Correspondence
particular usefulness in assisting clinicians who are not focused
on retinoblastoma, and may be unaware of risks for children in
families with retinoblastoma.
ALISON H. SKALET, MD, PHD1
DAN S. GOMBOS, MD2,3,4
BRENDA L. GALLIE, MD5
JONATHAN W. KIM, MD6
CAROL L. SHIELDS, MD7
BRIAN P. MARR, MD8,9,10
SHARON E. PLON, MD, PHD2,11
PATRICIA CHÉVEZ-BARRIOS, MD2,3,12,13,14,15
1
Casey Eye Institute, Department of Ophthalmology, Oregon Health and
Science University, Portland, Oregon; 2The Retinoblastoma Center of
Houston, Houston, Texas; 3Texas Children’s Cancer Center, Baylor
College of Medicine, Houston, Texas; 4Section of Ophthalmology,
Department of Head & Neck Surgery, MD Anderson Cancer Center,
Houston, Texas; 5Department of Ophthalmology and Visual Sciences,
The Hospital for Sick Children, Ontario, Canada; 6Vision Center,
Children’s Hospital Los Angeles, Los Angeles, California; 7Ocular
Oncology Service, Wills Eye Hospital, Thomas Jefferson University,
Philadelphia, Pennsylvania; 8Department of Surgery, Memorial Sloan
Kettering Cancer Center, New York, New York; 9Department of
Ophthalmology, Weill Cornell Medical College, New York Presbyterian
Hospital, New York, New York; 10Department of Ophthalmology, New
York-Presbyterian/Columbia University Medical Center, New York,
New York; 11Department of Molecular and Human Genetics, Baylor
College of Medicine and Texas Children’s Hospital, Houston, Texas;
12
Departments of Pathology and Genomic Medicine and
Ophthalmology, Houston Methodist Hospital, Houston, Texas;
13
Departments of Pathology and Laboratory Medicine and
Ophthalmology, Weill Cornell Medical College, New York;
14
Department of Pathology and Laboratory Medicine, University of
Texas MD Anderson Cancer Center, Houston, Texas; 15Center for Cell
and Gene Therapy, The Texas Children’s Cancer Center, and
Department of Ophthalmology, Baylor College of Medicine, Houston,
Texas
Financial Disclosures: The authors made the following disclosures:
B.M.: Consultant e Aura Bio science; Stock options, Travel expenses e
Aura Bio science, outside the submitted work.
B.G.: Expert testimony e Legal cases; Grants e Retina Society, all
outside the submitted work.
D.G.: Travel expenses e Children’s Oncology Group, American Board
of Ophthalmology; Consultant e American Board of Ophthalmology,
AbbVie, Aura Biosciences, American Society of Clinical Oncology;
Castle Biosciences; Grant e Housemann/Wilkins Foundation, Lois Kuss
Fund for Glaucoma Research, all outside the submitted work.
P.C.-B.: Grant e NASA; Lecture fees e UT Houston Medical School;
Travel expenses e Children’s Oncology Group, all outside the sub-
mitted work.
A.A.: Grant e Lloyd Research Endowment Faculty Grant; Consultant e
Castle Biosciences Inc, outside the submitted work.
S.P.: Board member e Baylor Genetics, American Society of Human
Genetics; Grant e National Institutes of Health; Royalties e Antibody
users; Travel expenses e American College of Medical Genetics and
Genomics, American Association of Cancer Research, all outside the
submitted work.
Available online: July 3, 2018.
Correspondence:
Patricia Chévez-Barrios, MD, Houston Methodist Hospital, 6565 Fan-
nin, M227, Department of Pathology & Genomic Medicine, Houston,
TX 77030. E-mail: pchevez-barrios@houstonmethodist.org.
References
1. Skalet AH, Gombos DS, Gallie BL, et al. Screening children at
risk for retinoblastoma: consensus report from the American
Association of Ophthalmic Oncologists and Pathologists.
Ophthalmology. 2018;125:453e458.
2. Kamihara J, Bourdeaut F, Foulkes WD, et al. Retinoblastoma
and neuroblastoma predisposition and surveillance. Clin Cancer
Res. 2017;23:e98ee106.
3. FDA Drug Safety Communication. FDA review results in new
warnings about using general anesthetics and sedation drugs in
young children and pregnant women. Available from: www.fda.
gov/Drugs/DrugSafety/ucm532356.htm.
4. Moll AC, Imhof SM, Meeteren AY, et al. At what age could
screening for familial retinoblastoma be stopped? A register
based study 1945-98. Br J Ophthalmol. 2000;84:1170e1172.
5. Rothschild PR, Lévy D, Savignoni A, et al. Familial retino-
blastoma: fundus screening schedule impact and guideline pro-
posal. A retrospective study. Eye (Lond). 2011;25:1555e1561.
Re: Wang et al.: Reversal of
glaucoma hemifield test results and
visual field features in glaucoma
(Ophthalmology. 2018;125:352-360)
TO THE EDITOR: We read with great interest the retrospective cohort
study by Wang et al1 describing a visual field (VF) feature model to
predict the reversal of glaucoma hemifield test to within normal
limits after 2 consecutive outside normal limits results. Although
the article was insightful and well-written, we seek clarification
on some concerns.
The authors stated that the 16 VF archetypes were identified by
an unsupervised machine learning method (archetypal analysis)
based on >13 000 reliable VFs, based on a database of 13 321 24-2
VFs previously published in literature.2 However, the reliability
criteria for the 24-s VFs were different between these 2 studies,
because Elze et al2 included all 24-s VFs with fixation losses of
<33% and false negatives of <20% (without mentioning false
positives),2 whereas fixation losses of
33% and false positives
and false negatives of
20% were the reliability inclusion criteria
for 24-2 VFs of the present study. We wonder whether this
difference could potentially lead to interpretation bias of the
results of the present study, because the accuracy of the
archetypal analysis in identifying VF archetypes has never been
tested and verified on SITA 24-2 VFs with the present study’s
more strict reliability criteria.
Second, the authors suggest that a model purely based on
parameters and features from 2 previous reliable 24-2 VFs can
predict the glaucoma diagnosis and can help clinicians in the
decision-making process, although clinicians naturally rely on other
clinical data to make glaucoma management decisions. However,
e65
 Ophthalmology Volume 125, Number 9, September 2018
because patients’ clinical data were available only from one of the
centers’ dataset, only 40 eyes were diagnosed with glaucoma based
on glaucomatous optic disc changes on fundus photographs, cor-
responding VF defects and OCT-retinal nerve fiber layer (RNFL)
images; thus, the authors’ conclusions might be not adequately
powered. It would be more useful and interesting to confirm in all
eyes of this study (or at least in a higher proportion of eyes) with
structural imaging whether the probable early glaucoma patients
(lower mean deviation, no glaucoma hemifield test reversal on 3
consecutive reliable 24-2 VFs þ VF archetypes suggesting early
glaucoma) had corresponding abnormalities confirmed by
OCT-RNFL.3 Also, the authors state that for the eyes from
Massachusetts Eye and Ear dataset, an assessment of glaucoma
status at the time of the third VF test was made based on the
consensus of 2 glaucoma specialists by reviewing fundus
photographs for glaucomatous optic disc changes and OCT
images for characteristic nerve fiber layer thinning closest to the
test date of the third VF. However, if this assessment was
performed several years from when the third VF was selected,
this remains the “closest” to the third VF date but may not be so
relevant. We also wonder how long after was the reliable VF that
postdated the third test (when structural data were equivocal) was
performed in order to confirm or exclude the diagnosis of
glaucoma.
Finally, the authors did not specify what was the interval be-
tween the various VFs, and we wonder whether an earlier glaucoma
diagnosis based on the correspondence between optic discs
appearances and VF defects suggesting glaucomatous damage, and
confirmed by correspondent changes at the OCT-RNFL would be
preferable to the model suggested by the authors (no glaucoma
hemifield test reversal on 3 consecutive reliable 24-2 VFs þ VF
archetypes suggesting early glaucoma), which can be significantly
influenced by the patients’ prior experience with VF testing and
might be considered obsolete at the present OCT-RNFL era.
We commend the author’s frank acknowledgement in the dis-
cussion of the limitations of the present study but seek clarification
of the points we have raised.
PIERGIACOMO GRASSI, MD
HENRIETTA HO, FRCOPHTH, FAMS
KIN SHENG LIM, FRCOPHTH, MD
Department of Ophthalmology, St Thomas’ Hospital, London, United
Kingdom
Financial Disclosures: The authors have no proprietary or commercial
interest in any materials discussed in this article.
Available online: July 3, 2018.
Correspondence:
Piergiacomo Grassi, MD, Department of Ophthalmology, St Thomas’
Hospital, Westminster Bridge Road, London, London SE1 7EH, United
Kingdom. E-mail: pjgrassi@libero.it.
References
1. Wang M, Pasquale LR, Shen LQ, et al. Reversal of glaucoma
hemifield test results and visual field features in glaucoma.
Ophthalmology. 2018;125:352e360.
e66
2. Elze T, Pasquale LR, Shen LQ, et al. Patterns of functional
vision loss in glaucoma determined with archetypal analysis.
J Royal Soc Interface. 2015;12(103).
3. Brusini P. OCT Glaucoma Staging System: a new method for
retinal nerve fiber layer damage classification using spectral-
domain OCT. Eye (Lond). 2018;32:113e119.
REPLY: We thank Grassi et al for their comments and for the
opportunity to clarify the implications of our work. In our
work,1 we statistically predicted the reversal of the glaucoma
hemifield test (GHT) to “within normal limits” after 2 consecutive
test results outside normal limits. Because it is commonly assumed
that glaucomatous visual field (VF) loss is irreversible, a GHT
reversal to normal at the third test can be considered to indicate
false-positive results of the 2 preceding outside the normal limits
tests. We combined a set of parameters, including representative
patterns of glaucomatous VF tests (so-called archetypes determined
by statistical learning2) to predict such reversals solely from the
appearance of the first 2 VF measurements. As we hypothesized,
combining patterns of glaucomatous damage to summary metrics
such as the GHT increased our certainty that functional loss was
likely glaucomatous, and not simply the result of long-term
testeretest variability. This finding was further validated in a
subsample with additional available clinical information.
Grassi et al correctly point out the reliability criteria to include
VFs in the present study were stricter than the criteria used to
statistically learn the archetypes in the earlier study.2 However, we do
not believe that this represents a bias in this study. If parameters of a
cross-validated model (i.e., a model trained and evaluated on
different data sets) improve model performance, these parameters are
meaningful. If, as in our case, the parameters were determined based
on data with a potentially higher noise level, the cross-validation error
may have increased, that is, we might have underestimated (instead
of overestimated) the true importance of the archetypes. Because the
inclusion of the archetypes yielded a substantial model evaluation
improvement, this demonstrates the archetypes are robust enough to
act as representative patterns of glaucomatous VF loss, even when
applied to new datasets.
After having evaluated the model performance, we concluded
our study with an additional analysis that compared the GHT and
our model with the diagnosis of 2 glaucoma specialists who
reviewed additional optic nerve imaging. Grassi et al express
concern that the 97 subjects in this clinician-diagnosed dataset
might be too small and that structural properties of all 16 604 eyes
should have been included. Apart from the fact that structure-based
diagnoses were not possible for most subjects in this large,
de-identified dataset, we would like to stress that our work entirely
focuses on the GHT, that is, a purely functional glaucoma impair-
ment measure. Although we acknowledge a clinical glaucoma
diagnosis should be based on various parameters, some of them
related to retinal structure, an evaluation of functional vision
separate from possible structural damage is helpful for clinicians to
interpret possibly confounding structural features and thereby to
improve diagnostic accuracy. Therefore, we decided not to add
optic nerve structure to predict GHT false positives. The additional
structure-based model evaluation was only included to demonstrate
our model not only optimized the prediction of a GHT reversal, but
performs well even when predicting a structure-based negative