Different Percentile Regression of Blood Glucose Among Adolescents Aged 12−20 — United States, 1999−2018 (2025)

As the burden of disease increases, there is a growing awareness of the dangers of elevated blood glucose (1). The incidence of diabetes among adolescents is increasing year over year, and the average annual growth rate of diabetes among children and adolescents in the United States is 2.3%. Elevated blood glucose has many hazards if not treated in time and regularly as it will not only affect the growth and development of children but also cause complications such as diabetic ketoacidosis and cataracts. In severe cases, it can lead to blindness and psychological disorders for children (2). Obesity is thought to be the catalyst for diabetes which can lead to an increase in blood glucose in individuals. A close relationship has been observed in recent decades between rising rates of obesity and an increased incidence of type-2 diabetes among adolescents (3-4). In the 2020 diabetes medical standard published by the American Diabetes Association (ADA), Body Mass Index (BMI) is the primary risk factor in the screening and diagnosis of adolescent type-2 diabetes (5). This paper used the blood glucose data of 7,786 adolescents aged 12–20 years from 1999 to 2018 of the National Health and Nutrition Examination Survey (NHANES). Quantile regression was used to analyze the blood glucose of adolescents to explore the influence of gender on blood glucose under different quantiles. At the same time, the results of model correction of BMI were also analyzed. The blood glucose of the female 15-year-old group was close to the normal distribution, and the blood glucose of adolescents of men and women of other ages did meet the normal distribution and was not suitable for typical linear regression analysis. The regression coefficients of gender factors in different ages and quantiles were obtained through the two models and revealed that male blood glucose was higher than female blood glucose and all age groups were statistically significant (P<0.05). A picture of regression coefficients based on different scales showed a downward trend in regression coefficient with the increase of age. There is an urgent need to set up standards for adolescent blood glucose according to various ages and genders.

The NHANES is a sustained survey project implemented by the US CDC since 1999, which uses a much more complex stage probability sampling to sample the American population (6-7). A cycle takes two years and is designed to assess the health and nutritional status of adults and children in the United States. The data of the project is free to the public, and no additional ethics application is required. In this study, the blood glucose data of 12–20 years old adolescents from 1999 to 2018 in NHANES was used for analysis, and 7,786 individuals had suitable data.

The statistical analysis was carried out with SAS software package (version 9.4, 100 SAS Campus Drive Cary, NC 27513). Quantile regression does not require the distribution of data but requires the minimization of residual error. Different estimators of regression coefficients under different quantiles reflect that explanatory variables have different effects on different levels of explanatory variables. Quantile regression integrates the concept of the quantile into ordinary linear regression. However, the conclusion no longer only reflects the central position but can reflect the whole distribution situation. Model one was a single factor quantile regression analysis without any controlling factors, and model two was a multivariate quantile regression analysis after controlling BMI.

According to ADA’s medical standards for diabetes in 2020, the normal range of adult blood glucose is 70–100 mg/dL. In this dataset, the blood glucose level in the 1st percentile was 72 mg/dL, the 85th percentile was 100 mg/dL, and their middle position was in the 42^nd percentile. Therefore, the regression equation was established at the five percentiles of P₁, P₄₂, P₅₀, P₈₅, and P₉₀, and the partial regression coefficients under different quantile regression were recorded. The inspection level α was set to 0.05.

The analysis showed that for adolescents aged 12–20 years, the distribution of blood glucose levels at different ages varied (Table 1). By analyzing the regression coefficients of gender factors at different ages and different quantiles, gender factors in each age group of 12–20 years old were found to be statistically significant except in the P₁, and the blood glucose level were higher in males than in females. After controlling for BMI, Model two found that the decrease in blood glucose levels was inconsistent with Model 1, such that for the 12-year-old age group, the regression coefficients for gender of Model 1 and Model 2 at P₄₂ were –2.7 and –3.1, respectively; at P ₅₀, the gender factor was –2.8 and –3.07, respectively; and at P ₈₅, the gender factor was –4.0 and –4.1, respectively ( Table 2). Other age groups also show the same phenomenon.