Skin Autofluorescence Based Decision Tree in Detection of Impaired Glucose Tolerance and Diabetes
Mulder DJ (2013) Skin Autofluorescence Based Decision Tree in Detection of Impaired Glucose Tolerance and
Diabetes. PLoS ONE 8(6): e65592. doi:10.1371/journal.pone.0065592
Skin Autofluorescence Based Decision Tree in Detection of Impaired Glucose Tolerance and Diabetes
Andries J. Smit 0 1
Jitske M. Smit 0 1
Gijs J. Botterblom 0 1
Douwe J. Mulder 0 1
Angelo Scuteri, INRCA, Italy
0 Results: 218 persons , age 56 yr, 128M/90F, 97 with previous CVE, participated. With OGTT 28 had DM, 46 IGT, 41 impaired fasting glucose, 103 normal glucose tolerance. SAF alone revealed 23 false positives (FP), 34 false negatives (FN) (sensitivity (S) 68%; specificity (SP) 86%). With SAF-DM, FP were reduced to 18, FN to 16 (5 with DM) (S 82%; SP 89%). HbA1c scored 48 FP, 18 FN (S 80%; SP 75%). Using HbA1c-defined DM-IGT/suspicion $6%/42 mmol/mol, SAF-DM scored 33 FP , 24 FN (4 DM) (S76%; SP72%), FPG 29 FP, 41 FN (S71%; SP80% ). FINDRISC$10 points as detection of HbA1c-based diabetes/suspicion scored 79 FP , 23 FN (S 69%; SP 45%)
1 1 Department of Medicine, University Medical Center Groningen, and University of Groningen , Groningen , The Netherlands , 2 Department of Medicine, Gelre Ziekenhuis , Apeldoorn , The Netherlands
Aim: Diabetes (DM) and impaired glucose tolerance (IGT) detection are conventionally based on glycemic criteria. Skin autofluorescence (SAF) is a noninvasive proxy of tissue accumulation of advanced glycation endproducts (AGE) which are considered to be a carrier of glycometabolic memory. We compared SAF and a SAF-based decision tree (SAF-DM) with fasting plasma glucose (FPG) and HbA1c, and additionally with the Finnish Diabetes Risk Score (FINDRISC) questionnaire6FPG for detection of oral glucose tolerance test (OGTT)- or HbA1c-defined IGT and diabetes in intermediate risk persons. Methods: Participants had $1 metabolic syndrome criteria. They underwent an OGTT, HbA1c, SAF and FINDRISC, in adition to SAF-DM which includes SAF, age, BMI, and conditional questions on DM family history, antihypertensives, renal or cardiovascular disease events (CVE). Conclusion: SAF-DM is superior to FPG and non-inferior to HbA1c to detect diabetes/IGT in intermediate-risk persons. SAFDM's value for diabetes/IGT screening is further supported by its established performance in predicting diabetic complications.
Funding: The study was partly funded by Diagnoptics Technologies, Groningen, the Netherlands. The study was also partly funded by the University Medical
Center Groningen. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: This study was partly funded by Diagnoptics Technologies. AJS is founder and shareholder, and JMS is also shareholder of Diagnoptics
Technologies B.V., the Netherlands. Diagnoptics Technologies is the company that developed the AGE Reader, which is used in the present article. AJS was also
the first inventor in the patent used for the AGE reader technology. He was inventor as employee of the University Medical Center Groningen (UMCG), which was
co-owner of it. Later the patent was sold to Diagnoptics Technologies, so the UMCG has no financial interests anymore. AJS now works 80% of his time as as
employee of the University Medical Center Groningen, but is also the scientific director of Diagnoptics technologies. As such, he has an interest in development
and validation of possible further future applications of the AGE reader, such as described in the present paper. The other authors have declared that no
competing interests exist; the personnel involved in the two participating hospitals in the study also had no competing interests. There are no further patents,
products in development or marketed products to declare. This does not alter the authors adherence to all the PLOS ONE policies on sharing data and materials,
as detailed online in the guide for authors.
Despite the major, woldwide increase in type 2 diabetes mellitus
(T2DM), and its preceding stages impaired fasting glucose (IFG)
and glucose tolerance, almost half of those affected are not aware
of having this condition (Eckardstein) . This leaves a long
clinically latent period in which T2DM and IGT can be detected,
for diabetes alone this is estimated to be 5 years. Diagnosis of
T2DM is still based on glycemic criteria, in the WHO criteria and
Europe on glucose levels. An International Expert Committee
(IEC) recently proposed new diagnostic criteria based on
measurement of A1C, with A1C$6.5%/48 mmol/mol for
diabetes and 6.06.4%/4246 mmol/mol for high risk of
progression to diabetes [4,5]. The American Diabetes Association
(ADA) subsequently proposed HbA1c$6.5% for the diagnosis of
diabetes and 5.76.4%/3946 mmol/mol for the highest risk to
progress to diabetes.
Early detection and screening for IGT and DM are warranted
because timely start of treatment in IGT may retard further
development to DM, and once DM is present may prevent
complications [6,7]. Many T2DM screening procedures and
scores have been developed, but several difficulties and limitations
make their broader use challenging: this includes matters like
limited response rates, limited availability and invasiveness of
required (confirmatory) tests (e.g. OGTT) which are not routinely
available and which also suffer from limited reproducibility . In
Europe, the use of questionnaires like the FINDRISC or
Cambridge Diabetes Risk score (CDRS) has been proposed as
first step in diabetes screening, followed by a plasma glucose step,
fasting or preferably with an OGTT, using WHO defined cut-off
values, as confirmatory test [8,9]. The importance, given in recent
European guidelines, to early interventions to prevent
development of diabetes also stresses the need for alternative technologies
for easy detection of IGT and DM [10,11].
Previously we showed in skin biopsy validation studies that
noninvasive SAF is a proxy for tissue accumulation of advanced
glycation endproducts (AGE) [12,13]. Holman et al proposed
AGE to be the prime candidate as carrier for the so-called
metabolic legacy effect [14,15]. This carrier effect of AGE has
been invoked as explanation for the prolonged and late beneficial
effects of an early, temporary period of intensified glycemic
control: accumulation of AGE in tissues with slow turnover,
persisting there for many years, may be prevented or retarded by
early improved glycemic control . In the preceding stages of
diabetes, intermittent periods of (post-meal) hyperglycemia in
impaired glucose tolerance might similarly result in persistent
increases in tissue AGE and skin AF levels. Thus, skin AF is a
candidate tool to detect IGT and diabetes in periods of
(intermittent) hyperglycemia. A study by Maynard et al in naive
persons indeed reported that noninvasive skin AF performed
better than glycemic criteria (FPG and HbA1c) in detecting
OGTT-confirmed diabetes . However, some caveats are
warranted concerning the specificity of skin AF: AGE may also be
rapidly formed during oxidative stress, and elevated SAF levels
have been reported after cardiovascular events, serious infections,
and in autoimmune disease [17,18]. In renal failure, diminished
renal excretion of AGE free adducts and AGE peptides is another
cause of increased plasma and tissue AGE levels and skin AF .
Accounting such alternative reasons for elevated skin AF might
improve specificity for diabetes and IGT detection especially in
intermediate risk persons with comorbidity. On the other hand,
decision trees (including age and systolic bloodpressure) have also
been proposed as simple and reliable tools for identifying
individuals with IGT or T2DM , while other well-known risk
factors for IGT/DM such as BMI, and family history of diabetes
are represented in most of the diabetes screening questionnaires
[8,9]. We assumed that the performance of noninvasive skin
AFbased DM/IGT decision tree might also be improved by
integrating such easily available items.
Therefore, we aimed to develop and validate a skin AF based
decision tree to detect IGT or diabetes in a group of subjects at
intermediate risk. Test characteristics were compared to those of
conventional diabetes diagnostic tools including FPG and HbA1c.
In addition, we also compared it to the FINDRISC questionnaire
alone or in combination with FPG.
The study protocol was approved by the IRB (METc) of the
University Medical Center Groningen (METc c2009-367), and
placed in the clinical trials.gov register NCT01406665. Informed
written study consent was obtained from all participants, and the
study was conducted according to the principles expressed in the
Declaration of Helsinki.
AGE Reader and Skin Autofluorescence
The AGE Reader (DiagnOptics Technologies BV, Groningen,
The Netherlands) is based on the non-invasive measurement of
skin autofluorescence (AF). Skin AF has been compared with
specific AGE in the dermis of skin biopsies from the measurement
site in patients ith diabetes, renal failure or healthypersons from a
wide age range. In a combined analysis of these three studies 76%
of variance in skin AF level could be explained by the variance of
dermal skin biopsy pentosidine levels. Lower, but still highly
significant relations were also found with CML (a non-fluorescent
AGE) and CEL . Measurements are performed at the volar
site of the forearm where the illuminating light enters the skin
almost perpendicularly over an area of ,4 cm2. This excitation
light is in the wavelength range of 350420 nm (maximum
intensity approximately at 370 nm). The AGE Reader uses a
spectrometer manufactured by Avantes BV (Eerbeek, The
Netherlands) to measure the light that is reflected and emitted
from the skin. Dark and white reference measurements take place
during each measurement to correct for detector properties and
background light, and to calculate skin reflectance. The device has
a completely automated measuring procedure of ,1 min, results
are presented immediately afterwards. The skin used for the
measurement has to be visibly normal, and not too dark of colour
(see also exclusion criteria), because irregularities and
pigmentation can influence the measurement. Reliable skin AF
measurements can be performed in persons with reflectance values .6%,
this covers Fitzpatrick skin color classes I-V. The AGE reader
automatically reports unreliable measurements in case of lower
reflectance levels; the handling of this result in evaluation of the
decision tree is described below. In persons with reflectance levels
between .610%, AGE reader software version 2.3 was used for
skin color independent assessment of skin AF as decribed by
Koetsier . AF is expressed in Arbitrary Units (AU). AF was
measured using the triple measurements setting with 3
measurements of approximately 20 seconds on 3 different lower arm sites
at one of the arms, and the mean value of these 3 AF
measurements was used for analysis. Reproducibility was earlier
tested by Meerwaldt et al in 25 healthy persons and 25 diabetic
patients, which showed a mean relative error in AF of 5.8% on a
single day .
Development of Decision Model
Use of skin AF age percentile cut-off levels alone. Age
percentiles are based on the reference value data by Koetsier et al
. Skin AF increases linearly with age, according to skin
AF = 0.0246age +0.83. This formula has been based on a study in
428 persons with normal renal function and glucose values,
without any known health problems, and without abnormalities
during screening physical exam. For persons .50 years, the choice
of cut-off level $70th age percentile for skin AF was based on data
distributions of skin AF levels previously reported by Lutgers et al
for persons with and without established diabetes . For persons
#50 years the choice of a higher cut-off level $80th age percentile
was not based on analysis of previously reported data distributions,
but on the lower risk of diabetes in this age group.
Decison model for detection of IGT/diabetes, based on
skin AF age percentile, BMI, and conditionally
questions. The summary design of this model is shown in
Figure 1. The model adds, depending on the measured skin AF
age percentile, question-based information which may be obtained
readily in a screening setting. In case of elevated skin AF age
percentile, maximally three questions are asked about conditions
possibly leading to increased skin AF levels: questions ask for a
recent history of serious (leading to hospital admission) infections
(,1 year) or CVE (,2 years), or a history of renal or autoimmune
disease. In case of documented and ongoing peripheral arterial or
coronary disease (stable angina pectoris) a recent admission was
not required. Further, in case skin AF is below age percentile
cutoff levels questions are asked on classical IGT/diabetes risk factors
apart from age: questions on weight/BMI, on 1st degree family
history of diabetes (and if so, the number), and on use of
antihypertensives. These questions were formulated and asked for
in the same manner as in the FINDRISC questionnaire .
Scoring in the developed decision model first depends on a .70th
(,50 years) or .80th ($50 years) age percentile cut-off level of
skin AF, plus information on the questions described above, and
on BMI with different cut-off levels: at .35 kg/m2, .32 kg/m2,
and .29 kg/m2, respectively. Persons with a BMI.35 kg/m2
were assumed to have diabetes/IGT, independent of other
information. For the latter two BMI groups, a combined weighing
of BMI, skin AF percentile and information on number of first
degree relatives with diabetes and on antihypertensive use
determined whether the decision tree outcome gave a positive
score for IGT/diabetes. This combined risk weighing was based
on the risk assumptions for the three variables BMI, family history
and skin AF derived from data of questionnaire-based risk
estimates (FINDRISC and Cambridge Diabetes Risk Score,
CDRS). In the persons with an elevated skin AF above the
defined age percentile cutoff, the then asked questions described
above, are accounted for as follows. If one of the questions on an
alternative condition in the last year was positively answered, the
decision tree result states that diabetes/IGT may be present, but
alternative explanations for the abnormal results are available. In
the decision tree performance analysis below such a response was
scored as negative for IGT/diabetes.
Finally, spectral information measured by the AGE reader in
addition to the skin AF is also used, especially skin reflectance
levels. As described above, if skin reflectance was #6%, and, thus,
the measurement rejected, the decision tree result states that
diabetes/IGT may be present, but alternative methods for
assessing diabetes-IGT risk are needed. Such a response was
scored as negative for IGT/diabetes in the Results.
Participants fulfilled the following inclusion criteria: males and
females aged 2080 years with an a priori intermediate risk for
IGT/diabetes defined by: presence of at least one criterion from
the NCEP definitions for metabolic syndrome (except obviously
high FPG) ; or at least once having had an increased glucose
or glycated hemoglobin value in the preceding 2 years, but these
below the range of diabetes/IGT. Exclusion criteria were known
diabetes mellitus, use of oral antidiabetics (for other purposes than
diabetes such as non-alcoholic fatty liver disease), local skin disease
of the lower arm obviating SAF measurement, known serious renal
insufficiency (s-creatinine .150 umol/l).
Participants were recruited from vascular and general internal
medicine outpatient clinics of the Gelre hospital Apeldoorn
(n = 101) and the University Medical Center Groningen (n = 117).
Study Procedures: Oral Glucose Tolerance Test (OGTT),
SAF, HbA1c, and FINDRISC
On the days preceding participation no important changes in
food habits were allowed, especially no carbohydrate restriction.
From the early evening before testing no alcohol or heavy meals
were allowed, only light snacks and low caloric drinks. The
participants came in a fasting state (at least 8 hours) in the
morning. Blood was drawn 15 min after arrival from an
antebrachial vein for venous plasma glucose and glycated
hemoglobin, and serum creatinine assessment. Then, the skin
AF measurement was performed. Thereafter the glucose load of
75 g of glucose solved in 150 ml water was given to drink. If
wished by the participants, they were allowed to drink or wash
their mouth with another 100 ml of water. No meals or drinks
were allowed in the subsequent two hours, at least until the final
blood samples had been drawn. Two hours after the oral glucose
administration another glucose sample was taken and a second
skin AF measurement performed. Participants avoided physical
exercise during this period.
Before or during the test body length, weight, waist
circumference, blood pressure, heart rate, family history for diabetes and for
cardiovascular disease, and smoking status were collected. All
participants filled in the Finnish version of the FINDRISC
questionnaire. A score $10 points was considered to signify
diabetes, or suspicion thereof, and was used in the analysis below
; an alternative approach used a cutoff level $7 points, followed
by a FPG.
Routine laboratory procedures were used for the assay of
venous plasma glucose, serum creatinine and glycated
Levels of glucose tolerance were defined according to
WHOGTT, or to 2009 IEC criteria: the IGT group according to
WHOGTT fulfilled the t = 120 min criterium of plasma glucose $7,8
11.0 mmol/L, with either a FPG level above or below 5.6 mmol/
l. For IEC criteria, an HbA1c$6.0%/42 mmol/mol defined
suspicion of diabetes . Diabetes definitions were in line with
reported WHO-GTT, or the 2009 IEC criteria, respectively.
For assessing test characteristics of the methods compared (FPG,
HbA1c, skin AF alone and skin AF-based decision tree,
FINDRISC) the categories diabetes, and IGT or suspicion, were
combined. For the HbA1c the WHO-GTT criteria were used, for
performance of FPG, the 2009 IEC criteria, using HbA1c levels
were used in comparison .
Sample Size and Statistical Analysis
Sample size. Using the in- and exclusion criteria above for
an intermediate risk group in an expected mainly middle aged to
elderly group, it was estimated that the risk for diabetes/IGT, as
defined by an OGTT, would be 2530%. In such an intermediate
risk group, misclassification was assumed to be 35% for cut-off
levels of fasting plasma glucose .6 mmol/l (in a previously
analysed pilot high risk sample it was 30%,unpublished
observations), and 35% for HbA1c$6%/42 mmol/mol . Assuming
this 35% misclassification rate for HbA1c in this setting, and 22%
for the decision tree, a sample size of 148 persons would be needed
to establish this with a p,0.05 and a power of 80%. After
estimated correction for drop-out, we aimed to include 180
Statistical and performance analysis. Mean and SD are
given in case of normal distribution, otherwise median and range.
The screening performance of FPG and HbA1C tests, of the
FINDRISC score $10 points, and of the skin AF alone and the
skin AF based decision tree, were assessed by comparing their
respective sensitivities and specificity to detect diabetes or IGT/
218 persons with an intermediate risk of having diabetes or
impaired glucose tolerance were included in the study.
Characteristics of this group are given in Table 1. Mean age was
55.9610,4 yr (33,45 yr, 23.70 yr). With OGTT-based WHO
criteria 28 had diabetes, 46 IGT, 41 IFG, 103 normal glucose
tolerance. Of the patients with diabetes, 2 had an HbA1c,6%/
42 mmol/mol, and 19 an HbA1c level between 6.0,6.5%/42
48 mmol/mol. Of the 46 persons with IGT, 16 had an
HbA1c,6%/42 mmol/mol. With HbA1c-based International
Figure 1. Summary design of skin AF based IGT/diabetes detection decision model (SAF-DM), using SAF level age-dependent
cutoff levels (shown here for age .50 years, different cut-off levels at lower age), using age percentiles. Effects of low reflectance have
not been included.
Expert Committee-2009 (IEC-2009) criteria, 13 had diabetes, 87
suspicon, and 118 normal glucose tolerance. 6 of these 13 diabetes
patients had normal glucose tolerance according to the
82 persons had a 1-st degree family member with DM (no data
on distribution type 1 vs type 2), 23 of them had .1 1-st degree
family members with DM. In 75 of the 97 persons with a CV event
this had occurred ,2 years previously or was still active. 29
persons had a history of kidney disease, 5 of them were known to
have CKD $ class III. 2 persons identified themselves to have an
27 persons had a FINDRISC score ,7 points, 61 had a score of
79 points, 130 had a score $10 points.
Mean AF was 2.4160.62 (range 14.87 AU), the study
participants were at a mean 61st percentile for their age. Mean
reflectance was at 1666 percent (range 343%; 4#6%).
Using AF alone to detect OGTT-based diabetes/IGT, 57 were
misclassified, with 23 false positives (FP), 34 false negatives (FN),
and a sensitivity (S) of 68% and specificity (SP) of 86%. Using AF
alone to detect IEC-2009 HbA1c-based diabetes/suspicion, 93
were misclassified, with 43 FP, 50 FN, with an S of 50% and a SP
of 64%. Using SAF-DM to detect OGTT-based diabetes/IGT, FP
was reduced to 18, FN to 16 (of which 5 with DM), with a
sensitivity (S) of 82% and specificity (SP) of 89%. Using SAF-DM
to detect IEC-2009 HbA1c-based diabetes/suspicion, FP was
reduced to 33, FN to 24 (4 for DM), resulting in a S of 76% and a
SP of 72%. For FPG, using IEC-criteria for DM/suspicion, 70
were misclassified (29 FP, 41 FN; S 59%, SP 75%). If information
on BMI and family history would have been added in a similar
way as in the SAF-DM decision tree, the number of FN would
have fallen with 5, but that of FP would have risen with 14 to 43,
with a resulting S of 80% and SP of 64% For HbA1c using OGTT
criteria, 66 were misclassified (48 FP, 18 FN; S 80%, SP 75%).
Again, if information on BMI and family history would have been
added in a similar way as in the SAF-DM decision tree, the
number of FN would have fallen with 3 (1 DM, 2 IGT), that of FP
would have risen with 24, with a resulting S of 80% and SP of
As for the FINDRISC score $10 points, 102 persons were
misclassified for detection of OGTT-based diabetes/IGT, with 79
FP and 23 FN, S 69% and SP 45%. With a FINDRISC score $10
points, using IEC-criteria for DM/suspicion, 119 were
misclassified (90 FP, 29 FN; S 71%, SP 24%). If, following a two-step
approach, a FINDRISC score $7 would qualify for a subsequent
FPG, for detection of OGTT-based diabetes/IGT, 77 persons
would be misclassified (32 FN, 45 FP) (S 57%; SP 69%). If,
following a two-step approach, a FINDRISC score $7 would
qualify for a subsequent FPG, for detection of IEC-based
diabetes/suspicion, 70 persons would be misclassified (48 FN, 22
FP) (S 52%; SP 81%).
DM/IGT according to OGTT (WHO criteria)
Smoking (N, yes/former/current)
Waist circumerence (cm)
Previous CV events
Family 1st degree diabetes (n;%)
Systolic BP (mmHg)
Diastolic BP (mmHg)
28 (13%)/46 (21%)
13 (6%)/87 (40%)
Values are mean 6 S.D (plus range or number) or percentage; AF:
Autofluorescence; DM: diabetes mellitus; Former smoker: smoked in the past 10
years, last year excluded; BP: blood pressure; BMI: Body Mass Index.
Part of these results are also shown, for the decision tree
(SAFDM) in comparison with FPG, HbA1c and FINDRISC, in table 2
and figures 2.
The current study shows that in persons with an intermediate
risk for IGT or diabetes mellitus, a skin AF-based decision model
performs better than the conventional diabetes detection tools
FPG, and than the FINDRISC questionnaire alone, or their
twostep combination, and equally to HbA1c in detecting IGT or
Previous results on the use of skin AF in detecting diabetes,
reported by Maynard et al, collected in naive persons, are now
confirmed . Risk prediction is even more accurate in our
higher risk group by using the SAF-DM decision tree. Several
differences exist between the studies, not only in our use of an
intermediate risk group instead of a naive group. The initial choice
for the intermediate risk group reflects the policy in many
countries to approach early diabetes detection not by population
screening but by selective or targeted screening in groups with
increased risk, or even opportunistic screening or case finding .
Age, and characteristics related to overweight or metabolic
syndrome, are usual criteria for such a selected screening. We
chose to use the presence of at least one metabolic syndrome
criterium, or a historical non-diabetic elevated glucose value as
entry criteria. These criteria are also used in some of the
questionnaires which are used as initial step in diabetes screening
programs. We agree that these study entry selection criteria form
one of the limitations of our study, and a follow-up study in lower
risk groups and in the general population should be performed.
Some of such studies are currently ongoing. However, our
approach resembles the selected screening approach currently
used in many countries.
A second major difference with the study by Maynard et al is
that in the present study skin AF was considered in conjunction
with some other known and easily retrievable risk factors for
diabetes, and resulted in an improvement in diagnostic
performance. In the Maynard study the measured skin AF is also
corrected for calendar age, but the procedure for this age
correction is not made explicit. We used a previously defined
70% or 80% age percentile cut-off value. Additionally and
probably more important, we incorporated other readily available
information on well-known other diabetes risk predictors such as
BMI, and conditionally family history of diabetes, and use of
antihypertensives in our skin AF based decision tree. These
predictors have been shown to have additional predictive value,
and are also part of initial diabetes screening tools like the
conventional diabetes risk questionnaires (such as FINDRISC or
Diabetes Risk Score) [8,9]. The improvement in diagnostic
performance is also confirmed in our study in the comparison
Figure 2. Sensitivity and specificity of skin AF alone, skin AF based decision tree (SAF-DM), FPG, HbA1c and FINDRISC for correct
classification of diabetes/IGT versus normal, using OGTT based WHO criteria, or for DM/suspicion DM using HbA1c based IEC 2009
with use of the age-corrected skin AF alone. The size of our group
was not large enough to reliably perform additional modeling on
the chosen cutoff levels for BMI subclasses, additional lower skin
AF age percentile subcategories, and answers to separate questions
on comorbidity. Future data from ongoing studies may allow to do
so. While obesity is unequivocally related to diabetes risk, in earlier
studies skin AF in persons both with and without diabetes was only
modestly related to BMI .
A third difference deals with alternative causes for high skin AF
levels. Elevated skin AF is found not only in diabetes or IGT, but
also in several other conditions such as infections, acute
cardiovascular events and autoimmune disease occurring in the
period (of at least months) preceding the skin AF measurement
[17,18]. In our decision tree we used additional questions in those
with an elevated skin AF (.70th or 80th percentile) level to detect
possible other factors than glycemic stress leading to increased skin
AF and AGE levels. Obviously, the aim is to diminish the number
of persons with a false positive result on screening for diabetes/
IGT defined by glycemic criteria. However, such an elevated skin
AF nevertheless has been shown to be a strong marker of increased
cardiovascular risk in several of these alternative conditions such as
renal failure, autoimmune disease, and previous cardiovascular
disease. Thus, the elevated skin AF may still be considered to be
valuable as a tool for CV risk prediction even when no diabetes/
IGT is present.
Our study confirms the well-known lack of concordance
between the conventional glycemic criteria [25,26]. A substantial
part of persons identified with one of the glycemic tests to have
diabetes or IGT scores normal glucose tolerance with the other
criteria. We obviously used the WHO-GTT criteria to test the
performance of HbA1c, while IEC criteria were used to score the
performance of FPG, since the FPG partly defines the criteria of
the gold standard OGTT. The low sensitivity for detection of
abnormal glucose tolerance or diabetes with FPG testing reported
is also not unexpected. In reviews on FPG or random glucose
screening for undiagnosed diabetes sensitivities range from 40 to
65% . Because persons tested negative on screening are not
subject to subsequent confirmatory testing, the high false negative
rate for FPG testing contributes to the growing number of
undiagnosed cases of type 2 diabetes.
The FINDRISC score had a good sensitivity of 69% at the
usually chosen 10 points cutoff level, but had poor specificity
(45%). A disadvantage of several criteria in the FINDRISC and
other questionnaires, like physical activity or use of fruits, is that
they are often not objectively measured,and may be dependent on
the availability of assistance by professionals . Nevertheless, the
FINDRISC has been suggested to be currently the best available
diabetes screening tool for use in clinical practice in Caucasian
populations . Other suggested cutoff levels than the $10 point
score (chosen for approximately equal sensitivity to the other tools)
leads in our study to an even worse balance between sensitivity (for
a suggested $14 point cutoff level falling far below that of the
other tools) and the already low specificity (for a $7 point cutoff).
Nevertheless, choosing a lower cut-off level might be considered
when using a second-step confirmatory test like the FPG. In our
study the specificity of this 2-step combination of FINDRISC .7
followed by FPG had a good specificity (81%), but still low
Our study was limited in size, and performed in a selected group
of persons with a moderately high risk of diabetes and IGT.
However, the higher specificity of skin AF alone for IGT/diabetes
detection than current detection tools, and the improvement in
sensitivity without loss in specificity by conditionally adding some
questions, shows the major potential of skin AF in early detection
of IGT/diabetes, even in a cohort with considerable
(cardiovascular) comorbidity (44%) when the specifity of skin AF is
challenged. As indicated above, confirmation in larger and lower
risk groups is warranted. This will also allow to define possible
alternative cutoff levels of skin AF, and the selection of other
characteristics included in the decision tree, depending on a priori
risk. A limitation of the decision model is that information on use
of antihypertensive medication, and on family history of diabetes
may be unreliable. Recent decision trees with good performance
for identifying individuals with IGT/diabetes have used systolic
blood pressure besides age .
One might argue that the predictive value of FPG or HbA1c
might similarly be improved by addition of questionnaire-based
information (like BMI, family history), but this was not the case.
Also the two-step approach with FINDRISC questionnaire first,
conditionally followed by FPG, lagged far behind in performance.
So, the added value of this information is higher for SAF-DM. In
several previously studied non-diabetic and diabetic cohorts the
relation between SAF and BMI was very modest [21,23].This may
be a factor explaining the observed additive value of BMI in
IGTdabetes risk prediction.
Skin AF and the derived decision tree may have potential for
clinical implementation in diabetes prevention programs. As for
monitoring effectiveness, of diabetes prevention programs, most
focus on lifestyle changes including weight control. Sofar, no data
is available on changes in skin AF over time in persons with IGT,
and their possible value in identifying higher risk of transition to
diabetes. Nevertheless, the dependence of the current decision tree
on BMI subclasses does already affect to some extent its change
over time. Future studies are needed tol reveal whether the
potential of the current skin AF-decision tree also extends to
monitoring effectiveness of diabetes prevention programs.
Our skin AF based decision tree has a diagnostic performance
for diabetes and IGT equal or superior to conventional risk
predictors, in at least intermediate risk groups. Validation of this
model in lower risk groups is still needed. The ease of use in a
point of care setting, its noninvasive character, and the
immediately available result qualify the skin AF based decision tree as a
new tool for detection of diabetes and IGT, with a potential for
clinical implementation in diabetes prevention programs. This is
also supported by the previously proven additive value of skin AF
in predicting complications of diabetes.
Conceived and designed the experiments: AJS DJM. Performed the
experiments: GJB JMS. Analyzed the data: AJS GJB DJM. Contributed
reagents/materials/analysis tools: AJS JMS. Wrote the paper: AJS JMS
DJM. Designed the decision tree: AJS JMS.
1. Noble D , Mathur R , Dent T , Meads C , Greenhalgh T ( 2011 ) Risk models and scores for type 2 diabetes: systematic review . BMJ 343: d7163 . doi: 10.1136/ bmj.d7163.
2. Von Eckardstein A , Schulte H , Assmann G ( 2000 ) Risk for diabetes mellitus in middle-aged Caucasian male participants of the PROCAM study: implications for the definition of impaired fasting glucose by the American Diabetes Association . Journal of Clinical Endocrinology and Metabolism 85 : 3101 - 3108 .
3. Christensen JO , Sandbaek A , Lauritzen T , Borch-Johnsen K ( 2004 ) Populationbased stepwise screening for unrecognised Type 2 diabetes is ineffective in general practice despite reliable algorithms . Diabetologia 47 : 1566 - 1573 .
4. International Expert Committee report on the role of the A1C assay in the diagnosis of diabetes (2009) Diabetes Care 32 : 1327 - 1334 .
5. American Association of Clinical Endocrinologists/American College of Endocrinology statement on the use of hemoglobin A1c for the diagnosis of diabetes (2010) Endocr Pract16 : 155 - 156 .
6. Knowler WC , Barrett-Connor E , Fowler SE , Hamman RF , Lachin JM , et al. ( 2002 ) Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin . Diabetes Prevention Program Research Group N Engl J Med 346 : 393 - 403 .
7. Tuomilehto J , Lindstrom J, Eriksson JG , Valle TT , Hamalainen H, et al. ( 2001 ) Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance . N Engl J Med 344 : 1343 - 1350 .
8. Silventoinen K , Pankow J , Lindstrom J, Jousilahti P , Hu G , et al. ( 2005 ) The validity of the Finnish Diabetes Risk Score for the prediction of the incidence of coronary heart disease and stroke, and total mortality . Eur J Cardiovasc Prev Rehabil 12 : 451 - 458 .
9. Griffin SJ , Little PS , Hales CN , Kinmonth AL , Wareham NJ ( 2000 ) Diabetes risk score: towards earlier detection of type 2 diabetes in general practice . Diabetes Metab Res Rev 164-171.
10. Paulweber B , Valensi P , Lindstrom J , Lalic NM , Greaves CJ , et al. ( 2010 ) A European evidence-based guideline for the prevention of type 2 diabetes . Horm Metab Res 42 Suppl 1 : S3 - S36 .
11. Lindstrom J , Neumann A , Sheppard KE , Gilis-Januszewska A , Greaves CJ ( 2010 ) Take action to prevent diabetes-the IMAGE toolkit for the prevention of type 2 diabetes in Europe . Horm Metab Res 42 Suppl 1 : S37 - s55 .
12. Smit AJ , Gerrits EG ( 2010 ) Skin autofluorescence as a measure of advanced glycation endproduct deposition: a novel risk marker in chronic kidney disease . Curr Opin Nephrol Hypertens 19 : 527 - 33 .
13. Meerwaldt R , Links TP , Graaff RG , Thorpe S , Baynes JW , et al. ( 2005 ) Simple noninvasive measurement of skin autofluorescence . Ann N Y Acad Sci 1043 : 290 - 298 .
14. Holman RR , Paul SK , Bethel MA , Neil HA , Matthews DR ( 2008 ) Long-term follow-up after tight control of blood pressure in type 2 diabetes . N Engl J Med 359 : 1565 - 1576 .
15. Holman RR , Paul SK , Bethel MA , Matthews DR , Neil HA ( 2008 ) 10-year follow-up of intensive glucose control in type 2 diabetes . N Engl J Med 359 : 1577 - 1589 .
16. Maynard JD , Rohrscheib M , Way JF , Nguyen CM , Ediger MN ( 2007 ) Noninvasive type 2 diabetes screening: superior sensitivity to fasting plasma glucose and A1C . Diabetes Care 30 : 1120 - 1124 .
17. Mulder DJ , van Haelst PL , Graaff R , Gans ROB , Zijlstra F , et al. ( 2009 ) Skin autofluorescence is elevated in acute myocardial infarction and is associated with the one-year incidence of major adverse cardiac events . Neth Heart J 17 : 162 - 168 .
18. Nienhuis HL , Westra J , Smit AJ , Limburg PC , Kallenberg CG , et al. ( 2009 ) AGE and their receptor RAGE in systemic autoimmune diseases: an inflammation propagating factor contributing to accelerated atherosclerosis . Autoimmunity 42 : 302 - 304 .
19. Meerwaldt R , Graaff R , Huisman RJ , Links TP , Thorpe SR , et al. ( 2005 ) Skin autofluorescence, a measure of cumulative metabolic stress and advanced glycation endproducts, predicts mortality in hemodialysis patients . J Am Soc Nephrol 16 : 3687 - 3693 .
20. Hische M , Luis-Dominguez O , Pfeiffer AF , Schwarz PE , Selbig J , et al. ( 2010 ) Decision trees as a simple-to-use and reliable tool to identify individuals with impaired glucose metabolism or type 2 diabetes mellitus . Eur J Endocrinol 163 : 565 - 571 .
21. Koetsier M , Lutgers HL , de Jonge C , Links TP , Smit AJ , et al. ( 2010 ) Reference values of skin autofluorescence . Diabetes Technol Ther 12 : 399 - 403 .
22. Koetsier M , Nur E , Chunmao H , Lutgers HL , Links TP , et al. ( 2010 ) Skin color independent assessment of aging using skin autofluorescence . Opt Express 18 : 14416 - 14429 .
23. Lutgers HL , Graaff R , Links TP , Ubink-Veltmaat LJ , Bilo HJG , et al. ( 2006 ) Skin autofluorescence as a noninvasive marker of vascular damage in patients with type 2 diabetes Diabetes Care 29 : 2654 - 2659 .
24. Lorenzo C , Williams K , Hunt KJ , Haffner SM ( 2007 ) The National Cholesterol Education Program - Adult Treatment Panel III, International Diabetes Federation, and World Health Organization definitions of the metabolic syndrome as predictors of incident cardiovascular disease and diabetes . Diabetes Care 30 : 8 - 13 .
25. Carson AP , Reynolds K , Fonseca VA , Huntner P ( 2010 ) Comparison of A1C and fasting glucose criteria to diagnose diabetes among U .S. adults. Diabetes Care 33 : 95 - 97 .
26. Olson DE , Rhee MK , Herrick K , Ziemer DC , Twombly JG , et al. ( 2010 ) Screening for diabetes and pre-diabetes with proposed A1C-based diagnostic criteria . Diabetes Care 33 : 2184 - 2189 .
27. Schwarz PE , Li J , Lindstrom J , Tuomilehto J ( 2009 ) Tools for predicting the risk of type 2 diabetes in daily practice . Horm Metab Res 41 : 86 - 97 .