213-LB: Developing a Prognostic Model to Assess Dysglycemia Risk in Canadians Aged 18-39

Abstract

In Canada, the prevalence of diabetes has seen the greatest relative increase in young adulthood, where the disorder is severely pathological compared to later-onset. Still, few prognostic models have been developed to screen young adults for dysglycemia risk and boost early identification and intervention. We sought to establish predictors of dysglycemia risk among young Canadian adults (aged 18-39) and evaluate their utility in identifying high-risk individuals. The Canadian Diabetes Risk Questionnaire (CANRISK) study collected questionnaire, anthropometric, and oral glucose tolerance test (OGTT) data from a large, multiethnic convenience sample of Canadians over two phases. Young adults with diagnosed diabetes, missing OGTT data, or pregnant were excluded. Potential factors that modestly predicted (p<0.20) dysglycemia status (FPG≥6.1mmol/L or 2h-PG≥7.8mmol/L) were entered into a lenient stepwise function, producing a young adult-specific model; risk scores were developed from adjusted odds ratios. Discriminatory ability was assessed by optimism-corrected area under the curve (AUC) via bootstrapping and goodness-of-fit by Hosmer-Lemeshow (H-L) test and calibration plot. More than half of the 3334 participants were female (62.4%), non-white (79.2%), less than 25kg/m2 (50.7%), and reported a family history of diabetes (55.4%); based on OGTT results, 7.3% were dysglycemic. The young adult-specific model displayed an adjusted AUC of 72.9%, and reasonable goodness-of-fit (H-L p=0.49). Model performance was similar when run sex-specifically (males: unadjusted AUC of 72.1%, H-L p=0.67; females: 73.6%, p=0.67). Employing a cut-point of 22, the tool displayed high sensitivity (78.8%) but low specificity (54.0%). Only 3% of those identified as low risk by the tool were misclassified. This young adult-specific risk score shows promise to identify high-risk individuals in a multiethnic Canadian sample. Additional studies are needed to assess its generalizability to new datasets. Disclosure S.A. Srugo: None. Y. Jiang: None. H.I. Morrison: None. M.M. deGroh: None.