F
c
f
i
a
t
m
m
r
d
c
t
i
i
r
c

1
t
m
T
E
E
R
l
d
t
m
c
f
l
c
t
fi
r
c
o
t

A

s

V
S

0
d

EDITORIALS

The Clinical Site-Reading Center Partnership in Clinical Trials
RONALD P. DANIS

t
e
t
a
p
q
O
M
U
R
a
a
e
a
f
e

c
v
r
t
e
i
d
c
p
R
s
b
o

s
o
o
m
s
n
s
f
g
e
t
b

ORMER U.S. PRESIDENT MARTIN VAN BUREN ONCE

wrote, “It is easier to do a job right than to explain why
you didn’t.” This dictum is applicable to all aspects of

linical trials. The clinical site-reading center partnership
ocuses on collection of high-quality, standardized, unbiased
maging data. High-quality data collection by clinical sites
ccording to standardized methodology has been central to
he majority of studies leading to the most exciting advance-
ents in ophthalmic patient care. The current model of
ulticenter clinical trials employing coordinating centers and

eading centers has evolved partly to promote a high-level of
ata quality through training, certification, and monitoring of
linical site data collection. This Editorial will briefly review
he roles of reading centers in clinical trials, the purpose of
maging protocols, certification, and image quality monitor-
ng, quality control within reading centers, clinical site
esponsibilities, and the impact of new technology in large
linical trials.

As residents in residents in ophthalmology during the
980s at the University of Wisconsin, we were steeped in
he importance of standardized independent evaluation to
inimize bias and variability in clinical trial assessments.
he seminal reports from the Diabetic Retinopathy Study,
arly Treatment Diabetic Retinopathy Study, Wisconsin
pidemiologic Study of Diabetic Retinopathy, Diabetic
etinopathy Vitrectomy Study, and Macular Photocoagu-

ation Study had radically changed the standard of care for
iabetic retinopathy and neovascular macular degenera-
ion, and they relied heavily upon photographic assess-
ents from reading centers. Later, however, in the role of

linical investigator recruiting patients for clinical trials
rom a busy retina subspecialty practice, I was shocked to
earn the minute details specified and required by reading
enters – photographic field definition, stereoscopy, film
ype (sometimes by film lot number), processing by certi-
ed labs, labeling and sorting of slides, etc. Meeting these
equirements was sometimes a frustrating burden upon the
linic staff and patients and sometimes seemed downright
bsessive. My liberation came with the gradual apprecia-
ion of the relationship between process and procedure at

ccepted for publication Jul 15, 2009.
From the Department of Ophthalmology and Visual Sciences, Univer-

ity of Wisconsin – Madison, Madison, Wisconsin.
Inquiries to Ronald P. Danis, Department of Ophthalmology and
e
isual Sciences, University of Wisconsin – Madison, 406 Science Drive,
uite 400, Madison, WI 53711; e-mail: rdanis@rc.ophth.wisc.edu

© 2009 BY ELSEVIER INC. A002-9394/09/$36.00
oi:10.1016/j.ajo.2009.07.017
he site and data quality. These minute requirements
xisted because the experience of the reading center taught
hat systems can fail without them. This renewed my
ppreciation for the clinical research coordinator and
hotographers who maintained the high-level of data
uality from our site despite my occasional impatience.
ur staff understood Van Buren’s quip better than I did.
y role has changed again and now, as Director of the
niversity of Wisconsin-Madison Fundus Photograph
eading Center, I have the opportunity to respect and
dmire the many clinical investigators and their staff that
re able to meet the requirements of clinical research with
nthusiasm. Such sites share in common the attributes of
dequate staff time dedicated to clinical research, thorough
amiliarity with the study protocol and procedures, and
xperience.

Ophthalmic reading centers are diverse and include
enters for interpretation of optic nerve head topography,
isual fields, corneal endothelial images, and a variety of
etinal images. Reading centers provide image evaluations
hat are uniform across clinical sites by evaluators who are
asily masked to treatment assignment and other clinical
nformation. They have the potential to employ more
etailed observations than what might be possible in
linics and to develop disease classifications from them; a
rime example has been the Early Treatment Diabetic
etinopathy Study (ETDRS) diabetic retinopathy severity

cale from stereoscopic color photographs,1 which has
een used as a major outcome in multiple important trials
f diabetes complications.
The requirements for image analysis in clinical trials are

pecific to each study. The importance of morphology
utcomes for a study weighs in the balance. If the primary
utcome is a functional assessment such as vision, there
ay be no need for a reading center to evaluate images. In

tudies where morphology is the primary outcome, the
eed for masked standardized independent assessment
hould be carefully considered. Imaging data may be useful
or the development of disease classifications, hypothesis
eneration, and subgroup analyses. Masked independent
valuation to control bias is particularly important when
he treatment benefit is small, a function which often can
e more easily and reliably performed in a reading center

nvironment.

LL RIGHTS RESERVED. 815

mailto:rdanis@rc.ophth.wisc.edu


p
u
u
t
a
a
c
m
c
i
j
R
i
p

a
c
w
p
t
a
o
p
i
d
s
c
m
a
b
B
i
i
r
c
w
i
r
T
c
r
p
m
o
q

m
a
v
d
“
t
s
m
c

a
d
A
u
i
f
i
i
f
d
h
t
i
h
s
s
t
t
m
p
c

m
a
i
n
p
h
s
i
q
L
t
i
t
i
m
a
d
t
m
i
m
m
c
a

f
f
O
d
r
e
s
r

8

The problem of variability in clinical assessments can be
artially overcome through the use of trained observers
sing a strict protocol in a centralized setting to promote
niformity.2,3 Historically, reading centers have excelled in
he interpretation of clinical images in a manner somewhat
nalogous to the physician observer but with greater detail
nd complexity than is possible in the clinic. The reading
enter evaluation may be repeated and the reproducibility
easured; an extremely difficult process to replicate in the

linic setting. Knowing the reproducibility of measurements is
mportant when statistically analyzing imaging data and in
udging significance of the data from a clinical perspective.2,3

eading centers continually test and report the reproducibil-
ty of grading for each data item. A systematic quality control
rogram is necessary as part of the grading process.

The reading center prepares protocols for image capture
nd submission, trains the site staff on these procedures, and
ertifies the imagers. The imaging protocol is detailed and
ritten clearly so as to standardize data collection as much as
ossible. Digital imaging requires the protocol to be specific to
he make and model of the equipment because most camera
nd optical coherence tomography (OCT) manufacturers in
phthalmology have developed proprietary software that
roduce unique non-standardized file formats. Certification
s the formal process by which the clinical site staff
emonstrate understanding and ability to adhere to the
tudy protocol before the first patient is enrolled. The
ertification process may be quite simple for some imaging
ethods, and very difficult for others. Inexperienced im-

gers may make multiple attempts and even then fail to
ecome certified – a source of frustration for all involved.
ecause of the pressures on modern clinical practices for

ncreased efficiency and productivity in the face of declin-
ng revenue, the clinical site staff member pressed into the
ole of imager is sometimes incompletely trained for
linical trial work. Clinical practice demands for imaging,
hich emphasize capture of the abnormalities of particular

nterest in an individual patient, are often different from the
igorous standard protocols typically used in clinical research.
he diligence required by the imager to become certified for
linical trial protocols, with feedback and advice from the
eading center, often improves the quality of the work
erformed for patient care purposes. Reading centers also
onitor and report upon image quality throughout the course

f a trial in order to identify and help correct problems
uickly should they occur.

Image quality issues are a common source of variability and
issing data in clinical trials. For instance, if images of an eye

t one visit show retinal neovascularization, but at the next
isit the image is poorly focused or the photographic field
oes not capture the area in question, there may be a spurious
disappearance” of the neovascularization that mimics the
herapeutic effect of an intervention. If the images are
everely flawed (fortunately, a rare event), the reading center
ay not be able to produce data for those visits. Missing data
an have a deleterious impact upon the statistical analyses i

AMERICAN JOURNAL OF16
nd results in a clinical trial, and, if there is enough missing
ata, the outcomes of the trial may be called into question.4

n error in obtaining baseline images may severely limit the
sefulness of imaging data on that patient. Of particular
nterest from an image quality standpoint is the transition
rom color film fundus photography to digital color fundus
mages in clinical trials. Because digital camera chips handle
llumination, contrast, and color balance quite differently
rom film,5 obtaining images similar to film quality with
igital cameras has been challenging. Post hoc image en-
ancement at the reading center can improve the illumina-
ion, color balance, and contrast of substandard digital
mages.5 This may help preserve continuity of grading with
istorical film data sets and minimize variability within
tudies using both film and digital photography. The best
olution is for the imagers at the clinical sites to educate
hemselves regarding the assessment and modification of
onal balance and how this is accomplished with the equip-
ent available at the clinic to produce the highest quality

hotographs – this is beneficial for both research and clinical
are purposes.

Technological advances have lead to instruments that
ake automated measurements that formerly were clinical

ssessments. For example, macular edema assessment by OCT
s clearly a better method for measurement of retinal thick-
ess at the center of the macula than stereoscopic color
hotographs or the eye of the clinician. In general, the
uman mind/eye remains (for the moment) superior to
oftware for purposes of lesion classification in complex
mages under a variety of quality conditions, for solving
uality issues, and for handling unusual disease presentations.
ooking forward, it is inevitable that automated lesion detec-
ion, classification, and measurement will become increas-
ngly reliable and used more frequently in the clinic setting. Is
here a future role for reading centers in the context of
ncreasing technologic innovation and automated measure-
ent? Actually, new imaging technology (eg, OCT, fundus

utofluorescence) seems to increase, rather than decrease, the
emand for reading center services. This is in part attributable
o the uncertainty of how best to classify disease with the new
ethodology, and to identify and rectify new imaging quality

ssues that present with new technology. Even if the only
orphologic outcome variable to be analyzed is an automated
easurement obtained directly from an instrument at the

linical site, a reading center may be of value for certification
nd quality control of imaging.

The evidence base upon which medical care of patients is
ounded is increasingly dependent upon the data from care-
ully designed and conducted multicenter clinical trials.6

cular imaging analysis in support of such trials must be of
emonstrable high quality. Without this assurance, a study
uns the risk of running afoul of Van Buren’s credo. In the
nd, it is the diligent efforts of clinical site investigators and
taff, and the patients that donate time and their own
esources, that create the foundation of ophthalmology clin-

cal research.

OPHTHALMOLOGY DECEMBER 2009



T
d
I

1

2

3

V

HE AUTHOR INDICATE NO FINANCIAL SUPPORT OR FINANCIAL CONFLICT OF INTEREST. THE AUTHOR WAS INVOLVED IN
esign and conduct of study; collection of data; management, analysis, and interpretation of data; and preparation, review, and approval of manuscript.
nstitutional Review Board approval was not needed for this study.

The author wishes to thank Dr Matthew D. Davis, University of Wisconsin-Madison, Madison, Wisconsin, for critical review of this manuscript.
4

5

6

REFERENCES

. Early Treatment Diabetic Retinopathy Study Research Group.
Grading diabetic retinopathy from stereoscopic color fundus photo-
graphs – an extension of the modified Airlie House classification.
ETDRS Report No. 10. Ophthalmology 1991;98:786 – 806.

. Lachin JM. The role of measurement reliability in clinical
trials. Clin Trials 2004;1:553–566.

. Ederer F. Methodological problems in eye disease epidemiol-

ogy. Epidemiol Rev 1983;5:51– 66.

EDITORIOL. 148, NO. 6
. Minckler DS, Vedula SS, Li TJ, Mathew MC, Ayyala RS,
Francis BA. Aqueous shunts for glaucoma. Cochrane Database
Syst Rev 2006:CD004918.

. Hubbard LD, Danis RP, Neider MW, et al. Brightness,
contrast, and color balance of digital versus film retinal images
in the age-related eye disease study 2. Invest Ophthalmol Vis
Sci 2008;49:3269 –3282.

. Ferris FL. Clinical trials – more than an assessment of
treatment effect: LXV Edward Jackson Memorial Lecture.

Am J Ophthalmol 2009;147:22–32.e21.

AL 817