The baseline characteristics of the participants with and without incident diabetes at 10-year follow-up are presented in . Of the 492 participants included in our analysis, 86 (17.5%) were found to have diabetes at follow-up. Diabetes was ascertained on the basis of fasting or 2-hour postload glucose levels, or both, in 72 (18.8%) of 383 participants. For 14 (12.8%) of 109 participants without follow-up blood samples, diabetes was ascertained on the basis of self-reported clinical diagnosis only. Compared with participants who had follow-up blood samples, those without samples were younger (p = 0.009) but were not different in terms of sex or body mass index (each p ≥ 0.05). When only self-report was considered for ascertainment of incident diabetes among participants with follow-up blood samples, 53 (13.8%) of 383 had diabetes.
The incidence of diabetes increased with age (p < 0.001). It was 10.5% among participants aged 10–19 years, 15.1% among those aged 20–29 years, 27.3% among those 30–39 years, 43.3% among those 40–49 years and 18.9% among those aged 50 and older.
Compared with individuals who did not have diabetes at the 10-year follow-up, those who did have incident diabetes had a higher body mass index, percent body fat, waist circumference and waist-to-height ratio at baseline (each p < 0.001). They also had a lower baseline HDL cholesterol level (p = 0.02) and higher baseline LDL cholesterol, triglyceride, fasting plasma glucose, 2-hour postload glucose and fasting insulin levels, and higher baseline systolic and diastolic blood pressures (each p < 0.001). Participants with incident diabetes were also more likely than those without diabetes to have had hypertension, impaired glucose tolerance and metabolic syndrome at baseline (each p ≤ 0.001) ().
In the multiple logistic regression analysis, we found that the following clinical variables at baseline were significantly associated with an increased risk of incident diabetes in the age-and sex-adjusted models: high body mass index, percent body fat, waist circumference, waist-to-height ratio; high fasting plasma glucose, 2-hour postload glucose and fasting insulin levels; high systolic and diastolic blood pressures; high LDL cholesterol and triglyceride levels; a low HDL cholesterol level; and hypertension (each p ≤ 0.03) ().
Figure 2: Age-and sex-adjusted risk of type 2 diabetes associated with readily accessible clinical measurements.
When we adjusted the models further for individual components of metabolic syndrome, excluding the main effect, we found that the direction of the associations of clinical variables with incident diabetes remained the same (data not shown), except for certain variables that were no longer significantly associated with incident diabetes. These variables were systolic blood pressure (OR 1.18, 95% CI 0.98–1.41), HDL cholesterol (OR 0.51, 95% CI 0.18– 1.44), LDL cholesterol (OR 1.24, 95% CI 0.84– 1.84) and fasting insulin level (OR 1.05, 95% CI 0.99– 1.12). Active smoking was not independently associated with incident diabetes in model 1 () or model 2 (data not shown). There were no statistically significant sex interactions with clinical variables in predicting incident diabetes (all interactions p ≥ 0.05), except with baseline fasting plasma glucose level (p = 0.03) (data not shown). When we stratified data by sex, we found a significant association between fasting plasma glucose level and incident diabetes among males after adjustment for components of metabolic syndrome (OR 2.05, 95% CI 1.31– 3.19). This was not the case among females (OR 1.06, 95% CI 0.76– 1.48).
When we assessed the diagnostic performance of impaired glucose tolerance at baseline in predicting incident diabetes, we found that it had high specificity (91%) and high negative predictive value (85%), yet low sensitivity and low positive predictive value (). We found similar performance results for metabolic syndrome regardless of which set of defining criteria were used; however, we observed a slightly lower specificity and higher sensitivity for the syndrome as defined by the International Diabetes Federation criteria (). For the individual components of metabolic syndrome, we found that, in general, the specificity and negative predictive values were within a moderate to high range, and the sensitivity and positive predictive values were low (). As an exception, waist circumference as defined by the International Diabetes Federation had a high sensitivity and negative predictive value but a low specificity and positive predictive value ().
We found that metabolic syndrome at baseline was associated with incident diabetes after adjustment for age and sex regardless of whether we used the National Cholesterol Education Program definition (OR 2.03, 95% CI 1.10–3.75) or the International Diabetes Federation definition (OR 2.14, 95% CI 1.29– 3.55) (). There was no significant difference between the 2 definitions in their ability to discriminate participants with incident diabetes from those who did not have diabetes (p ≥ 0.05 for comparison of C statistics).
Figure 3: Age-and sex-adjusted risk of type 2 diabetes associated with impaired glucose tolerance and with metabolic syndrome. C statistics for metabolic syndrome, whether defined by the National Cholesterol Education Program criteria (0.668) or the International (more ...)
When we lowered the fasting plasma glucose cut-off to 5.6 mmol/L or greater in the National Cholesterol Education Program criteria, we found that the modified definition of metabolic syndrome was significantly associated with incident diabetes to the same degree as the original definition (p
= 0.54 for comparison of C statistics) and as the International Diabetes Federation definition (Appendices 1 and 2, available at www.cmaj.ca/cgi/content/full/180/6/617/DC2
). The age-and sex-adjusted model for impaired glucose tolerance (C statistic 0.681) was not better at detecting future diabetes than the metabolic syndrome models, whether defined by the National Cholesterol Education Program criteria (C statistic 0.668) or the International Diabetes Federation criteria (C statistic 0.682) (comparison of C statistics: p
= 0.57 for the National Cholesterol Education Program definition v. impaired glucose tolerance, and p
= 0.96 for the International Diabetes Foundation definition v. impaired glucose tolerance) (). In addition, the capability of individual components of the metabolic syndrome (i.e., waist circumference, triglyceride level, HDL cholesterol level, blood pressure or fasting plasma glucose level) did not differ significantly from the capability of impaired glucose tolerance in predicting incident diabetes (each p
≥ 0.05 for comparison of C statistics) (Appendices 3 and 4, available at www.cmaj.ca/cgi/content/full/180/6/617/DC2