Practical Approach in data discussion

advertisement
Ves-Matic Cube 200 vs. Test-1
• We would like to compare the results of two papers recently
issued on two international scientific journals:
Let’s start with Linear Regressions
• Linear Regression is a way to
study the strength of the
correlation between two
methods, not their agreement
• the closer to 1,0 is the value of
correlation coefficient - that is the
closer the slope of the regression
line is to 45° (line of equivalence)
- the more two methods are in
agreement
• In comparing method it is
important to detect the bias
Let’s start with Linear Regressions
Ves-Matic Cube 200 vs Westergren
Linear Regression according to Passing – Bablok
(ρ = 0.946; P < 0.001; y = - 0,0435x + 1,0435)
Test-1 vs Westergren
Linear Regression
(r = 0.386; P < .0005; y = 0.547x + 14.881)
2
By comparing the values it is clear that the Ves-Matic Cube correlates with the
Westergren method better than the Test-1 (ρ = 0.946 compared with r2 = 0.386).
Let’s start with Linear Regressions
LET’S SHOW THE TRICK
Please note that in Figure 1 - Test-1 graph there are points (left side of the Test-1 vs
Westergren graph) which show that the
Test-1 has provided either very high ESR
values, which resulted normal or slightly
high with Westergren method, and falsely
low values (around 20 mm), while with the
Westergren method they were high (over 40
mm). Consider that different ESR levels may
have a different clinical significance and thus
address clinical decisions to one or another
direction
a graphic trick has been used to try and
minimize the visual impact of the
overestimation of ESR low values in the Test-1/Westergren graph: point 20 has a
different proportion on x and y axis (Figure 1
- Test-1 vs Westergren chart).
And now Bland and Altman’s plot
•
The Bland-Altman plot is used for the comparison among methods because,
compared to the correlation coefficient, it is able to detect if one of the methods is
affected by systematic errors of inaccuracy or imprecision.
•
Systematic error means that a measure system is prone to “constantly”
overestimate or underestimate the true object of the measure.
•
A systematic difference between the reported measures means that in the scatter
diagram the points line up along a straight line at approximately 45° without giving
any indications about the inaccuracy direction or entity.
And now Bland and Altman’s plot
•
•
The Bland and Altman method compares
the difference between the paired
measures performed with the two
methods (Westergren - Ves-Matic Cube),
with the average of these measures
((Westergren + Ves-Matic Cube)/2)
starting from the premise that the more
the two measures will be similar, the
lower the difference between them will
be, and therefore the more restricted will
be the limits of agreement (dashed lines)
At the limit, when the two measurements
coincide, the difference is = 0.
And now Bland and Altman’s plot
Limits of agreement (-13,9 to 12,9 mm/h)
Bias: - 0,5 mm/h
Limits of agreement (-29,9 to 51,8 mm/h)
Bias: -10,95 mm/h
Take home messages
• The analysis of scattergraphs (that "measure" the
strength of the correlation between two
methods, not their agreement, i.e. if the two
methods give very similar data) suggests that the
Ves-Matic Cube 200 correlates much better than
the Test-1 with the Westergren method. This can
be deduced because the value of the coefficient
of correlation is equal to 0,946, while the value of
coefficient of correlation of the Test-1 value is
equal to 0,386. Important: the closer to 1 is the
value of the correlation coefficient, the more two
methods are correlated.
Take home messages
•
The Bland-Altman method "measures" the real agreement between two methods. The lesser is the
difference between the two compared measures, the higher is the two data agreement (in fact, if
the two measures are equal, their difference is equal to zero). he Bland-Altman method indicates
the limits of agreement between the two methods: the narrower is the interval between the two
limits (i.e. the closer to zero is the difference between the two measures) the greater will be the
agreement between methods.
•
Ves-Matic Cube 200 limits of agreement:
mm/h)
Test 1 limits of agreement:
mm/h)
•
•
•
•
•
•
•
-13.9 - 12.9 mm/h (total 26,8
-29.9 - 51.8 mm/h (total 81,7
As you can see, the Ves-Matic Cube 200 limit of agreement is closer than the Test-1’s.
In addition to the agreement, the Bland-Altman method gives also indication about the bias or
deviation between the real value and the one measured by the system under evaluation:
Ves-Matic Cube 200 Bias :
Test-1 Bias :
- 0,50 mm/h
- 10,95 mm/h
The average deviation of the Ves-Cube values from the real one is only – 0,5 mm/h, while the Test-1
is about -11 mm/h, which means that the instrument has a general tendency to underestimation.
Take home messages
• In the Korean paper it is stated that the results of
the test-1 have a better correlation with
inflammation markers than the Westergren ESR
does
• If indeed the Test-1 data correlate differently
from ESR according to Westergren (which is the
original and reference method used as clinical
datum by physicians for decades) with
inflammation markers, this means that Test-1
datum is NOT ESR, and therefore it is unfairly
reimbursed.
Download