A Colombian Diabetes Risk Score For Detecting Undiagnosed Diabetes and Impaired Glucose Regulation
A Colombian Diabetes Risk Score For Detecting Undiagnosed Diabetes and Impaired Glucose Regulation
Original research
a r t i c l e i n f o a b s t r a c t
Article history: Aims: (i) To develop a diabetes mellitus risk score model for the Colombian population
Received 5 December 2015 (ColDRISC); and (ii) to evaluate the accuracy of the ColDRISC unknown Type 2 diabetes
Received in revised form mellitus
31 August 2016 Methods: Cross-sectional screening study of the 18–74 years-old population of a health-care
Accepted 17 September 2016 insurance company (n = 2060) in northern Colombia. Lifestyle habits and risk factors for
Available online 7 October 2016 diabetes mellitus were assessed by an interview using a questionnaire consisting of informa-
tion regarding sociodemographic factors, history of diabetes mellitus, tobacco consumption,
Keywords: hypertension, nutritional and physical activity habits. Anthropometric measurements and
Screening an oral glucose tolerance test were taken. The sensitivity and the specificity, receiver-
Impaired glucose regulation operating characteristic (ROC) curves, were calculated for the ColDRISC and FINDRISC.
Colombia Results: The area under the ROC curve for unknown Type 2 diabetes mellitus was 0.74 (95%
Risk score CI: 0.70–0.79) for the ColDRISC and 0.73 for the FINDRISC (95% confidence intervals [CI]
Type 2 diabetes mellitus 0.69–0.78). Using the risk score cutoff value of 4 in the ColDRISC to detect Type 2 diabetes
mellitus resulted in a sensitivity of 73% and specificity of 67%.
Conclusions: The characteristics of the ColDRISC show that it can be used as a simple, safe,
and inexpensive test to identify people at high risk for Type 2 diabetes mellitus in Colombia.
© 2016 Primary Care Diabetes Europe. Published by Elsevier Ltd. All rights reserved.
∗
Corresponding author at: Herbert Wertheim College of Medicine, Florida International University, 11200 SW 8th Street, AHC2, Miami, FL
33199, USA.
E-mail address: [email protected] (N.C. Barengo).
http://dx.doi.org/10.1016/j.pcd.2016.09.004
1751-9918/© 2016 Primary Care Diabetes Europe. Published by Elsevier Ltd. All rights reserved.
p r i m a r y c a r e d i a b e t e s 1 1 ( 2 0 1 7 ) 86–93 87
% (n)
Sex
Men 38 (783)
Women 62 (1277)
2.3. Statistical analysis Age group
<45 years 40,9 (842)
The data was analyzed using IBM SPSS statistics version 19.0 45–54 years 27,4 (565)
for Windows. Logistic regression models were used for assess- 55–64 years 14,4 (297)
>64 years 17,3 (356)
ing the association between socio-demographic factors, risk
factors and lifestyle habits of Type 2 diabetes mellitus and Body mass index
the outcome variables. The outcome for the prognostic model < 25 kg/m2 41,6 (857)
was IGR-based on the collective results of the OGTT. The 25–30 kg/m2 34,8 (717)
>30 kg/m2 23,5 (484)
Hosmer–Lemeshow summary statistics were used to assess
the goodness-of-fit of the model. The level of statistical signif- Waist cirumference LA FINDRISC
icance was set to 0.05. <94 cm (men)/<90 cm (women) 53,8 (1108)
≥94 cm (men)/≥90 cm (women) 46,2 (952)
First, a univariate logistic regression analysis was per-
formed in order to evaluate which variables were statistically Waist cirumference FINDRISC
significantly related to the outcome variable. In a second <94 cm (men)/<80 cm (women) 36,8 (759)
94–102 cm (men)/80–88 cm (women) 23,6 (486)
step, the variables that showed statistical significance were
>102 cm (men)/>88 cm (women) 39,6 (815)
included in the multivariate models to develop the final model
of the ColDRISC. Following the methodology of the develop- 30 min daily physical activity
ment of the original FINDRISC, logistic regression was used to No 94,7 (1950)
Yes 5,3 (110
compute ˇ-coefficients for known risk factors for diabetes [10].
Coefficients (ˇ) of the model were used to assign a score value Daily fruit or vegetable intake
for each variable, and the composite diabetes risk score was No 28,2 (577)
Yes 71,8 (1468)
calculated as the sum of those scores. The diabetes risk score
value was defined using the full model, from the ˇ coefficient Use of blood pressure medication
as follows: for ˇ = 0.01–0.49, the score is 1; for ˇ = 0.50–0.9, the No 80,7 (1662)
Yes 19,3 (398)
score is 2; and for ˇ = 0,9–1,2, the score is 3. The lowest category
(reference) of each variable was given a score of 0. The total Past history of hyperglycemia (in a medical check-up, during an
diabetes risk score was calculated as the sum of the individual illness or pregnancy)
scores and varied from 0 to 9. No 97,8 (2015)
Yes 2,2 (45)
The ColDRISC was tested against the original FINDRISC [12]
and a modified FINDRISC [11]. The only difference between Family history of diabetes mellitus
the original and modified FINDRISC is that the original FIND- No 63 (1297)
Yes: grandparent, uncle, aunt or cousin 16,5 (340)
RISC uses three categories for waist circumference (men:
Yes: biological father, mother or sibling 20,5 (423)
<94 cm (0 risk points), 94–102 cm (3 risk points), >102 cm (4
Unknown Type 2 diabetes mellitus 5,1 (105)
risk points); women: <80 cm (0 risk points), 80–88 cm (3 risk
Impaired glucose regulation 27,4 (565)
points), >88 cm (4 risk points)), whereas the modified FIND-
RISC, recently renamed as the LA-FINDRISC, uses only two
categories (men: <94 cm (0 risk points), ≥94 cm (4 risk points);
women: <90 cm (0 risk points), ≥90 cm (4 risk points)) to
identify abdominal obesity in the Latin American population 2.4. Ethical considerations
[17]. The modified FINDRISC has been validated in people in
Colombia and Venezuela [11] and most recently in 521 peo- This study followed the Good Clinical Practice guidelines and
ple from a population based cross-sectional, cluster sampling the guidelines of the Helsinki Declaration. All the data have
study in Venezuela [18]. been collected using previously tested questionnaires and
The sensitivity and the specificity were assessed for methods. Besides blood samples, no invasive methods were
the ColDRISC, original and modified FINDRISC. In addition, used. The study protocol was approved by the Research Ethics
receiver-operating characteristic (ROC) curves, sensitivity, Committee of the Central Military Hospital, Bogotá, Colombia.
specificity, positive and negative predictive values were cal- All participants gave the written informed consent prior their
culated for each score for several cut-off points. participation in the study.
p r i m a r y c a r e d i a b e t e s 1 1 ( 2 0 1 7 ) 86–93 89
Table 2 – Logistic regression models with impaired glucose regulation as the dependent variable.
Full model (n = 2060)
Age group
45–54 years 0,37 1,45 (1,10–1,90) 1
55–64 years 0,729 2,07 (1,51–2,85) 2
>64 years 1,118 3,06 (2,25–4,16) 3
Waist circumference
≥94 cm (men)/≥90 cm (women) 0,854 2,35 (1,90–2,90) 2
1.00
0.75
Sensitivity
0.50
0.25
0.00
Fig. 1 – Receiver operating characteristics (ROC) curves for the prevalence of unknown Type 2 diabetes mellitus for the
ColDRISC, modified FINDRISC and original FINDRISC.
power to detect a possible statistically significant association next step is the implementation of the ColDRISC in the popu-
with this variable in the model. This also shows the impor- lation of Colombia as a tool to detect unknown Type 2 diabetes
tance to weight the individual components of a risk score in mellitus. Currently, there are three approaches for early detec-
a given population as their frequency differ remarkably from tion of Type 2 diabetes mellitus and IGR in the population. The
the original Finnish population for FINDRISC. The important first one consists of fasting glucose or HbA1c measurements to
1.00
0.75
Sensitivity
0.50
0.25
0.00
Fig. 2 – Receiver operating characteristics (ROC) curves for the prevalence of impaired glucose regulation for the ColDRISC,
modified FINDRISC and original FINDRISC.
p r i m a r y c a r e d i a b e t e s 1 1 ( 2 0 1 7 ) 86–93 91
Table 3 – Characteristics of the ColDRISC and the modified FINDRISC using different cutoff values for unknown Type 2
diabetes and impaired glucose regulation.
Sensitivity Specificity Positive predictive value Negative predictive value % of study sample
Unknown Type 2 diabetes mellitus
ColDRISC
Cutoff value 3 0,86 0,5 0,085 0,985 52
Cutoff value 4 0,73 0,67 0,106 0,979 35
Cutoff value 5 0,52 0,77 0,109 0,968 25
Modified FINDRISC
Cutoff value 9 0,82 0,52 0,08 0,989 58
Cutoff value 10 0,72 0,6 0,084 0,984 50
Cutoff value 12 0,56 0,77 0,096 0,971 33
Modified FINDRISC
Cutoff value 8 0,78 0,5 0,36 0,877 64
Cutoff value 9 0,71 0,58 0,372 0,864 58
Cutoff value 12 0,43 0,82 0,44 0,87 33
determine explicitly IGR. The second proposed strategy uses their lifetime, and they will benefit from lifestyle interventions
demographic and clinical characteristics as well as previous [26,27,31,32].
laboratory tests to determine the likelihood for Type 2 dia- It has been shown that the most cost effective method for
betes mellitus. The last approach collects questionnaire-based Type 2 diabetes mellitus screening in the general population is
information that provides information on the presence of eti- the use of a non-invasive tool for risk stratification followed by
ological risk factors Type 2 diabetes mellitus [20]. The last two a blood test for glycaemia [33]. Furthermore, it has been sug-
strategies do not clearly determine the glycaemia and blood gested that strategies for early detection of persons with Type 2
glucose testing is necessary in all three approaches to accu- diabetes mellitus are only cost-effective when combined with
rately define whether IGR exist. However, the results from a lifestyle interventions in those identified with IGR [34–36]. In
simple first-level screening can remarkably decrease the num- addition, recent evaluation of four different screening strate-
ber of people who should to be referred for further testing gies for unknown diabetes in Colombia revealed that the most
of glycaemia and appropriately target laboratory measure- cost-effective strategy was screening by using the FINDRISC
ments to the segment of the population that has high risk followed by fasting blood glucose and OGTT if necessary [37].
of Type 2 diabetes mellitus. The second approach mentioned At the present, the evidence in regard screening and the
above is particularly suited for those with pre-existing cardio- benefits of lifestyle intervention in people at high risk have
vascular disease (CVD) and women with previous gestational not yet been established. Therefore, it is important to imple-
diabetes, while the third option is best for the general popu- ment a screening tool for Type 2 diabetes mellitus and IGR
lation including overweight or obese people. The guideline of such as the ColDRISC in order to reduce the burden of Type 2
the European Society of Cardiology and European Association diabetes mellitus as soon as possible in the current situation
for the Study of Diabetes recommends that the appropriate in the Colombian population where half of all cases of Type 2
screening strategy in the general population and people with diabetes mellitus are undetected [1,38,39].
assumed abnormalities is to start with a Type 2 diabetes mel- Naturally, our study had some limitations. The partici-
litus risk score and to ask for an OGTT or a combination of pants in our study are not from a representative sample of
HbA1c and FPG in individuals with risk score above the set the Colombian population; thus, the results may not reflect
cut-off value [21]. In patients with CVD, no diabetes risk score the proportions of IGR categories in the Colombian popula-
is needed but an OGTT is indicated if HbA1c and/or FPG are tion at large. However, the study participants were randomly
inconclusive, as people belonging to these groups may often selected from the sampling frame using the phone number
have Type 2 diabetes mellitus revealed only by elevated 2- registered at the health-care insurance company. Moreover,
h plasma glucose [22–25]. The importance to identify people the study participants was recruited geographically in order
with IGR within the Colombian population is justified by sev- to provide adequate numbers from various municipalities at
eral intervention studies that have clearly shown that Type the Colombian Atlantic coast to provide adequate statistical
2 diabetes mellitus can be prevented or at least be delayed power for developing a risk factor model. Therefore, the sam-
in people with IGT by lifestyle interventions targeting physi- ple developed for this study represents the intended target
cal activity and nutritional changes [26–30]. Thus, the optimal groups. We found that fewer variables have the same perfor-
Type 2 diabetes mellitus prevention strategy should be to iden- mance as the original or modified FINDRISC. However, it may
tify people with IGT since at least 50% of them will develop be argued that when the questionnaire is simplified, the possi-
Type 2 diabetes mellitus in 10 years and most of them during bility to explore and educate about other variables such as BMI,
nutrition and physical activity is lost. This may be valid from
92 p r i m a r y c a r e d i a b e t e s 1 1 ( 2 0 1 7 ) 86–93
a health promotion point-of-view when a risk score is applied [2] DECODE Study Group, Age- and sex-specific prevalences of
in community intervention programs. For rapid screening pur- diabetes and impaired glucose regulation in 13 European
poses, however, a shorter, less time consuming version of the cohorts, Diabetes Care 26 (1) (2003) 61–69.
[3] L.N. Pani, L. Korenda, J.B. Meigs, et al., Effect of aging on A1C
risk score may be preferable. Furthermore, it has been recog-
levels in individuals without diabetes: evidence from the
nized that there is no evidence that applying a simple risk Framingham Offspring Study and the National Health and
score will as such improve health outcomes such as phys- Nutrition Examination Survey 2001–2004, Diabetes Care 31
ical activity or nutritional behavior on short or long-term. (10) (2008) 1991–1996.
Finally, the prevalence of people engaged in daily physical [4] American Diabetes Association Position Statement,
activity of 30 min or more was very low compared to popu- Diagnosis and classification of diabetes mellitus, Diabetes
lation estimates within the region and may reflect a potential Care 33 (2010) S62–69.
[5] World Health Organization (WHO) Consultation, Definition
to misclassification bias. However, none of the physical activ-
and Diagnosis of Diabetes and Intermediate
ity variables whether measures as a dichotomous variable or Hyperglycaemia, 2006.
in metabolic equivalent obtained by the IPAQ were statisti- [6] M. Davidson, D. Schriger, A. Peters, B. Lorber, Relationship
cally significant predictors of IGR. Thus, any potential bias of between fasting plasma glucose and glycosylated
physical activity did not affect our risk score model. hemoglobin: potential for false positive diagnoses of type 2
In conclusion, the characteristics, accuracy and discrimi- diabetes using new diagnostic criteria, JAMA 281 (1999)
1203–1210.
nation of the ColDRISC show that it can be used as a simple,
[7] A. Abbasi, L.M. Peelen, E. Corpeleijn, et al., Prediction models
safe, and inexpensive screening test to identify unknown dia-
for risk of developing type 2 diabetes: systematic literature
betes mellitus in Colombia. The effect of sampling in the search and independent external validation study, BMJ 345
validation of a risk score is unknown. However, some obser- (2012) e5900.
vations have shown that the cut-off of the risk score can vary [8] K.G. Alberti, P. Zimmet, J. Shaw, International diabetes
according the sample considered even within the same pop- federation: a consensus on Type 2 diabetes prevention,
ulation [10,11,18]. Thus, it may be difficult to find a single Diabet Med. 24 (5) (2007) 451–463.
[9] P.A. Kengne, J. Beulens, L.M. Peelen, et al., Non-invasive risk
cut-off point for an entire population and any threshold cho-
scores for prediction of type 2 diabetes (EPIC-InterAct): a
sen may affect sensitivity and specificity of a risk score. In the validation of existing models, Lancet Diabetes Endocrinol. 2
end, population-wide strategies may use the best available evi- (1) (2014) 19–29.
dence in choosing a cut-off point for a risk score according to [10] Diego Gomez-Arbelaez, Laura Alvarado-Jurado, Miguel
their economic and administrative possibilities for the man- Ayala-Castillo, Leonardo Forero-Naranjo, Paul Anthony
agement of people labeled to be at a high risk of the disease. Camacho, Patricio Lopez-Jaramillo, Evaluation of the Finnish
Diabetes Risk Score to predict type 2 diabetes mellitus in a
Colombian population: a longitudinal observational study,
World J. Diabetes 6 (17) (2015) 1337–1344.
Conflict of interest
[11] P. Aschner, R. Nieto-Martinez, A. Marín, M. Rios, Evaluation
of the FINDRISC score as a screening tool for people with
The authors state that they have no conflict of interest. impaired glucose regulation in Latin America using modified
score points for waist circumference according to the
validated regional cutoff values for abdominal obesity,
Financial support Minerva Endocrinol. 37 (4) (2012) 114.
[12] J. Lindström, J. Tuomilehto, The diabetes risk score: a
practical tool to predict type 2 diabetes risk, Diabetes Care
The ColDRISC study was financed by Novo Nordisk® Colombia.
26 (2003) 725–731.
[13] C.L. Craig, A.L. Marshall, M. Sjöström, A.E. Bauman, M.L.
Booth, B.E. Ainsworth, et al., International physical activity
Acknowledgment questionnaire: 12-country reliability and validity, Med. Sci.
Sports Exerc. 35 (8) (2003) 1381–1395.
The development of Diabetes Observatory of Colombia activi- [14] U. Ekelund, H. Sepp, S. Brage, et al., Criterion-related validity
ties is supported in the infrastructure, staff and scientific and of the last 7-day, short form of the International Physical
Activity Questionnaire in Swedish adults, Public Health
logistical support of the Organization for Health Excellence.
Nutr. 9 (2) (2006) 258–265.
Novo Nordisk ® provides funding for this initiative; it does [15] WHO, Chronic diseases and health promotion, STEPwise
not entail any involvement in the processing, management approach to chronic disease risk factor surveillance (STEPS)
and dissemination of information and knowledge produced by 2013, [Last cited on 2013 Apr 15], Available from:
the Diabetes Observatory of Colombia. In addition, we would http://www.who.int/chp/steps/riskfactor/en/index.html.
like to thank to Mutual SER EPSS, Clínica Crecer and the Fun- [16] WHO Consultation, Definition, Diagnosis and Classification
of Diabetes Mellitus and Its Complications. Part 1: Diagnosis
dación Ser Social for their support in implementing the project
and Classification of Diabetes Mellitus, World Health
activities.
Organization, 1999, Report No 99.2.
[17] Guía de práctica clínica para el diagnóstico, tratamiento y
references seguimiento de diabetes tipo 1, diabetes tipo 2 en mayores
de 18 años y diabetes gestacional [Clinical practice guideline
for the diagnosis, treatment and monitoring of type 1
diabetes, type 2 diabetes and gestational diabetes in people
[1] International Diabetes Federation, IDF Diabetes Atlas, 6th aged 18 and older]. Ministry of Health and Social Protection,
edn., International Diabetes Federation, Brussels, Belgium, Bogotá, Colombia. Alianza CINETS; 2015. Available from:
2013.
p r i m a r y c a r e d i a b e t e s 1 1 ( 2 0 1 7 ) 86–93 93