See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/310658635

Development and Validation of a Predictive Model for Chronic Kidney Disease

Progression in Type 2 Diabetes Mellitus Based on a 13-year study in Singapore

Article  in  Diabetes research and clinical practice · November 2016

DOI: 10.1016/j.diabres.2016.11.008

CITATIONS READS

4 123

8 authors, including:

Serena Low Shiyi Zhou

Nanyang Technological University Beijing Genomics Institute
44 PUBLICATIONS   1,146 CITATIONS    13 PUBLICATIONS   179 CITATIONS   

SEE PROFILE SEE PROFILE

Allen Yan Lun Liu Subramaniam Tavintharan

Khoo Teck Puat Hospital Khoo Teck Puat Hospital
41 PUBLICATIONS   498 CITATIONS    117 PUBLICATIONS   2,325 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Achieving Increase in Patient Uptake to Peritoneal Dialysis Program – An Interventional Nephrology-Urology Collaborative Approach on Peritoneal Dialysis Catheter
Insertion View project

All content following this page was uploaded by Allen Yan Lun Liu on 23 January 2018.

The user has requested enhancement of the downloaded file.

 diabetes research and clinical practice 1 2 3 (2 0 17 ) 4 9–54

Contents available at ScienceDirect

Diabetes Research
and Clinical Practice
journal homepage: www.elsevier.com/locat e/dia bre s

Development and validation of a predictive model

for Chronic Kidney Disease progression in Type 2
Diabetes Mellitus based on a 13-year study in
Singapore

Serena Low a,*, Su Chi Lim b, Xiao Zhang a, Shiyi Zhou a, Lee Ying Yeoh c, Yan Lun Liu c,
Tavintharan Subramaniam b, Chee Fang Sum b
a
Khoo Teck Puat Hospital, Clinical Research Unit, 90 Yishun Central, Singapore 768828, Singapore
b
Khoo Teck Puat Hospital, Diabetes Centre, 90 Yishun Central, Singapore 768828, Singapore
c
Khoo Teck Puat Hospital, Department of General Medicine, 90 Yishun Central, Singapore 768828, Singapore

A R T I C L E I N F O A B S T R A C T

Article history: Aims: This study aims to develop and validate a predictive model for Chronic Kidney Dis-
Received 29 June 2016 ease (CKD) progression in Type 2 Diabetes Mellitus (T2DM).
Received in revised form Methods: We conducted a prospective study on 1582 patients with T2DM from a Diabetes
29 October 2016 Centre in regional hospital in 2002–2014. CKD progression was defined as deterioration
Accepted 11 November 2016 across eGFR categories with P25% drop from baseline. The dataset was randomly split into
Available online 22 November 2016 development (70%) and validation (30%) datasets. Stepwise multivariable logistic regression
was used to identify baseline predictors for model development. Model performance in the
two datasets was assessed.
Keywords:
Results: During median follow-up of 5.5 years, 679 (42.9%) had CKD progression. Progres-
Diabetes mellitus
sion occurred in 467 (42.2%) and 212 patients (44.6%) in development and validation data-
Chronic kidney disease
sets respectively. Systolic blood pressure, HbA1c, estimated glomerular filtration rate and
Progression
urinary albumin-to-creatinine ratio were associated with progression. Areas under receiv
ing-operating-characteristics curve for the training and test datasets were 0.80 (95%CI,
0.77–0.83) and 0.83 (95%CI, 0.79–0.87). Observed and predicted probabilities by quintiles
were not statistically different with Hosmer-Lemeshow v2 0.65 (p = 0.986) and 1.36
(p = 0.928) in the two datasets. Sensitivity and specificity were 71.4% and 72.2% in develop-
ment dataset, and 75.6% and 72.3% in the validation dataset.
Conclusions: A model using routinely available clinical measurements can accurately pre-
dict CKD progression in T2DM.
Ó 2016 Elsevier Ireland Ltd. All rights reserved.

* Corresponding author. Fax: +65 66023772.

E-mail addresses: [email protected] (S. Low), [email protected] (S.C. Lim), zhang.xiao@alex-
andrahealth.com.sg (X. Zhang), [email protected] (S. Zhou), [email protected] (L.Y. Yeoh), liu.allen.
[email protected] (Y.L. Liu), [email protected] (S. Tavintharan), sum.chee.fang@alexandra-
health.com.sg (C.F. Sum).
http://dx.doi.org/10.1016/j.diabres.2016.11.008
0168-8227/Ó 2016 Elsevier Ireland Ltd. All rights reserved.
50 diabetes research and clinical practice 1 2 3 ( 2 0 1 7 ) 4 9 –5 4

1. Introduction history of glomerulonephritis). For the purpose of analysis,

subjects were also excluded if they had fewer than three mea-
Diabetes Mellitus (DM) is a global health challenge. According surements of estimated glomerular filtration (eGFR), shorter
to the International Diabetes Federation, the number of peo- than two years of follow-up, and eGFR < 15 ml/min/1.75 m2
ple with DM is expected to increase from 382 million in 2013 at baseline. There were 1582 patients who were eligible for
to 592 million in 2035 [1]. Approximately 60% of the world’s this study. Ethics approval was obtained from National
population with diabetes is found in Asia [2]. The myriad of Healthcare Group Domain Specific Review Board in Singapore.
complications engendered by DM therefore poses a consider- All patients who participated had given informed written
able health burden. Of note, diabetic nephropathy is one of consent.
the major complications of DM; it was reported that kidney
complications occurred in about 25–40% of individuals with 2.2. Data collection
Type 2 DM (T2DM) [3]. In Singapore, the prevalence of DM rose
from 8.2% in 2004 to 11.3% in 2010 [4,5]. Given the rising Demographics and clinical characteristics were obtained by
prevalence of DM and the fact that diabetic nephropathy is trained nurses from patients’ case records and standard ques-
the leading cause of new patients with end-stage kidney dis- tionnaire administered to the patients. Blood pressure (BP)
ease in Singapore, it is imperative to identify patients with T2 was measured using a standard automated sphygomo-
DM early for targeted management in order to prevent pro- manometer in the sitting position after resting for at least
gression of kidney disease in DM. 5 min (HEM-C7011-C1, OMRON Corp., Kyoto, Japan). The sphy-
There is emerging literature on the development and vali- gomomanometer has been validated according to European
dation of risk models that predict progression of Chronic Kid- Society of Hypertension (ESH) Protocol and Association for
ney Disease (CKD). Such models can potentially be used for the Advancement of Medical Instrumentation (AAMI) Protocol
risk communications to patients to improve lifestyle and [7]. The hospital laboratory accredited by the Royal College of
health behaviour, tailoring management for patients at differ- the American Pathologists carried out measurements of blood
ent risk strata, and prognosticating patients for preparation of and urine samples. Serum creatinine (intra-assay coefficient
renal replacement therapy. The predictors included age, of variation (CV) was 0.6–1.1%; inter-assay CV was 1.1–1.4%),
gender, body mass index, systolic blood pressure, serum low density lipoprotein cholesterol (LDL-cholesterol) (intra-
creatinine, measure of proteinuria, urinary albumin-to- assay CV was 0.7–1.2%; inter-assay CV was 1.9–2.5%) and
creatinine ratio (uACR), estimated glomerular filtration rate triglycerides (TG) (intra-assay CV was 0.7–1.1%; inter-assay
(eGFR) and even novel biomarkers [6]. CV was 1.6–2.0%) were quantitated using enzymatic colori-
However, there were some limitations of the models. metric test (Roche cobasÒ c 501) [8–10], Haemoglobin A1c
Firstly, the models were predominantly derived from Cau- (HbA1c) (intra-assay CV was 1.0–1.6%; inter-assay CV was
casian populations and general population [6]. Their validity 1.4–2.0%) using Tina-quant Hemoglobin A1c Gen.3 (Roche
in Asians with T2DM remains unclear due to limited data cobasÒ c 501) [11], and urinary albumin (intra-assay CV was
on this population. Secondly, the predicted outcomes were 0.7–1.6%; inter-assay CV was 1.2–2.8%) using immunoturibid-
occurrence of end-stage renal disease (ESRD) and doubling mimetric assay (Roche cobasÒ c 501) [12]. We used the Modi-
of serum creatinine which were the extreme ends of the kid- fication of Diet in Renal Disease (MDRD) formula to calculate
ney disease spectrum [6]. Given the variable outcomes of CKD the eGFR rate [13].
such as progression to renal replacement therapy, stability
and death, there is a need to develop a prediction model that 2.3. Definition of outcome
aims to arrest progression at the intermediate stage prior to
the occurrence of ESRD. The outcome was CKD progression. This was defined as
To address these gaps, we aimed to develop and validate a eGFR decline which involved worsening of eGFR categories
predictive model for progression of CKD in Singapore, a multi- (stage 1, P90 ml/min/1.73 m2; stage 2, 60–89 ml/min/1.73 m2;
ethnic society in Southeast Asia. The presence of this ‘inter- stage 3a, 45–59 ml/min/1.73vm2; stage 3b, 30–44 ml/
mediate’ outcome offers an opportunity to identify patients min/1.73 m2;stage 4, 15–29 ml/min/1.73 m2; and stage 5,
at risk early so as to prevent or slow down CKD progression <15 ml/min/1.73 m2), coupled with a 25% or more reduction
to end-stage. in eGFR from baseline according to the 2013 Kidney
Disease: Improving Global Outcomes (KDIGO) Clinical
2. Materials and methods Practice Guideline for the Evaluation and Management of
Chronic Kidney Disease KDIGO [14]. Patients with improve-
2.1. Study population ment in eGFR, same eGFR category, or less than 25% reduc-
tion in eGFR from baseline were deemed as having no CKD
This was a prospective cohort study involving patients with progression.
T2DM who were enrolled at a Diabetes Centre in a regional
hospital in Singapore in 2002–2014. Subjects were excluded 2.4. Statistical analysis
if they were less than 21 years old, pregnant, have active
infections, active cancer and autoimmune disease, involve- The dataset was randomly split into two datasets – training
ment of other suspected causes of renal diseases (e.g. urinary dataset for development of the prediction model (70%) and
tract infection, polycystic kidney disease, haematuria or test dataset for validation of the model (30%). Categorical
 diabetes research and clinical practice 1 2 3 ( 2 0 1 7 ) 4 9 –5 4 51

variables were presented as number (percentage), and contin- 2.4.1. Model validation
uous variables as means (standard deviation) or median A series of methods was employed to validate the model.
(interquartile range) as appropriate. Differences in demo- These included the following.
graphic and clinical characteristics stratified by the two
groups and progression were examined by Chi-Square test 2.4.1.1. Discrimination. This was assessed from the area
for categorical variables and student-t test or Mann- under the receiving operating characteristics curve (AUC; C
Whitney test for continuous variables. static). Discrimination is deemed perfect, good, moderate
In the training dataset, we first tested for univariate asso- and poor if the corresponding AUC is 1, >0.8, 0.6–0.8 and
ciations between potential variables and CKD progression <0.6 [15,16].
using logistic regression. These variables, which were pro-
posed as risk factors for CKD progression based on literature 2.4.1.2. Calibration (or goodness-of-fit). This was assessed
[6] and biological plausibility, included age, gender, ethnicity, by the Hosmer-Lemeshow Ĉ-test. This involved dividing the
smoking, duration of DM, body mass index, HbA1c, systolic cohort into quintiles of predicted risks of progression and
blood pressure, eGFR, uACR, LDL-cholesterol, TG and renin- comparing with actual outcomes. Calibration is poor if there
angiotensin antagonist. Those with p < 0.10 were selected is significant difference between the predicted and observed
for backward stepwise multivariable logistic regression risks (p < 0.05) [15,16].
(p < 0.10 for entry and p < 0.05 for stay). The choice of back-
ward selection was premised on the need to refine selection 2.4.1.3. Summary of classification. Sensitivity refers to the
from a modest-sized group of potential variables, which were proportion of patients with CKD progression correctly classi-
first identified in the univariate logistic regression, by elimi- fied as such by the model. Specificity refers to the proportion
nating a few variables. The candidate baseline variables for of patients without CKD progression correctly classified as
inclusion in the final model in the training dataset were such by the model. Positive predictive value refers to the pro-
age, HbA1c, systolic blood pressure (SBP), uACR and eGFR at portion of patients identified by the model with positive result
baseline. as truly having CKD progression. Negative predictive value

Table 1 – Baseline characteristics of patients at baseline in the training and test data sets.

Variables All (n = 1582) Training data set (n = 1107) Test data set (n = 475) P

Age (years) 57.3 ± 11.6 57.6 ± 11.4 56.5 ± 12.1 0.080

Gender (%) 0.997
Male 916 (57.9) 641 (57.9) 275 (57.9)
Female 666 (42.1) 466 (42.1) 200 (42.1)
Ethnicity (%) 0.385
Chinese 1097 (69.5) 762 (68.8) 335 (70.5)
Malay 258 (16.3) 189 (17.1) 69 (14.5)
Indian 224 (14.2) 153 (13.8) 71 (15.0)
Duration of DM (years) 10.0 (5.0–18.0) 10.0 (5.0–18.0) 10.0 (5.0–19.0) 0.179
BMI (kg/m2) 26.7 ± 4.9 26.7 ± 4.9 26.6 ± 4.9 0.662
Smoking (%) 0.667
Non-smoker 974 (72.6) 699 (73.3) 275 (70.9)
Ex-smoker 210 (15.7) 145 (15.2) 65 (16.8)
Current smoker 158 (11.8) 110 (11.5) 48 (12.4)
HbA1c (%) (mmol/mol) 8.3 ± 1.9 (67 ± 21) 8.3 ± 1.9 (67 ± 21) 8.3 ± 1.9 (67 ± 21) 0.930
SBP (mmHg) 135.9 ± 19.2 135.9 ± 19.4 135.7 ± 18.8 0.814
Urinary ACR (mg/g) 37.0 (12.0–176.0) 37.0 (12.0–171.0) 36.0 (11.0–184.0) 0.590
eGFR (ml/min/1.73 m2) 81.4 ± 34.0 79.7 ± 34.0 85.4 ± 33.6 0.003
LDL-cholesterol (mmol/l) 2.8 ± 0.9 2.8 ± 0.8 2.8 ± 0.9 0.215
TG (mmol/l) 1.9 ± 1.6 1.9 ± 1.6 1.9 ± 1.4 0.901
RAS-antagonist (%) 0.384
No 579 (36.8) 397 (36.1) 182 (38.4)
Yes 995 (63.2) 703 (63.9) 292 (61.6)
Follow-up period (years) 5.5 (4.3–7.0) 5.5 (4.0–6.9) 5.7 (4.5–7.4) 0.006
Progression (%) 0.368
No 903 (57.1) 640 (57.8) 263 (55.4)
Yes 679 (42.9) 467 (42.2) 212 (44.6)
DM, Diabetes Mellitus; BMI, Body Mass Index; HbA1c, Haemoglobin A1c; HbA1c-CV, Haemoglobin A1c Coefficient of Variation; SBP, Systolic
Blood Pressure; Urinary ACR, Urinary Albumin-to-Creatinine Ratio; eGFR, estimated Glomerular Filtration Rate; LDL-Cholesterol, Low Density
Lipoprotein Cholesterol; TG, Triglycerides; RAS-antagonist, Renin-angiotensin system antagonist.
52 diabetes research and clinical practice 1 2 3 ( 2 0 1 7 ) 4 9 –5 4

refers to the proportion of patients identified by the model 90.0

Risk of CKD Progression (%)

with negative result as truly having no CKD progression. A 80.0
cut-off of 0.40 was selected as the threshold beyond which 70.0
an observation has a predicted positive outcome. Using lens 60.0
χ2=0.65, p=0.986
50.0
command, the graphs of sensitivity and specificity versus
40.0
probability cut-off were produced. The cut-off point was cho-
30.0 Observed
sen at the intersection of the sensitivity and specificity 20.0 Predicted
curves. 10.0
Statistical analysis was performed using STATA version 0.0
1 2 3 4 5
14.0 (STATA Corporation, College Station, Texas).
Observed 11.2 23.4 36.7 56.1 84.6
Predicted 11.5 23.3 37.2 57.2 82.8
3. Results Predicted Risk Quinle

During a median follow-up period of 5.5 years (IQR 4.3–7.0), Fig. 1 – Observed versus predicted risk of CKD progression
679 (42.9%) had CKD progression. Progression occurred in during a 6-year follow-up in the training data set.
467 patients (42.2%) in training dataset, and 212 patients
(44.6%) in the test dataset. The patients in the training dataset
and the test dataset were similar in terms of age, gender, eth-
nicity, body mass index (bmi), duration of diabetes, smoking (p = 0.986) (Fig. 1). Using a cut-off of p = 0.40, the model had
status, HbA1c, LDL-Cholesterol, TG, uACR and use of renin- a sensitivity of 71.4%, specificity of 72.2%, positive predictive
angiotensin system antagonists at baseline. The patients in value of 65.3% and negative predictive value of 77.4%
the training dataset had a lower eGFR at baseline than those (Table 3).
in the validation dataset (p = 0.003) (Table 1). All the variables in the final model using the test dataset
The final variables selected in the final multivariable logis- were significantly associated with progression (Table 2). The
tic regression after backward selection were log uACR (mg/g), model in the test dataset shows good discrimination with
SBP (per 10 mmHg), HbA1c (%), eGFR (per 5 ml/min/1.73 m2), AUC of 0.83 (95%CI, 0.79–0.87). The observed and predicted
LDL-Cholesterol (mmol/l) and age (per 10 years) (Table 2). probabilities of CKD progression were not significantly differ-
The AUC of the final model was 0.80 (95% Confidence Interval ent with Hosmer-Lemeshow v2 of 1.36 (p = 0.928) (Fig. 2). Using
(CI), 0.77–0.83). The predicted probability was not significantly a cut-off of p = 0.40, the model had a sensitivity of 75.6%,
different from the observed probability of CKD progression specificity of 72.3%, positive predictive value of 68.9% and
over 6 years of follow-up with Hosmer-Lemeshow v2 of 0.65 negative predictive value of 78.5% (Table 3).

Table 2 – Univariable and final multivariable models for CKD progression in the training data set.
Variable Univariable Multivariablea
OR (95% CI) P OR (95% CI) P

Age (per 10-year increase) 1.41 (1.26–1.57) <0.001 1.20 (1.04–1.39) 0.014
Gender (male vs female) 0.73 (0.58–0.93) 0.012
Ethnicity
Malay (vs Chinese) 1.88 (1.36–2.60) <0.001
Indian (vs Chinese) 0.64 (0.44–0.93) 0.019
Duration of DM (per 5-year increase) 1.28 (1.19–1.38) <0.001
BMI (per 5-kg/m2) 1.17 (1.03–1.33) 0.016
Smoking
Ex-smoker (vs non-smoker) 0.92 (0.64–1.33) 0.670
Current smoker (vs non-smoker) 0.70 (0.46–1.07) 0.102
HbA1c (per 1% increase) 1.15 (1.08–1.23) <0.001 1.12 (1.04–1.21) 0.004
SBP (per 10-mmHg increase) 1.32 (1.23–1.41) <0.001 1.14 (1.05–1.23) 0.002
Log urinary ACR (mg/g) 1.87 (1.71–2.03) <0.001 1.69 (1.54–1.85) <0.001
eGFR (per 5-ml/min/1.73 m2 increase) 0.92 (0.90–0.94) <0.001 0.97 (0.94–0.99) 0.008
LDL-cholesterol (per mmol/l increase) 1.33 (1.15–1.54) <0.001 1.31 (1.10–1.57) 0.003
TG (per mmol/l increase) 1.13 (1.04–1.23) 0.002
RAS-antagonist (yes vs no) 1.87 (1.45–2.42) <0.001
DM, Diabetes Mellitus; BMI, Body Mass Index; HbA1c, Haemoglobin A1c; SBP, Systolic Blood Pressure; Urinary ACR, Urinary Albumin-to-
Creatinine Ratio; eGFR, estimated Glomerular Filtration Rate; LDL-Cholesterol, Low Density Lipoprotein Cholesterol; TG, Triglycerides; RAS-
antagonist, Renin-angiotensin system antagonist.
a
Final model includes age, HbA1c, SBP, eGFR, urinary ACR and LDL-cholesterol at baseline.
 diabetes research and clinical practice 1 2 3 ( 2 0 1 7 ) 4 9 –5 4 53

Our findings also partially overlap with those in earlier

Table 3 – Performance of the model in the training and test
data sets. studies on patients with T2DM, albeit on different outcomes.
For example, in the Action in Diabetes and Vascular Disease:
Training Test Preterax and Diamicron MR Controlled Evaluation (ADVANCE)
data set data set
study, the prediction model on major kidney related out-
Total number of subjects 1107 475 comes (doubling of serum creatinine, renal replacement ther-
apy or renal death) included gender, diabetic retinopathy and
Discrimination
AUC (95% CI) 0.80 0.83 age at completion of formal education, in addition to four of
(0.77–0.83) (0.79–0.87) the variables found in our final model (eGFR, uACR, SBP and
P <0.001 <0.001 HbA1c) [17]. In the Reduction of Endpoints in NIDDM with
Calibration the Angiotensin II Antagonist Losartan (RENAAL) study, uACR,
Hosmer-Lemeshow Ĉ test P 0.986 0.928 serum albumin, serum creatinine and hemoglobin were iden-
Sensitivity (%) 71.4 % 75.6% tified as predictors of ESRD in patients with T2DM and
Specificity (%) 72.2% 72.3% nephropathy [18]. Another study included eGFR, uACR and
PPV (%) 65.3% 68.9% haematocrit in their risk model for predicting ESRD in Hong
NPV (%) 77.4% 78.5%
Kong Chinese patients with T2DM [19].
AUC, area under the receiver operating characteristics curve; PPV, One of the strengths of our study is the choice of definition
positive predictive value; NPV, negative predictive value. of CKD progression in line with the KDIGO guidelines. To the
best of our knowledge, this is the first study using this defini-
tion as the outcome for developing prediction model. Previous
100.0
studies have developed or validated models on major end-
90.0
Risk of CKD Progression (%)

80.0 points such initiation of chronic dialysis, kidney transplanta-

70.0 tion or doubling of serum creatinine [17–25]. Only one study
χ2=1.36, p=0.928
60.0 defined progressive CKD as having the most renal function
50.0 decline (top 20% of the total population) and eGFR < 60 ml/
40.0
30.0 Observed min/1.73 m2 using follow-up [26]. In view of the lengthy
20.0 Predicted course of kidney disease in DM, the use of prediction models
10.0 in both early and late stages of kidney disease provides a wide
0.0 window of opportunity for prevention. Our prediction tool can
1 2 3 4 5
Observed 10.9 20.7 40.7 66.3 85.7 potentially help clinicians to identify patients early before
Predicted 9.7 23.5 39.9 63.2 87.9 CKD progression starts, thereby allowing intensive treatment
Predicted Risk Quinle for those at high risk. Second, our study is based on a cohort
with T2DM exclusively. This reduced the heterogeneity of the
Fig. 2 – Observed versus predicted risk of CKD progression
study population in assessing the contribution of different
during a 6-year follow-up in the test data set.
risk factors to CKD progression and addressed an important
unmet need of a population at extremely high risk for CKD.
Our results will also add to the findings from the current lim-
ited pool of research which focused on T2DM [17–19]. The
4. Discussion
third strength is the use of easily available clinical and labora-
We have developed and validated a prediction model for CKD tory information which are routinely used in clinical practice.
progression in patients with T2DM. The AUC or C-statistic in Furthermore, the prediction tool is practical and can be easily
our model was 0.80 in the training dataset and 0.83 in the test incorporated into the clinical information system. It was
dataset, showing good discriminatory performance. In a sys- reported that the presence of risk prediction model and inte-
tematic review by Echouffo-Tcheugi on risk models to predict gration into clinical practice guidelines have resulted in
CKD and its progression, it was reported that the C-statistic improved compliance to treatment guidelines and promoted
ranged from 0.56 to 0.94, indicating modest to good perfor- individual decision-making.
mance [6]. Our model has performed relatively well in com- There are a few limitations in this study. First, our sample
parison with these studies. In addition, our model showed size was moderate in comparison with other studies. Second,
good calibration with p > 0.90. our cohort came from Diabetes Centre in a regional hospital
Our study demonstrated that lower eGFR, higher uACR, and hence was not fully representative of the entire popula-
higher SBP, higher HbA1c, higher LDL-Cholesterol and older tion of individuals with T2DM in Singapore. This limits the
age are significantly associated with CKD progression in generalizability of findings to other cohorts such as patients
T2DM. This reflects the influence of adverse metabolic profile in primary care. Third, there is a potential problem of mea-
on CKD progression. These metabolic risk factors, apart from surement error using MDRD equation to estimate GFR. It
age, are potentially modifiable and the importance of their was, however, shown that MDRD had better or similar perfor-
control should be emphasized in patient education and clini- mance than CKD-EPI equation in patients with DM [27]. Lastly,
cal management. The model incorporating these risk factors there may be other risk factors which may influence CKD pro-
can potentially be used as a simple medical calculator gression but not explored in this study such as genetic fac-
(Supplementary Fig. 1). tors, socio-economic status, diet and exercise.
54 diabetes research and clinical practice 1 2 3 ( 2 0 1 7 ) 4 9 –5 4

5. Conclusions [12] Tina-quant Albumin Gen. 2: CobasÒ. Mannheim (Germany):

Roche Diagnostics GmbH; 2014.
[13] Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A
We have developed and validated a model to predict CKD pro-
more accurate method to estimate glomerular filtration rate
gression in Asians with T2DM. The model demonstrated that from serum creatinine: a new prediction equation. Ann Int
eGFR, uACR, SBP, HbA1c, LDL-Cholesterol and age influenced Med 1999;130:461–70.
the risk of CKD progression. The model performed well [14] Kidney Disease: Improving Global Outcomes (KDIGO) CKD
among individuals who share characteristics similar to our Work Group. KDIGO 2012 clinical practice guideline for the
source-population and can be a practical tool for easy imple- evaluation and management of chronic kidney disease.
mentation in routine clinical practice. Kidney Int Suppl 2012;2013(3):1–150.
[15] Guzder RN, Gatling W, Mullee MA, Mehta RL, Byrne CD.
Prognostic value of the Framingham cardiovascular risk
Conflict of interest statement engine and the UKPDS risk engine for coronary heart disease
in newly diagnosed type 2 diabetes: results from a United
The authors declare that they have no conflict of interest. Kingdom study. Diab Med 2005;22:554–62.
[16] Davis WA, Colagiuri S, Davis TM. Comparison of the
Framingham and United Kingdom prospective diabetes study
Acknowledgements cardiovascular risk equations in Australian patients with
type 2 diabetes from the fremantle diabetes study. Med J Aust
This study was supported by Alexandra Health Grant (AHPL 2009;190:180–4.
SIGII/11001; SIG/11029 and SIG/12024). [17] Jardine MJ, Hata J, Woodward M, Perkovic V, Ninomiya T,
Arima H, et al. Prediction of kidney-related outcomes in
patients with type 2 diabetes. Am J Kidney Dis 2012;60:770–8.
Appendix A. Supplementary material [18] Keane WF, Zhang Z, Lyle PA, Cooper ME, de Zeeuw D,
Grunfeld P, et al. Risk scores for predicting outcomes in
Supplementary data associated with this article can be found, patients with type 2 diabetes and nephropathy: the RENAAL
in the online version, at http://dx.doi.org/10.1016/j.diabres. study. Clin J Am Soc Nephrol 2006;1:761–7.
2016.11.008. [19] Yang XL, So WY, Kong AP, Clarke P, Ho CS, Lam CW, et al. End-
stage renal disease risk equations for Hong Kong Chinese
patients with type 2 diabetes: Hong Kong diabetes registry.
Diabetologia 2006;49:2299–308.
R E F E R E N C E S
[20] Tangri N, Stevens LA, Griffith J, Tighiouart H, Djurdjev O,
Naimark D, et al. A predictive model for progression of
chronic kidney disease to kidney failure. JAMA
[1] International Diabetes Federation. IDF diabetes atlas. 7th 2011;305:1553–9.
ed. Brussels, Belgium: International Diabetes Federation; [21] Peeters MJ, van Zuilen AD, van den Brand JA, Bots ML,
2015. Blankestijn PJ, Wetzels JF. Validation of the kidney failure risk
[2] Ramachandran A, Snehalatha C, Shetty AS, Nanditha A. equation in European CKD patients. Nephrol Dial Transplant
Trends in prevalence of diabetes in Asian countries. World J 2013;28:1773–9.
Diab 2012;3:110–7. [22] Desai AS, Toto R, Jarolim P, Uno H, Eckardt KU, Kewalramani
[3] Remuzzi G, Schieppati A, Ruggenenti P. Clinical practice. R, et al. Association between cardiac biomarkers and the
Nephropathy in patients with type 2 diabetes. N Engl J Med development of ESRD in patients with type 2 diabetes
2002;346:1145–51. mellitus, anemia, and CKD. Am J Kidney Dis 2011;58:717–28.
[4] Ministry of Health, Singapore. National health survey [23] Hallan SI, Ritz E, Lydersen S, Romundstad S, Kvenild K, Orth
2004. Singapore: Ministry of Health; 2005. SR. Combining GFR and albuminuria to classify CKD
[5] Ministry of Health, Singapore. National health survey improves prediction of ESRD. J Am Soc Nephrol
2010. Singapore: Ministry of Health; 2013. 2009;20:1069–77.
[6] Echouffo-Tcheugui JB, Kengne AP. Risk models to predict [24] Landray MJ, Emberson JR, Blackwell L, Dasgupta T, Zakeri R,
chronic kidney disease and its progression: a systematic Morgan MD, et al. Prediction of ESRD and death among
review. PLoS Med 2012;9:e1001344. people with CKD: the chronic renal impairment in
[7] dablÒ Educational Trust. Sphygmomanometers for self- Birmingham (CRIB) prospective cohort study. Am J Kidney Dis
management of blood pressure (SBPM). Upper arm devices 2010;56:1082–94.
for self-measurement of blood pressure. Dublin, Ireland: [25] Johnson ES, Thorp ML, Platt RW, Smith DH. Predicting the risk
dablÒ Educational Trust, 2014. Available at: <http://www. of dialysis and transplant among patients with CKD: a
dableducational.org/sphygmomanometers/devices_2_sbpm. retrospective cohort study. Am J Kidney Dis 2008;52:653–60.
html#OmronHEM7011C1>. [26] Halbesma N, Jansen DF, Heymans MW, Stolk RP, de Jong PE,
[8] Roche. Creatinine plus ver. 2: cobasÒ. Mannheim Gansevoort RT. Development and validation of a general
(Germany): Roche Diagnostics GmbH; 2016. population renal risk score. Clin J Am Soc Nephrol
[9] Roche. LDL-cholesterol gen. 3: cobasÒ. Mannheim 2011;6:1731–8.
(Germany): Roche Diagnostics GmbH; 2016. [27] Rognant N, Lemoine S, Laville M, Hadj-Aı̈ssa A, Dubourg L.
[10] Roche. Triglycerides: cobasÒ. Mannheim (Germany): Roche Performance of the chronic kidney disease epidemiology
Diagnostics GmbH; 2016. collaboration equation to estimate glomerular filtration rate
[11] Tina-quant Hemoglobin A1c Gen. 3 – Hemolysate and whole in diabetic patients. Diab Care 2011;34:1320–2.
blood application: cobasÒ. Mannheim (Germany): Roche
Diagnostics GmbH; 2015.

View publication stats