A clinically practicable diagnostic score for metabolic syndrome improves its predictivity of diabetes mellitus: The Gruppo Italiano per lo Studio della Sopravvivenza nell'Infarto miocardico (GISSI)–Prevenzione scoring

Article Outline

• Abstract
• Material and methods
  • Multivariate analysis
  • Global risk assessment score
  • Validation of the algorithm
• Results
  • Final risk model
  • Composite risk assessment score
  • MS-Score as a diagnostic tool for MS
  • MS-Score and clinical events during follow-up
• Discussion
• Conclusion
• Appendix. 
• References
• Copyright

Background

Metabolic syndrome (MS) is associated with late-onset diabetes. However, diagnostic criteria for individual components of MS are based on categorical/arbitrary cut points and, therefore, do not exploit the information yield of each factor. We aimed to generate a diagnostic score for MS (MS-Score), aimed at predicting diabetes by giving appropriate weight to the individual components of MS.

Methods

Of 11323 patients with prior myocardial infarction and followed up for 3.5 years in the GISSI-Prevenzione study, 3855 subjects with diabetes at baseline or missing information for relevant variables were excluded. A Cox proportional hazards model including age, sex, glycemia, high-density lipoprotein cholesterol, triglycerides, hypertension, and body mass index was fitted to create a diagnostic score. A cutoff point of 28 of the score was the best compromise between sensitivity and specificity for MS diagnosis (MS-Score). The prognostic performance of the MS-Score was compared with that of the diagnostic criteria of MS, as defined by National Cholesterol Education Program Adult Treatment Panel III (MS-ATP).

Results

Of 7468 patients, 940 developed diabetes. The risk of getting diabetes significantly and progressively increased in the quintiles of the score reaching >6-fold higher risk in the last one. The predictive capability of MS-Score was significantly higher than that of the MS-ATP (AUC = 0.650 vs 0.587, sensitivity 67% vs 52%, specificity 63% vs 66%, P = .0002). The MS-Score, but not the MS-ATP, was significantly associated with mortality.

Conclusion

MS-Score improves the prediction of diabetes development by using the full informative content of individual components for diagnosis of MS.

Abnormal glucose metabolism precedes type 2 diabetes mellitus, which is a well-known risk factor for cardiovascular disease,¹, ², ³, ⁴, ⁵ and develops over a protracted period during which individuals are at high risk of cardiovascular events.⁶, ⁷, ⁸, ⁹, ¹⁰ This period is characterized by a progressive resistance to the action of insulin, which usually clusters with several cardiovascular risk factors.¹¹, ¹², ¹³ Early identification of such subjects to prevent diabetes as well as cardiovascular events during this “prediabetes” status is a major task in clinical practice. The recommended diagnostic procedure is to perform an oral glucose tolerance test, although the feasibility and the cost/benefit of such strategy in clinical practice has been questioned.¹⁴ A pragmatic surrogate of oral glucose tolerance test to diagnose insulin resistance (IR) is the diagnosis of metabolic syndrome (MS).¹⁵ However, there is no agreement on the criteria and cutoff points for the variables used to diagnose MS.¹⁵, ¹⁶, ¹⁷, ¹⁸, ¹⁹ The various definitions share a categorical use of the diagnostic components of MS, making the syndrome more a repackaging of cardiovascular risk factors—contributing in the same fashion to the diagnosis—into a more attractive entity that hopefully spurs physicians to treat high-risk patients than a clear-cut nosological entity.²⁰

The GISSI-Prevenzione database,²¹ including patients with recent myocardial infarction who were followed up for 3.5 years after the index event, allowed us to exploit the informative yield of the individual diagnostic components used for the diagnosis of MS by creating a diagnostic score (MS-Score) aimed at predicting the risk of late-onset diabetes to be easily used in clinical practice.

Back to Article Outline

Material and methods 

A detailed description of the study population has been reported previously.²¹ Briefly, 11323 patients with recent (≤3 months) myocardial infarction were enrolled into GISSI-Prevenzione, a multicenter, open-label clinical trial on the efficacy of n-3 polyunsaturated fatty acids (PUFA) 1 g daily and vitamin E 300 mg daily, both and neither drug. Clinical and laboratory evaluation were carried out at the follow-up visits scheduled at 6, 12, 18, 30, and 42 months after randomization.

Patients with at least 3 of the following 5 criteria at baseline were considered to have MS according to National Cholesterol Education Program (NCEP) Adult Treatment Panel III (ATP III) report (MS-ATP) criteria¹⁵: hypertension (history of hypertension, systolic blood pressure ≥130 mm Hg, diastolic blood pressure ≥85 mm Hg, or antihypertensive treatment), triglycerides ≥150 mg/dL, high-density lipoprotein (HDL) cholesterol <40 mg/dL for men and <50 mg/dL for women, fasting glucose (FG) level ≥110 mg/dL. Because we did not measure waist circumference, we used the median value of body mass index (BMI ≥26) as a proxy of abdominal obesity. We did a sensitivity analysis using BMI cutoffs of 28 and 29 (corresponding to the upper quartile and quintile of the distribution) with no meaningful changes in the prognostic capacity of MS-ATP (data not shown).

Patients with diabetes at baseline (no. 2139, either history of diabetes, FG ≥126 mg/dL, or hypoglycemic treatment), those with missing data for MS-ATP components or diabetes status at baseline (no. 939), and those with missing information as to glycemic status during follow-up (no. 777) were excluded from this analysis. The final analysis was carried out on 7468 patients. The main outcome measure was development of diabetes during follow-up, defined as an FG ≥126 mg/dL or current hypoglycemic pharmacological treatment. We also assessed the predictive value of the MS-Score on total mortality and on the composite end point of death or development of diabetes.

Multivariate analysis 

The robustness of the final multivariate time-to-event analysis was confirmed by fitting a logistic model that gave similar results and was used to calculate the Hosmer-Lemeshow statistics and the c index.²² Potential confounders (eg, cardiovascular risk factors, left ventricular ejection fraction, residual coronary ischemia, etc) were initially included in the multivariate model but finally withdrawn because they were not associated with the risk of developing diabetes during follow-up.

Triglycerides, HDL cholesterol, FG, BMI, systolic (SBP) and diastolic blood pressure (DBP) were categorized as statistical quintiles. Categories of variables showing similar relative risks were collapsed together to improve the usability of the equations, as well as to avoid overfitting. Because SBP and DBP quintiles did not discriminate the risk of late-onset diabetes, hypertension was entered into the final model as the same categorical variable used in the ATP III classification. Although highly significant, the risk of diabetes did not increase proportionally with age, therefore we adopted a cutoff point at 50 years. The population attributable risk (PAR) for diabetes mellitus of each factor was calculated by a method based on unconditional logistic regression and, therefore, is adjusted for confounders.²³, ²⁴, ²⁵

Global risk assessment score 

Details of the statistical procedures are given in the appendix. To produce a global risk assessment score, we assigned points for each risk factor, which were weighted proportionally to the value of the β coefficients of the multivariate analysis. The score was obtained by summing all the individual points. The risk predicted by a model with the total score as a continuous explanatory variable was used to compute quintiles of score indicating progressively increasing risk of diabetes.

We calculated sensitivity, specificity, positive predictive value, negative predictive value, and receiver-operating characteristic (ROC) curves for the score with their 95% CIs.²⁶, ²⁷ The optimal cut point being the peak of the curve for the total score at 28 was the best compromise between sensitivity and specificity for MS diagnosis (MS-Score).

We classified the patients in 4 groups according to the MS-ATP and MS-Score criteria, and compared the ROC curves of the 2 diagnostic tools to compare their predictive ability of late-onset diabetes. Statistical differences in the area under the ROC curves (AUCs) and 95% CIs were compared using the method of DeLong et al.²⁸ We used likelihood ratio tests to globally compare the various models.

Validation of the algorithm 

Calibration of the model was assessed by evaluating expected to observed outcomes across deciles of risk, whereas discrimination was measured as the AUC.²⁹, ³⁰, ³¹, ³², ³³ As to generalizability, the predictive performance of the model and the extent of overfitting were assessed through internal validation, that is, crossvalidation and bootstrapping,³⁴ and used to estimate the parameter “overoptimism,” that is, a measure of the extent to which the predictive accuracy of the model based on the full set of data was overoptimistic. All P values were 2-sided. All computations have been carried out using the SAS statistical package (SAS Institute Inc., Cary, NC).³⁵

Back to Article Outline

Results 

Subjects in GISSI-Prevenzione appeared to be a broad, relatively low-risk population of acute myocardial infarction survivors recruited early (median 16 days) after the index event (Table I).

Table I. Baseline characteristics of 7468 patients of the GISSI-Prevenzione database

Age	58.7 (10.6)
Age ≥70 y	1116 (14.9)
Men	6496 (87.0)
Diagnosis of MS according to ATP III	2731 (36.6)
No. of MS diagnostic components
0	544 (7.3)
1	1761 (23.6)
2	2432 (32.6)
3	1908 (25.6)
4	750 (10.0)
5	73 (1.0)
MS diagnostic criteria of NCEP/ATP III
Blood pressure	4228 (56.6)
Triglycerides	3380 (45.3)
HDL cholesterol	3748 (50.2)
Glucose	661 (8.9)
BMI	3697 (49.5)
Systolic blood pressure (mm Hg)	123 (15)
Diastolic blood pressure (mm Hg)	77 (9)
Total blood cholesterol (mg/dL)	212 (42)
Triglycerides (mg/dL)	158 (79)
HDL cholesterol (mg/dL)	42 (11)
Fasting glucose level (mg/dL)	91 (12)
Smokers before index event	3325 (44.5)
Peripheral arterial disease	256 (3.4)
Creatinine (mg/dL)	1.08 (0.34)
NYHA II	627 (8.4)
Ejection fraction ≤40%	766 (11.8)
Ejection fraction (%)	53 (10)

For continuous variables values are mean (SD).

For categorical variables: percentage values are given in parenthesis.

Diagnosis of MS according to ATP III criteria was present in 37% patients and was based on the presence of 3, 4, and 5 diagnostic components in 70%, 27%, and 3% of subjects, respectively. Hypertension was the most frequent individual contributor to the diagnosis of MS (81% of patients) and FG ≥110 mg/dL was the lesser one (18% of patients).

Thirteen percent of patients developed diabetes during follow-up, 454 (9.6%) of 4737 patients without MS-ATP diagnosis and 486 (17.8%) of 2731 patients with MS-ATP diagnosis (RR 1.95, 95% CI 1.71-2.21, P < .0001).

Final risk model 

Table II shows the statistically significant predictors of diabetes during follow-up. Fasting glucose was the most important and statistically significant contributor to the score, levels from 90 to 99, 100 to 109, and ≥110 mg/dL being associated with >2-, 3-, and 6-fold greater relative risk of diabetes than patients with FG <80 mg/dL, respectively. Triglycerides were the second most important contributor to the score, with values between 100 to 159 and ≥200 mg/dL being associated with a 43% and 96% increased relative risk of diabetes, respectively. Body mass index was associated with 22% and 51% increased relative risk of diabetes in patients with moderate overweight and with BMI ≥28, respectively.

Table II. Final predictive model: distribution, relative risk, 95% CI, P value, and risk points for independent predictors contributing to the MS-Score to predict late-onset diabetes

	%	Diabetes (no. of patients)	Relative risk	95% CI	P	PAR	Risk points
Men	87.0	835/6496 (12.9)	1.22	1.00-1.51	.0563	17%	3
Age >50 y	77.7	764/5804 (13.2)	1.30	1.09-1.53	.0028	19%	4
Hypertension	56.6	594/4228 (14.1)	1.19	1.04-1.36	.0136	11%	2
BMI						19%
<26	50.5	368/3771 (9.8)	1.00				0
26-27	22.6	221/1685 (13.1)	1.22	1.03-1.45	.0195		3
≥28	26.9	351/2012 (17.5)	1.51	1.30-1.75	<.0001		6
Triglycerides (mg/dL)						35%
<100	19.7	119/1471 (8.1)	1.00				0
100-159	40.8	366/3046 (12.0)	1.43	1.16-1.76	.0008		5
160-199	18.5	195/1382 (14.1)	1.63	1.29-2.05	<.0001		7
≥200	21.0	260/1569 (16.6)	1.96	1.57-2.44	<.0001		9
HDL blood cholesterol (mg/dL)						19%
≥50	22.0	161/1640 (9.8)	1.00				0
30-49	66.4	642/4956 (13.0)	1.23	1.03-1.46	.0227		3
<30	11.7	137/872 (15.7)	1.47	1.16-1.85	.0012		5
Fasting glucose level (mg/dL)						56%
<80	14.8	63/1105 (5.7)	1.00				0
80-89	30.7	187/2292 (8.2)	1.43	1.07-1.90	.0144		5
90-99	30.3	269/2266 (11.9)	2.03	1.54-2.67	<.0001		10
100-109	15.3	212/1144 (18.5)	3.27	2.47-4.34	<.0001		16
≥110	8.9	209/661 (31.6)	6.20	4.67-8.23	<.0001		28

PAR is for the top higher categories versus the lowest one (reference). Collectively, these 7 risk factors accounted for 89% of the PAR.

Together, age, sex, hypertension, BMI, triglycerides, HDL cholesterol, and FG accounted for 89% of the PAR of getting diabetes mellitus (Table II). However, FG was, by far, the most important factor (PAR = 56%), whereas hypertension was the factor with the lowest PAR (11%).

Composite risk assessment score 

The risk assessment score was obtained by summing all the individual points (shown in the right hand column of Table II) and ranged from 0 to 57 risk points.

Absolute observed risk of diabetes over 3.5 years and relative risk estimates (ie, the ratio of the absolute risk of a patient to that of a low-risk one) for each quintile of the score are given in Table III. The score discriminated subjects in the lowest quintile who were at low risk of diabetes (absolute risk 1.4 per 100 person-years) from those at very high risk of diabetes in the highest quintile (absolute risk 9.3 per 100 person-years). As compared with patients placed into the first quintile of the score, those who were in the other quintiles had a statistically significant and progressive increase of the relative risk of diabetes up to a 6-fold increase for the last quintile (Table III).

Table III. Incidence of diabetes mellitus during follow-up, according to score quintiles

Pretest probability of diabetes in the whole population was 13%.

The distribution of the score in the GISSI-Prevenzione population was skewed toward high-risk values, with most patients having a low to intermediate-low risk of getting diabetes (Figure 1). The final model had a high predictive capability, as illustrated by the plotting of observed and expected proportions of risk for diabetes (Figure 1), and the AUC was 0·696 (95% CI 0·678-0·714, P < .0001), which is a measure of the ability of the model to discriminate between patients who developed or did not develop diabetes within 3.5 years. The ROC curve of the global risk assessment score (Figure 2) used as a continuous variable was nearly identical to the one of the final model with the original variables, the AUC associated with the global risk assessment score being the same (test of DeLong et al,²⁸ P = .9992).

• 
  • View Large Image
  • Download to PowerPoint

• Figure 1. 

  Observed, predicted risk of diabetes, and distribution of patients according to the score for diagnosis of MS. Solid circles (•) and the gray line indicate the distribution of patients according to progressively increasing values of the score. Solid (♦) diamonds indicate the observed proportion of patients with diagnosis of diabetes during follow-up. The black line indicates the expected incidence of diabetes using the score points.

• 
  • View Large Image
  • Download to PowerPoint

• Figure 2. 

  Receiver-operating characteristic curves showing the performance of the various diagnostic tools in predicting diabetes. Separate plots were used to compare the final model with the original variables to the global risk assessment score used as a continuous variable, the MS-Score and the MS-ATP diagnostic criteria. □ = final model; ▴ = global risk assessment score; ♦ = MS-ATP; • = MS-Score.

The predictive performance of the full model was internally validated through cross validation and bootstrapping. The correction coefficient to the ROC curve (overoptimism) with the bootstrap technique was only 0.006 (from 0.696 to 0.690) for the final model with the original variables. Similar results were obtained with crossvalidation. As a consequence, the final predictive models had no need to be adjusted for overoptimism. The mathematical equations are given in the appendix, along with an example for their use.

MS-Score as a diagnostic tool for MS 

A cutoff point of 28 for the score was found to be the best compromise between sensitivity and specificity for MS diagnosis (MS-Score). Patients with a low MS-Score (<28) were at low risk of getting diabetes (7%), as compared with those with a high MS-Score (≥28) who were at high risk of diabetes (20.7%, RR 3.22, 95% CI 2.81-3.68, P < .0001). Figure 3 compares pictorially the predictive ability of the 2 diagnostic tools for MS in terms of diabetes-free survival curves. As compared with normal subjects according to both diagnostic criteria (47%), those who were positive to both diagnostic criteria (25%) had the higher risk of getting diabetes; those who were positive only to the MS-Score (16%) had a significantly higher risk of diabetes, whereas no difference in the risk of diabetes was found for patients who were positive only to the MS-ATP (12%).

• 
  • View Large Image
  • Download to PowerPoint

• Figure 3. 

  Diabetes-free survival of patients attributed to MS-ATP and MS-Score risk groups predicted by a multivariate model. As compared with 3523 patients without MS according to both MS-ATP and MS-Score criteria (incidence of diabetes 6.8%, RR 1.00), 896 were diagnosed as having MS according to ATP-III criteria but not with the MS-Score (incidence of diabetes 8.0%, RR 1.20, 95% CI 0.92-1.56, P = .18), 1214 were diagnosed as having MS according to MS-Score but not with the ATP III criteria (incidence of diabetes 17.8%, RR 2.84, 95% CI 2.36-3.42, P < .0001), and 1835 had an MS diagnosis according to both diagnostic criteria (incidence of diabetes 22.6%, RR 3.68, 95% CI 3.14-4.32, P < .0001).

Table IV shows the distribution of the diagnostic factors according to the concordance between the MS-ATP and MS-Score in the diagnosis of MS. Plasma FG and hypertension were the factors that differed most, 99.8% of subjects positive to MS-ATP and negative to MS-Score having FG <100 mg/dL and being likely to be hypertensive (79.9%). The reverse was true for subjects who were negative to MS-ATP and positive to MS-Score: 62.2% of them had FG ≥100 mg/dL and only 44.4% were hypertensive. As compared with subjects without MS according to both definitions, triglycerides and BMI levels were more likely to be high in subjects who were positive to MS-ATP and negative to MS-Score.

Table IV. Concordance between the MS-ATP and MS-Score in the diagnosis of MS according to the distribution of the various diagnostic factors

	MS-Score(−) MS-ATP (−) (n = 3523) (47.2%)	MS-Score(−) MS-ATP (+) (n = 896) (12.0%)	MS-Score (+) MS-ATP (−) (n = 1214) (16.3%)	MS-Score (+) MS-ATP (+) (n = 1835) (24.6%)
Age >50 y	2639 (74.9)	594 (66.3)	1054 (86.8)	1517 (82.7)
	58.4 ± 11.1 (58.9)	56.4 ± 11.0 (56.4)	60.3 ± 9.4 (60.6)	59.3 ± 9.7 (59.7)
Men	3027 (85.9)	727 (81.1)	1138 (93.7)	1604 (87.4)
Hypertension	1468 (41.7)	716 (79.9)	539 (44.4)	1505 (82.0)
SBP	120.1 ± 14.8 (120.0)	127.0 ± 14.8 (130.0)	121.7 ± 14.5 (120.0)	128.2 ± 14.9 (130.0)
DBP	75.2 ± 8.6 (80.0)	78.7 ± 8.6 (80.0)	76.0 ± 7.9 (80.0)	78.7 ± 8.7 (80.0)
Triglycerides (mg/dL)	131.3 ± 64.5 (120.0)	188.0 ± 80.2 (173.0)	156.9 ± 79.2 (135.0)	196.9 ± 84.2 (180.0)
<100	1121 (31.8)	77 (8.6)	184 (15.2)	89 (4.9)
100-159	1649 (46.8)	258 (28.8)	618 (50.9)	521 (28.4)
160-199	420 (11.9)	285 (31.8)	150 (12.4)	527 (28.7)
≥200	333 (9.5)	276 (30.8)	262 (21.6)	698 (38.0)
HDL (mg/dL)	44.8 ± 11.9 (44.0)	36.9 ± 9.1 (36.0)	43.6 ± 10.8 (43.0)	36.7 ± 9.2 (36.0)
≥50	1134 (32.2)	81 (9.0)	272 (22.4)	153 (8.3)
30-49	2129 (60.4)	676 (75.5)	833 (68.6)	1318 (71.8)
<30	260 (7.4)	139 (15.5)	109 (9.0)	364 (19.8)
BMI (kg/cm2)	24.8 ± 2.8 (24.7)	27.3 ± 3.0 (27.1)	26.5 ± 3.3 (25.9)	28.7 ± 4.0 (28.3)
<26	2625 (74.5)	231 (25.8)	624 (51.4)	291 (15.9)
26-27	537 (15.2)	398 (44.4)	209 (17.2)	541 (29.5)
≥28	361 (10.3)	267 (29.8)	381 (31.4)	1003 (54.7)
Glycemia (mg/ dL)	85.7 ± 8.9 (86.0)	81.8 ± 8.3 (83.0)	101.4 ± 8.5 (101.0)	100.8 ± 10.9 (99.0)
<80	781 (22.2)	322 (35.9)	0 (0.0)	2 (0.1)
80-89	1523 (43.2)	474 (52.9)	50 (4.1)	245 (13.4)
90-99	1087 (30.9)	98 (10.9)	409 (33.7)	672 (36.6)
100-109	132 (3.8)	2 (0.2)	575 (47.4)	435 (23.7)
≥110	0 (0.0)	0 (0.0)	180 (14.8)	481 (26.2)
Incidence of diabetes during follow-up	238 (6.8)	72 (8.0)	216 (17.8)	414 (22.6)

For continuous variables, values are mean ± SD (median). For categorical variables, percentage values are given in parenthesis.

As shown in Figure 2, the predictive capability of the MS-Score (AUC = 0.650) was significantly higher than that of the MS-ATP (AUC = 0.587) (test of DeLong et al,²⁸ P = .0002, sensitivity 67% vs 52%, specificity 63% vs 66%, positive predictive value 21% vs 18%, negative predictive value 93% vs 90%).

The significance of the differences between areas under the ROC curves were also validated by the bootstrap procedure for the following 3 models: (1) the score as continuous variable and quintiles, (2) the MS-score, and (3) the MS-ATP definition adjusted for age and sex. In the replications obtained with the bootstrap procedures, the median difference in the areas under the ROC curves for the models that included the MS-Score exceed those of the MS-ATP definition by 14.4%, the first and 99th percentiles for such difference being 9.8% and 19.1%.

MS-Score and clinical events during follow-up 

Figure 4 shows the capability of NCEP/ATP III diagnostic criteria for MS and of MS-Score to predict major clinical events during follow-up. As compared with patients who were negative to the MS-Score, those with a positive MS-Score had a 30% higher risk of fatal events (P < .0075), whereas the risk of death or diabetes during follow-up was more than doubled (P < .0001).

• 
  • View Large Image
  • Download to PowerPoint

• Figure 4. 

  Incidence, relative risk, and 95% CI for risk of diabetes, death, and death plus diabetes according to the MS-Score and the MS-ATP diagnostic criteria.

The ability of MS-ATP to predict deaths (loglikelihood χ² = 1.324, P = .2499) was significantly lower than that of the MS-Score (loglikelihood χ² = 7.084, P = .0078). Similar results were obtained for death plus diabetes (MS-ATP: loglikelihood χ² = 85.677, P < .0001) (MS-Score: loglikelihood χ² = 260.139, P < .0001).

Back to Article Outline

Discussion 

The MS is a complex nosological entity characterized by the clustering of several cardiovascular risk factors, the risk of developing diabetes and cardiovascular events, and that has at least 3 potential etiological categories: (1) obesity and disorders of adipose tissue, (2) IR, and (3) a constellation of other factors (eg, molecules of vascular, hepatic, and immunologic origin; hormonal changes; proinflammatory state; etc).³⁶ Accordingly, no agreement exists as to the definition of the MS diagnostic criteria. The ATP III considered the obesity epidemic¹⁵, ³⁵ and the association between MS and cardiovascular risk as the founding characteristics of the syndrome.¹⁵ The World Health organization considered IR as a mandatory component for MS diagnosis, and therefore, no diagnosis of MS can be done in the absence of a glycemic dysmetabolism.¹⁶ The American Association of Clinical Endocrinologists proposed diagnostic criteria for the Insulin Resistance Syndrome based on clinical judgment, which somewhat fall in the midline between those of ATP III and the World Health Organization.¹⁷

Such complexity is hardly dealt with by assuming that having 3 factors of 5, each of them contributing equally, can establish the diagnosis of MS. Our analysis shows that a weighted use of the individual components for MS diagnosis, which can be easily measured in clinical practice, allows to improve the prediction of late-onset diabetes and is significantly associated with the risk of hard clinical events. To diagnose MS, FG, BMI, HDL cholesterol, and triglycerides should be given weights progressively increasing with their levels. FG was the major determinant for the diagnosis of MS as well as of the risk of late-onset diabetes, although hypertension contributed slightly to the diagnosis of MS. It is worth noting that FG levels considered to be normal (eg, 90-100 mg/dL) were associated to 2- to 3-fold increase of the risk of getting diabetes. Although provocative, this fact is in line with the recent decision to lower the FG cutoff for the diagnosis of diabetes and to recommend to pay attention to high-normal glucose levels as possible indicators of a prediabetic status.³⁷ If we consider MS as an independent nosological entity associated with a dysmetabolic status (ie, IR), it is likely that we should consider a differential use of the variables, with those more strictly associated with IR receiving greater weight than those that cluster with the components of the MS but are less strictly associated with its pathophysiological mechanisms (eg, hypertension). According to the aforementioned, the NCEP/ATP III definition of MS seems more to point to creating an attractive framework that could catch the attention of physicians and move them to pay attention to the clustering of often overlooked cardiovascular risk factors to improve cardiovascular prevention strategies than as a means of uncovering a new nosological condition.²⁰ Such possibility is sustained by the results of new analyses of the Framingham study suggesting that no advantage can be gained in cardiovascular risk assessment by adding the unique risk factors of the MS-ATP to the usual Framingham risk factors in risk assessment. This finding was probably due to the fact that most of the cardiovascular risk associated with MS is already captured by age, blood pressure, total cholesterol, diabetes, and HDL cholesterol.³⁸

We found a change of the AUC from 0.587 with ATP III to 0.650 with MS-Score. Such change was statistically significant and allowed to increase the sensitivity for the diagnosis of MS by 15%. In absolute numbers, this means that among 1214 (16.3%) patients who had been considered as without MS according to ATP III criteria but with MS using our score, 216 (17.8%) of them developed diabetes mellitus during follow-up (see Table IV). On the other side, among the 896 (12%) patients diagnosed as affected by MS according to the ATP III criteria but not using our score, only 72 (8%) subjects developed diabetes. Because the study was closed after 3 years of follow-up, it is not unlikely that a relevant proportion of other MS-Score–positive subjects who had not yet become diabetic (currently flagged as false-positive) will develop diabetes mellitus. If so, it is likely that the rate of false-positive subjects of our MS-Score has been overestimated.

This allows a more self-assurance risk stratification of both patients classified as “false” low risk and those classified as false high risk according to MS-ATP.

We did not perform an oral glucose testing that was a recommended, validated, and useful tool to identify subjects at risk for late-onset diabetes. However, oral glucose testing is time consuming, costly, and relatively inconvenient, thereby, it is unlikely that it can have a widespread application in clinical practice. In addition, Stern³⁹ et al demonstrated that patients at high risk of diabetes could be identified using a clinical prediction model that was based on a complex mathematical formula not including a 2-hour oral glucose test.

The MS-Score was significantly associated with the risk of hard clinical events during follow-up, thus underlining the importance of MS also in a high-risk setting like the one after myocardial infarction and could constitute an improvement by providing an easy and ready-to-use tool to assess, in clinical practice, the probability of getting diabetes.

Regardless of diagnostic criteria used, there is full agreement that the first-line therapy for MS is lifestyle change with weight reduction as its therapeutic cornerstone. In a recent analysis on the GISSI-Prevenzione database, we have shown that subjects with a diagnosis of MS according to the NCEP/ATP III criteria and who had a clinically meaningful weight loss during the first 6 months after myocardial infarction had a statistically significant lower risk of becoming diabetic.⁴⁰ This finding confirms previous evidence suggesting that diet, increased physical exercise, and weight loss are able to reduce the risk of diabetes by >50%.⁴¹, ⁴², ⁴³ It has been reported that weight loss and control of classic risk factors could prevent >80% coronary events in patients with MS.⁴⁴ Although the proportion of preventable events could have been overestimated, it is undisputable that the more aggressive the control of all risk factors, the greater the overall benefit.

Some possible limitations of our study must be acknowledged. Because of the characteristics of the GISSI-Prevenzione patients (ie, post–myocardial infarction Mediterranean patients), the generalizability of the results to specific populations could be limited, and a validation of the MS-score should be considered. We used the most modern techniques of internal validation,³⁴ and the “optimism” score was quite promising; therefore, it is unlikely that an external validation would have given different results.

Back to Article Outline

Conclusion 

As type II diabetes is preventable condition, the MS-Score provides a pragmatic diagnostic tool to identify subjects at high risk of diabetes who could benefit more from specific lifestyle interventions.

Back to Article Outline

Appendix. 

Validation of the algorithm

As to cross validation, data were split into 10 equal-sized parts, then the model using 90% of data was fitted, and the resulting model was tested (calculating the ROC curve) on the remaining 10% of data. This process was repeated excluding progressively each of the selected parts. The difference between the ROC curve estimated using 100% of data and the mean value of the 10 measures of the ROC curves after cross-validation estimated the parameter “overoptimism,” that is, a measure of the extent to which the predictive accuracy of the model based on the full set of data was overoptimistic. With bootstrapping (sampling of individual subjects with replacement), 200 bootstrap samples with replacement each constituted by 7468 subjects were generated from the full set of data. The ROC curve and the β coefficients were calculated from each sample; the difference between mean values of these ROC curves, and mean values of the ROC curves obtained when the β coefficients for each bootstrap sample were applied to the original sample, giving the bootstrap estimate of overoptimism.

Application of the final model using original variables

The β coefficients given in Table II are used to compute the 3.5-year probability of late-onset diabetes: P = 1 − S(t)^B where S(t) is the survival function at time t, and B is the relative odds for diabetes. In GISSI-Prevenzione, S(t) at 3.5 years is equal to 0.89169. The relative odds for diabetes can be calculated as B = exponential (L − G). The quantities L and G for a given patient can be calculated as follows:

Example

A patient with the following characteristics: male, 49 years old, hypertensive, BMI 31, triglycerides 185 mg/dL, HDL cholesterol 36 mg/dL, and FG 100 mg/dL.

We can calculate the 3.5-year probability of diabetes predicted by the full model with the following steps:

An estimate of the probability of diabetes can be calculated by using the scoring system shown in Table II, where S = 0.89079, G = 1.75585, and β = 0.06640 for each point increase of the score.

The global score for the patient of the example is equal to 37, which corresponds, therefore:

Because the global score is ≥28, the patient is affected by MS and his probability of developing diabetes during the next 3.5 years is equal to 20.6%.

Back to Article Outline

References 

1. Haffner SM, D'Agostino R, Mykkanen L, et al. Insulin sensitivity in subjects with type 2 diabetes. Relationship to cardiovascular risk factors: the Insulin Resistance Atherosclerosis Study. Diabetes Care. 1999;22:562–568
   • View In Article
   • MEDLINE
   • CrossRef
2. Brand FN, Abbott RD, Kannel WB. Diabetes, intermittent claudication, and risk of cardiovascular events. The Framingham Study. Diabetes. 1989;38:504–509
   • View In Article
   • MEDLINE
3. Pan WH, Cedres LB, Liu K, et al. Relationship of clinical diabetes and asymptomatic hyperglycemia to risk of coronary heart disease mortality in men and women. Am J Epidemiol. 1986;123:504–516
   • View In Article
   • MEDLINE
4. Stamler J, Vaccaro O, Neaton JD, Wentworth D. Diabetes, other risk factors, and 12-yr cardiovascular mortality for men screened in the Multiple Risk Factor Intervention Trial. Diabetes Care. 1993;16:434–444
   • View In Article
   • MEDLINE
5. Balkau B, Shipley M, Jarrett RJ, et al. High blood glucose concentration is a risk factor for mortality in middle-aged nondiabetic men. 20-year follow-up in the Whitehall Study, the Paris Prospective Study, and the Helsinki Policemen Study. Diabetes Care. 1998;21:360–367
   • View In Article
   • MEDLINE
   • CrossRef
6. Goldstein BJ. Insulin resistance: from benign to type 2 diabetes mellitus. Rev Cardiovasc Med. 2003;4(Suppl 6):S3–S10
   • View In Article
7. Warram JH, Martin BC, Krolewski AS, et al. Slow glucose removal rate and hyperinsulinemia precede the development of type II diabetes in the offspring of diabetic parents. Ann Intern Med. 1990;113:909–915
   • View In Article
   • MEDLINE
8. Charles MA, Fontbonne A, Thibult N, et al. Risk factors for NIDDM in white population. Paris prospective study. Diabetes. 1991;40:796–799
   • View In Article
   • MEDLINE
9. Sicree RA, Zimmet PZ, King HOM, et al. Plasma insulin response among Nauruans: prediction of deterioration in glucose tolerance over 6 years. Diabetes. 1987;36:179–186
   • View In Article
   • MEDLINE
10. Lillioja S, Mott DM, Spraul M, et al. Insulin resistance and insulin secretory dysfunction as precursors of non-insulin dependent diabetes mellitus. Prospective studies of Pima Indians. N Engl J Med. 1993;329:1988–1992
    • View In Article
    • MEDLINE
    • CrossRef
11. Hunter SJ, Garvey WT. Insulin action and insulin resistance: diseases involving defects in insulin receptors, signal transduction, and the glucose transport effector system. Am J Med. 1998;105:331–345
    • View In Article
    • Full Text
    • Full-Text PDF (597 KB)
    • CrossRef
12. Cefalu WT. Insulin resistance: cellular and clinical concepts. Exp Biol Med. 2001;226:13–26
    • View In Article
13. Liese AD, Mayer-Davis EJ, Haffner SM. Development of the multiple metabolic syndrome: an epidemiologic perspective. Epidemiol Rev. 1998;20:157–172
    • View In Article
    • MEDLINE
14. Stern MP, Williams K, Haffner SM. Do we need the oral glucose tolerance test to identify future cases of type 2 diabetes?. Diabetes Care. 2003;26:940–941
    • View In Article
    • MEDLINE
    • CrossRef
15. National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) . Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) final report. Circulation. 2002;106:3143–3421
    • View In Article
16. Alberti KG, Zimmet PZ. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. Diabet Med. 1998;15:539–553
    • View In Article
    • MEDLINE
    • CrossRef
17. Einhorn D, Reaven GM, Cobin RH, et al. American College of Endocrinology position statement on the insulin resistance syndrome. Endocr Pract. 2003;9:237–252
    • View In Article
    • MEDLINE
18. Laaksonen DE, Lakka HM, Niskanen LK, et al. Metabolic syndrome and development of diabetes mellitus: application and validation of recently suggested definitions of the metabolic syndrome in a prospective cohort study. Am J Epidemiol. 2002;156:1070–1077
    • View In Article
    • MEDLINE
    • CrossRef
19. Lorenzo C, Okoloise M, Williams K, et al. The metabolic syndrome as predictor of type 2 diabetes: the San Antonio heart study. Diabetes Care. 2003;26:3153–3159
    • View In Article
    • MEDLINE
    • CrossRef
20. Mitka M. Metabolic syndrome recasts old cardiac, diabetes risk factors as a “new” entity. JAMA. 2004;291:2062–2063
    • View In Article
    • CrossRef
21. GISSI-Prevenzione Investigators. Dietary supplementation with n-3 polyunsaturated fatty acids and vitamin E after myocardial infarction: results of the GISSI-Prevenzione trial. Lancet. 1999;354:447–455
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (104 KB)
    • CrossRef
22. Hosmer DW, Lemeshow S. Applied logistic regression. New York (NY): John Wiley & Sons, Inc; 1989;
    • View In Article
23. Bruzzi P, Green SB, Byar DP, et al. Estimating the population attributable risk for multiple risk factors using case-control data. Am J Epidemiol. 1985;122:904–914
    • View In Article
    • MEDLINE
24. Walter SD. The distribution of Levin's measure of attributable risk. Biometrika. 1975;62:371–374
    • View In Article
    • CrossRef
25. Engel LS, Chow WH, Vaughan TL, et al. Population attributable risks of esophageal and gastric cancers. J Natl Cancer Inst. 2003;95:1404–1413
    • View In Article
    • CrossRef
26. Lowry R. Clinical Calculator 1 from an observed sample: estimates of population prevalence, sensitivity, specificity, predictive values, and likelihood ratios. Available from http://faculty.vassar.edu/lowry/clin1.html.2001;
    • View In Article
27. Silman AJ. Epidemiological studies: a practical guide. Cambridge (UK): Cambridge University Press; 1995;
    • View In Article
28. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristics curves: a nonparametric approach. Biometrics. 1988;44:837–845
    • View In Article
    • MEDLINE
    • CrossRef
29. Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of prognostic information. Ann Intern Med. 1999;130:515–524
    • View In Article
    • MEDLINE
30. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143:29–36
    • View In Article
    • MEDLINE
31. Neter J, Wasserman W. Multiple regression. In: Applied linear statistical models. Homewood (Ill): Irwin; 1974;p. 214–272
    • View In Article
32. Cox DR. Regression models and life tables. J R Stat Soc B. 1972;34:187–220
    • View In Article
33. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issue in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387
    • View In Article
    • MEDLINE
    • CrossRef
34. Efron B, Tibshirani R. An introduction to the Bootstrap. New York (NY): Chapman and Hall; 1993;
    • View In Article
35. SAS Institute Inc., Cary, NC, U.S.A. SAS Institute Inc. . SAS/STAT Users Guide, Version 8. Cary (NC): SAS Institute Inc; 1999;
    • View In Article
36. Grundy SM, Brewer HB, Cleeman JI, et al. Definition of metabolic syndrome: report of the National Heart, Lung, and Blood Institute/American Heart Association conference on scientific issues related to definition. Circulation. 2004;109:433–438
    • View In Article
    • CrossRef
37. Genuth S, Alberti KG, Bennett P, et al. Follow-up report on the diagnosis of diabetes mellitus. Diabetes Care. 2003;26:3160–3167
    • View In Article
    • MEDLINE
    • CrossRef
38. Wilson PW, D'Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97:1837–1847
    • View In Article
    • MEDLINE
39. Stern MP, Williams K, Haffner SM. Identification of persons at high risk for type 2 diabetes mellitus: do we need the oral glucose tolerance test?. Ann Intern Med. 2002;136:575–581
    • View In Article
40. Levantesi G, Macchia A, Marfisi RM, et al. Metabolic syndrome and risk of cardiovascular events after myocardial infarction. J Am Coll Cardiol. 2005;46:277–283
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (153 KB)
    • CrossRef
41. Knowler WC, Barrett-Connor E, Fowler SE, et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N Engl J Med. 2002;346:393–403
    • View In Article
    • CrossRef
42. Hu FB, Manson JE, Stampfer MJ, et al. Diet, lifestyle, and the risk of type 2 diabetes mellitus in women. N Engl J Med. 2001;345:790–797
    • View In Article
    • MEDLINE
    • CrossRef
43. Tuomilehto J, Lindstrom J, Eriksson JG, et al. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. N Engl J Med. 2001;344:1343–1350
    • View In Article
    • MEDLINE
    • CrossRef
44. Watkins LL, Sherwood A, Feinglos M, et al. Effects of exercise and weight loss on cardiac risk factors associated with syndrome X. Arch Intern Med. 2003;163:1889–1895
    • View In Article
    • MEDLINE
    • CrossRef