Normal weight obesity and the risk of diabetes in Chinese people: a 9-year population-based cohort study

We evaluated the risk of developing diabetes in Chinese individuals with normal weight obesity (NWO). This 9-year population-based cohort study was based on the China National Diabetes and Metabolic Disorders Survey. A total of 1128 subjects without diabetes were included. Body fat percentage (BF%) was assessed by electrical bioimpedance. NWO was defined as subjects with a normal BMI (< 24 kg/m2) and an excess BF% (≥ 24% in men; ≥ 33% in women). Of 1128 individuals, 528 individuals were normal weight non-obese (NWNO), 118 (10.5%) were normal weight obese (NWO), 63 were overweight non-obese (OWNO), and 419 were overweight obese (OWO). During a follow-up of 9.0 years (interquartile range: 8.9–9.3), 113 (10.0%) individuals developed diabetes. The incidence rates of diabetes in NWNO, NWO, OWNO and OWO people were 5.69 (27 cases), 11.30 (12 cases), 3.53 (2 cases) and 19.09 (72 cases) per 1000 person-years, respectively. Cox regression analyses indicated multivariate-adjusted hazard ratios of diabetes in NWO, OWNO and OWO people were 2.110 (95% CI 1.026–4.337, p = 0.025), 0.441 (95% CI 0.101–1.928, p = 0.232) and 3.465 (95% CI 2.163–5.551, p < 0.001), respectively, relative to NWNO people. Chinese people with NWO are at increased risk of developing diabetes. We strongly suggest the incorporation of BF% measurement into the regular physical examination in Chinese medical practice.


Results
The study sample included 1128 individuals without diabetes, with 42.3% male, a mean age of 41.66 years, a mean BMI of 23.65 kg/m 2 , and a mean BF% of 29.17%. During the 9-year (interquartile range: 8.9-9.3) follow-up, 113 (10.0%) individuals developed diabetes, of whom 92 were diagnosed based on the oral glucose tolerance test (OGTT) and 21 diagnoses were based on self-reported diabetes history or use of hypoglycemic drugs (Fig. 1).
Within the study sample, 528 individuals were considered normal weight non-obese (NWNO), 118 (10.5%) were classified as normal weight obese (NWO), 63 were classified as overweight non-obese (OWNO), and 419 were diagnosed as overweight obese (OWO). Compared with NWNO individuals, NWO individuals were more likely to be older, male, to have higher blood pressure, and higher levels of glucose, total cholesterol, LDL cholesterol and triglycerides. Baseline measures are provided according to BMI and obesity phenotypes in Table 1. Compared with individuals who did not develop diabetes, individuals who developed diabetes during follow-up had a higher baseline age, BMI, BF%, blood pressure, glucose level and triglyceride levels ( Table 2).

Discussion
We previously reported that the optimal foot-to-foot BIA-measured BF% cutoff in Chinese people is 24% for men and 33% for women. The prevalence of NWO is 7.39% based on these criteria 11,12 . In the present study, we conduct a 9-year longitudinal survey and show that Chinese people with NWO had an approximately two times greater risk of developing diabetes (HR: 2.110, 95% CI 1.026-4.337, p = 0.025) compared to NWNO controls. Overweight people with normal BF% did not have an increased risk of diabetes (HR: 0.441, 95% CI 0.101-1.928, p = 0.232). The strengths of this study include that it is a population-based sample, has a long-term follow-up period, diagnosed diabetes based on OGTT, excluded individuals with diabetes at baseline, and adjusted for several potentially confounding factors.
Our results showed that people with NWO have a significantly increased risk of developing diabetes. Only a few cross-sectional studies have previously explored the association between NWO and insulin resistance, hyperglycemia or diabetes, but the results were equivocal 8,9 . For example, Madeira et al. studied 1222 males and females aged 23-25 years and showed that NWO was significantly associated with a homeostasis model assessment-insulin resistance (OR = 3.81, 95% CI 1.57-9.82) but not with high blood glucose (OR: 1.60, 95% CI 0.54-4.76, p = 0.395) 8 . In contrast, a cross-sectional study conducted by Marques-Vidal et al. with 3213 women and 2912 men aged 35-75 years in Switzerland, suggested that NWO women had a higher prevalence of fasting   10 . Differences in RRs between Zhao et al. and our study might be due to differences in study design, duration of follow up, control group, and baseline metabolic state of the subjects included.
In addition, we found that overweight people with normal BF% did not have an increased risk of diabetes (HR: 0.441, 95% CI 0.101-1.928, p = 0.232). The results agree with those of Zhao et al. 10 , who found that those with stable normal BF% and abnormal BMI did not have a significantly higher risk of developing diabetes (RR = 2.838, 95% CI 0.293-27.537). The possible existence of healthy obesity has been a subject of debate for many years because some overweight and obese individuals are found to have a metabolically healthy status 13 . However, meta-analyses indicates that adults with metabolically heathy obesity (MHO) have substantially increased risks of developing diabetes and cardiovascular events compared with metabolically healthy normal-weight adults 14,15 , and prospective evidence does not indicate that healthy obesity is a harmless condition. Thus, data corroborate the concept that MHO is one step in the process of obesity whereby individuals are likely to become metabolically unhealthy 16 . Nonetheless, the key point is that obesity judged by BMI is problematic, because BMI does not take into account a person's bone, muscle, or fat proportions, and thus fat mass and fat-free mass are not differentiated with BMI 17 . In fact, a person with a higher BMI but normal BF% is more likely to take part in strength training or be an athlete. For example, in the National Football League (NFL), 97% of all players have a BMI of ≥ 25.0 kg/ m 2 , but based on BF% only 8.9% of these athletes are obese and 14.5% have below-normal BF% 18 . Thus, BF% is a more accurate measure for identification of obesity-related diabetes risk in populations such as athletes, and clinical focus should be on BF% rather than BMI.
A previous nationwide population-based cohort study with a mean follow-up of 8.8 years showed that NWO was associated with high prevalence of cardiometabolic dysregulation, metabolic syndrome, and cardiovascular risk factors. In women, NWO is independently associated with increased risk for cardiovascular mortality 7 . Those findings, together with our own, drew our attention to NWO, as a special phenotype of obesity. Since obesity is defined as a condition where there is an excess of body fat 19 , as opposed to excess weight (in fact, excess BMI),   Table 3. Cox regression analysis of obesity phenotypes in predicting type 2 diabetes mellitus in total, male, and female participants. CI confidence interval. *Hazard ratio and 95% CI were calculated using a backward stepwise method. The covariables were age, (gender), cigarette smoking, alcohol drinking, physical activities, family history of diabetes and baseline blood glucose levels. We would emphasize that NWO is of greater concern than MONW, a condition that has been studied for some decades. In contrast to subjects with MONW, who present major metabolic changes that result in signs and symptoms as well as subsequent diagnosis 21 , individuals with NWO are likely undiagnosed and thus unaware of the cardiometabolic risks they face. NWO screening using simple methods (e.g., foot-to-foot BIA) is thus strongly suggested in clinical practice.
Several limitations of the study should be addressed. First, although dual energy X-ray absorptiometry is considered the gold standard for BF% measurement 22 , we used bioelectrical impedance in our study based on its acceptable accuracy, simplicity, lack of radiation, and relatively low cost 23 . Nevertheless, bioelectrical impedance may underestimate upper-body obesity 24 , we had a relatively low follow-up rate (41.0%) for the entire sample. However, after excluding 177 individuals in the low-rate sites, the follow-up rate was 72.6% with a sample size of 1173, which is sufficient for a high-quality cohort study. We also conducted sensitivity analysis by excluding sites with low follow-up rate and showing results were similar. Third, misclassification may have happened in our study, since people could have had changes in their body composition during the follow-up period. This concern is true for most longitudinal studies which use baseline information on the exposure variable. Fourth, unmeasured and residual confounding effects cannot be eliminated although a range of potentially confounding factors were included. For example, information on dietary intake was not collected. Fifth, cases of incident diabetes and overweight non-obesity were small, so relevant conclusions need to be confirmed by future studies with larger sample sizes. Last, our findings from a single-province survey might not be generalizable to the whole country.
In conclusion, we demonstrate that Chinese people with NWO are at increased risk of developing diabetes. Our study shows that BF% constitutes a better measure of obesity assessment than BMI. We strongly suggest incorporation of BF% measurement into the regular physical examination, and NWO screening using simple methods in clinical practice in the Chinese population.

Methods
Study sample. This cohort study with a follow-up of 9 years was based on the China National Diabetes and Metabolic Disorders Study (CNDMDS). CNDMDS was a nationwide population-based cross-sectional survey, with aims to investigate the prevalence of diabetes and associated metabolic risk factors in China. The current cohort study included only individuals of CNDMDS in Shaanxi province, Northwestern China. During June 2007 and May 2008, we used a multi-stage stratified sampling method to select a representative sample from Shaanxi province, narrowing down the sample pool in four stages, as detailed previously 25 . Briefly, in the first stage, two cities (Xi'an as capital city and Yan'an as a non-capital city) were selected non-randomly. In the second stage, city districts from cities and rural townships from countryside were randomly selected. In the third stage, street districts from city districts and rural villages from townships were randomly selected. In the last stage, adults ≥ 20 years were selected using stratified sampling methods according to the sex and age distribution. Table 4. Sensitivity analyses. RR relative risk, CI confidence interval. *Hazard ratio and 95% CI were calculated using a backward stepwise method. The covariables were age, gender, cigarette smoking, alcohol drinking, physical activities, family history of diabetes, and baseline blood glucose levels. www.nature.com/scientificreports/ In total, 3296 subjects (2009 in Xi'an city and 1287 in Yan'an city) were selected as the baseline study sample and were invited to participate in the 9-year follow-up evaluation from October 2016 to November 2017. Of these, 1350 subjects (1221 in Xi'an city and 129 in Yan'an city) agreed and participated in the follow-up evaluation; 972 had a face-to-face evaluation and 378 were evaluated by telephone (21 were deceased). The total follow-up rate was 41.0% in all study sites; after excluding study sites with a low follow-up rate, the rate was 72.6% with a study sample size of 1173. The details of follow-up rate for each study site are described in Table S1. Comparison of baseline data between participants who did and did not undergo follow-up examinations is provided in Table S2. We excluded subjects with diabetes (n = 122) diagnosed by oral glucose tolerance test (OGTT) or a history of antidiabetic medications at baseline. We also excluded 98 subjects with missing data for BF% and two subjects with BF% > 60% or < 5%. Ultimately, a total of 1128 subjects were included as our study sample for analysis (Fig. 1).
This study was approved by the Ethics Committee of Xijing Hospital, Affiliated Hospital of Fourth Military Medical University. All participants signed written informed consent prior to data collection. All methods were performed in accordance with the relevant guidelines and regulations.
Baseline data collection. During 2007During -2008, using a standard questionnaire, trained physicians and nurses collected baseline data on demographic characteristics, lifestyle risk factors, personal medical history and family history of diseases 26 . Body size and obesity phenotype were evaluated including body height, body weight, waist circumference and BF% measurement. Body weight was measured in kilograms on a calibrated digital scale and body height was measured using a stadiometer in centimeters with subjects wearing light clothes and without shoes. BF% was assessed by electrical bioimpedance analysis (BIA) using the Tanita TBF-300 body composition analyser (Tanita Corporation, Tokyo, Japan) at baseline during 2007-2008. An OGTT was performed on all subjects. Subjects without history of diabetes were administered a standard 75-g glucose solution after at least 10 h of overnight fasting. For security reasons, subjects with a self-reported history of diabetes were given a steamed bun with approximately 80 g of complex carbohydrates. A hexokinase enzymatic method was used to measure serum glucose. Plasma lipid profiles, including total cholesterol, serum triglyceride and low-density lipoprotein (LDL) cholesterol were also measured from fasting blood samples by the enzymatic method. All laboratory measurements were conducted under conditions of a standardization and certification program 25 . Assessment of obesity phenotype. Subjects were stratified into four groups according to baseline BMI and BF%. BMI was calculated by dividing body weight in kilograms by the square of body height in meters. The normal weight obese (NWO) group contained subjects with a normal BMI (< 24 kg/m 2 ) and an excess BF% (≥ 24% in men and ≥ 33% in women) 11 . Subjects with a normal BMI (< 24 kg/m 2 ) and normal BF% (< 24% in men and < 33% in women) were placed in the normal weight non-obese (NWNO) group as controls. Subjects with a BMI ≥ 24 kg/m 2 were considered overweight, and included both overweight (24 kg/m 2 ≤ BMI < 28 kg/m 2 ) and obese (BMI ≥ 28 kg/m 2 ) subjects. Overweight subjects with a normal BF% were placed in the overweight non-obese (OWNO) group, and overweight subjects with an excess BF% were in the overweight obese (OWO) group.

Assessment of covariates.
Alcohol drinking was defined as consuming at least 30 g of alcohol per week for at least a year. Individuals that had consumed at least 100 cigarettes during their lifetime were classified as cigarette smoking. Regular leisure-time physical activity was defined as participation in moderate to vigorous activity for 30 min or more per day on at least 3 days a week. Education level was categorized as (1) below college or (2) college and above. Family history of disease was defined as at least one of parents or siblings diagnosed with a disease in their lifetime and based on self-reporting 25 . Blood pressure was measured using a mercury sphygmomanometer in a sitting position in the morning after at least 5 min of rest. Two consecutive measurements were performed and the mean of the measurements recorded.
Follow-up evaluation. The primary outcome was the incidence of diabetes with a follow-up of 9 years. For subjects receiving face-to-face evaluation at the follow-up survey, OGTT was performed to evaluate diabetes status, as described above. According to World Health Organization diagnostic criteria 27 , diabetes is defined as the use of antidiabetic medications, a fasting glucose level ≥ 7.0 mmol/l, or a 2-h glucose level ≥ 11.1 mmol/l after a 75 g OGTT. Pre-diabetes was defined as fasting glucose level between 6.1 and 6.9 mmol/l, and/or a 2-h glucose level between 7.8 and 11.0 mmol/l after a 75 g OGTT. For individuals receiving follow-up interviews by telephone evaluation, diabetes was diagnosed by self-reporting. Statistical analysis. Sample size requirement was estimated in advance using PASS 11.0 (NCSS, LLC, Utah, USA). Based on a previous study that indicated NWO individuals had a 4.8-fold increased risk of diabetes (95% CI 1.1-21.6) 10 , a sample size of 106-425 participants was planned to provide an 80% power with 95% confidence intervals (α = 0.05), which corresponds to HRs of 2.0-4.0 (P0 was estimated as 5%).
Data were summarized as the mean ± standard deviation (SD) for normally distributed data, the median with interquartile range for skewed variables, or percentages, as suitable. Differences between groups in continuous variables were analyzed by a t-test or analysis of variance (ANOVA) as appropriate. Differences for enumeration data were assessed by the chi-square test of independence.
The incidence of diabetes was calculated for the NWNO, NWO, OWNO and OWO groups. To identify the impact of NWO on the risk of diabetes, a cox regression analysis was done and hazard ratios (HRs) with corresponding 95% CI were obtained using a backward stepwise method. In addition, sensitivity analyses were conducted to assess the robustness of the results by re-running all the models with a subset of individuals excluded.