Content uploaded by Noel Brennan
Author content
All content in this area was uploaded by Noel Brennan on May 09, 2019
Content may be subject to copyright.
MYOPIA CONTROL INTERVENTION PRODUCES ABSOLUTE, RATHER THAN
RELATIVE, TREATMENT EFFECT ACROSS THE PROGRESSION RANGE
© Johnson & Johnson Vision, Inc. 2019. All rights reserved.
Purpose
Method
Results
Conclusion
References
4344
Noel A. Brennan1MScOptom, PhD, FAAO Xu Cheng1MD, PhD, FAAO Mark A. Bullimore2MCOptom, PhD, FAAO, FARVO
1. Johnson & Johnson Vision 2. University of Houston
ARVO 2019
•Myopia control treatment efficacy is commonly presented as a
‘percentage’ (see refs 1-8 for a small sample)
•However, normalized cumulative frequency progression data for
treated and untreated eyes in myopia control trials are fit with lines
of near equivalent gradient, strongly suggesting consistent absolute
reduction in axial elongation across the progression range, rather
than percentage reduction.9,10 (see figure 1) This approach relies on
the assumption that progression at a given percentile in the control
group serves as an adequate match for progression at the same
percentile in the treatment group.
•A clinical trial where intervention was applied monocularly offers the
opportunity to confirm the validity of this assumption, by using
progression data from untreated eyes as proxies for propensity to
progress in treated eyes.11
Figure 1: Standardized cumulative frequency of 6 and 12 month axial elongation for test and
control lens populations from Cheng et al.12 The parallel nature of the lines suggests absolute
rather than relative treatment effect across the progression range.
•Axial elongation (see boxout below) for the first 10-mo period of Anstice &
Phillips3(their fig. 5B) were digitized using ImageJ.13 Significant correlation
between treated and untreated eyes (R=0.54, P < 0.001) indicates that the
untreated eye can be used for the propensity of the treated eye to progress.
•While Anstice & Phillips perform a simple regression to examine the relation
between the treated eye and the contralateral control eye, this type of
regression does not enable accurate prediction of ‘y’ values from ‘x’ values if
there is significant variance in the ‘x’ measurement (see boxout below).
•Deming regression was performed to account for variance in measures of both
treated and control eyes with equal variance in both measures assumed.
•A ‘Bland-Altman-like’ plot was also constructed to aid visualization.
Figure 2: Reproduction of figure 5B from Anstice & Phillips3plotting treated eye
axial elongation versus contralateral control eye with overlaid red dots indicating
our digitized data points. The broken line is the line of equivalence and the black
line is their simple regression. The red line is our Deming regression line showing
near parallel slope to the line of equivalence (p=0.60), demonstrating constant
absolute treatment effect across the progression range.
Figure 3: Further illustration of the consistency of absolute treatment effect across
the progression range using a Bland-Altman-like plot. Fast progressors are to the
right and slow progressors to the left. Treatment effect is shown on the y-axis. Note
that considerable stochastic variation is apparent.
•This analysis corroborates our previous observation of myopia
control intervention providing a consistent absolute, rather than
relative, treatment effect across the progression range.
•Further research may demonstrate a relative component but
current evidence shows that the absolute effect predominates.
•Expressing myopia control treatment efficacy as a percentage
erroneously inflates expectations for treatment effect in faster
progressors.
•There is a considerable literature outside the ophthalmic field
warning of the pitfalls of using ‘percentage’ effects.14-17
•This observation has major implications for projected efficacy of
myopia control interventions.
1. Cooper J et al. Eye & Contact Lens. 2018;44: e16.
2. Sankaridurg P et al. Asia Pac J Ophthalmol. 2018;7:405.
3. Kang P. Clin Exp Optom. 2018;101:321.
4. Wildsoet CF et al. Invest Ophthalmol Vis Sci. 2019;60:M106.
5. Cooper J, Tkatchenko AV. Eye & Contact Lens. 2018;44;231.
6. Lipson MJ et al. Eye & Contact Lens. 2018;44:224.
7. Leo SW. Curr Opin Ophthalmol. 2017;28:267.
8. Walline JJ. Eye & Contact Lens. 2016;42:3.
9. Brennan NA, Cheng X. Ophthalmic Physiol Opt https://online
library.wiley.com/action/downloadSupplement?doi=10.1111%2Fopo.12459
&file=opo12459-sup-0002-DataS2.docx O028.
10. Brennan NA, Cheng X, Eye & Contact Lens. 2018 PAP doi:
10.1097/ICL.0000000000000566
11. Anstice N, Phillips J. Ophthalmology 2011:118:1152
12. Cheng X et al. Optom Vis Sci
13. https://imagej.nih.gov/ij/
14. Agarwal A et al. J Clin Epidemiol. 2017;81:3.
15. Faraone SV. P & T 2008;33:700.
16. Heneghan C et al. Trials. 2017;18:122.
17. King NB et al. BMJ. 2012;345:e5774.
Why Axial Length?
Most papers that refer to the need to slow myopia
progression state that the principal purpose is to
reduce disease risk by limiting axial elongation.
•Atropine appears to slow refractive progression,
at least in part, through changes to the crystalline
lens as well as axial length (see figure).
•It is reasonable to assume that reducing refractive
progression without analagous effect on axial
length partially corrects refractive error but does
not reduce disease risk correspondingly.
Interferometric axial length measurement is
relatively more sensitive than refractive error
measurement by a factor of at least 3 times.
Refractive error can not be accurately assessed
during orthokeratology.
What is Deming Regression? Future Directions (or ‘sneak peek’!)
See Poster 4345
The top figure shows a set of randomly generated data points with
loose correlation (R 0.50) The black line is simple regression of y
on x. This line is calculated by minimizing the sum of the squared
vertical distance from the points to the line. This approach
assumes that all variance is in the ‘y’ measures and that the ‘x’
values are known exactly. The red line is the regression of x on y. In
this case, the horizontal distances from the points to the line
would be used. Note that the two lines are different.
Where there is variance in the ‘x’ measures and we wish to predict
one set of measures from the other, a Deming regression is
appropriate. In our example (see bottom figure), equal variance is
assumed and the distances used to calculate the sum of squares,
and therefore the line of best fit, is at 45°to the vertical.
If treatment effect is absolute across the
progression range, this should be evident across
age. The figure below demonstrates this (and
explains why Aller et al. got 79% efficacy as well).
The graph below plots absolute versus relative
treatment effect, showing again the folly of using
‘percentage’ effect. For this and more, come and
see us at IMC in Tokyo, September 2019.
Disclosure:
Noel Brennan and Xu Cheng are employees of Johnson & Johnson Vision.
Mark Bullimore consults for Johnson & Johnson Vision, Alcon Laboratories,
Inc., CooperVision, Inc., Essilor of America, Eyenovia, Inc., Genentech, Inc.,
Innovega, Inc., jCyte, Inc., Novartis Pharma AG, and Tear Film Innovations.
