Professional Documents
Culture Documents
Dr Alberto Corrias
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 1 / 39
PollEv game Standings
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 2 / 39
Laparoscopy versus open surgery
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 3 / 39
The Da Vinci system
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 4 / 39
The Da Vinci system - doctor’s opinion
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 5 / 39
Statistics of the Da Vinci system
Approved for
Totally Endoscopic 64
Atrial Septal Defect (ASD) countries where
Gynecological Laparoscopic Procedures system is in use
Thoracoscopically-Assisted
Cardiotomy Procedures
Mitral valve repair surgery
General Laparoscopic Surgery
∼10K
publications on
(gallbladder, gastroesophageal reflux and
the DaVinci
gynecologic surgery)
system since 1998
Laparoscopic Radical Prostatectomy
†
Sources:
Intuitive Surgical Website: http://www.intuitivesurgical.com/
FDA website: www.fda.gov
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 6 / 39
Our first scenario
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 7 / 39
Our fist scenario: the 4 protocols
Protocol G1 (cone game) Protocol G2 (rings game)
†
Video link http://bcove.me/pi91olz5
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 8 / 39
Our first scenario: training 40 students
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 9 / 39
Our first scenario: evaluation of performance
Participants evaluated using the Global Evaluative Assessment of
Robotic Skills (GEARS) on an in vivo operation on a pig.
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 10 / 39
The data
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 11 / 39
The problem in statistical terms
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 12 / 39
Looking at the data using boxplots
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 13 / 39
Looking at the data using histograms
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 14 / 39
Performing ANOVA calculations
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 15 / 39
Performing ANOVA calculations
X G1 + X G2 + X G3 + X S
XX = = 43.4815
4
4 Calculate (sX2 )
(X G 1 − X X )2 + (X G 2 − X X )2 + ...
sX2 = = 96.297
4−1
5 2 and s 2
Compute key quantities swit bet
2
sbet = nsX2 = 10(96.297) = 962.97
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 16 / 39
More on the role of the sample size
You are analyzing data from 4 groups, each with n = 10 subjects. You have
2 2
computed swit and sbet . At this point you are told that actually the experiment was
run on 20 subjects per group and they give you the additional data that was
missing. Now n = 20 per group. Assuming that all the additional measurements
Xi just happened to be exactly of the same value as the sample means X you had
computed initially,which of the following is true about the new analysis with
n = 20
2 2
1 swit will become bigger, but sbet will become smaller
2 2
2 swit will become smaller, but sbet will become bigger
2 2
3 swit and sbet will become bigger
2 2
4 swit and sbet will become smaller
ANSWER:
Pn 2
2 2 i=1 (Xi −X )
swit will be smaller. swit include terms like n−1
2 2
sbet will be bigger. sbet = nsX2
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 17 / 39
Performing the actual test at α = 0.05 significance
2
sbet
F = 2
swit
= 6.518 This light blue
0.4
Numerator degrees area is 0.05
of freedm:
m−1=3
0.2
Denominator
degrees of freedom:
m(n − 1) = 36
fcrit
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 18 / 39
The F distribution
Suppose you are doing 2 ANOVA analysis
ANOVA A: m = 3 groups with n = 4 subjects per group
ANOVA B: m = 3 groups with n = 40 subjects per group
Both analysis are conducted using the same 95% significance. For ANOVA A, you
A B
compute fcrit , while for ANOVA B, you compute fcrit . Intuitively, which is true?
A B
1 fcrit > fcrit
A
2 fcrit B
< fcrit
A
3 fcrit B
= fcrit
A B
ANSWER: fcrit > fcrit . Intuitively, the smaller n, the more uncertainty, the larger
the tail area, the greater fcrit will be for the same area.
0.8
0.6 A : ν1 = 2, ν2 = 9
0.4 B : ν1 = 2, ν2 = 127
0.2
0
0 1 2 3 4
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 19 / 39
Determining fcrit using tables
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 20 / 39
Performing the actual test at α = 0.05 significance
∼ 2.9 6.518
In statistical terms
The differences in scores among the 4 groups are statistically
significant (at 95% level). This means that, if all groups were drawn
from a single population, the probability of obtaining samples
displaying the observed differences are less than 5%. We still don’t
know which specific group(s) is (are) responsible for the observed
differences.
How it is reported
We observed statistically significant differences (p < 0.05) in
performance scores among the 4 groups who underwent different
training protocols.
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 22 / 39
Statistical interpretations
Our analysis found F = 6.518 which is greater than the Fcrit = 2.9.
Imagine, for a moment, that, instead, the calculations ended up in
computing F = 1.9. In such a case, can we conclude that there is no
difference among the 4 protocols? ANSWER: NO! ANOVA is still a
form of hypothesis testing. When we fail to reject the NULL
hypothesis, we can’t conclude that the NULL hypothesis is true. All
we can say is that the data failed to support any difference among the
protocol. The hypothesis that the training protocol does not matter is
still plausible (NOTE: ”plausible”, NOT ”true”!)
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 23 / 39
Pairwise t-tests: the Bonferroni method
X1 − X2
Six possible pairwise Student t t=p 2 2
swit /n1 + swit /n2
tests
1 G1 vs G2 The Student t distribution
2 G1 vs G3 will have m(n − 1) degrees of
3 G1 vs S P (N − m where
freedom
4 G2 vs G3 N = ni , if groups have
different sizes)
5 G2 vs S
If we choose a significance
6 G3 vs S level of 95%, then we will use
a tail area of 0.05
6 = 0.0083 to
compute the tcrit value
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 24 / 39
The 6 Bonferroni t tests (e.g., 95%)
H0 : µ1 = µ2 ,H1 : µ1 6= µ2
G1vsG2 G1vsG3 G1vsS G2vsG3 G2vsS G3vsS
tstat -0.97 -0.96 -4.14 0.01 -3.17 -3.18
BONF
tcrit 2.79 2.79 2.79 2.79 2.79 2.79
tcrit 2.028 2.028 2.028 2.028 2.028 2.028
Reject? NO NO YES NO YES YES
p 0.34 0.35 0.0002 0.99 0.003 0.003
Original significance
value. Area = 0.05/2
When doing multiple
New significance value. comparisons after ANOVA,
Area = 0.0083/2 you should use tcrit BONF , not
2 0.06
3 0.05
4 0.04
5 0.03
6 0.02
7 0.01
8 0.005
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 26 / 39
How results are reported
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 27 / 39
How results are reported: examples from BME papers
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 28 / 39
The ANOVA table
n((X G 1 − X X )2 + (X G 2 − X X )2 + (X G 3 − X X )2 + (X S − X X )2 )
2
sbet
2
swit
P G1
(Xi − X G 1 )2 + (XiG 2 − X G 2 )2 + (XiG 3 − X G 3 )2 + (XiS − X S )2
P P P
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 29 / 39
Our second scenario: a more realistic exercise
Link to video
http://intuitivesurgical.com/products/skills_simulator/
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 30 / 39
Calculation question (3 points)
GEARS scores Question:
Medicine Urology Junior Use ANOVA to determine
Students Interns Doctors whether there is any
65.14 70.22 74.86 difference in GEARS scores
39.70 75.18 57.78 among the 3 groups. Provide
the values of F , Fcrit and your
43.68 66.30 65.17
answer using a 95%
72.17 73.53 66.29 confidence level.
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 31 / 39
Solution
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 32 / 39
Solution
2
sbet = nsX2 = 4(67.67) = 270.68
2
sstudent 2
+ sintern 2
+ sjunior
2
swit = = 105.864
3
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 33 / 39
Solution
fcrit
No Rejection Region Rejection Region
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 34 / 39
Determining fcrit using tables
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 35 / 39
Performing the actual test at α = 0.05 significance
2.557 4.26
No Rejection Region Rejection Region
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 36 / 39
ANOVA: assumptions check
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 37 / 39
FAQ: why do we fail to reject the NULL hypothesis when
F ∼ 1?
2
sbet
F = 2
swit
2 (X G 1 − X X )2 + (X G 2 − X X )2 + ...
sbet = n(sX2 ) = n( )
m−1
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 38 / 39
Different types of ANOVA
BN2102 Bioengineering Data Analysis ANOVA case study: the Da Vinci system 39 / 39