Development of a Weighted Cardiometabolic Disease Staging (CMDS) System for the Prediction of Future Diabetes

Abstract

Context:

Metabolic syndrome traits are important risk factors for diabetes; however, each trait has different predictive power for future diabetes. Additionally, the impact of insulin resistance on metabolic profile can differ by gender and racial group, suggesting that gender-race specific prediction algorithms for diabetes may be warranted.

Objective:

To develop a quantitative scoring system based on weighting of risk components in the cardiometabolic disease staging (CMDS) system for the prediction of future diabetes.

Design, Setting, and Participants:

We derived the CMDS score in 2857 participants with valid follow-up information on incident diabetes from the Coronary Artery Risk Development in Young Adults study and validated it in 6425 older participants from the Atherosclerosis Risk in Communities study. We assigned a simple integer value for each CMDS risk factor component.

Main Outcome Measures:

Incident diabetes.

Results:

Fasting glucose, 2-hour glucose, waist circumference, and blood pressure components contributed similarly for the prediction of future diabetes (CMDS scores, 23, 21, 26, and 20, respectively). The area under the receiver operating characteristic curve was 0.7158 for the CMDS scoring system, whereas it was 0.7053 for the Framingham diabetes score. The CMDS components performed differently for prediction of future diabetes in Black and White men and women. The components with the highest predictive power for diabetes were waist circumference in Black men, 2-hour glucose in Black women, and fasting glucose in both White men and White women.

Conclusions:

The weighted CMDS score has high model discrimination power for diabetes and can be used clinically to identify patients for weight loss therapy based on differential risk for future diabetes.

Obesity is associated with elevated risk for diabetes and cardiovascular diseases (1). Recent approval of new weight loss medications (2, 3) has enabled a pharmaceutical approach for obesity therapy. Although an average weight loss of approximately 10% will often not suffice to meet the cosmetic goals of patients or even bring many patients below the body mass index (BMI) threshold for obesity, it is sufficient to exert powerful benefits regarding weight-related complications (4, 5). However, with nearly 70% of US adults being overweight or obese (6), and due to concerns of safety and cost, it is impracticable to treat all overweight and obese subjects with medical or surgical therapy. The patients who will benefit most from treatment with medications or surgery have obesity-related comorbidities that can be categorized into two general classes: insulin resistance and cardiometabolic disease, and the mechanical consequences of excess body weight (5).

BMI is widely used as a population-based tool to assess adiposity (7) and is used as a base in clinical guidelines for obesity management (8). The problem is that BMI is a poor indicator of cardiovascular disease and overall mortality risk, and the risks of cardiometabolic disease largely exist independent of BMI as a general measure of adiposity (9–11). From this perspective, baseline BMI can be less important than the existence and severity of complications at baseline in identifying patients who will benefit most from weight loss (11). To better categorize the risk levels of people with excess body weight, we have established five stages of cardiometabolic disease risk—the cardiometabolic disease staging (CMDS) system (12) based on Adult Treatment Panel III metabolic syndrome risk factors (13, 14)—to guide decision making for selection of treatment modality and intensity in the management of obesity.

Metabolic syndrome traits, such as waist circumference, blood glucose, blood pressure, and lipid profile, are all important risk factors for type 2 diabetes mellitus (T2DM) (15); however, each trait may have different prediction power for future diabetes (16, 17). Additionally, the impact of insulin resistance on the metabolic profile can differ by gender and racial group (18–21), suggesting that gender- and race-specific prediction algorithms for diabetes may be warranted.

To enhance the application of the CMDS system in clinical settings, we sought to develop a weighted scoring system based on risk factor components in the CMDS system for the prediction of future diabetes by separate identification and weighting of those risk components. We quantified the relative contribution of metabolic syndrome traits to overall diabetes risk.

Materials and Methods

CMDS components

CMDS system components include the following: 1) fasting blood glucose ≥ 100 mg/dL; 2) 2-hour glucose from oral glucose tolerance test (OGTT) ≥140 mg/dL; 3) waist circumference ≥102 cm in men, ≥ 88 cm in women; 4) systolic blood pressure ≥ 130 mm Hg and/or diastolic blood pressure ≥85 mm Hg, or on antihypertensive medication; 5) high-density lipoprotein (HDL) cholesterol <40 mg/dL in men, <50 mg/dL in women; and 6) fasting triglycerides ≥150 mg/dL or on antihyperlipidemia medication. We only assessed the relative contributions of those risk factors in our score algorithms. Because the primary goal of this study is to enhance the application of the CMDS system in clinical settings and to develop a weighted scoring system based on risk factor components in the CMDS system for the prediction of future diabetes, we did not include other risk factors for diabetes in our score system, such as age, family history, dietary factors, and physical activity. We derived a weighted CMDS score for predicting incident diabetes based on the six components above using data from the Coronary Artery Risk Development in Young Adults (CARDIA) study and validated it in participants from the Atherosclerosis Risk in Communities (ARIC) study. Analyses were also performed after elimination of the 2-hour glucose risk factor. Obesity was defined according to BMI (weight/height²): obese, ≥30 kg/m²; and overweight, 25–29.9 kg/m².

CARDIA study

The CARDIA study (22) began in 1985–1986 and is a large, ongoing population study of 5115 young Black and White adults aged 18–30 years at intake from four sites in the United States, including Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; and Oakland, California. Participants received a physical examination at study entry in 1986 (year 0) and attended follow-up examinations during 1987–1988 (year 2), 1990–1991 (year 5), 1992–1993 (year 7), 1995–1996 (year 10), 2000–2001 (year 15), 2005–2006 (year 20), and 2009–2010 (year 25). Site institutional review committee approval and informed consent were obtained.

Standardized blood pressures were obtained by sphygmomanometer. Glucose and lipids were assayed using the fasting blood sample. OGTTs were performed at year 10, year 20, and year 25 examinations. Glucose and lipids were measured at central laboratories. Incident diabetes was defined as fasting plasma glucose ≥126 mg/dL and/or 2-hour glucose ≥200 mg/dL (23), a self-reported diagnosis of diabetes, or taking antidiabetic medication. Baseline for this study was set at the year 10 examination because 2-hour glucose values were not available in examinations before year 10.

ARIC study

The ARIC study is a large, ongoing prospective cohort study that began in 1987 (24). ARIC included 15 792 Black and White men and women, aged 45–64 years at entry, from four US communities: Jackson, Mississippi; Forsyth County, North Carolina; Minneapolis, Minnesota; and Washington County, Maryland. The first examination of participants (visit 1) was conducted during 1987–1989, with three follow-up visits occurring 3 years apart. OGTT was first performed on participants attending the visit 4 examination (1996–1998) after overnight fasting. Therefore, we included participants attending the visit 4 examination as the baseline cohort.

Analyses of fasting serum and plasma specimens were performed at central laboratories. Institutional Review Boards at each clinical site approved the study, and written informed consent was obtained from all participants. For incident diabetes, we used standard outcome definitions in the ARIC study. Incident diabetes was defined by criterion-based abstractions of hospital records (obtained through surveillance of hospital admissions), and from responses to annual postal or telephone questionnaires distributed during follow-up years through 2009 as self-reported diagnosis of diabetes or on antidiabetic medication (25).

Statistical analysis

We derived the CMDS score based on verified incident diabetes cases from the CARDIA study and validated it in participants from the ARIC study. We fitted Cox models and derived CMDS scores for diabetes accordingly. The contribution of each CMDS component to the overall score was estimated from the β coefficient of the Cox model specific to each gender-race, as described previously (16, 26). We rounded each point score assignment to the nearest integer and set the maximum of the sum of scores for all six CMDS components to 100. We also estimated the 15-year risk for diabetes for the CMDS score. Considering that OGTTs are not routinely performed in clinical settings, we also developed a modified CMDS scoring system by removing the 2-hour glucose component. To account for the gender and racial differences in the predictive power of diabetes by those risk factors, we also developed gender-race specific CMDS scores.

Validation of the CMDS scoring system was based on the 12-year diabetes incidence rates estimated from Kaplan-Meier survival curves. We compared the prediction power of the CMDS scoring system with the Framingham diabetes risk score by constructing receiver operating characteristic (ROC) curves (27) and calculating the area under the ROC curves (AUCs) on the basis of incident diabetes from the ARIC study. The Framingham score (www.framinghamheartstudy.org/risk-functions/diabetes/index.php) included the following risk factors: parental history of diabetes mellitus, overweight or obese based on BMI, fasting glucose, blood pressure, HDL-cholesterol, and fasting triglycerides (16). The proportional hazards assumption for Cox models was assessed using Schoenfeld residuals. Statistical analyses were carried out with SAS for Windows version 9.4 (SAS Institute). A two-sided P < .05 was determined to be statistically significant.

Results

Study population

Baseline characteristics of the study subjects are presented in Supplemental Table 1. Flowcharts of the numbers of participants included in the final analyses in the CARDIA study and in the ARIC study are presented in Supplemental Figures 1 and 2. Pregnant participants and participants with diabetes or coronary heart diseases at baseline were excluded from analysis. In the CARDIA study, 2857 subjects (mean age, 35 y) received examinations at year 10 and had valid information for the six CMDS components and follow-up information to assess incident diabetes. There were 6425 participants (mean age, 64.3 y) attending ARIC visit 4, and they had valid follow-up information for incident diabetes. In both the ARIC and CARDIA studies, Black women had the highest prevalence of obesity. During a median follow-up of 15 years, there were 303 cases of diabetes ascertained among 2857 subjects from the CARDIA study; during a median follow-up of 12 years, there were 973 cases of diabetes in 6425 subjects from the ARIC study.

CMDS score

We developed a weighted CMDS score for diabetes based on six risk factors in 2857 subjects from the CARDIA study, including fasting glucose, 2-hour glucose, waist circumference, blood pressure, HDL-cholesterol, and triglycerides. The β coefficients from gender-race specific Cox models are shown in Supplemental Table 2. The score for each CMDS component was estimated from the corresponding β coefficient and is presented in Table 1. Fasting glucose, 2-hour glucose, waist circumference, and blood pressure components had similar contributions for the prediction of future diabetes (CMDS scores, 23, 21, 26, and 20, respectively). The estimated 15-year diabetes risk for CMDS scores is shown in Table 2.

Table 1.

Scoring Sheet for CMDS Score and Modified CMDS Score

Items	CMDS Score	Modified CMDS Score
Fasting glucose ≥100 mg/dL	23	30
2-Hour glucose ≥140 mg/dL	21
Waist circumference (≥102 cm in men; ≥88 cm in women)	26	33
Blood pressure (systolic ≥130 mm Hg and/or diastolic ≥85 mm Hg) or on antihypertensive medication	20	22
HDL-cholesterol (<40 mg/dL in men; <50 mg/dL in women)	5	7
Fasting triglycerides ≥150 mg/dL or on antihyperlipidemia medication	5	8
Total (range 0 to 100 possible) =

Table 2.

Predicted 15-Year Diabetes Risk According to CMDS Score and Modified CMDS Score in Participants From CARDIA Study

Score	15-Year Risk of Diabetes, %
	CMDS Score	Modified CMDS Score
≤10	≤7	≤7
11–20	≤11	≤10
21–30	≤16	≤13
31–40	≤23	≤18
41–50	≤32	≤24
51–60	≤44	≤33
61–70	≤59	≤43
>70	>59	>43

We removed the 2-hour glucose component as a risk factor to develop a modified CMDS score system for use when 2-hour glucose was not available in clinical settings (Supplemental Table 2). Fasting glucose and waist circumference represented the two highest scores, followed by blood pressure (Table 1). The estimated 15-year diabetes risk for modified CMDS scores is shown in Table 2.

Gender-race specific CMDS score

The risk factor components performed differently for prediction of future diabetes in Black and White men and women (Supplemental Table 3). Among Black men, the score for the waist circumference component had the highest score of 31; blood pressure, HDL-cholesterol, and triglycerides components had similar scores (18, 16, and 18, respectively; Table 3). Among White men, the fasting glucose and blood pressure components had the highest scores (25 and 22, respectively). Among both Black and White women, fasting glucose, 2-hour glucose, and waist circumference had similarly high scores. The estimated 15-year diabetes risk for the gender-race specific CMDS scores is shown in Table 4. Black men and Black women had similar estimated risks at the same CMDS score, and so did White men and White women. When the scores were ≤ 50, the White subjects had lower risks for future diabetes than the Black subjects at the same score. When the scores were greater than 60, the White and Black subjects had similar risks for future diabetes.

Table 3.

Scoring Sheet for Gender-Race Specific CMDS Score

Items	Points
	Black Men	Black Women	White Men	White Women
Fasting glucose ≥100 mg/dL	10	28	25	24
2-Hour glucose ≥140 mg/dL	7	30	16	20
Waist circumference (≥102 cm in men; ≥88 cm in women)	31	26	18	19
Blood pressure (systolic ≥130 mm Hg and/or diastolic ≥85 mm Hg) or on antihypertensive medication	18	15	22	17
HDL-cholesterol (<40 mg/dL in men; <50 mg/dL in women)	16	1	7	11
Fasting triglycerides ≥ 150 mg/dL or on antihyperlipidemia medication	18	−8	12	9
Total (range 0 to 100 possible) =

Table 4.

Predicted 15-Year Diabetes Risk According to Gender-Race Specific CMDS Score in Participants From CARDIA Study

CMDS Score	15-Year Risk of Diabetes, %
	Black Men	Black Women	White Men	White Women
≤10	≤11	≤11	≤3	≤5
11–20	≤15	≤15	≤6	≤7
21–30	≤20	≤20	≤10	≤11
31–40	≤27	≤27	≤16	≤18
41–50	≤36	≤36	≤25	≤26
51–60	≤47	≤47	≤38	≤39
61–70	≤59	≤59	≤55	≤54
>70	>59	>59	>55	>54

The modified gender-race specific CMDS score system with 2-hour glucose removed is presented in Supplemental Tables 4 and 5. Similar patterns were observed as in the gender-race specific CMDS score system that contained the 2-hour glucose risk factor component. The fasting glucose component was the strongest predictor for diabetes, except in Black men. Among Black men, the waist circumference component was the strongest predictor. We also presented the estimated 15-year diabetes risk for the modified gender-race specific CMDS scores in Supplemental Table 6.

Validation of CMDS score system

We validated the CMDS score in 6425 subjects from the ARIC study and constructed the ROC curves for the CMDS score, modified CMDS score, and Framingham diabetes score (Figure 1), as well as the gender-race specific CMDS score, modified gender-race specific CMDS score, and the original CMDS system (Supplemental Figure 3). The AUC for the CMDS scoring system was 0.7158, which was superior in performance when compared with the Framingham diabetes score with an AUC of 0.7053. The modified CMDS score with the 2-hour glucose component removed also differentiated the risks for incident diabetes well (AUC, 0.7013). The gender-race specific CMDS score system slightly increased the highest discrimination power for future diabetes (AUC, 0.7199). The original unweighted CMDS system had substantial discriminative power for predicting future diabetes (AUC, 0.6981) but did not perform as well as weighted CMDS.

Figure 1.

ROC curves for CMDS score, modified CMDS score, and Framingham score in the ARIC cohort. The CMDS score included the following risk factors: fasting glucose, 2-hour glucose, waist circumference, blood pressure, HDL-cholesterol, and fasting triglycerides. The modified CMDS score removed the 2-hour glucose component. The Framingham score included the following risk factors: parental history of diabetes mellitus, overweight or obese classification based on BMI, fasting glucose, blood pressure, HDL-cholesterol, and fasting triglycerides.

Diabetes risk increased exponentially across CMDS score tertiles in participants from both the ARIC study and the CARDIA study (Figure 2). In the ARIC study, compared to participants in the first (lowest) tertile of CMDS scores system, the hazard ratio for incident diabetes was 2.11 (95% confidence interval, 1.71–2.60) in the second tertile and 5.14 (95% confidence interval, 4.23–6.24) in the third tertile.

Figure 2.

Risk for incident diabetes according to CMDS score tertiles. The cumulative diabetes incidence was derived from Kaplan-Meier survival curves. Hazard ratio for incident diabetes was calculated for multivariable-adjusted Cox models, when controlling for age, gender, and race. Tertiles of the CMDS score in the ARIC study: 0–30, 30–56, 57+. Tertiles of the CMDS score in the CARDIA study: 0, 5–20, 21+.

Discussion

Using data from two large national cohorts, the CARDIA study and the ARIC study, we developed and validated a weighted CMDS scoring system for the prediction of future diabetes based on six CMDS risk factor components. Our CMDS scoring system was effective in quantitatively stratifying differential risks for incident diabetes among both young adult participants in the CARDIA study and older participants in the ARIC study. We modified the CMDS score by removing the 2-hour glucose component from the OGTT test for practical application in clinical settings when OGTT data were not available. The modified CMDS score also had a high model discrimination of future diabetes. Our CMDS scoring system and modified CMDS scoring system both adopted integer point scores that can sum up to a maximum of 100 and can be easily calculated.

This is also the first time that a gender-race specific diabetes prediction algorithm simplified through the use of integer point values has been developed. The integer score for each risk factor component quantifies the relative contribution of that risk factor for diabetes. Clearly, metabolic syndrome traits exhibited differences in predictive power across gender and racial subgroups in our study. For example, the waist circumference component was the strongest predictor for diabetes in Black men. The blood pressure component had higher predictive power in men than in women. Low HDL-cholesterol, on the other hand was a trivial predictor for diabetes in Black women but an important predictor for diabetes in Black men. Although other diabetes prediction algorithms included a blood pressure component (17, 26, 28), the underlying mechanism for the linkage between insulin resistance and elevated blood pressure remains to be fully elucidated. For example, it is unclear whether elevated blood pressure contributes to the development of insulin resistance and subsequently T2DM, or whether it just reflects the shared underlying humoral and pathophysiological changes (26). Low HDL-cholesterol is an important risk factor for T2DM (15) because it promotes cellular cholesterol efflux and has direct antioxidative and anti-inflammatory properties. However, again, the underlying mechanism for our observed differences in Black men and Black women requires further investigation.

The 2-hour glucose component was the most important contributor in the diabetes prediction model in Black women and the second most important contributor in White women. Removing the 2-hour glucose component from our CMDS scoring system increased the contribution of fasting glucose to diabetes risk but also decreased the discrimination power of the model for future diabetes. The proportion of subjects with prediabetes on the basis of 2-hour glucose alone is quite high and represents the clear majority of prediabetics among older adults (29–31). Thus, it is important to consider performing OGTTs when evaluating high-risk patients. Elevated 2-hour glucose has previously been used in several diabetes prediction algorithms (16, 32). Nevertheless, OGTTs are not routinely assessed in clinical practice. Therefore, we modified our CMDS scoring system by removing the 2-hour glucose component to facilitate application of this CMDS scoring system in a clinical setting for the guidance of treatment modality and intensity in patients with obesity. Using hemoglobin A1c (ie, values of 5.7–6.4%) to identify prediabetes (33) can also be problematic because hemoglobin A1c has high specificity but low sensitivity when identifying prediabetes diagnosed using fasting and 2-hour glucose values (34). Although the modified CMDS scoring system (not including the 2-hour glucose component) had decreased model discrimination power, it continued to effectively identify patients at high risk of future diabetes.

Clinical implications

Our CMDS scoring system and the modified score system without the 2-hour OGTT glucose component both have high model discrimination power for future diabetes and can differentiate diabetes risk well in young and older people. Those score systems also use simple integer point values and can be readily calculated in clinical settings using basic information readily available to health care professionals. This is consistent with the complications-centric algorithm for obesity management (35) as advocated by the American Association of Clinical Endocrinologists that targets more aggressive therapy to those patients who will benefit most from the intervention based on risk and severity of weight-related complications (36). Additionally, our score systems are the first to provide for gender-race specific quantification of diabetes risk. Metabolic risk traits perform differently for the prediction of future diabetes across gender-racial groups, calling for specific diabetes predication algorithms for those groups.

Strength and limitations

The main strength of this study is the development and validation of a weighted diabetes algorithm using data from two large prospective national cohorts, the CARDIA study and the ARIC study. Those cohorts included men and women, Blacks and Whites. CARDIA participants were young adults, and ARIC participants were older adults. This has enabled the generalization of our diabetes prediction algorithm. Additionally, incident diabetes in the CARDIA study was based on rigorous measures of fasting glucose and 2-hour OGTT glucose, such that the ascertained diabetes cases provide a solid basis for the development of our weighted CMDS scoring system.

A limitation in this study was that blood glucose measures were not available for ARIC participants after visit 4, and diabetes cases were based on hospital records and self-reports. In addition, ARIC and CARDIA mainly included Blacks and Whites; therefore, our CMDS score system needs to be validated in other racial and ethnic groups, such as Asians and Hispanics. Also, we only included risk factors for diabetes from the original CMDS system in our scoring algorithm. We did not assess the contribution of other risk factors for diabetes, such as age, parental history of diabetes, dietary factors, physical activity, or sleeping patterns, in the scoring system.

Conclusions

The weighted CMDS scoring system and modified CMDS scoring system for prediction of diabetes were derived in participants from the CARDIA study and validated in the older participants from the ARIC study. The CMDS score systems have great model discrimination power for diabetes and can effectively differentiate risks for future diabetes. The CMDS score systems use simple integer point values and can be readily adopted in clinical settings to aid the selection of treatment options for obesity management. Thus, targeting more aggressive therapy to those patients with high risk scores for diabetes will optimize the benefit-risk ratio and cost effectiveness of these interventions.