Assessing Risk for Development of Diabetes in Young Adults

INTRODUCTION

Considerable evidence has been presented on the increased prevalence of diabetes in the United States,¹ This prevalence has become so large that diabetes has been termed an epidemic.^1,2 In particular, diabetes is increasingly diagnosed among adolescents and younger adults.^3,4 One factor thought to be driving the diabetes epidemic is the increase in obesity.^5–7

Prediction of chronic conditions that have a definable onset in adults can help to guide interventions and health policy development. Prediction is an important issue, given that diabetes leads to considerable morbidity and mortality, which can be mitigated through early recognition and treatment.⁸ Major risk factors for diabetes have been identified and are currently used by the American Diabetes Association to guide screening strategies. Risk scores for diabetes fall into 2 primary categories that are conceptually distinct. Although risk scores are usually thought to quantify an individual’s risk of developing disease, as with the Framingham Risk Score for coronary heart disease, most self-identified diabetes risk scores do not assess the risk of developing disease; rather, they assess the likelihood of having undiagnosed diabetes.^9–14 There are few measures that assess the risk of developing diabetes.^15,16

Some risk scores for the likelihood of having undiagnosed diabetes have been tested in populations other than the ones in which they were created and have unfortunately not worked as well.^17,18 Considering the importance of identifying individuals at risk for developing diabetes, a strategy for assessing risk of developing diabetes in young adults has many benefits, including targeted interventions for young adults at high risk. Thus, the purpose of this study was to evaluate how well a risk score for developing diabetes that was created with a middle-aged population performs in a cohort of young adults.

METHODS

This study is based on an analysis of the Coronary Artery Risk Development in Young Adults (CARDIA), a population-based observational study of participants aged 18 to 30 years recruited in 1985–1986. Participants were recruited in 4 communities: Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; and Oakland, California. Recruitment was stratified by race (black and white), age (18 to 24 years, and 25 to 30 years), and education (less than high school, and high school or more). Second (1987–1988), third (1990–1991), fourth (1992–1993), fifth (1995–1996), and sixth examinations (2000–2001) have been completed in the cohort. The public use data set used for this study, however, only includes information from the first 5 examinations.

For the progression to diabetes analyses, all individuals had no indication of diabetes at baseline. This cohort was comprised of 2,543 persons. A total of 100 persons out of 2,543 developed diabetes within the 10 years.

RESULTS

Table 1⇓ displays the characteristics of the cohort. Only 2.1% of this young adult cohort was classified at baseline as being at high risk for developing diabetes, whereas 3.9% developed diabetes within 10 years. Further, 32.9% of the cohort was overweight or obese at baseline (BMI ≥25). The proportion of individuals with a family history of diabetes had a moderate increase when the definition of family history was changed from parents to parents and siblings.

The area under the ROC curve for the diabetes risk score in this young adult cohort is not optimal at .70 compared with .78 found in the middle-aged cohort in the ARIC study (Table 2⇓). A diabetes risk score of 4 or greater had a sensitivity of 15.0% and a specificity of 98.4% in the CARDIA participants.

The area under the ROC curve does not increase significantly when family history is added to the diabetes risk score. Moreover, the 3-category BMI variable is not significantly different from the multivariable diabetes risk score in the area under the ROC curve, nor is BMI plus family history of diabetes significantly better than BMI alone (P = .08).

Table 3⇓ shows the analyses within racial groups. Blacks (.72; 95% CI, .69–.74) and whites (.68; 95% CI, .66–.71) do not significantly differ in the area under the ROC curve for the risk score, as indicated by the overlapping 95% confidence intervals. The area under the ROC curve for BMI is significantly larger for blacks (.69; 95% CI, .66–.72) than for whites (.63; 95% CI, .60–.65). Similarly, the area under the ROC curve for BMI plus family history is significantly larger for blacks (.73; 95% CI, .70–.76) than for whites (.64; 95% CI, .61–.66).

DISCUSSION

As the prevalence of diabetes rises, and more young adults and adolescents develop diabetes, it is crucial from a clinical and public health perspective to be able to identify high-risk populations. The findings of this study indicate that a risk score for the development of diabetes created from a middle-aged population is a less successful predictor of the development of diabetes in a younger population.

We chose to assess the diabetes risk score developed in the ARIC cohort because it is one of the few measures designed to assess the risk of developing diabetes rather than the likelihood of having undiagnosed diabetes (eg, Cambridge Risk Score). An alternative measure to the ARIC diabetes risk score was considered for evaluation, but it included in the model “history of high blood glucose,” which was defined as “Have you ever been told by a health-care professional that you have diabetes or latent diabetes?”¹⁶ Including a previous diagnosis of diabetes as a predictor of the development of diabetes did not seem to be a logical strategy for identifying individuals at high risk for developing diabetes. Thus, we believed that the Lindstrom score was not as useful for evaluation as the ARIC diabetes risk score.

Multivariable risk scores that are diagnostically helpful should be clinically less burdensome in the age of personal digital assistants and electronic health records and should therefore allow the clinician to go beyond assessing risk factors singly for development of disease. In this case, the multivariable diabetes risk score does not predict the development of diabetes any better than simply using the BMI. Because the diabetes risk score includes BMI in addition to 5 other variables, it would be expected to perform better than BMI alone. With an area under the ROC curve of .67, however, BMI as a predictor of the development of diabetes in young adults is not optimal. Further, the addition of family history to either BMI or the diabetes risk score did not significantly improve prediction of the development of diabetes. These findings suggest that more work is needed to create an effective strategy for identifying young adults at high risk for developing diabetes.

Recent evidence has suggested that risk assessment strategies may need to differ depending on which racial or ethnic population is being evaluated.¹⁸ The results reported in this study indicate that although the diabetes risk score did not differ significantly between young black and white adults in the prediction of diabetes, BMI and BMI plus family history did differ between the 2 groups: BMI plus family history was a significantly more predictive strategy for identifying risk for the development of diabetes among blacks than among whites. These racial differences in the relationship of BMI and the development of diabetes may be due to the interaction of race and diet, as Pereira et al²² found that fast-food habits varied by race and sex and were related to insulin resistance in the CARDIA study. This finding indicates the need to be more aware of racial and ethnic differences in diabetes risk and the need to include that awareness in the development of diabetes risk assessment strategies. Further evaluation of the novel factors, including biomarkers, underlying these differences is also necessary.

There are several limitations to this study. First, the biomarker diagnosis of diabetes in the CARDIA data is based on a single fasting glucose test. This strategy, although common in epidemiological studies, could potentially underestimate the prevalence of diabetes associated with isolated postchallenge hyperglycemia, which occurs more commonly in women and lean populations. It could also overestimate diabetes prevalence, because a clinical diagnosis of diabetes in asymptomatic persons requires 2 abnormal fasting glucose levels. Second, racial differences in the predictive utility of the risk assessment strategies suggest that evaluating the risk score and other markers may be enhanced by having a diverse sample of ethnic groups. The CARDIA study is limited to blacks and whites and thus does not allow for evaluation with other racial or ethnic groups. Third, not only is it inherently difficult to improve on conventional risk factors when developing a scoring system as a prognostic tool, as shown by Wang et al²³ and discussed by Ware,²⁴ it is also difficult to improve on conventional risk factors when developing a prognostic tool. Hence, we compared the ARIC metabolic syndrome (augmented) model with BMI alone and BMI plus family history of diabetes to determine whether multivariate diabetes risk factors performed better than more general risk factors.

In conclusion, an established risk score for the development of diabetes among middle-aged persons had limited utility in a younger population. The diabetes risk score had no advantage compared with BMI alone. Neither BMI nor the risk score, however, had optimal predictive ability, suggesting that future research needs to focus on identifying novel factors that may improve the risk stratification for diabetes development among young adults.