www.eurordis.

org

Introductory statistics for

medical research

Julia Saperia
European Medicines Agency

Barcelona, 18 June 2012

 What does statistics mean to you?

percentage
spread
mean
measurement
information
graphs
confidence
variability
data
average
evidence
probability median

2
 An example

• The idea of (clinical) research is to provide answers to questions

• Let’s ask a question…
 How tall are Eurordis summer school 2012
participants?

3
 An example

• Thinking about one (silly) question tells us some important

things about data.
 people are different variability in data
 the reliability of an average depends on the
population being studied and the sample drawn
from that population  valid inference

4
 A sillier example

• Let’s say I find out 5 heights (in cm)

165, 166, 167, 168, 169
mean = 167.0 cm
standard deviation = 1.6 cm

• Let’s say I find out 5 different heights (in cm)

154, 165, 168, 172, 183
mean = 168.4 cm
standard deviation = 10.5 cm

• So the means are the quite similar…but the two groups look
different

5
 A sillier example

• what we learn from the second example is that we get more

information from data if we know how variable they are

 with these pieces of information, we can draw a picture of

the data (provided some assumptions are true)

6
 The histogram

•Taken from http://www.mathsisfun.com/data/histograms.html on 28 May 2012

7
 Picturing “normal” data

Johann Carl Friedrich Gauss 1777-1855

~99%,
~95,5%
68,3%

mean - 2sd mean - sd mean mean + sd mean + 2sd

8
 Biostatistics

… is the method of gaining empirical knowledge in the life sciences

… is the method for quantifying variation and uncertainty

1. How do I gain adequate data?

Thorough planning of studies
2. What can I do with my data?
Descriptive and inferential statistics
If (1) is not satisfactorily resolved, (2) will never be adequate.

9
 (a population is a set of all
Population conceivable observations of a
certain phenomenon)

Inferential statistics

Draw conclusions for the

whole population based
draw a on information gained
sample from a sample
sample
Target of study design
representative sample
= ~ all "typical"
Descriptive representatives are
statistics included
sample data
10
 Descriptive statistics (1)

• For continuous variables (e.g. height, blood pressure,

body mass index, blood glucose)
• Histograms
• Measures of location
–mean, median…
• Measures of spread
–variance, standard deviation, range, interquartile
range

11
 Descriptive statistics (2)

• For categorical variables (e.g. sex, dead/alive,

smoker/non-smoker, blood group)
• Bar charts, (pie charts)…
• Frequencies, percentages, proportions
–when using a percentage, always state the number
it’s based on
–(e.g. 68% (n=279))

12
 Describing data

• Going back to the heights example

–we didn’t have time to measure the heights of the
population (all participants), so we took a sample
–we calculated the mean of the sample
–we can use the mean of the sample as an estimate
of the mean in the population
–but seeing as it’s only an estimate, we want to know
how reliable it is; that is, how well it represents the
true (population) value
–we calculated the standard deviation of the sample
–this is a fact (like the sample mean) about the
spread of the data in the sample
13
 Inferential statistics – confidence intervals

• the standard deviation also helps us to understand how precise our

estimate of the mean is
• the precision depends on the number of data points in the sample
– we can calculate the standard error of the mean
standard deviation
standard error of the mean
number in sample

– (in our more realistic example, the standard error is 4.7)

– the larger the sample, the smaller the standard error, the
more precise the estimate
• so now we have an estimate of the population mean and a measure
of its precision
• so what?
• we can use the standard error of the mean and the properties of the
normal distribution to gain some level of confidence about our
estimate
14
 Confidence intervals

• How confident do you need to be about the outcome of a race in order to

place a bet?
• In statistics it’s typical to talk about 95% confidence
• we can use the standard error of the mean to calculate a 95%
confidence interval for the mean

95% confidence interval mean 2 standard error

measure of precision of
estimate of population mean the sample mean

relates to the normal distribution

• in our example: 95%CI for the mean ≈ 168.4 2 x 4.7

≈ [159.0, 177.8]

15
 Confidence intervals

• What does a 95% confidence interval mean?

• remember that we have drawn a sample from a population
• we want to know how good our estimate of the population mean (the
truth) is
• the 95% CI tells us that the true mean lies in 95 out of 100 confidence
intervals calculated using different samples from this population
 true mean lies between 159 cm and 177.8 cm with 95% probability
• remember: the larger the number in the sample, the smaller the standard
error
• the smaller the standard error, the narrower the confidence interval
 the larger the number in the sample, the more precise the estimate

16
 A real example

Lateral wedge insoles for medial knee osteoarthritis: 12 month

randomised controlled trial (BMJ 2011; 342:d2912 )
Participants 200 people aged 50 or more with clinical and radiographic
diagnosis of mild to moderately severe medial knee osteoarthritis.
Interventions Full length 5 degree lateral wedged insoles or flat control
insoles worn inside the shoes daily for 12 months.
Main outcome measure Change in overall knee pain (past week)
measured on an 11 point numerical rating scale (0=no pain, 10=worst
pain imaginable).
Research question:
Are full length 5 degree lateral wedged insoles better than flat control
insoles at reducing pain after 12 months in patients with mild to
moderately severe medial knee osteoarthritis?

17
 A real example

• Assume 5 degree lateral wedged insoles and flat control insoles reduce
pain to the same extent (this is our null hypothesis)
• Collect data in both groups of patients and calculate:
the mean reduction from baseline in the 5 deg lateral insole group (0.9)
the mean reduction from baseline in the flat insole group (1.2)
the difference between the two groups in mean change from baseline (0.9
– 1.2 = -0.3)
• We can see that there was some reduction in pain in both groups and
that the reduction was greater in the flat insole group
• The 95% confidence interval tells us how important that difference is

-1.0 - 0.3 0 0.3

flat insole better 5 deg lateral insole better

18
 A real example

• Conclusion: with the available evidence, there is no reason

to suggest that one treatment is better than the other in
reducing pain scores after one year.
• Or: there is no statistically significant difference in
treatments (because the 95% confidence interval crosses
the null value)

19
 Statistical significance vs clinical significance

• Remember: the larger the number of data points in a sample, the

more precise the estimate
• Let’s imagine that instead of enrolling 200 patients in their study, the
researchers enrolled 1500.

-0.4 -0.25 -0.1 0

flat insole better 5 deg lateral insole better

• This time, the confidence interval does not cross the null value
• So we can conclude that the flat insole reduces pain by 0.25 points
more than the 5 degree lateral insole.
• Hooray?
• Is a difference of 0.25 points clinically relevant?

20
 Quantifying statistical significance – the p-value

• In the last example, we looked at the 95% confidence interval to decide

whether we had a statistically significant result
• (the confidence interval is useful because it gives clinicians and patients a
range of values in which the true value is likely to lie)
• another measure of statistical significance is the p-value
• Remember our null hypothesis: 5 degree lateral wedged insoles and
flat control insoles reduce pain to the same extent
• We go into our experiment assuming that the null hypothesis is true
• After we’ve collected the data, the p-value helps us decide whether our
results are due to chance alone

21
 The p-value

• the p-value is a probability  it takes values between 0 and 1

• the p-value is the probability of observing the data (or data more
extreme) if the null hypothesis is true
 (if the null hypothesis is true, the two treatments are the same)
 a low p-value (a value very close to zero) means there is low
probability of observing such data when the null hypothesis is true
 reject the null hypothesis (i.e. conclude that there is difference in
treatments)

22
 Significance level

• p-values are compared to a pre-specified significance level, alpha

• alpha is usually 5% (analogous to 95% confidence intervals)
 if p ≤ 0.05 reject the null hypothesis (i.e., the result is statistically significant
at the 5% level)  conclude that there is a difference in treatments

 if p > 0.05 do not reject the null hypothesis  conclude that there is not
sufficient information to reject the null hypothesis
(See Statistics notes: Absence of evidence is not evidence of absence:
BMJ 1995;311:485)
The authors of the insole paper did not cite the p-value for the result we looked
at but we can conclude that p>0.05 because the 95%CI crosses the null value
and so we cannot reject the null hypothesis (of no difference).

23
 Statistical mistakes/errors
True state of nature
Hypothesis accepted H0 H1
Test decision

H0 Correct
1-
H1 False positive

Type I error:
Error of rejecting a null hypothesis when it is actually true
( error, error of the first kind)
represents a FALSE POSITIVE decision: A placebo is declared to be more
effective than another placebo!)
24
 Statistical mistakes/errors
True state of nature
Hypothesis accepted H0 H1
Test decision

H0 Correct False negative

1-
H1 False positive Correct
1-

Type II error:
The alternative hypothesis is true, but the null hypothesis is erroneously
not rejected. (β error, error of the second kind)
FALSE NEGATIVE decision: an effective treatment could not be
significantly distinguished from placebo
25
 Which sorts of error can occur when performing statistical tests?
Statistical mistakes/errors
True state of nature
Hypothesis accepted H0 H1
Test decision

H0 Correct False negative

1-
H1 False positive Correct
1-

Significance level is usually predetermined in the study protocol, e.g.,

=0.05 (5%, 1 in 20), = 0.01 (1%, 1 in 100), = 0.001 (0.1%, 1 in 1000)

POWER 1- : The power of a statistical test is the probability that the test will reject
a false null hypothesis.

26
 Statistical mistakes/errors
True state of nature
Hypothesis accepted H0 H1
Test decision

H0 Correct False negative

1-
H1 False positive Correct
1-

Consumer‘s risk?

Producer‘s risk?

27
 Compare a statistical test to a court of justice
Statistical tests vs court of law
True state of nature
H0 H1
Innocent Not Innocent
Decision of the court

H0 Correct False negative

“Not innocent" "innocent"

(i.e. guilty but not caught)

1-
Test decision

Judged

H1 False positive Correct

(i.e. innocent but convicted) 1-
Judged

False positive: A court finds a person guilty of a crime that they did not actually commit.
False negative: A court finds a person not guilty of a crime that they did commit.

28
 Minimising statistical errors

Remember:
How do I gain adequate data?
Thorough planning of studies
• Defining acceptable levels of statistical error is key to the planning
of studies
• alpha (in clinical trials) is pre-defined by regulatory guidance
(usually)
• beta is not, but deciding on the power (1-beta) of the study is
crucial to enrolling sufficient patients
• the power of a study is usually chosen to be 80% or 90%
• conducting an “underpowered” study is not ethically acceptable
because you know in advance that your results will be inconclusive

29
 Deciding how many patients to enrol

• the sample size calculation depends on:

– the clinically relevant effect that is expected (1)
– the amount of variability in the data that is expected (2)
– the significance level at which you plan to test (3)
– the power that you hope to achieve in your study (4)
• if you knew (1) and (2), you wouldn’t need to conduct a study
 picking your sample size is a gamble
• the smaller the treatment effect, the more patients you need
• the more variable the treatment effect, the more patients you need
• the smaller the risks (or statistical errors) you’re prepared to take,
the more patients you need

30
 Conclusion

• amount of variability in data defines conclusions from

statistical tests
• clinical trial methodology is underpinned by statistics
• valid statistical inference, and therefore decision-
making, depends on robust planning
• understanding statistics isn’t as hard as it looks (I
hope)

31
 Further reading

• www.consort-statement.org
 standards for reporting clinical trials in the
literature
• Statistical Principles for Clinical Trials ICH E9
 useful glossary
• http://openwetware.org/wiki/BMJ_Statistics_Notes_series
 coverage of a number of topics related to
statistics in clinical research, mostly by Douglas
Altman and Martin Bland

32