Type 2 diabetes development among individuals with different weight status

Abstract

Diabetes affects more than 10% of adults worldwide. Although obesity is a well-documented risk factor for type 2 diabetes, many nonobese individuals develop diabetes, and only a few studies have assessed weight-related disparities in diabetes epidemiology, risk factors, and cardiometabolic risks.

This thesis reports studies that estimated weight-specific time trends in diabetes prevalence and incidence, developed screening tools for nonobese diabetes based on risk factor exploration using machine-learning approaches, and examined weight-related disparities in cardiometabolic risks and glucose trajectories before diabetes diagnosis.

Data from five population-based surveys, including the National Health and Nutrition Examination Survey (NHANES), Korea NHANES (KNHANES), China Health and Nutrition Survey (CHNS), China Health and Retirement Longitudinal Study (CHARLS), and English Longitudinal Study of Ageing (ELSA), were used.

First, we conducted a time-trend analysis and found that diabetes prevalence has significantly increased among obese and nonobese Chinese (CHNS 1997–2015) and Korean (KNHANES 2001–2021) adults. The proportion of nonobese diabetes decreased from 69.2% to 37.8% in China but remained around 50% in Korea.

Because incidence is more reliable for describing disease burden, a meta-analysis was performed to assess weight-specific diabetes incidences and related trends. Ninety-four prospective cohorts of 3.4 million adults from 22 countries reporting diabetes incidence by body mass index (BMI) categories were included. The pooled incidence in adults with normal weight and overweight/obesity was 2.7 and 10.5 per 1000 person-years, respectively. Diabetes incidence has tripled from 1985 to 2010 in both individuals with normal weight and overweight/obesity.

Existing diabetes screening tools overestimate diabetes risk in nonobese individuals. We explored risk factors for nonobese diabetes using multi-cross-sectional data from CHNS/CHARLS/KNHANES/ELSA/NHANES, including classical risk factors such as age, sex, education levels, smoking, family history, BMI, waist circumference (WC), and blood pressure, as well as novel risk factors, sleep duration, pulse rate, and grip strength.

We then developed and validated screening tools specifically for nonobese diabetes using LogR, LASSO, random forest, and CatBoost machine-learning algorithms. The CatBoost-modeled tool performed well in discrimination (C-statistic = 0.838) and calibration (Hosmer–Lemeshow p-value >0.05).

We used CHNS/CHARLS/KNHANES/ELSA/NHANES dataset to compare cardiometabolic risks between obese and nonobese diabetes. Obese diabetes presents with higher hemoglobin A1c (HbA1c, group difference Δ = 0.1%), total cholesterol (Δ = 4.36 mg/dL), and triglyceride (Δ = 40.36 mg/dL) and lower high-density lipoprotein (Δ = −4.87 mg/dL) levels. However, these differences became non-significant when WC was adjusted, reflecting that central obesity is a major risk factor for cardiometabolic health in both obese and nonobese diabetes.

Finally, we used CHARLS/ELSA longitudinal data to assess weight-related disparities in fasting plasma glucose (FPG) and HbA1c trajectories before diagnosis by multilevel modeling. HbA1c increased more steeply and to a higher level in obese incident diabetes than in their nonobese counterparts; however, the FPG trajectory between them was similar.

In conclusion, the prevalence and incidence of diabetes in nonobese individuals are increasing, particularly in Asia, and weight-related disparities in cardiometabolic risks and glucose trajectories highlight the need for more tailored diabetes risk screening and prevention strategies for individuals with different weight status.