



Bias (Systematic error)


must be considered in the design of the study
Confounding


statistics are used to reduce it by appropriate design of
the study
statistics are used to estimate the probability that the
observed results are due to chance
can be dealt with during both the design and the analysis
of the study
True association
DETERMINATION OF CAUSATION

The general QUESTION: Is there a cause and
effect relationship between the presence of
factor X and the development of disease Y?
Nature of Evidence:
1. Replication of Findings –

consistent in populations
2. Strength of Association –

significant high risk
3. Temporal Sequence –

exposure precede disease
Nature of Evidence:
4. Dose-Response –

higher dose exposure, higher risk
5. Biologic Credibility –

6.
exposure linked to pathogenesis
Consideration of alternative explanations –

the extent to which other explanations have been
considered.
Nature of Evidence
7. Cessation of exposure (Dynamics) –

removal of exposure – reduces risk
8. Specificity

specific exposure is associated with only one
disease
9. Experimental evidence

What is the appropriate test?

Scales
Nominal
Ordinal
Interval
Ratio
Normal Distribution
Skewed curve
Greenhalgh, T. BMJ 1997;315:364-366
Copyright ©1997 BMJ Publishing Group Ltd.


Parametric versus non-parametric tests
Transformation

Subgroup analysis?


Analyses showed that the drug was especially
effective in women above 35 who were unable to
say supercalifragilisticexpialidocious.
We divided the study population according to sex,
then each group were divided to 10 age groups,
each age group was subdivided according to
educational background and whether they were
left-handed or right-handed.
Scenario
Subgroup analysis
Statistical Tests
Type of Data
Goal
Measurement
(from
Gaussian
Population)
Rank, Score, or
Measureme
nt (from
NonGaussian
Population)
Binomial
(Two
Possible
Outcomes)
Survival Time
Describe one
group
Mean, SD
Median,
interquartile
range
Proportion
Kaplan Meier
survival
curve
Compare one
group to a
hypothetica
l value
One-sample t test
Wilcoxon test
Chi-square
or
Binomial
test **
Compare two
unpaired
groups
Unpaired t test
Mann-Whitney test
Fisher's test
(chi-square
for large
samples)
Log-rank test or
MantelHaenszel*
Compare two
paired
groups
Paired t test
Wilcoxon test
McNemar's test
Conditional
proportional
hazards
regression*
Statistical Tests
Compare three or
more
unmatched
groups
One-way ANOVA
Kruskal-Wallis test
Chi-square test
Cox proportional
hazard
regression**
Compare three or
more
matched
groups
Repeated-measures
ANOVA
Friedman test
Cochrane Q**
Conditional
proportional
hazards
regression**
Quantify
association
between
two
variables
Pearson correlation
Spearman
correlation
Contingency
coefficients*
*
Predict value from
another
measured
variable
Simple linear
regression
or
Nonlinear
regression
Nonparametric
regression**
Simple logistic
regression*
Cox proportional
hazard
regression*
Predict value from
several
measured
or binomial
variables
Multiple linear
regression*
or
Multiple
nonlinear
regression**
Multiple logistic
regression*
Cox proportional
hazard
regression*