Key words
metabolic syndrome - prediction model - risk factors - general population
Abbreviations
ALB:
Albumin
ALP:
Alkaline phosphatase
ALT:
Alanine aminotransferase
AST:
Aspartate aminotransferase
AUROC:
The area under the receiver operating characteristic curve
BMI:
Body mass index
CI:
Confidence interval
Cr:
Creatinine
DBP:
Diastolic blood pressure
FPG:
Fasting plasma glucose
GGT:
γ-Glutamyl transferase
HDL-C:
High-density lipoprotein cholesterol
HR:
Hazard ratio
LDL-C:
Low-density lipoprotein cholesterol
MetS:
Metabolic syndrome
NAFL:
Non-alcoholic fatty liver
PLT:
Platelet count
ROC:
Receiver operator characteristic
SBP:
Systolic blood pressure
STB:
Serum total bilirubin
TC:
Total cholesterol
TG:
Triglyceride
UA: Uric acid
WBC:
White blood cell count
Introduction
Metabolic syndrome (MetS), characterized by a group of metabolic disturbances including
central obesity, glucose intolerance, hypertension, and hyperlipidemia [1] is a growing public health problem worldwide [2] with an increasing prevalence [3]. Additionally, MetS is associated with the risk of diabetes, cardiovascular disease,
and even death [4]
[5]
[6]. Therefore, more attention should be paid to the prevention of MetS. A prediction
model, which is able to identify the individuals with higher risk of MetS development,
is urgently needed to estimate the risk of MetS leading to proper interventions at
an earlier stage.
Previous studies have reported many related risk factors associated with the incidence
of MetS [7]
[8]
[9], such as uric acid (UA), γ-glutamyl transpeptidase (GGT), and alanine aminotransferase
(ALT). However, there are few studies grouping these factors to develop a model and
predict the risk of MetS. Only one study [10] published in 2015 has identified several independent risks and created a composite
score to predict the incidence of MetS. Based on a Japanese employees database, a
model was developed to discriminate MetS from healthy individuals and evaluated the
predictive potential for recovery from MetS. However, the model contains nine variables
including five used to diagnose MetS plus four independent factors, limiting its application.
In this study, we identified four items related to the MetS, constructed and validated
a clear model based on routine laboratory and anthropometric parameters to predict
a 3-year risk of MetS.
Subjects and Methods
Study population
In this study, we screened 4395 initially MetS-free patients who underwent an annual
health examination at Wenzhou Medical Center of Wenzhou People’s Hospital from 2010
to 2014. The examination includes anthropometric measurements, blood tests, and a
physical examination. The entire database was then randomly divided in a 2:1 ratio
into training (1365 males and 1565 females, age 41.75±14.70 years) and validation
cohort (681 males and 781 females, age 41.95±14.89 years). In addition, verbal informed
consent was obtained from each participant and the protocol of the study was approved
by the ethics committee of the Wenzhou People’s Hospital, respectively.
Event criteria
MetS was defined according to the guidelines as proposed by the China Diabetes Federation
[11] , which indicates that MetS is present if at least three of the following parameters
are present: 1) Central obesity: BMI≥25 in both genders; 2) Hypertriglyceridemia:
TG ≥1.7 mmol/l; HDL-C <0.9 mmol/l in males and <1.0 mmol/l in females; 3) Hypertension:
SBP≥140 mmHg or DBP≥90 mmHg or previously diagnosed; and 4) Hyperglycemia: FPG≥6.1 mmol/l,
or hyperglycemia previously diagnosed.
Data collection
All subjects were instructed to refrain from exercise prior to their examination with
clinical examination and data recording conducted the next morning. Medical history,
lifestyle, and drug regimes were recorded by well-trained medical staff. Anthropometric
measurements included body height and weight. The body mass index (BMI), was calculated
by dividing weight (kg) by the square of height (m). Blood pressure, including systolic
blood pressure (SBP) and diabolic blood pressure (DBP), was measured automatically
(OMRON, Japan), lege artis. Fasting blood samples were collected from each subject
and were used for the analysis of biochemical laboratory test. The laboratories were
certified according to International Organization Standardization. Laboratory parameters
included albumin, white blood cell counts (WBC), blood urea nitrogen (BUN), triglyceride
(TG), total cholesterol (TC), low density lipoprotein-cholesterol (LDL-C), high density
lipoprotein-cholesterol (HDL-C), serum total bilirubin (STB), fasting plasma glucose
(FPG), serum creatinine (sCr), serum uric acid (sUA), alkaline phosphatase (ALP),
aspartate aminotransferase (AST), alanine aminotransferase (ALT), and γ-glutamyl transferase
(GGT). All values were subsequently analyzed by an automated analyzer (Abbott AxSYM,
Park, IL, USA) using standard methods.
Statistical analysis
Continuous variables were summarized as mean±standard deviation (SD), and categorical
variables were expressed as percentages (%). The characteristics of the study population
according to database were assessed using one way analysis of variance (ANOVA) and
χ2 test for categorical variables. Univariate and multivariate logistic regression analyses
were used to determine the risk factors for MetS. Additionally, a stepwise multivariable
logistic regression model was employed to develop a predictive model from the training
cohort. For all analysis, two-tailed p-value <0.05 were considered statistically significant
and a p-value <0.1 was considered indicative of a statistical trend. Data analyses
were conducted using SPSS statistics software (version 22; IBM Corp.) and MedCalc
version 12.7 (MedCalc Software).
Variable selection
First, several potential MetS risk factors based on recent literature were selected
for evaluation. Univariate association analysis including potential risk factors was
conducted and variables with p <0.05 were considered significant. Finally we retained,
the four variables that performed well both in the univariate and multivariate analysis
for the final model.
Construction of the MetS risk score
For the training cohort, we translated continuous risk factor variables into categorized
variables first, and then performed stepwise multivariable logistic regression analysis
to compute β-coefficients for the four variables. For the analysis, the conditional
probabilities used for the entry and removal of a factor were 0.05 and 0.10, respectively.
Then, we established a scoring system that assigned risk scores to each variable based
on the magnitude of its β-coefficient in the multivariable logistic regression model.
A sum score, which was named the MetS risk score, was calculated for each participant
by adding the scores for four variables together. The mean 3-year risk of all possible
combinations of risk factors for a specific total score was computed to obtain 3-year
risk values ([Fig. 2]). To assess the predictive potential of the model to discriminate “events” from
“nonevents”, the area under the receiver-operating characteristic curve (AUROC) was
calculated.
Fig. 2 Score sheet for estimating 3-year risk of MetS incidence.
Validation of the MetS risk score
The performance of the risk score was evaluated in the validation cohort and entire
sample. The predictive performance of the MetS risk score was evaluated using the
AUROCs.
Results
Baseline characteristics of cohort population
A total of 10419 individuals were initially recruited into the study. Only 4395 (2930
in training cohort and 1465 in validation cohort) individuals were enrolled according
to exclusion criteria ([Fig. 1]). Baseline clinical and biochemical parameters of training and validation cohort
are summarized in Table 1S. No significant difference was found between the training (n=2930) and the validation
cohorts (n=1465).
Fig. 1 Study flow diagram.
Table 2S shows that patients who developed MetS in 3 years had an older age (44.28 vs. 41.52,
p=0.006), a higher BMI (22.98 vs. 21.97, p<0.001), FPG (5.60 vs. 5.19, p<0.001), SBP
(123.74 vs. 120.22, p=0.001), DBP (74.32 vs. 77.03, p<0.001), UA (293.74 vs. 278.49,
p=0.029), GGT (32.72 vs. 25.66, p<0.001), and lower HDL-C (1.31 vs. 1.38, p<0.001)
in the training cohort while LDL-C, TC and Cr showed no significant difference. The
patients’ conditions were similar to the validation cohort.
Development of the MetS risk score
To identify independent predictors of MetS, the univariate logistic regression analysis
was performed to test the relationships between the potential risk factors and the
incidence of MetS in the training cohort (Table 3S). In the univariate analysis, we found that age, BMI, SBP, DBP, TG, TC, FPG, GGT,
HDL-C, UA, and WBC were significantly associated with MetS development (all p<0.05).
The above variables that were significantly associated with the risk of MetS were
consequently entered into the multivariable logistic regression analysis to select
independent predictors. Finally, as presented in [Table 1], BMI (HR=0.091, 95% CI: 1.040–1.155), FPG (HR=1.507, 95% CI: 1.305–1.714, p<0.001),
DBP (HR=0.016, 95% CI: 1.002–1.031), HDL (HR=0.539, 95% CI: 0.303–0.959), were identified
as the independent risk factors.
Table 1 Multivariate analysis of the association between incidence
and clinical and biochemical characteristics in the training cohort.
|
Variables
|
B
|
HR
|
95 % CI
|
p-Value
|
|
BMI
|
0.091
|
1.096
|
1.040–1.155
|
0.001
|
|
FPG
|
0.410
|
1.507
|
1.305–1.741
|
˂0.001
|
|
DBP
|
0.016
|
1.016
|
1.002-1.031
|
0.021
|
|
HDL-C
|
− 0.617
|
0.539
|
0.303–0.959
|
0.035
|
B: Intercept. For abbreviations, see text
[Table 2] illustrates the results of the multivariate logistic regression performed on the
4 variables and the method to calculate the MetS score. For these 4 variables, the
cut-off value of each parameter to distinguish two severity categories with maximum
Youden Index in order to predict the risk of MetS was calculated. Then we derived
an integer or half-integer score for prediction based on the multivariable logistic
regression coefficients. The lowest value of β-coefficients was chosen to obtain a
score of 1. Other β-coefficient values were then divided to the lowest value and resulted
as score for each variable. Each point was rounded to an integer and half-integer.
As a result, we assigned 1.5 points to the predictor BMI (kg/m2)≥24 and FPG (mmol/l)≥5.2, –1.5 points to the predictor HDL-C (mmol/l)≥ 1.4 and 1
point to the predictor DBP (mmHg)≥73.5. The final score per subject ranged from –1.5
to 4. A risk estimation chart based on combinations of different points of the four
predictors in the MetS risk score was drawn for individual risk prediction ([Fig. 4]).
Fig. 4 Risk estimation chart of MetS risk score.
Table 2 MetS risk score based on multivariable analysis of risk factors for MetS in the training
cohort.
|
Categorical variable
|
Range
|
β
|
p-Value
|
HR (95% CI)
|
Risk score
|
|
BMI (kg/m2)
|
<24
|
Reference
|
–
|
1.00
|
0
|
|
≥24
|
0.790
|
<0.001
|
2.204 (1.661–2.924)
|
1.5
|
|
FPG (mmol/l)
|
<5.2
|
Reference
|
–
|
1.00
|
0
|
|
≥5.2
|
0.695
|
<0.001
|
2.004 (1.524–2.637)
|
1.5
|
|
HDL-C (mmol/l)
|
<1.4
|
Reference
|
–
|
1.00
|
0
|
|
≥1.4
|
−0.780
|
<0.001
|
0.458 (0.312–0.612)
|
–1.5
|
|
DBP (mmHg)
|
<73.5
|
Reference
|
–
|
1.00
|
0
|
|
≥73.5
|
0.467
|
0.001
|
1.595 (1.199–2.123)
|
1
|
Performance of MetS score in the training cohort
[Fig. 3] illustrates the ROC curves in the training cohort, validation cohort and entire
sample. The area under the receiver operating characteristic curve (AUROC) was 0.674
for training cohort, 0.690 for validation cohort and 0.680 for the entire database,
implying that the discrimination of the model was good. Observed and predicted ratios
of 3-year incidence/risk (%) of MetS was detailed in Fig. 1S, which was calculated according to risk values revealed in [Fig. 2], suggesting no significant difference between observed and predicted ratios in both
training cohort (8.98% vs. 8.16%) and validation cohort (8.66% vs. 8.73%).
Fig. 3 Receiver operating characteristic curves analysis of the discriminative ability of
MetS risk score to predict 3-year MetS development risk in the general population.
Discussion
In this study, we established and validated a new MetS risk score to predict the risk
of MetS within the next three years. To our knowledge, this is the first study that
developed a prediction score for incident MetS using the components of MetS. According
to the result of the AUROC, the model showed a good predictability. Physicians can
employ the MetS Risk Score to make individual predictions easily, and identify those
potential patients and inform them to promote healthier behaviors to prevent the disease.
As we know, there are many diagnostic criteria of MetS around the world [11]
[12]
[13]. such as the criteria from the Joint Interim Statement of the International Diabetes
Federation Task Force of Epidemiology and Prevention American Heart Association, World
Heart Federation, International Atherosclerosis Society, American Heart Association,
International Association for the Study of Obesity and China Diabetes Federation.
Finally, considering the composition of the population and the popularity of the criteria,
MetS was diagnosed according to the China Diabetes Federation, which was composed
of Central Obesity, Hypertriglyceridemia, Hypertension and Hyperglycemia.
Several risk factors of MetS have been identified in our study. First, we found that
abdominal obesity (BMI) (β=0.790, HR=2.204) was the most important contributor to
the incidence of MetS among the four items included in the score, which is in line
with others [14] indicating that intra-abdominal fat is a major determinant for the metabolic syndrome.
Recent researches [15]
[16] showed that GGT is also associated with the risk of developing MetS, which was also
identified in our study. The mechanisms have not been fully elucidated, but a study
found that GGT may contribute to incident MetS via inflammation, oxidative stress
pathways and insulin resistance [17]. However, due to the poor performance in the multivariable logistic regression analysis,
it was not included in the score. Although the effect of gender difference on the
incidence of the syndrome remains uncertain, we discovered that female sex could also
be regarded as a risk score as a previous study demonstrated that female sex is an
independent item to predict MetS in the United States [18]. Therefore, further research is required to validate if female sex can also be regarded
as a risk factor in Asian population. Additionally, non-alcoholic fatty liver (NAFL)
demonstrated a tight connection with the MetS [19]. Its components have also been reported to independently predict the risk of NAFL
in a northern urban Han Chinese population [20]. However, we found no significant association between MetS and NAFL in our study.
One possible explanation is that NAFL may gradually disappear in a 3-year follow up
not directly predicting the appearance of MetS. Elevated uric acid was also reported
to be associated with MetS, which was confirmed in several studies [8]
[21]
[22] . The exclusion of this important factor may have influenced the difference of diagnostic
criteria. Finally, considering the performance both in multivariate analysis and AUROC,
we identified four risk factors, BMI, FPG, HDL-C and DBP in our model.
There are several strengths shown in our study. First, this is the first clean model
based on routine laboratory and anthropometric parameters to predict a 3-year risk
of MetS. The items included in the model are easy to obtain facilitating implementation.
Individuals, who were identified at high-risk for MetS, can initiate a healthier lifestyle.
Despite the strengths of the present study, there are some limitations that need to
be addressed. First, diagnoses of MetS in our study were based on the China Diabetes
Federation, which is not universally implemented but considered most suitable for
our study. Second, because of insufficient information retrieval, we could not evaluate
lifestyle related parameters such as smoking, drinking and physical activity in this
study [23]. Third, related with a limited dataset, the scoring model was not validated in different
centers. We randomly divided the data into two parts: one to build the model and another
to validate its performance. Validation could be optimized by X-validation. Further
research is required to validate the scoring model in other multicenter studies.
Conclusions
In conclusion, we have developed a clear scoring model, the MetS risk score, for evaluating
the 3-year risk of MetS individually. Our model identified 4 predictors to score the
risk and performed well both in the training cohort and validation cohort.
Authors’ Contributions
Tian-Tian Zou: designed the study, did the statistical analyses, interpreted data,
and wrote the manuscript. Yu-Jie Zhou: did the statistical analyses and collected
data. Xiao-Dong Zhou and Wen-Yue Liu: interpreted data and revised the manuscript.
Sven Van Poucke: revised the manuscript. Wen-Jun Wu and Dong-Chu Zhang: data collection.
Ji-Na Zheng and Xue-Mei Gu: interpreted data. Ming-Hua Zheng and Xiao-Yan Pan: designed
the study, obtained funding, reviewed the results, and helped to write the manuscript.
All authors saw and approved the final version of the paper.
Funding
This work was supported by grants from the National Natural Science Foundation of
China (81500665), Scientific Research Foundation of Wenzhou (Y20160223), High Level
Creative Talents from Department of Public Health in Zhejiang Province, and Project
of New Century 551 Talent Nurturing in Wenzhou.
