ArticlePDF Available

Abstract

Our study aimed to validate a model to determine a personalised screening frequency for diabetic retinopathy. A model calculating a personalised screening interval for monitoring retinopathy based on patients' risk profile was validated using the data of 3,319 type 2 diabetic patients in the Diabetes Care System West-Friesland, the Netherlands. Two-field fundus photographs were graded according to the EURODIAB coding system. Sight-threatening retinopathy (STR) was considered to be grades 3-5. Validity of the model was assessed using calibration and discrimination measures. We compared model-based time of screening with time of STR diagnosis and calculated the differences in the number of fundus photographs using the model compared with those in annual or biennial screening. During a mean of 53 months of follow-up, 76 patients (2.3%) developed STR. Using the model, the mean screening interval was 31 months, leading to a reduced screening frequency of 61% compared with annual screening and 23% compared with biennial screening. STR incidence occurred after a mean of 26 months after the model-based time of screening in 67 patients (88.2%). In nine patients (11.8%), STR had developed before the model-based time of screening. The discriminatory ability of the model was good (C-statistic 0.83; 95% CI 0.74, 0.92). Calibration showed that the model overestimated STR risk. A large reduction in retinopathy screening was achieved using the model in this population of patients with a very low incidence of retinopathy. Considering the number of potentially missed cases of STR, there is room for improvement in the model. Use of the model for personalised screening may eventually help to reduce healthcare use and costs of diabetes care.
ARTICLE
Validation of a model to estimate personalised screening
frequency to monitor diabetic retinopathy
Amber A. W. A. van der Heijden &Iris Walraven &Esther van tRiet&
Thor Aspelund &Sigrún H. Lund &Petra Elders &Bettine C. P. Polak &
Annette C. Moll &Jan E. E. Keunen &Jacqueline M. Dekker &Giel Nijpels
Received: 14 November 2013 /Accepted: 28 March 2014
#Springer-Verlag Berlin Heidelberg 2014
Abstract
Aims/hypothesis Our study aimed to validate a model to de-
termine a personalised screening frequency for diabetic
retinopathy.
Methods A model calculating a personalised screening inter-
val for monitoring retinopathy based on patientsrisk profile
was validated using the data of 3,319 type 2 diabetic patients
in the Diabetes Care System West-Friesland, the Netherlands.
Two-field fundus photographs were graded according to the
EURODIAB coding system. Sight-threatening retinopathy
(STR) was considered to be grades 35. Validity of the model
was assessed using calibration and discrimination measures.
We compared model-based time of screening with time of
STR diagnosis and calculated the differences in the number of
fundus photographs using the model compared with those in
annual or biennial screening.
Results During a mean of 53 months of follow-up, 76 patients
(2.3%) developed STR. Using the model, the mean
screening interval was 31 months, leading to a reduced
screening frequency of 61% compared with annual screen-
ing and 23% compared with biennial screening. STR
incidence occurred after a mean of 26 months after the
model-based time of screening in 67 patients (88.2%). In
nine patients (11.8%), STR had developed before the
model-based time of screening. The discriminatory ability
of the model was good (C-statistic 0.83; 95% CI 0.74,
0.92). Calibration showed that the model overestimated
STR risk.
Conclusions/interpretation A large reduction in retinopathy
screening was achieved using the model in this population of
patients with a very low incidence of retinopathy. Considering
the number of potentially missed cases of STR, there is room
for improvement in the model. Use of the model for
personalised screening may eventually help to reduce
healthcare use and costs of diabetes care.
Keywords Diabetic retinopathy .Retinal screening .
Screening intervals
Abbreviations
DCS Diabetes Care System West-Friesland
STR Sight-threatening retinopathy
A. A. W. A. van der Heijden and I. Walraven contributed equally to this
study.
Electronic supplementary material The online version of this article
(doi:10.1007/s00125-014-3246-4) contains peer-reviewed but unedited
supplementary material, which is available to authorised users.
A. A. W. A. van der Heijden (*):I. Walraven :E. van tRiet:
P. Elders :B. C. P. Polak :A. C. Moll :J. M. Dekker :G. Nijpels
EMGO Institute for Health and Care Research, VU University
Medical Center, van der Boechorststraat 7, 1081 BT Amsterdam,
The Netherlands
e-mail: a.vanderheijden@vumc.nl
A. A. W. A. van der Heijden:P. Elders :G. Nijpels
Department of General Practice, VU University Medical Center,
Amsterdam, the Netherlands
I. Walraven :A. C. Moll
Department of Ophthalmology, VU University Medical Center,
Amsterdam, the Netherlands
I. Walraven :E. van t Riet :B. C. P. Polak :J. M. Dekker
Department of Epidemiology and Biostatistics, VU University
Medical Center, Amsterdam, the Netherlands
T. Aspe l u nd :S. H. Lund
Faculty of Medicine, University of Iceland, Reykjavik, Iceland
J. E. E. Keunen
Department of Ophthalmology, University Medical Center
St Radboud, Nijmegen, the Netherlands
Diabetologia
DOI 10.1007/s00125-014-3246-4
Introduction
With the growing prevalence of type 2 diabetes, diabetic
retinopathy is set to become the leading cause of visual
impairment and blindness [1]. Screening for diabetic retinop-
athy among diabetic patients is an effective method to de-
crease the disease burden by early diagnosis and timely med-
ical treatment [1]. Many countries have therefore adopted an
annual or biennial screening programme which is usually
integrated within regular diabetes care [25]. Nevertheless, a
one size fits allapproach is costly and time-consuming, as
only a minority of type 2 diabetic patients eventually develop
sight-threatening retinopathy (STR), which is mainly depen-
dent on several well-known risk factors such as diabetes
duration, HbA
1c
, blood pressure and early stages of retinopa-
thy [69]. Consequently, patients at low risk of STR may be
over-screened by using a predefined screening interval, while
those at high risk of developing STR may be missed. A
personalised screening interval in which the individual risk
of developing STR is taken into account might therefore be a
more appropriate or cost-effective method to screen for STR
in type 2 diabetic patients [2,3,1016].
Aspelund et al recently developed a model, based on
diabetes-related risk factors, in which they aimed to achieve
a more personalised approach to screening for STR in diabetes
[17]. However, assessment of the accuracy and validity of a
model should be performed before applying it to other popu-
lations. The Diabetes Care System West-Friesland (DCS) is a
large prospective study in which diabetes-related risk factors
and complications are extensively measured in patients with
type 2 diabetes, offering a unique opportunity to validate the
model. The aim of this study was therefore to validate the
personalised screening model of Aspelund et al [17]by
assessing the accuracy of predicting STR.
Methods
Study population The data of type 2 diabetic patients included
in the dynamic cohort of the DCS were used. The DCS
coordinates regional care for type 2 diabetes using a centrally
organised database available to all involved care givers.
Diabetes nurses and dietitians perform annual follow-up
examinations, including glucose control, cardiovascular
risk profiling and identifying the presence of micro- and
macrovascular complications [18].
In the study period between 1998 and 2010, the DCS
cohort consisted of 8,308 type 2 diabetic patients. The cohort
is dynamic, meaning that each year newly diagnosed diabetic
patients as well as patients with a longer duration of diabetes
enter the DCS cohort. For each patient, the year of entry was
considered as baseline. Because current type 2 diabetes guide-
lines recommend annual or biennial screening for retinopathy
[25], in most patients diagnostic fundus photography was
performed within 2 years after baseline. We therefore included
patients in whom the results of graded fundus photographs
and variables used to estimate screening frequency were pres-
ent at baseline (n=5,483). In patients with missing fundus
photograph results at baseline but in whom fundus photograph
results were present 1 or 2 years after baseline (n= 1,030), the
measurement taken 1 or 2 years after baseline was considered
to be the baseline measurement. Of the 6,513 patients, we
excluded patients in whom STR was already present at base-
line (EURODIAB grade 3, 4 or 5) (n=61). For 3,133 patients,
no graded fundus photographs within 1 year after model-
based time of screening were available. Patients who had no
graded fundus photographs available within 1 year after
model-based time of screening had a significantly higher age
(+1.4 years), shorter diabetes duration (0.3 years), lower
HbA
1c
(0.7%), lower systolic blood pressure (5.5 mmHg)
and lower diastolic blood pressure (5.1 mmHg) at baseline.
Furthermore, those patients had a significantly lower preva-
lence of grade 1 retinopathy (3.5%) and a comparable preva-
lence of grade 2 retinopathy (1.4%) at baseline and were
appointed a significantly longer (+9 months) screening inter-
val compared with those included in the current analyses.
Thus, 3,319 type 2 diabetic patients remained for the cur-
rent analyses. Patients who were included but censored due to
incomplete follow-up (n=991) had a significantly higher age
(+6.5 years), longer diabetes duration (+2.4 years), lower
HbA
1c
(0.21%) and lower diastolic blood pressure
(3.7 mmHg) compared with patients with complete 60-month
follow-up data. Mortality rates did not significantly differ be-
tween patients with complete (17.5%) or incomplete (16.3%)
follow-up (p=0.2).
Confirmation of diabetes and diabetes duration Type 2 dia -
betes was considered confirmed if at least one of the following
was reported by patientsgeneral practitioner: (1) one or more
classic symptoms (excessive thirst, polyuria, weight loss, hun-
ger or pruritis) and fasting plasma glucose 7.0 mmol/l or
random plasma glucose 11.1 mmol/l; (2) at least two
elevated plasma glucose levels on different occasions
(fasting glucose 7.0 mmol/l or random plasma glucose
11.1 mmol/l) in the absence of symptoms [19]. Diabetes
duration was calculated from the date of diabetes diagnosis
until the date of the baseline measurement.
Measurements Annual physical examinations were per-
formed following a standardised protocol. Weight and height
were measured with patients barefoot and wearing light
clothes. HbA
1c
was measured using HPLC. Fasting plasma
glucose was measured by means of a hexokinase method
(Roche Diagnostics, Mannheim, Germany). Levels of total
cholesterol, HDL-cholesterol and triacylglycerol were mea-
sured using enzymatic techniques (Boehringer-Mannheim,
Diabetologia
Mannheim, Germany). Systolic and diastolic blood pressures
were measured on the right arm after 5 min resting in a seated
position using a random-zero sphygmomanometer
(Hawksley-Gelman, Lancing, UK). Information on year of
onset of diabetes and on country of birth of the patient and
the patients parents was obtained by self-report [9]. Patients
were categorised into two ethnic groups: Western (European
countries [except for Turkey], Indonesia, USA and Oceania)
and non-Western (Turkey, Morocco, Surinam, Aruba, Neth-
erlands Antilles, Africa, Asia and Latin America).
Diabetic retinopathy From 1998 until 2000, fundus photog-
raphy of both eyes was performed with a Kowa Pro Fundus
camera fitted with a green filter (Kowa Optical Industry,
Torrance, CA, USA). From the beginning of 2000 until
2004, fundus photography of both eyes was performed with
a non-mydriatic Canon CR5 camera (Canon, Tokyo, Japan).
From 2004, fundus photography of both eyes was performed
with a non-mydriatic Topcon TRC NW 100 camera (Topcon,
Tokyo, Japan) [9]. All participants were examined with 45°
fundus photographs. One photograph was centred on the
macula and the other nasally, with the optic disc one disc
diameter from the temporal edge. Mydriasis with 0.5%
tropicamide and 2.5% phenylephrine eye drops was per-
formed when the non-mydriatic photograph was not gradable.
All photographs were graded by an experienced ophthalmol-
ogist who is trained as a retinal specialist. Each patientsgrade
of retinopathy was based on the grading of the worst eye. All
photographs were graded according to the EURODIAB clas-
sification score, in which grade 0 is no retinopathy,grade1
is minimal non-proliferative retinopathy(one or a few
scattered haemorrhages or microaneurysms), grade 2 is
moderate non-proliferative retinopathy,grade3issevere
non-proliferative or preproliferative retinopathy,grade4is
photocoagulated retinopathyand grade 5 is proliferative
retinopathy[20]. Because patients with grades 35retinopa-
thy are usually referred to an ophthalmologist for assessment
and/or treatment, grades 35 were included in the outcome
measure and considered as STR. If abnormalities were seen
near to the macula and maculopathy was suspected, the
EURODIAB grading was minimally set on 3 and the patient
was referred to an ophthalmologist.
Statistical analyses Variables are presented as percentages,
means (±SD) or medians (interquartile range) in case of a
skewed distribution. Differences in baseline characteristics
between patients with and without grades 12retinopathyat
baseline were tested by Studentsttest for normal distributed
variables and the MannWhitney Utest for variables with a
skewed distribution. Differences in proportions were tested by
the χ
2
test. Using the model of Aspelund et al, STR risk and an
accompanying screening interval ranging from 6 to 60 months
were calculated for each patient based on sex, HbA
1c
,systolic
blood pressure, presence of retinopathy and diabetes
duration [17]. The model is described briefly in the electronic
supplementary material.
In patients with incident STR during follow-up, we
checked whether STR occurred before or after the model-
based time of screening. Outcomes of omitted fundus photo-
graphs according to the model and potentially missed cases of
retinopathy grades 35 were checked. For the total population,
we calculated the reduction in screening frequency by com-
paring the mean screening interval of the population assigned
by the model to annual and biennial screening. The predictive
accuracy of the model was estimated using calibration and
discrimination techniques. Calibration is the ability of the
model to predict the number of observed cases of STR during
follow-up and it was visually checked by plotting the predict-
ed risk of developing STR against the observed incidence of
STR. Participants were grouped into quintiles of predicted
STR risk during 60 months of follow-up, which was calculat-
ed in the first step of the model. Using Poisson regression for
survival data, the observed incidence of STR was calculated
within each quintile, taking into account censored data [21].
Within each quintile, observed STR incidence was plotted
against predicted risk. Discrimination is the ability to distin-
guish between those who develop STR during follow-up from
those who do not. Discriminatory ability was estimated by
calculating Harrells C-statistic (similar to the area under the
receiver operating characteristic curve), censoring missing
data from patients who did not have complete 60 month
follow-up data [22]. Discriminatory power is graded as poor
for a C-statistic below 0.6, moderate between 0.6 and 0.8, and
good for >0.8.
Finally, we investigated whether updating the model for the
observed risk of developing STR within our study population
improved calibration and discrimination of the model.
Updating the model was performed by calculating the calibra-
tion factor: (1 the proportion of patients without STR after
60 months)/mean risk calculated by the model. The calibration
factor was then used to update the risk function calculated by
the model: calibration factor × mean risk calculated by
the model.
All statistical analyses were conducted using SPSS version
20 (SPSS, Chicago, IL, USA) and R version 3.0.2 (Vienna,
Austria) for Windows.
Results
A total of 3,319 type 2 diabetic patients were included in the
study, of whom 339 (10.2%) presented grade 1 or 2 retinop-
athy at baseline, mostly mild non-proliferative retinopathy
(grade 1) (9.1%). During a mean of 53 months of follow-up,
76 patients (2.3%) developed STR. Patients with prevalent
Diabetologia
grade 1 or 2 retinopathy at baseline were significantly older,
had longer diabetes duration and higher HbA
1c
levels com-
pared with patients without retinopathy at baseline (Table 1).
Tab le 2shows the model-based screening interval for the
total population (mean [SD] 31.0 [20.0] months), stratified for
patients with different grades of retinopathy at baseline. With
progressing grades of retinopathy at baseline, the model-based
screening interval shortened and the incidence of STR
increased.
Using the model, 2,468 patients (74.4%) were appointed a
screening interval longer than 12 months and 1,755 patients
(52.9%) were assigned a screening interval longer than
24 months. When using the model, screening frequency could
be reduced by 23% compared with biennial screening, and a
reduction of 61% could be achieved compared with annual
screening.
A total of 76 patients developed STR during follow-up. Of
these, 67 patients (88.2%) developed STR after the model-
based time of screening (mean [SD] 25.5 [22.2] months),
meaning that the model-based screening interval was safe in
these patients. Of these, five patients (7.5%) developed STR
within 1 year after baseline; according to the model, screening
should have been performed after 6 months. Of all the patients
who developed STR during a mean of 53 months of follow-
up, nine (11.8%) developed STR before the model-based time
of screening (mean [SD] 24.3 [13.2] months), which is later
than current care. Patients who were potentially missed by the
model had significantly lower systolic blood pressure levels
(127 vs 146 mmHg) at baseline compared with patients
screened in time. HbA
1c
level was not significantly different
between the two patient groups.
The C-statistic of the personalised screening model was
0.83 (95% CI 0.74, 0.92), indicating that the discriminatory
power of the model is a good fit. Calibration of the model is
depicted in Fig. 1a. The figure shows the observed incidence
of STR within each quintile of the predicted risk of developing
STR. The model overestimates the risk of developing grades
35 retinopathy. Figure 1b shows the calibration plot of the
observed incidence of STR against the predicted incidence of
STR after updating the model to the observed risk in our study
population. Compared with the initial calibration plot, the
observed incidence within quintiles of the updated predicted
risk lies closer to the diagonal line, meaning that updating the
model improved calibration of the model in our study popu-
lation. When using the updated model, the screening frequen-
cy could be reduced by 29% compared with biennial screen-
ing and a reduction of 65% could be achieved compared with
annual screening. Of all the patients who developed STR
during a mean of 53 months of follow-up, 13 developed
STR before the updated model-based time of screening (mean
[SD] 25.5 [14.3] months), which is later than current care.
Discussion
This study is the first to validate the model for personalised
screening of diabetic retinopathy of Aspelund et al [17].
Compared with conventional annual or biennial screening
intervals, use of the personalised screening model can attain
a substantial reduction in the number of screenings, while
safety and efficacy are barely compromised. Furthermore,
we showed that the model is a good fit and that it overesti-
mates the risk of developing STR.
Strengths of the present study include the large sample size
and its long-term follow-up with measurement of type 2
diabetes-related risk factors and complications. Furthermore,
fundus photographs were graded using the internationally
Tabl e 1 Baseline characteristics
of the type 2 diabetic patient
population stratified by
retinopathy grade (EURODIAB)
at baseline
Data are presented as proportions,
means (SD) or median
(interquartile range) in case of a
skewed distribution
*Significantly (p<0.05) different
from reference category
(no retinopathy)
Characteristic No retinopathy Grade 1 retinopathy Grade 2 retinopathy
n, % 2,980 (89.8) 302 (9.1) 37 (1.1)
Age, years 60.3 (11.3) 62.6 (12.0)* 62.4 (10.2)
Men, % 53.7 57.6 67.6
Ethnicity, % Western 91.4 87.6 85.7
BMI, kg/m
2
30.2 (5.4) 29.3 (5.1)* 30.1 (5.0)
Diabetes duration, years 1.1 (0.33.7) 2.8 (0.98.5)* 9.6 (2.714.4)*
HbA
1c
, % 7.6 (2.0) 7.8 (1.8) 8.4 (1.7)*
HbA
1c
, mmol/mol 60 (21) 61 (20) 68 (18)*
Systolic blood pressure, mmHg 145.3 (22.0) 146.5 (22.5) 148.8 (18.3)
Diastolic blood pressure, mmHg 83.2 (10.8) 83.6 (11.7) 82.3 (10.8)
Total cholesterol, mmol/l 5.3 (1.2) 5.3 (1.0) 5.7 (1.1)
LDL-cholesterol, mmol/l 3.2 (1.3) 3.3 (0.9) 3.5 (0.9)
HDL-cholesterol, mmol/l 1.2 (0.3) 1.2 (0.3) 1.1 (0.3)
Triacylglycerol, mmol/l 2.1 (1.7) 1.9 (1.0) 2.7 (2.3)
Diabetologia
accepted EURODIAB classification system [20], which en-
hances generalisability of our results to other populations.
Using the model, a reduction in screening frequency rang-
ing from 23% to 61% could be achieved compared with
biennial or annual screening, respectively. These reductions
are comparable to those of Aspelund et al, who showed in a
Danish population used to test the fit of the model that 59% of
the screenings could be safely omitted. Other studies have
already shown that annual screening intervals can be safely
prolonged for individuals at low risk of developing STR [2,3,
1016]. The model of Aspelund et al also integrated a shorter
(<12 months) screening interval for patients at high risk of
STR, perhaps diminishing the time to diagnosis of STR.
We showed that the model overestimated the risk of devel-
oping retinopathy in the present study cohort, which was also
true in the Danish cohort used to test the fit of the model [17].
Overestimation of the model may be due to several reasons.
First, the present study cohort consists of a well-treated group
of type 2 diabetic patients, in whom glucose and blood pres-
sure control are very well maintained. This may result in less
precise estimates of the relative contribution of the indepen-
dent risk factors in the model. Second, baseline prevalence of
grades 12 retinopathy in our cohort is lower than in the
cohorts used by Aspelund et al. Third, estimation of the risk
could have been less precise due to the relatively low inci-
dence of grades 35 retinopathy (2.3%). Updating the model
to the observed risk of STR in our population improved the
calibration of the model but did not lead to increased predic-
tive accuracy. Updating the model before using it in popula-
tions with a different prevalence of STR is therefore not
recommended.
Patients who developed STR during follow-up and were
missed by the model had significantly lower baseline systolic
blood pressure levels, which led to an underestimation of STR
risk. Taking into account previous trends in risk factors, med-
ication use and more specific information on the presence of
early stages of retinopathy (grades 12) might increase the
accuracy of the model in estimating patientsrisk of STR.
In the present analyses, the development of STR was
determined using the follow-up data until the first fundus
photograph after the model-based screening interval. This
might have led to a less precise estimation of the predictive
accuracy of the model. However, the time between model-
based time of screening and actual fundus photograph results
was restricted to a maximum of 1 year.
Mainly due to large model-based screening intervals and
the dynamic nature of the cohort, a considerable number of
patients had incomplete follow-up. The estimated risk of STR
in patients with incomplete follow-up was lower than in
patients with 5 years of follow-up. The predictive accuracy
of the model was determined by estimating calibration and
Tabl e 2 Model-based screening interval and cases of STR for the total population, stratified by grade of retinopathy (EURODIAB) at baseline
Variable Total population Retinopathy grade at baseline
012
N(%) 3,319 2,980 (89.8) 302 (9.1) 37 (1.1)
Incidence of STR during a mean of 53 months of follow-up, n(%) 76 (2.3) 31 (1.0) 27 (8.9) 18 (48.6)
Screening interval, months, mean (SD) 31.0 (20.0) 32.8 (19.8) 15.9 (14.4) 10.9 (10.2)
Screening interval, months, median (interquartile range) 26 (1253) 29 (1556) 10 (619) 6 (611)
Cases of STR missed if the model was applied
a
, mean (SD) 9 (0.3) 9 (0.3) 0 0
a
STR incidence before model-based time of screening
0
0.05
0.10
0.15
0.20
0 0.05 0.10 0.15 0.20
Observed incidence of sight
threatening retinopathy
Predicted risk of sight threatening retinopathy
a
0
0.05
0.10
0.15
0.20
0 0.05 0.10 0.15 0.20
Observed incidence of sight
threatening retinopathy
Predicted risk of si
g
ht threatenin
g
retino
p
ath
y
b
Fig. 1 Observed incidence of STR (EURODIAB grades 35) within
quintiles of predicted risk according to the model (a) and according to the
updated model based on STR incidence in the study population (b)
Diabetologia
discrimination of the model while censoring patients with
incomplete follow-up. Taking into account patients with in-
complete follow-up might have led to a more precise estimate
of the accuracy of the model.
We validated the model in a mainly white population.
Before extrapolation of the results to patients of other ethnic
origins, the model should first be validated in these groups.
The current study was performed in a real-life setting and
the fundus photographs were graded by only one grader. We
therefore have no information on inter- and intra-rater reliabil-
ity. The fundus photographs were graded by an experienced
ophthalmologist who was trained as a retinal specialist, en-
hancing reliable grading in a real-life setting [23]. Still, some
cases of STR might have been misclassified by the grader, due
to the possibility of over-grading background retinopathy as
preproliferative retinopathy and due to the enhanced possibility
of missing fine exudates by retinal specialists [24]. Differences
in grading protocols may also affect the validity of the
personalised screening model. Validation before implementing
the model in populations where other grading protocols are
used is therefore recommended.
A limitation of personalised screening is that it is dependent
on the accuracy and punctuality of the care system and care
professionals. The use of personalised screening intervals
makes the system more prone to errors: for example, missing
a fundus screening. It is therefore recommended to use a
computerised system when applying the model in clinical care
so that reminders for screening are automatically generated to
ensure safe follow-up procedures.
To conclude, we have validated a new model for
personalised diabetic retinopathy screening and demonstrated
that a large reduction in screenings for retinopathy was
achieved in this well-managed population with a relatively
short duration of diabetes and an extremely low incidence of
STR. Further improvement of the model might decrease the
number of missed cases of STR. Use of the model for
personalised screening may eventually help to reduce
healthcare use and costs of diabetes care.
Acknowledgements We gratefully acknowledge M. S. A. Suttorp-
Shulten (Department of Ophthalmology, OLVG, Amsterdam, the
Netherlands) for grading of the fundus photographs.
Funding This study was funded byZonMw, the Netherlands organisation
for health research.
Duality of interest The authors declare that there is no duality of
interest associated with this manuscript.
Contribution statement AAWAvdH was involved in the conception
and design of the study, analysis and interpretation of the data and drafting
of the manuscript. IW was involved in the analysis and interpretation of
the data and drafting of the manuscript. EvtR, TA, SHL, PE, BCPP,
ACM, JEEK, JMD and GN were involved in the interpretation of the
data and revision of the manuscript. GN was involved in the conception
of the study, interpretation of the data and revision of the manuscript. All
authors approved the final version of the manuscript. AAWAvdH and IW
are the guarantors of this work, had full access to the data in the study and
take responsibility for the integrity of the data and the accuracy of the data
analysis.
References
1. Ding J, Wong TY (2012) Current epidemiology of diabetic retinop-
athy and diabetic macular edema. Curr Diab Rep 12:346354
2. Chalk D, Pitt M, Vaidya B, Stein K (2012) Can the retinal screening
interval be safely increased to 2 years for type 2 diabetic patients
without retinopathy? Diabetes Care 35:16631668
3. Olafsdottir E, Stefansson E (2007) Biennial eye screening in
patients with diabetes without retinopathy: 10-year experience.
Br J Ophthalmol 91:15991601
4. Rodbard HW, Blonde L, Braithwaite SS et al (2007) American
Association of Clinical Endocrinologists medical guidelines
for clinical practice for the management of diabetes mellitus.
Endocr Pract 13(Suppl 1):168
5. Scanlon PH (2008) The English national screening programme for
sight-threatening diabetic retinopathy. J Med Screen 15:14
6. Cheung N, Mitchell P, Wong TY (2010) Diabetic retinopathy. Lancet
376:124136
7. Mehlsen J, Erlandsen M, Poulsen PL, Bek T (2011) Identification of
independent risk factors for the development of diabetic retinopathy
requiring treatment. Acta Ophthalmol 89:515521
8. Stratton IM, Kohner EM, Aldington SJ et al (2001) UKPDS 50: risk
factors for incidence and progression of retinopathy in Type II
diabetes over 6 years from diagnosis. Diabetologia 44:156163
9. Zavrelova H, Hoekstra T, Alssema M et al (2011) Progression and
regression: distinct developmental patterns of diabetic retinopathy in
patients with type 2 diabetes treated in the Diabetes Care System
West-Friesland, the Netherlands. Diabetes Care 34:867872
10. Agardh E, Tababat-Khani P (2011) Adopting 3-year screening inter-
vals for sight-threatening retinal vascular lesions in type 2 diabetic
subjects without retinopathy. Diabetes Care 34:13181319
11. Kohner EM, Stratton IM, Aldington SJ, Holman RR, Matthews DR
(2001) Relationship between the severity of retinopathy and progres-
sion to photocoagulation in patients with Type 2 diabetes mellitus in
the UKPDS (UKPDS 52). Diabet Med 18:178184
12. Looker HC, Nyangoma SO, Cromie DT et al(2013) Predicted impact
of extending the screening interval for diabetic retinopathy: the
Scottish Diabetic Retinopathy Screening programme. Diabetologia
56:17161725
13. Porta M, Maurino M, Severini S et al (2013) Clinical characteristics
influence screening intervals for diabetic retinopathy. Diabetologia
56:21472152
14. Stratton IM, Aldington SJ, Taylor DJ, Adler AI, Scanlon PH (2013)
A simple risk stratification for time to development of sight-
threatening diabetic retinopathy. Diabetes Care 36:580585
15. Vijan S, Hofer TP, Hayward RA (2000) Costutility analysis of
screening intervals for diabetic retinopathy in patients with type 2
diabetes mellitus. JAMA 283:889896
16. Younis N, Broadbent DM, Vora JP, Harding SP (2003) Incidence of
sight-threatening retinopathy in patients with type 2 diabetes in the
Liverpool Diabetic Eye Study: a cohort study. Lancet 361:195200
17. Aspelund T, Thornorisdottir O, Olafsdottir E et al (2011) Individual
risk assessment and information technology to optimise screening
frequency for diabetic retinopathy. Diabetologia 54:25252532
18. Welschen LM, van Oppen P, Dekker JM, Bouter LM, Stalman WA,
Nijpels G (2007) The effectiveness of adding cognitive behavioural
therapy aimed at changing lifestyle to managed diabetes care for
Diabetologia
patients with type 2 diabetes: design of a randomised controlled trial.
BMC Public Health 7:74
19. Expert Committee on the Diagnosis and Classification of Diabetes
Mellitus (2003) Report of the expert committee on the diagnosis and
classification of diabetes mellitus. Diabetes Care 26(Suppl 1):S5S20
20. Aldington SJ, Kohner EM, Meuer S, Klein R, Sjolie AK (1995)
Methodology for retinal photography and assessment of diabetic
retinopathy: the EURODIAB IDDM complications study.
Diabetologia 38:437444
21. Crowson CS, Atkinson EJ, Therneau TM (2013) Assessing calibra-
tion of prognostic risk scores. Stat Methods Med Res. doi:10.1177/
0962280213497434
22. Harrell FE Jr, Lee KL, Mark DB (1996) Multivariable prognostic
models: issues in developing models, evaluating assumptions
and adequacy, and measuring and reducing errors. Stat Med
15:361387
23. Ruamviboonsuk P, Teerasuwanajak K, Tiensuwan M, Yuttitham K
(2006) Interobserver agreement in the interpretation of single-
field digital fundus images for diabetic retinopathy screening.
Ophthalmology 113:826832
24. Healy R, Sallam A, Jones V et al (2013) Agreement between photo-
graphic screening and hospital biomicroscopy grading of diabetic
retinopathy and maculopathy. Eur J Ophthalmol. doi:10.5301/ejo.
5000404
Diabetologia
... In those that do develop STR, a latent stage of diabetic retinopathy, symptoms are clearer and have a significant impact on patients' quality of life [10,11], and usually require active treatment. Notably, the risk of developing STR is highly variable and 'one size fits all' approaches may therefore not only lead to late screening in high-risk patients, but also over-screening in lowrisk patients [12,13]. Moreover, from a health economics perspective, 'one size fits all' approaches will pose unnecessary costs for both healthcare systems and patients. ...
... Aspelund et al proposed a model to compute personalised screening intervals based on individual patient characteristics [12]. This model was validated in different cohorts, including the cohort used in the present study, showing good model performance [13,[16][17][18]. Nevertheless, the costeffectiveness of this screening strategy has not been assessed. ...
... The retinopathy grades were reported according to the EURODIAB scale, which ranges from 0 to 5 [20]: grade 0 means no retinopathy; grade 1 is 'minimal non-proliferative retinopathy'; grade 2 is 'moderate non-proliferative retinopathy'; grade 3 is 'severe non-proliferative retinopathy'; grade 4 is 'photocoagulated retinopathy'; and grade 5 is 'proliferative retinopathy'. According to the Dutch guideline, grades 3-5 were considered STR, and usually, patients with these grades were referred to an ophthalmologist for treatment [13,16]. The study has been approved by the Medical Ethical Review Committee of the VU University Medical Center, Amsterdam. ...
Article
Full-text available
Aims/hypothesis: In this study we examined the cost-effectiveness of three different screening strategies for diabetic retinopathy: using a personalised adaptive model, annual screening (fixed intervals), and the current Dutch guideline (stratified based on previous retinopathy grade). Methods: For each individual, optimal diabetic retinopathy screening intervals were determined, using a validated risk prediction model. Observational data (1998-2017) from the Hoorn Diabetes Care System cohort of people with type 2 diabetes were used (n = 5514). The missing values of retinopathy grades were imputed using two scenarios of slow and fast sight-threatening retinopathy (STR) progression. By comparing the model-based screening intervals to observed time to develop STR, the number of delayed STR diagnoses was determined. Costs were calculated using the healthcare perspective and the societal perspective. Finally, outcomes and costs were compared for the different screening strategies. Results: For the fast STR progression scenario, personalised screening resulted in 11.6% more delayed STR diagnoses and €11.4 less costs per patient compared to annual screening from a healthcare perspective. The personalised screening model performed better in terms of timely diagnosis of STR (8.8% less delayed STR diagnosis) but it was slightly more expensive (€1.8 per patient from a healthcare perspective) than the Dutch guideline strategy. Conclusions/interpretation: The personalised diabetic retinopathy screening model is more cost-effective than the Dutch guideline screening strategy. Although the personalised screening strategy was less effective, in terms of timely diagnosis of STR patients, than annual screening, the number of delayed STR diagnoses is low and the cost saving is considerable. With around one million people with type 2 diabetes in the Netherlands, implementing this personalised model could save €11.4 million per year compared with annual screening, at the cost of 658 delayed STR diagnoses with a maximum delayed time to diagnosis of 48 months.
... DR [11][12][13][14][15][16][17][18], CKD [19][20][21][22][23][24][25] and ESRD models [12,[26][27][28][29]) using various statistical methods. A lot of prognostic models were externally validated [12,17,19,20,22,26,29,30], whilst other models were not [13,15,16,21,27,28]. Nonetheless, the best prognostic model for each complication was still inconclusive. ...
... Followup time ranged varied from 1.0 [56] to 20 [11] years with a median of 5 years. Only 5 [18,24,30,50,58] studies reported percent loss to follow-up which ranged from 2.4 to 31.3%. Eighteen [11-13, 15, 30, 41-45, 48, 50, 51, 54, 55, 57, 59, 61] studies used various methods for dealing with missing data, in which about a half of them used multiple imputations (Table S1). ...
... As for phase of prediction, 4 [12,17,41,42], 2 [11,45], 4 [30,43,44,46], 15 [13,15,16,18,[48][49][50][51][52][53][54][56][57][58][59] and 6 [14,24,55,60,62,63] studies were respectively determined as derived-internal-external (D/I/E), derived-external (D/E), external (E), derived-internal (D/I) and only derived (D) phases (Table S2). Amongst 8 [12,17,30,[41][42][43][44]46] external-validation studies, 5 [12,17,41,42,44] validated their own derived models in the same ethnicity (i.e. ...
Article
Full-text available
Background Many prognostic models of diabetic microvascular complications have been developed, but their performances still varies. Therefore, we conducted a systematic review and meta-analysis to summarise the performances of the existing models. Methods Prognostic models of diabetic microvascular complications were retrieved from PubMed and Scopus up to 31 December 2020. Studies were selected, if they developed or internally/externally validated models of any microvascular complication in type 2 diabetes (T2D). Results In total, 71 studies were eligible, of which 32, 30 and 18 studies initially developed prognostic model for diabetic retinopathy (DR), chronic kidney disease (CKD) and end stage renal disease (ESRD) with the number of derived equations of 84, 96 and 51, respectively. Most models were derived-phases, some were internal and external validations. Common predictors were age, sex, HbA1c, diabetic duration, SBP and BMI. Traditional statistical models (i.e. Cox and logit regression) were mostly applied, otherwise machine learning. In cohorts, the discriminative performance in derived-logit was pooled with C statistics of 0.82 (0.73‑0.92) for DR and 0.78 (0.74‑0.83) for CKD. Pooled Cox regression yielded 0.75 (0.74‑0.77), 0.78 (0.74‑0.82) and 0.87 (0.84‑0.89) for DR, CKD and ESRD, respectively. External validation performances were sufficiently pooled with 0.81 (0.78‑0.83), 0.75 (0.67‑0.84) and 0.87 (0.85‑0.88) for DR, CKD and ESRD, respectively. Conclusions Several prognostic models were developed, but less were externally validated. A few studies derived the models by using appropriate methods and were satisfactory reported. More external validations and impact analyses are required before applying these models in clinical practice. Systematic review registration PROSPERO CRD42018105287
... The algorithm by Aspelund et al., developed from both Type 1 and Type 2 diabetes subjects in Iceland, based on six traditional DR risk factors, accurately predicted STDR outcomes in the FIELD substudy of T2D subjects from Australia, New Zealand and Finland. To date, the algorithm has also been validated in the Netherlands, Spain, and England in both Type 1 and Type 2 diabetes subjects [12][13][14][15]. Two studies were conducted in the Netherlands. ...
... Two studies were conducted in the Netherlands. van der Heijden et al. [15] validated the algorithm in 3,319 T2D subjects with an average of 4.4 years follow-up, and determined a C-statistic/AUC of 0.83 (95% CI 0.74-0.92), while Schreur et al. trial [13] involved 268 subjects with Type 1 diabetes with an average of 4.6 years follow-up, and obtained a C-statistic/AUC of 0.82 (95% CI 0.74-0.90). ...
Article
Aims To evaluate the risk algorithm by Aspelund et al. for predicting sight-threatening diabetic retinopathy (STDR) in Type 2 diabetes (T2D), and to develop a new STDR prediction model. Methods The Aspelund et al. algorithm was used to calculate STDR risk from baseline variables in 1012 participants in the Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) ophthalmological substudy, compared to on-trial STDR status, and receiver operating characteristic analysis performed. Using multivariable logistic regression, traditional risk factors and fenofibrate allocation as STDR predictors were evaluated, with bootstrap-based optimism-adjusted estimates of predictive performance calculated. Results STDR developed in 28 participants. The Aspelund et al. algorithm predicted STDR at 2- and 5-years with area under the curve (AUC) 0.86 (95% CI 0.77-0.94) and 0.86 (0.81-0.92), respectively. In the second model STDR risk factors were any DR at baseline (OR 24.0 [95% CI 5.53-104]), HbA1c (OR 1.95 [1.43-2.64]) and male sex (OR 4.34 [1.32-14.3]), while fenofibrate (OR 0.13 [0.05-0.38]) was protective. This model had excellent discriminatory ability (AUC=0.89). Conclusions The algorithm by Aspelund et al. predicts STDR well in the FIELD ophthalmology substudy. Logistic regression analysis found DR at baseline, male sex, and HbA1c were predictive of STDR and, fenofibrate was protective.
... A risk calculation engine has been produced to estimate the risk of developing STDR at 1 year, 5 years, or 10 years for people in Iceland [34], based on eight risk factors. This model has been validated in different populations [28,59,60], and other RCEs are in development. Recently, a new RCE has been developed using data from Liverpool, UK [61]. ...
... Their model suggested that optimal screening intervals range from 6 to 60 months, dependent on an individual's level of risk. Van [59]. The authors found that the model enabled a 23% reduction in screening frequency compared with biennial screening, and a 61% reduction compared with annual screening, and suggested that the use of such a model could help reduce the costs of diabetes care. ...
Thesis
Full-text available
Publicly provided health screening programmes tend to offer standardised screening for a fixed eligible population. Recently, the development of risk calculation engines has introduced the potential for the stratification of screening based on individuals' risks of disease onset. This possibility raises practical, methodological, and ethical challenges. To date, no such programme has been the subject of an economic evaluation. In this thesis we present reason and basis for the allocation of screening based on individual risk. The research is conducted in the context of screening for diabetic eye disease in the UK. Diabetic retinopathy is a common complication of diabetes that can lead to blindness, substantial detriments to quality of life, and significant health care resource use. Our study is linked to a programme of research that includes a cohort study and randomised controlled trial in the city of Liverpool. We review and further develop the evidence base to inform the evaluation of a risk-based screening programme for diabetic eye disease. Specifically, we generate new evidence on the costs and health outcomes associated with the screening and treatment of diabetic retinopathy. We report on a cross-sectional study of health-related quality of life for people attending screening for diabetic retinopathy and find that people with pre-symptomatic disease tend to report poorer quality of life than people with no disease, with EQ-5D-5L index values of 0.733 on average compared with 0.787 for people with no disease. A meta-analysis of published health state utility values for diabetic eye disease shows a negative impact on health-related quality of life before progression to blindness. Our meta-regression found a utility index decrement of 0.024 for people with proliferative retinopathy. The costs of screening are low at the individual level, estimated to be £32.03 in our costing study. But the overall budget impact of changes in the frequency of screening can be significant. We analyse a large data set of hospital and community screening activity to identify key treatment pathways for diabetic eye disease. We find that these have changed in recent years, with the introduction of more expensive interventions. The evidence generated by our work is used to inform the development of a decision analytic model. The model is designed to estimate the cost-effectiveness of risk-based screening for diabetic eye disease, compared with current practice. We find that risk-based screening is likely to be more cost-effective than standardised screening programmes. Evaluating a programme that allocates screening according to individuals' levels of risk raises theoretical and ethical challenges. To this end, we develop a simple framework for individualised cost-effectiveness analysis that can be used to inform the design of a risk-based screening programme. We also explore the ethics of risk-based screening, developing the notion of screening need as distinct from treatment need. Risk-based screening is likely to be cost-effective in the context of diabetic eye disease. The evidence presented in this thesis can be used to support the evaluation of new programmes, which can be designed in order to optimise cost-effectiveness using the methods that we describe. Such an approach is consistent with equitable policy objectives.
... In addition, we included renal function (microalbuminuria and glomerular filtration rate) and body mass index (BMI) values. The algorithm yielded good results for identifying those patients who might develop STDR, [18][19][20] but not those who might develop the early stages of DR. ...
Article
Full-text available
Aim: The aim of the present study was to build a clinical decision support system (CDSS) that can predict the presence of diabetic retinopathy (DR) in type 1 diabetes (T1DM) patients. Material and method: We built two versions of our CDSS to predict the presence of any-type DR and sight-threatening DR (STDR) in T1DM patients. The first version was trained using 324 T1DM and 826 T2DM patients. The second was trained with only the 324 T1DM patients. Results: The first version achieved an accuracy (ACC) = 0.795, specificity (SP) = 83%, and sensitivity (S) = 65.7% to predict the presence of any-DR, and an ACC = 0.918, SP = 87.1% and S = 87.8% for STDR. The second model achieved ACC = 0.799, SP = 87.5% and S = 86.3% when predicting any-DR and ACC = 0.937, SP = 90.9% and S = 83.0% for STDR. Conclusion: The two models better predict STDR than any-DR in T1DM patients. We will need a larger sample to strengthen our results.
... Based on a biennial screening model, the following risk variables have been included to improve risk predictions for each individual patient: age, gender, diabetes duration, type of diabetes, HbA 1 c level, blood pressure, and retinopathy stage. An European collaborative network has used this model to calculate the most appropriate interval between examinations for each patient, the outcome of which was a reduction of 17-23% in the screening visits needed, compared to the biennial screening model [29,30]. A personalized screening approach would have the advantage of recommending more frequent screening intervals to high-risk patients and less frequent to lowrisk patients. ...
Article
Full-text available
Purpose. Diabetic retinopathy (DR) is the leading cause of vision loss among active adults in industrialized countries. We aimed to investigate the prevalence of diabetes mellitus (DM), DR and its different grades, in patients with DM in the Csongrád County, South-Eastern region, Hungary. Furthermore, we aimed to detect the risk factors for developing DR and the diabetology/ophthalmology screening patterns and frequencies, as well as the effect of socioeconomic status- (SES-) related factors on the health and behavior of DM patients. Methods. A cross-sectional study was conducted on adults (>18 years) involving handheld fundus camera screening (Smartscope Pro Optomed, Finland) and image assessment using the Spectra DR software (Health Intelligence, England). Self-completed questionnaires on self-perceived health status (SPHS) and health behavior, as well as visual acuity, HbA1c level, type of DM, and attendance at healthcare services were also recorded. Results. 787 participants with fundus camera images and full self-administered questionnaires were included in the study; 46.2% of the images were unassessable. T1D and T2D were present in 13.5% and 86.5% of the participants, respectively. Among the T1D and T2D patients, 25.0% and 33.5% had DR, respectively. The SES showed significant proportion differences in the T1D group. Lower education was associated with a lower DR rate compared to non-DR (7.7% vs. 40.5%), while bad/very bad perceived financial status was associated with significantly higher DR proportion compared to non-DR (63.6% vs. 22.2%). Neither the SPHS nor the health behavior showed a significant relationship with the disease for both DM groups. Mild nonproliferative retinopathy without maculopathy (R1M0) was detected in 6% and 23% of the T1D and T2D patients having DR, respectively; R1 with maculopathy (R1M1) was present in 82% and 66% of the T1D and T2D groups, respectively. Both moderate nonproliferative retinopathy with maculopathy (R2M1) and active proliferative retinopathy with maculopathy (R3M1) were detected in 6% and 7% of the T1D and T2D patients having DR, respectively. The level of HbA1c affected the attendance at the diabetology screening ( associated with >50% of all quarter-yearly attendance in DM patients, and with 10% of the diabetology screening nonattendance). Conclusion. The prevalence of DM and DR in the studied population in Hungary followed the country trend, with a slightly higher sight-threatening DR than the previously reported national average. SES appears to affect the DR rate, in particular, for T1D. Although DR screening using handheld cameras seems to be simple and dynamic, much training and experience, as well as overcoming the issue of decreased optic clarity is needed to achieve a proper level of image assessability, and in particular, for use in future telemedicine or artificial intelligence screening programs. 1. Introduction Diabetes mellitus (DM) is a major medical and societal challenge due to its rapid increase in global prevalence and devastating late complications [1, 2]. The global occurrence of DM among adults (>18 years of age) was 8.5% in 2014, and this has nearly doubled from its 4.7% level in 1980 [3]. In 2016, 1.6 million deaths were directly attributed to DM, with more than half of them occurring in the lower- and middle-income countries. According to the WHO forecast, DM will be the seventh leading cause of death in 2030, while diabetic retinopathy (DR) will be the leading cause of vision loss among active adults in industrialized countries [4]. DR is the most common late complication of DM in people aged 20 to 64 years—the working-age population, and except for where effective screening programs have been implemented, it is the leading cause of blindness and reduced vision in this group in the developed world [5, 6]. In a study comparing data from 35 populations, the global prevalence of sight-threatening retinopathy (STR) was estimated at 10.2% for all DM patients [6]. In Hungary, a total of 865 069 patients (9.5% of the population) suffered from DM among adults (>18 years of age) in 2011 [7], and some degree of DR could be observed among 19% of the patients with type 1 DM (T1D) and 24% in those suffering from type 2 DM (T2D) for 3 or 4 years [8]. Systematic DR screening and monitoring has been proven to be cost-effective in reducing blindness and visual impairment in patients having DM. Screening enables optimized timing of laser and medical therapy that may halt disease progression [9]. The WHO guidelines [10] for DR screening state that “annual eye examinations are recommended for patients with diabetes (and every other year for persons with excellent glycemic control and no retinopathy at the previous examination...).” “Such programs need systematic evaluation for their impact on health outcomes, cost effectiveness and health equity.” The WHO recommendation further states “Member States should choose the most appropriate interval between examinations” [10]. The development of optimized and effective DR screening programs is becoming eminent. The aim of this study was to investigate the prevalence of DR and its different grades in patients with DM in the Csongrád County—a South-Eastern region in Hungary, using for the first time in this country a handheld fundus camera (Smartscope Pro Optomed, Finland). Moreover, we aimed to detect the risk factors for developing DR and the diabetology/ophthalmology screening patterns and frequencies, as well as the effect of socioeconomic status- (SES-) related factors on the health and behavior of DM patients. 2. Patients and Methods 2.1. Physical Examination A cross-sectional study was conducted between the Departments of Ophthalmology and Internal Medicine Diabetology Unit, University of Szeged, Szeged, Hungary, between November 2015 and December 2016. All examinations were voluntary and free of charge to the participants, and the patients were recruited consecutively from the Diabetology Outpatient Clinic. Written informed consent was obtained from all participants. The study was approved by the local ethical committee of the University of Szeged (No.197/2015). The detection of DR was based upon examination with a handheld fundus camera (Smartscope Pro Optomed, Finland) in a dark room by qualified professionals. The results were directly evaluated by a qualified specialist without the need to do data/file transfer. In the case of constricted pupil, another image was taken after ensuring normal intraocular pressure level and applying cyclopentolate (5 mg/mL) eye drops to achieve mydriasis. The assessment of the fundus images was performed using the Spectra DR software (Health Intelligence, UK). The recordings were safely deposited and kept inaccessible to third parties for 10 years at a designated server, so that later they can be used in further comparative studies on DR. The images acquired with the Optomed Smartscope Pro digital handheld camera included two pictures from the participants’ eyes—one with the macula—and another with the optic nerve—in the center—which is in line with the English screening requirements [11]. In case of presence of amblyopia or nontransparent media (e.g., cataract and corneal or visual axis obstructing conditions), the patients were excluded from the study. During image evaluation, the graders (A.F./G.P./G.R.) classified the signs and stages of DR and maculopathy in the standardized English-based software Spectra DR and graded the images in alignment with the English standard grading protocols [12]. Each image was evaluated in two stages: first, the referral outcome graders/ROGs (D.E./G.R.) evaluated them, and then a supervisor/ophthalmic consultant confirmed the diagnosis (A.F./G.P.). At the end, an expert opinion regarding the grade of retinopathy was provided, which included the stage of retinopathy (R0/1/2/3A) and the absence or existence of maculopathy (M0/1). Other discovered abnormalities were not diagnosed in this study, although they were recorded, as they can provide further information about other symptoms, which may have occurred in the past, and therefore may require medical attention over a specified period of time. The classification of the DR has been described before [13]—in brief: (R0) no clinical anomaly—repeated screening was recommended one year later; (R1) mild nonproliferative—presence of microaneurysms, dot- or blot- like hemorrhages, or exudates—control examination was recommended one year later; (R2) moderate or severe nonproliferative—presence of major bleeding(s) and intraretinal microvascular abnormalities (IRMAs)—control examination was required within one month; (R3A) active proliferative—presence of neovascularization of the optic disc (NVD) or elsewhere (NVE) or preretinal bleeding(s), vitreous bleeding, preretinal fibrosis, and tractional retinal detachment—immediate medical examination was required within two weeks. All the stages were combined with sight-threatening maculopathy which was determined by the presence of exudates regardless of visual acuity (VA), or red lesions with a VA of 6/12 or worse after pinhole correction, that is within 1 disc diameter of the center of the fovea, and/or a group of exudates where the area of exudates that is greater than or equal to half the disc area, and this area is all within the macular area (as defined by the ETDRS macular grid) when medical examination was required within a month (M1). 2.2. Self-Completed Questionnaire Participants were asked to fill out a self-administered questionnaire which was based upon the European Health Interview Survey 2009—it included demographic characteristics such as gender, age, and place of residency. From the place of residency, the distance to the healthcare facility was calculated as <10 km or ≥10 km. The marital status was categorized as married or lives with a partner, single, separated or divorced and widowed; due to the low sample size, categories were merged together as living alone or living in partnership. SES of the study participants was examined: education and economic status. The economic status was characterized as working—full time and working—part-time, unemployed, retired, temporarily laid off, and student; due to the lack of data between each category, the categories were allocated and merged as inactive or active. The level of education was measured as primary, secondary, or higher education (college, university, or higher). Data were collected about self-perceived health status (SPHS) and characterized as bad satisfactory, and good. Information was also collected about “Perception of what the subject can do for his/her health status,” and the information was categorized as almost nothing (nothing/little) or much more (much/very much). Health behavior was assessed by alcohol consumption, smoking, physical activity, and diet (no/yes). Smoking was classified as yes/quit/never smoking, while alcohol consumption was classified as no/yes. Physical activity was defined according to the amount or occasions spent in the previous month in cycling, walking: daily/weekly more time, weekly, once/no activity at all (inactive). Information was also collected about the DM-related and other health conditions, for example, if the study participant has/had hypertension: no/yes. If yes, data were collected about the duration of the hypertension (years). If the participant attended blood pressure controls, a recording was made about the last measurement of the systolic and diastolic blood pressures in millimeters of mercury (mmHg). Information was further collected about other health conditions, for example, VA (<0.3 or ≥0.3), HbA1c level (normal <7% or elevated ≥7%), type of diabetes mellitus (T1DM or T2DM), use of medications, DM in the family or occurrence of diabetic maculopathy. In addition, data about the attendance at healthcare services like diabetology (monthly, every 3rd month, every 6th month, yearly, more than a year, or no attendance) were also collected. 2.3. Statistical Analysis The analysis of the data was performed by descriptive statistics; percentage distribution, mean and standard deviation (SD), and in case of nonnormality of continuous variables, median and interquartile range (IQR) and range (minimum, maximum) are shown. Normality of the continuous variables was tested on a histogram, Q-Q- plot, and by Shapiro-Wilk and Kolmogorov-Smirnov test. The Independent Sample -test was used to compare the means of the continuous, numerical variables, when the normality assumption was satisfied; otherwise, Mann–Whitney test was used. Homogeneity of variance was analyzed with the Levene test. Chi-square () and Fisher test were used to test the differences of the distribution of categorical variables; for multiple comparisons, the 2-sample -test with Bonferroni correction was applied to detect the differences in the proportions between the studied groups. If the sample within each column was 1 or less, then the -test could not be used. The significance limit was set at . The statistical analysis of the data was performed by IBM SPSS Statistics Version 24 software. 2.4. Ethical Issues The Regional and Institutional Human Medical Biological Research Ethics Committee of the Szent-Györgyi Albert Clinical Center, University of Szeged approved the study protocol (No. 197/2015). The research provided anonymity to the participants. Before the beginning of a test, the participants signed a voluntary written consent form in which they agreed to permit the use of data for research purposes. 3. Results The data were collected from a total of 848 participants with known DM in the Csongrád County, South-Eastern region in Hungary (Figure 1). Out of the initial participants, 787 (92.8%) had available fundus camera images and answered the self-administered questionnaire. T1D was present in 13.5% () of participants, while T2D was present in 86.5% () of the participants. Among the T1D and T2D patients, 25.0% () and 33.5% () had DR, respectively. A large portion of the participants had unassessable fundus camera images/results 46.2% () when using the handheld camera, and therefore excluded from the further analysis (Figure 1).
Article
Full-text available
Objective The aim of present study was to evaluate our clinical decision support system (CDSS) for predicting risk of diabetic retinopathy (DR). We selected randomly a real population of patients with type 2 diabetes (T2DM) who were attending our screening programme. Methods and analysis The sample size was 602 patients with T2DM randomly selected from those who attended the DR screening programme. The algorithm developed uses nine risk factors: current age, sex, body mass index (BMI), duration and treatment of diabetes mellitus (DM), arterial hypertension, Glicated hemoglobine (HbA1c), urine–albumin ratio and glomerular filtration. Results The mean current age of 67.03±10.91, and 272 were male (53.2%), and DM duration was 10.12±6.4 years, 222 had DR (35.8%). The CDSS was employed for 1 year. The prediction algorithm that the CDSS uses included nine risk factors: current age, sex, BMI, DM duration and treatment, arterial hypertension, HbA1c, urine–albumin ratio and glomerular filtration. The area under the curve (AUC) for predicting the presence of any DR achieved a value of 0.9884, the sensitivity of 98.21%, specificity of 99.21%, positive predictive value of 98.65%, negative predictive value of 98.95%, α error of 0.0079 and β error of 0.0179. Conclusion Our CDSS for predicting DR was successful when applied to a real population.
Conference Paper
Diabetic Retinopathy (DR) is a leading cause of visual impairment but its pathophysiology is not well understood. Moderate/severe non-proliferative DR (NPDR) is characterised by the presence of three features: deep haemorrhages (DH), venous beading (VB) and intraretinal microvascular abnormalities (IRMA). They are grouped together as risk factors for progression to sight threatening DR. It remains unclear whether these individual features have similar pathophysiologies, and whether they respond equally to anti-VEGF, a new therapy for NPDR. Optomap images of 504 NPDR eyes were examined to evaluate the distribution and prevalence of these three features. DNA samples from 199 patients with NPDR and 397 diabetic patients with no DR were collected. The genotype of specific candidate genes were evaluated in patients with DR, VB or IRMA vs no DR. Optical coherence tomography angiography (OCTA) images of 30 patients were examined for focal ischemia adjacent to VB and IRMA. The responses of these three features to anti-VEGF treatment were also re-examined in the images from the CLARITY trial. DH were present in most cases of NPDR. VB and IRMA did not always co-exist in the same eye and when they do, were often in different locations. VEGF, TGFb-1 and ARHGAP22 polymorphisms (ischaemia-related genes) were more common in patients with DR and IRMA, but not VB. Areas of focal ischaemia were more frequently adjacent to IRMA than to VB. DH and IRMA responded to anti-VEGF therapy but VB did not. These findings suggest that VB and IRMA do not share the same pathophysiology, and that IRMA are more likely to be ischaemic driven. Nonetheless, some IRMA may not be driven by ischaemia as they have no adjacent ischaemia on OCTA, do not carry the specific genotype, and do not respond to anti-VEGF. Furthermore, patients with VB may not benefit from anti-VEGF therapy.
Article
Full-text available
Background /aims: To evaluate the performance of existing prediction models to determine risk of progression to referable diabetic retinopathy (RDR) using data from a prospective Irish cohort of people with type 2 diabetes (T2D). Methods: A cohort of 939 people with T2D followed prospectively was used to test the performance of risk prediction models developed in Gloucester, UK, and Iceland. Observed risk of progression to RDR in the Irish cohort was compared with that derived from each of the prediction models evaluated. Receiver operating characteristic curves assessed models' performance. Results: The cohort was followed for a total of 2929 person years during which 2906 screening episodes occurred. Among 939 individuals followed, there were 40 referrals (4%) for diabetic maculopathy, pre-proliferative DR and proliferative DR. The original Gloucester model, which includes results of two consecutive retinal screenings; a model incorporating, in addition, systemic biomarkers (HbA1c and serum cholesterol); and a model including results of one retinopathy screening, HbA1c, total cholesterol and duration of diabetes, had acceptable discriminatory power (area under the curve (AUC) of 0.69, 0.76 and 0.77, respectively). The Icelandic model, which combined retinopathy grading, duration and type of diabetes, HbA1c and systolic blood pressure, performed very similarly (AUC of 0.74). Conclusion: In an Irish cohort of people with T2D, the prediction models tested had an acceptable performance identifying those at risk of progression to RDR. These risk models would be useful in establishing more personalised screening intervals for people with T2D.
Article
Full-text available
Aims/hypothesis Using variable diabetic retinopathy screening intervals, informed by personal risk levels, offers improved engagement of people with diabetes and reallocation of resources to high-risk groups, while addressing the increasing prevalence of diabetes. However, safety data on extending screening intervals are minimal. The aim of this study was to evaluate the safety and cost-effectiveness of individualised, variable-interval, risk-based population screening compared with usual care, with wide-ranging input from individuals with diabetes. Methods This was a two-arm, parallel-assignment, equivalence RCT (minimum 2 year follow-up) in individuals with diabetes aged 12 years or older registered with a single English screening programme. Participants were randomly allocated 1:1 at baseline to individualised screening at 6, 12 or 24 months for those at high, medium and low risk, respectively, as determined at each screening episode by a risk-calculation engine using local demographic, screening and clinical data, or to annual screening (control group). Screening staff and investigators were observer-masked to allocation and interval. Data were collected within the screening programme. The primary outcome was attendance (safety). A secondary safety outcome was the development of sight-threatening diabetic retinopathy. Cost-effectiveness was evaluated within a 2 year time horizon from National Health Service and societal perspectives. Results A total of 4534 participants were randomised. After withdrawals, there were 2097 participants in the individualised screening arm and 2224 in the control arm. Attendance rates at first follow-up were equivalent between the two arms (individualised screening 83.6%; control arm 84.7%; difference −1.0 [95% CI −3.2, 1.2]), while sight-threatening diabetic retinopathy detection rates were non-inferior in the individualised screening arm (individualised screening 1.4%, control arm 1.7%; difference −0.3 [95% CI −1.1, 0.5]). Sensitivity analyses confirmed these findings. No important adverse events were observed. Mean differences in complete case quality-adjusted life-years (EuroQol Five-Dimension Questionnaire, Health Utilities Index Mark 3) did not significantly differ from zero; multiple imputation supported the dominance of individualised screening. Incremental cost savings per person with individualised screening were £17.34 (95% CI 17.02, 17.67) from the National Health Service perspective and £23.11 (95% CI 22.73, 23.53) from the societal perspective, representing a 21% reduction in overall programme costs. Overall, 43.2% fewer screening appointments were required in the individualised arm. Conclusions/interpretation Stakeholders involved in diabetes care can be reassured by this study, which is the largest ophthalmic RCT in diabetic retinopathy screening to date, that extended and individualised, variable-interval, risk-based screening is feasible and can be safely and cost-effectively introduced in established systematic programmes. Because of the 2 year time horizon of the trial and the long time frame of the disease, robust monitoring of attendance and retinopathy rates should be included in any future implementation. Trial registration ISRCTN 87561257 Funding The study was funded by the UK National Institute for Health Research. Graphical abstract
Article
Full-text available
Aims/hypothesis The aim of our study was to identify subgroups of patients attending the Scottish Diabetic Retinopathy Screening (DRS) programme who might safely move from annual to two yearly retinopathy screening. Methods This was a retrospective cohort study of screening data from the DRS programme collected between 2005 and 2011 for people aged ≥12 years with type 1 or type 2 diabetes in Scotland. We used hidden Markov models to calculate the probabilities of transitions to referable diabetic retinopathy (referable background or proliferative retinopathy) or referable maculopathy. Results The study included 155,114 individuals with no referable diabetic retinopathy or maculopathy at their first DRS examination and with one or more further DRS examinations. There were 11,275 incident cases of referable diabetic eye disease (9,204 referable maculopathy, 2,071 referable background or proliferative retinopathy). The observed transitions to referable background or proliferative retinopathy were lower for people with no visible retinopathy vs mild background retinopathy at their prior examination (respectively, 1.2% vs 8.1% for type 1 diabetes and 0.6% vs 5.1% for type 2 diabetes). The lowest probability for transitioning to referable background or proliferative retinopathy was among people with two consecutive screens showing no visible retinopathy, where the probability was <0.3% for type 1 and <0.2% for type 2 diabetes at 2 years. Conclusions/interpretation Transition rates to referable diabetic eye disease were lowest among people with type 2 diabetes and two consecutive screens showing no visible retinopathy. If such people had been offered two yearly screening the DRS service would have needed to screen 40% fewer people in 2009.
Article
Full-text available
Most guidelines recommend annual screening for diabetic retinopathy (DR) but limited resources and the slow progression of DR suggest that longer recall intervals should be considered if patients have no detectable lesions. This study aimed to identify the cumulative incidence and time of development of referable DR in patients with no DR at baseline, classified by clinical characteristics. Analysis was performed of data collected prospectively over 20 years in a screening clinic based in a teaching hospital according to a consensus protocol. The cumulative incidence, time of development and relative risk of developing referable retinopathy over 6 years following a negative screening for DR were calculated in 4,320 patients, stratified according to age at onset of diabetes (<30 or ≥30 years), being on insulin treatment at the time of screening and known duration of diabetes (<10 or ≥10 years). The 6 year cumulative incidence of referable retinopathy was 10.5% (95% CI 9.4, 11.8). Retinopathy progressed within 3 years to referable severity in 6.9% (95% CI 4.3, 11.0) of patients with age at onset ≥30 years, who were on insulin treatment and had a known disease duration of 10 years or longer. The other patients, especially those with age at onset <30 years, on insulin and <10 years duration, progressed more slowly. Screening can be repeated safely at 2 year intervals in any patient without retinopathy. Longer intervals may be practicable, provided all efforts are made to ensure adherence to standards in procedures and to trace and recall non-attenders.
Article
Full-text available
OBJECTIVE The American Diabetes Association and the English NHS Diabetic Eye Screening Program recommend annual screening for diabetic retinopathy (DR) with referral to ophthalmology clinics of patients with sight-threatening DR (STDR). Using only longitudinal data from retinal photographs in the population-based NHS Diabetic Eye Screening Program in Gloucestershire, we developed a simple means to estimate risk of STDR.RESEARCH DESIGN AND METHODS From 2005, 14,554 patients with no DR or mild nonproliferative DR only at two consecutive annual digital photographic screenings were categorized by the presence of DR in neither, one, or both eyes at each screening and were followed for a further median 2.8 years.RESULTSOf 7,246 with no DR at either screening, 120 progressed to STDR, equivalent to an annual rate of 0.7%. Of 1,778 with no DR in either eye at first screening and in one eye at second screening, 80 progressed to STDR, equivalent to an annual rate of 1.9% and to a hazard ratio (HR) of 2.9 (95% CI 2.2-3.8) compared with those with no DR. Of 1,159 with background DR in both eyes at both screenings, 299 progressed to STDR equivalent to an annual rate of 11% and an HR of 18.2 (14.7-22.5) compared with individuals with no DR.CONCLUSIONS Combining the results from 2 consecutive years of photographic screening enables estimation of the risk of future development of STDR. In countries with systematic screening programs, these results could inform decisions about screening frequency.
Article
Full-text available
With increasing global prevalence of diabetes, diabetic retinopathy (DR) is set to be the principle cause of vision impairment in many countries. DR affects a third of people with diabetes and the prevalence increases with duration of diabetes, hyperglycemia, and hypertension-the major risk factors for the onset and progression of DR. There are now increasing data on the epidemiology of diabetic macular edema (DME), an advanced complication of DR, with studies suggesting DME may affect up to 7 % of people with diabetes. The risk factors for DME are largely similar to DR, but dyslipidemia appears to play a more significant role. Early detection of DR and DME through screening programs and appropriate referral for therapy is important to preserve vision in individuals with diabetes. Future research is necessary to better understand the potential role of other risk factors such as apolipoproteins and genetic predisposition to shape public health programs.
Article
Full-text available
In the U.K., people with diabetes are typically screened for retinopathy annually. However, diabetic retinopathy sometimes has a slow progression rate. We developed a simulation model to predict the likely impact of screening patients with type 2 diabetes, who have not been diagnosed with diabetic retinopathy, every 2 years rather than annually. We aimed to assess whether or not such a policy would increase the proportion of patients who developed retinopathy-mediated vision loss compared with the current policy, along with the potential cost savings that could be achieved. We developed a model that simulates the progression of retinopathy in type 2 diabetic patients, and the screening of these patients, to predict rates of retinopathy-mediated vision loss. We populated the model with data obtained from a National Health Service Foundation Trust. We generated comparative 15-year forecasts to assess the differences between the current and proposed screening policies. RESULTS The simulation model predicts that implementing a 2-year screening interval for type 2 diabetic patients without evidence of diabetic retinopathy does not increase their risk of vision loss. Furthermore, we predict that this policy could reduce screening costs by ~25%. Screening people with type 2 diabetes, who have not yet developed retinopathy, every 2 years, rather than annually, is a safe and cost-effective strategy. Our findings support those of other studies, and we therefore recommend a review of the current National Institute for Health and Clinical Excellence (NICE) guidelines for diabetic retinopathy screening implemented in the U.K.
Article
Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.
Article
Purpose: To examine the level of agreement and reasons for disagreement between grading of diabetic retinopathy and maculopathy using mydriatic digital photographs in a diabetic retinopathy screening service (DRSS) and hospital eye service (HES). Methods: English NHS Diabetic Eye Screening Programme grades for diabetic retinopathy prospectively recorded on a hospital electronic medical record were compared to the grades from the DRSS event that prompted referral. In cases of disagreement, images were reviewed. Results: Data for 1,501 patients (3,002 eyes) referred between 2008 and 2011 were analyzed. The HES retinopathy grades were R0 (no retinopathy) in 341 eyes, R1 (background retinopathy) in 1,712 eyes, R2 (pre-proliferative retinopathy) in 821 eyes, and R3 (proliferative retinopathy) in 128 eyes. The DRSS grades were in agreement in 2,309 eyes (76.9%), recorded a lower grade in 227 eyes, and recorded a higher grade in 466 eyes. Agreement was substantial (κ = 0.65). The commonest cause for disagreement was overgrading of R1 as R2 by hospital clinicians. The HES maculopathy grades were M0 (no maculopathy) in 2,267 eyes and M1 (maculopathy) in 735 eyes. The DRSS were in agreement in 2,111 eyes (70.2%), recorded a lower grade in 106 eyes, and recorded a higher grade in 785 eyes. Agreement was fair (κ = 0.39). The commonest cause for disagreement was hospital clinicians missing fine exudates. Conclusions: This study establishes a benchmark standard for agreement between HES and DRSS grading. Review of DRSS and grading reports images for newly referred patients is likely to improve levels of agreement, particularly for diabetic retinopathy, and should be strongly encouraged.
Article
Current methods used to assess calibration are limited, particularly in the assessment of prognostic models. Methods for testing and visualizing calibration (e.g. the Hosmer-Lemeshow test and calibration slope) have been well thought out in the binary regression setting. However, extension of these methods to Cox models is less well known and could be improved. We describe a model-based framework for the assessment of calibration in the binary setting that provides natural extensions to the survival data setting. We show that Poisson regression models can be used to easily assess calibration in prognostic models. In addition, we show that a calibration test suggested for use in survival data has poor performance. Finally, we apply these methods to the problem of external validation of a risk score developed for the general population when assessed in a special patient population (i.e. patients with particular comorbidities, such as rheumatoid arthritis).
Article
Multivariable regression models are powerful tools that are used frequently in studies of clinical outcomes. These models can use a mixture of categorical and continuous variables and can handle partially observed (censored) responses. However, uncritical application of modelling techniques can result in models that poorly fit the dataset at hand, or, even more likely, inaccurately predict outcomes on new subjects. One must know how to measure qualities of a model's fit in order to avoid poorly fitted or overfitted models. Measurement of predictive accuracy can be difficult for survival time data in the presence of censoring. We discuss an easily interpretable index of predictive discrimination as well as methods for assessing calibration of predicted survival probabilities. Both types of predictive accuracy should be unbiasedly validated using bootstrapping or cross-validation, before using predictions in a new data series. We discuss some of the hazards of poorly fitted and overfitted regression models and present one modelling strategy that avoids many of the problems discussed. The methods described are applicable to all regression models, but are particularly needed for binary, ordinal, and time-to-event outcomes. Methods are illustrated with a survival analysis in prostate cancer using Cox regression.