Statistics . . . . .

1

• Statistics . . . . .

. . . . .So Whats the Big Deal?
2
 

• Case Presentation
• Symptom clusterSweaty palms Pale Increased
  heart rateGlassy-eyed stareLoss of affect

• Diagnosis Photonumerophobia The fear that ones
  fear of numbers will come to light
• (DSM IV)
• -- attributed to David Sackett

3
Goals for Today

• Communicate the purpose of statistics a way to
  approximate the truth
• Present some easy definitions of common
  statistical terms
• Differentiate between statistical and clinical
  significance
• Practice interpreting statistics through several
  case studies

4
A Far-fetched Analogy
This is not a pipe
Rene Magritte. The Air and the Song, 1928.
5
Arcane point 1. Statistics are an approximation
of the Truth

• This painting is a representation of a pipe, not
  the pipe itself
• Statistical significance is not the truth, but an
  approximation of the truth
• The Truth
• What we do for people helps them to live longer
  or live better
• Research helps us get closer to understanding the
  Truth
• Our goal figure out how close the statistics
  represent the truth

6
Arcane point 2. Users of statistics dont have
to be statisticians

• I am a user of statistics, not a statistician (I
  have friends, however, who are statisticians)
• You dont have to know a lot about statistics to
  effectively use statistics
• Dont focus on whether the statistics are right
• Learn to figure out what the statistics are
  trying to tell you

7
PThe Shrine of StatisticsThe Sacred P-Value
8
P Value

• "Probability" level
• The likelihood that the difference observed
  between two interventions could have arisen by
  chance
• Arbitrarily set at 5 risk (P 0.05)

9
Riboflavin and Migraine Headache
10
P Value

• Depends on several factors
• How large the effect was
• How consistent the effect was
• How many patients were studied
• As all of these factors increase, the likely of
  finding statistical significance increases
• Once weve decide the difference was NOT due to
  chance, we have to decide on clinical significance

11
This aint physics

• Unfortunately, the publication standards of
  medical journals are quite low compared to other
  science fields such as physics . . . Presumably
  necessary to assure that possible helpful
  therapies are not kept from needy patients for
  far too long.
• No respectable physics journal would publish a
  result with a p-value of a few percents. In fact,
  the publication standard in physics is typically
  a p-value of 0.0001 . . .
• Victor Stenger, PhD
• Professor Emeritus of Physics, U. of Hawaii
• Discoverer that the neutrino has mass

12
Wringing out information from P values

• "Highly Significant" P
• If the number of patients is small, P value tells
  us that the effect was either large or consistent
  (or both)
• If the number of patients is large, the size of
  the effect may not be that large

13
Wringing out information from P values

• Not Significant P 0.05 (i.e., 0.15)
• If the number of patients is small, there might
  not have been enough patients to find a
  difference that truly does exist
• If the number of patients is large, we can be
  confident that either there is no difference
  between treatments, or the treatment effect is
  not consistent

14
Wringing out information from P values

• "Borderline Significance" P 0.08 ????
• May have reached significance had their been more
  patients in the study
• The effect size may be small or inconsistent
• No conclusions can be drawn, except that more
  study is needed

15
Number Needed to Treat

• The number of patients that need to be treated
  for one additional patient to receive benefit
• The number of patients that need to be treated to
  prevent one additional outcome
• Takes into account the relative risks as well as
  the absolute risk of no treatment
• NNT 100
• in treatment group - in control group

16
Riboflavin and Migraine Headache

• A 50 response rate
• Attack frequency
• NNT 100 2.7
• 56 19
• Migraine days
• NNT 100 2.28
• 59 - 15
• Migraine index
• NNT 100 3.03
• 41 - 8

17
NNTs for Prevention
18
NNTs for Treatment
19
Other statistics you need to understand research
20
 Relative Risk

• The risk of harm with one treatment as compared
  with another
• The risk of benefit with one treatment as
  compared with another
• If RR 1, then there is no difference between
  the two treatments
• Depends only on the relative difference between
  the two treatments
• Does not take into account the risk of no
  treatment -- the "absolute risk"

5
21
Randomised trial of cholesterol lowering in 4444
patients with coronary heart disease the
Scandinavian Simvastatin Survival Study (4S).
Scandinavian Simvastatin Survival Study Group

• Summary 4444 patients with high cholesterol
  and CHD were given either simvastatin or placebo
  for a
• median of 5.4 years.
• Results 256 (12) in the placebo group
  died
• 182 (8) in the simvastatin
  group died
• Relative risk 0.70
• Risk reduction 30
• NNT ?

22
Examples of whererelative risk can be misleading
23
Meta-analysis Statins to prevent stroke and MI

• Meta-analysis of 29 studies, 10,000 patients
• Statins vs. control (usual care)
• Stroke risk 0.82 (18 decrease)
• MI risk 0.74 (26 decrease)
• 18 from what? 26 from what?
• Briel M. Effects of statins on stroke prevention
  in patients with and without coronary heart
  disease A meta-analysis of randomized controlled
  trials. Am J Med 2004117

24
Meta-analysis Statins to prevent stroke and MI

• Stroke Risk
• Low risk 0.2
• High risk (CHD) 0.9/year
• MI risk
• Low risk 0.9
• High risk 3.7/year
• Briel M. Effects of statins on stroke prevention
  in patients with and without coronary heart
  disease A meta-analysis of randomized controlled
  trials. Am J Med 2004117

• NNT to prevent 1 stroke/1 yr
• Low risk 2,778 patients
• High risk 617 patients
• NNT to prevent 1 MI/1 year
• Low risk 427 patients
• High risk 104 patients

25
Reformulation of Clarithromycin
66 decrease Relative risk 0.34
26
Confidence Interval

• "A statistic of a statistic"
•  Statistics are estimates
• Confidence intervals tells us the upper and lower
  possibilities of our statistical estimates

27
Example Results from the UKPDS
28
Since the 95 CI crosses 1.0, the difference is
not significant
95 C.I.
1.1
0.94
0.80
Risk could be this high
Risk could be this low
1.0
29
Dont Be Afraid of Statistics

• Statistical significance is a requirement for
  determining clinical significance, but is not
  enough to signify a clinical difference
• The P value tells us the risk that the difference
  between two treatments was due to chance
• Relative risk tells part, but not all of the
  story NNT does it better
• Confidence intervals help us to understand how
  close our estimate is to the "truth"

30
Diarrhea was somewhat more common in the
amoxicillin-clavulanate group (29.9 vs 19.6 P
.045, number needed to treat to harm ?
NNTH 100 10 30 - 20
31
Using intention-to-treat analysis, 59.2 of the
patients assigned to antidepressants survived
compared with 36.4 who received placebo (P
.03 number needed to treat 4). How many
people wont receive a benefit when treated?
32
What is the clinical relevance of this research?
Whats missing?
Osteoporotic fractures, in general, are
important, but the dangerous fractures that must
be prevented are hip fractures.