Figure 2: Smoothed plot for odds ratios (ORs) of the diabetes risk according to age at enrollment. Subjects were divided into four phenotypes according to overweight/obese phenotype and metabolic health status. The ORs were estimated by using the restricted cubic spline logistic regression models with knots placed at the 5th, 50th, and 95th percentiles of age at enrollment. The models were adjusted for the probability of sampling using weights taking into consideration clustering effects arising from the same census block and household. The linear correlation between age at enrollment and the risk of diabetes in each phenotype was significant (, 0.0001, 0.04, and <0.0001 for phenotypes with metabolically healthy normal weight, metabolically healthy overweight/obesity, metabolically unhealthy normal weight, and metabolically unhealthy overweight/obesity, resp.).