What should pediatricians know about statistical communication

1
What should pediatricians know about statistical
communication?

• Practical issues of statistics in clinical
  practice

Warunee Punpanich Infectious Disease
Division Queen Sirikit National Institute of
Child Health
2

• Why is a physician held in much higher esteem
  than a statistician?
• A physician makes an analysis of a complex
  illness
• whereas a statistician makes you ill with a
  complex analysis!

3
Statistical thinking will one day be as
necessary for efficient citizenship as the
ability to read and write HG. Wells
4
How to understand risk or benefit?
5
Which intervention will you apply?
6
From a scale 1-5 1 Definitely less likely to
treat 2 Less likely to treat 3 Not sure or no
preference 4 More likely to treat 5 Definitely
more likely to treat
7
1) 0.1 95 CI -0.5, 0.3 decrease in the
incidence of fatal myocardial infarction (0.3
vs. 0.4) 2) 1.4 95 CI -2.5, -9
decrease in the incidence of fatal and nonfatal
myocardial infarction (2.5 vs. 3.9) 3) 26
95 CI -74, 114 relative decrease in the
incidence of fatal myocardial infarction. 4)
34 95 CI -55, -9 relative decrease in the
incidence of fatal or nonfatal myocardial
infarction (0.3 vs. 0.4) 5) 77 persons must be
treated for an average duration of 5 years to
prevent 1 myocardial infarction.
8
What is your chance to develop breast cancer?
9

• The probability that a women age 40 has breast CA
  (BCA) is about 1.
• If she has BCA, the probability that she tests
  positive on a screening mammogram is 99.
• If she does not have BCA, the probability that
  she nevertheless tests positive is 9.
• What are the chances that a woman who test
  positive actually has BCA?

10 50 80 95 99
10
Prozacs side effects
11

• A psychiatrist prescribing Prozac to depressed
  patients told them that they had a 30-50 chance
  of developing a sexual problem
• Patients interpret that Something would go awry
  in 30-50 of their sexual encounters.
• While the psychiatrist means out of every 10
  people to whom he prescribes Prozac, 3-5
  experience a sexual problem.

12

• Mathematically, these numbers are the same as
  the percentages he used before.
• Communicating by using natural frequency can
  make a difference psychologically.
• Patients can ask questions such as what to do if
  they were among the 3-5 people.
• Patients found this explanation much more
  understandable, even though they are
  mathematically equivalent.

13
What is your chance to develop breast cancer?
14
The probability that a women age 40 has breast CA
(BCA) is about 1. If she has BCA, the
probability that she tests positive on a
screening mammogram is 99. If she does not have
BCA, the probability that she nevertheless tests
positive is 9. What are the chances that a
woman who test positive actually has BCA.
10 50 80 95 99
15
Prevalence ac/(abcd) P(veD) a/(ac)
Sensitivity P(Dve) a/(ab) Positive
predictive value
16
Bayes Theorem
P(Dve) Prev x Sn Prev x
Sn (1-prev) x (1-Sp)
.01x.99 .01 x .99 .99 x.09
17

• Think of 100 women. One has BCA, and she will
  probably test positive.
• Of the 99 who do not have BCA, 9 will also test
  positive.
• A total of 10 women will test positive.
• How many of those who test positive actually have
  BCA.

18
Conditional Probability
19

• Prob of A given B vs. Prob of B given A
• We can reduce this confusion by replacing
  conditional probabilities with natural
  frequencies

20
P(veD) a/(ac) Sensitivity P(Dve)
a/(ab) Positive Predictive value
21
Solving a problem simply means representing it
so as to make the solution transparent.
Herbert A. Simon, The sciences of the
Artificial
22
Natural Frequency
Probability
100 people
P(D) 0.01 P(D) 0.99 P(nonD) 0.09
1 disease
99 no disease
P(Dve) Prev x Sn Prev x
Sn (1-prev) x (1-Sp) .01x.99
0.099 .01 x .99 .99 x.09
0.99 ve
0.01 -ve
9 ve
90 -ve
P(Dve) 0.99/(0.999) 0.099 (9-10)
23
What would you tell the patient who has a
positive mammogram?
24
When thinking and talking about risks, use
frequencies rather than probabilities. ..
25

• You have 30 chance of developing a sexual
  problem left the reference class unclear
• - Each person chose the reference class based on
  his or her own perspective.
• Frequencies, such as 3 out of 10 patients, make
  the reference class clear, reducing the
  possibility of miscommunication.

26
Which intervention will you apply?
27
1) 0.1 95 CI -0.5, 0.3 decrease in the
incidence of fatal myocardial infarction (0.3
vs. 0.4) 2) 1.4 95 CI -2.5, -9
decrease in the incidence of fatal and nonfatal
myocardial infarction (2.5 vs. 3.9) 3) 26
95 CI -74, 114 relative decrease in the
incidence of fatal myocardial infarction. 4)
34 95 CI -55, -9 relative decrease in the
incidence of fatal or nonfatal myocardial
infarction (0.3 vs. 0.4) 5) 77 persons must be
treated for an average duration of 5 years to
prevent 1 myocardial infarction.
28
1) 0.1 95 CI -0.5, 0.3 decrease in the
incidence of fatal myocardial infarction (0.3
vs. 0.4) 2) 1.4 95 CI -2.5, -9
decrease in the incidence of fatal and nonfatal
myocardial infarction (2.5 vs. 3.9) 3) 26
95 CI -74, 114 relative decrease in the
incidence of fatal myocardial infarction. 4)
34 95 CI -55, -9 relative decrease in the
incidence of fatal or nonfatal myocardial
infarction (0.3 vs. 0.4) 5) 77 persons must be
treated for an average duration of 5 years to
prevent 1 myocardial infarction.
29
How to understand risk
30
Tx 1 Tx 2

• Overall risk ac/ abcd
• Risk for Tx 1 (R1) a/(ab)
• Risk for Tx 2 (R2) c/(cd)
• Risk Difference R1 R2
• Risk Ratio (Relative Risk) R1/R2
• Relative Risk Difference (R1-R2)/R2

31

• What is the benefit of a cholesterol-lowering
  drug on the risk of coronary heart disease?
• In 1995, the results of the west of Scotland
  Coronary Prevention Study found People with
  high cholesterol can rapidly reduce their risk of
  death by 22 by taking pravastatin sodium.

32

• What does 22 mean?
• Study indicated that a majority of people think
  that
• Out of 1000 people with high cholesterol, 220
  can be prevented from becoming heart attack
  victims.

33
Reduction in total mortality for people who take
pravastatin. (From Skolbekken, 1998)

• Relative Risk Difference (R1-R2)/R2
• 22 (41-32)/41 x 100

34
3 ways to present the benefit
35
Absolute risk reduction (ARR) - Proportion of
those who die w/o Tx those w/ Tx -
41/1000 32/1000 0.9 Relative risk reduction
(RRR) - ARR/Proportion of those who die w/o
Tx - 0.9/4.1 22.1 Number needed to treat
- No. of people who must participate in the
treatment to save 1 life - 1/ARR 111
36
Is a treatment effective?
37
The rarer the event, the higher the RRR. RRR of
25 means many lives saved if the disease is
frequent, but only a few if the disease is
rare This can refer to ARR from 40 to 30 NNT
(1/0.1) 10 0.04 to 0.03 NNT (1/0.0001)
10,000
38
If combine with the duration of treatment of 5
years this could be NNT 10,000 x 5 50,000
person-years If combine with the cost of
treatment of 20,000 Bh/year for duration of
treatment of 5 years ? NNT 20,000 x 10,000 x
5 1 billion Bh to save 1 life
39

• Transparency can also be achieved by expressing
  the benefit by means of ARR, NNT

40
(No Transcript)
41
Professor of Business Administration at Baruch
42
Bottom Line
When communicating about risk Use natural
frequency rather than probability To understand
the impact of treatment Use natural frequency
NNT or ARR rather than RRR
43
.Statistical thinking will one day be as
necessary for efficient pediatricianship as the
ability to diagnose and to treat.
44
Education's purpose is to replace an empty mind
with an open one. Malcolm S. Forbes
45
(No Transcript)
46
Low Prevalence/Risk Disease
P(Dtest ve)
P(TestveD)
P (dis) or Prev or Risk
P (testve)
47
High Prevalence/Risk Disease
P(Dtest ve)
P(TestveD)
P (dis) or Prev or Risk
P (testve)
48
(No Transcript)
49
(No Transcript)