The Prevalence of Metabolically Healthy and Unhealthy Obesity according to Different Criteria

The prevalence of obesity is increasing worldwide, and now it has become a major public health problem in China [1]. Obesity is an identifiable important risk factor for type 2 diabetes, hypertension, coronary heart disease, stroke, and several types of cancers and is thus linked to premature death [2, 3]. Additionally, obese individuals are more likely to have osteoarthritis, chronic pain, asthma, gallbladder disease, metabolic disorders, and poor quality of life [4]. Although maintaining a healthy weight to reduce the incidence of these diseases or adverse health consequences related to obesity has been a global consensus, it is increasingly recognized that the disease risks may not be uniform among all obese subjects. In an obese population where metabolic abnormalities are absent (e.g., hypertension, diabetes, dyslipidemia, insulin resistance [IR], or inflammatory factors), a better fitness with a 30–50% lower mortality was observed than in obese peers with these risk factors [5]. This risk difference indicates that a heterogeneous risk profile might exist among obese subjects. Therefore, a subset of obesity has been further defined as metabolically healthy obesity (MHO), including obese subjects who seem to be protected against obesity-related metabolic complications and lack a set of cardiometabolic abnormalities. Conversely, obesity with metabolic risk factors is called metabolically unhealthy obesity (MUO) [6]. MHO includes subjects who are considered as having large quantities of fat mass or body weight but exhibit a healthy metabolic profile; thus, it is referred to as a benign condition [7, 8]. However, a meta-analysis including 8 longitudinal studies showed that MHO individuals are at increased risk for all-cause mortality over the long term (≥10 years), which indicates that MHO might be an intermediate stage of MUO [9].

Currently, the prevalence of the MHO phenotype in different populations varies from 1.1 to 28.5%, while this variation might result from the diverse definitions of metabolic abnormalities adopted in different studies or from other methodological issues, such as small sample size or limiting to a special subgroup [7]. Until now, there has been no standard definition of metabolic abnormality worldwide when defining obesity phenotypes. It also remains unclear which components of metabolic disorder should be included when defining MHO and how to identify the detailed cutoffs of the selected metabolic components to differentiate metabolically healthy from unhealthy subjects. Furthermore, some criteria added IR and high-sensitivity C-reactive protein (hsCRP) as the basic metabolic components (elevated blood pressure, dysglycemia, and dyslipidemia) to define obesity phenotypes [10]. Using different definitions of metabolic disorder to estimate the prevalence of obesity types within a certain population will diminish or reduce the variation from study design and ensure internal validity. Therefore, the aims of the current study were to (1) compare the prevalence of obesity phenotypes using different adiposity measures and criteria of metabolic abnormality; (2) provide evidence if MHO is an intermediate state of MUO; (3) assess if IR and hsCRP are essential components to define metabolic abnormality in obesity.

This study included only the Beijing site derived from the China Hypertension Survey, a nation-wide survey conducted during 2013–2015. The method and design of the survey were described elsewhere, and the specific protocol was strictly followed at each study site including Beijing [11, 12]. Briefly, a 4-stage stratified random procedure was used to recruit a total of 500,000 permanent residents aged 15 years and older from 262 urban cities and rural counties over 31 metropolitan/provinces in mainland China. In the 1st-stage random sampling, we used the probability proportional to size method to randomly select 2–4 cities in urban areas and 2–4 counties in rural areas from each metropolitan/province. We then used a simple random sampling method in the 2nd to 3rd stage random procedure to select 2 districts (urban) or townships (rural) and 3 communities or villages within each district or township, respectively. In the final stage, by using a simple random sampling method, a given number of participants from each of the 14 gender/age strata (male/female and aged 15–24, 25–34, 35–44, 45–54, 55–64, 65–74, and ≥75 years) based on the government-registered data of households were selected. All selected participants were invited to complete a questionnaire and anthropometric measurement. In addition, blood and urine samples were collected from those aged ≥35 years. Beijing as one of the metropolitan cities in China contributed a total of 13,057 subjects aged ≥15 years to the nationwide survey, and 6,906 subjects were ≥35 years old. The blood and urine samples were collected from 4,757 participants and used for this analysis.

To ensure the uniformity of the survey, a manual of operations was compiled to standardize the methods and procedures, including the standard approaches to recruiting subjects, determining the sample size, interviewing the subjects, performing the physical examinations, defining the cardiovascular events, and quality control. All research staff participating in the survey was required to accept standardized training before the survey was conducted [11].

The study was approved by the Ethics Committee of both Fuwai Hospital and Xuanwu Hospital (Beijing, China). A written informed consent was obtained from every subject before data collection.

All participants answered the standardized questionnaire administered by well-trained investigators, including information on sociodemographic factors, history of hypertension, diabetes mellitus, dyslipidemia, cardio- and cerebrovascular diseases, as well as current medications [11]. The medical record was reviewed carefully.

After the interview, all subjects underwent measurements according to a standardized protocol. Briefly, body weight was measured in standing position without shoes and in light clothing using an OMRON body fat and weight measurement device (V-body HBF-371, OMRON, Kyoto, Japan). Height was measured using a standard right-angle device and a fixed measurement tape (to the nearest 5 mm). Waist circumference (WC) was defined as mid-way between the lowest rib and iliac crest and measured under steady breaths using a cloth tape directly on the participant’s skin.

Sitting blood pressure was measured on the right arm 3 times with a 30-s interval with a Professional Portable Blood Pressure Monitor (OMRON). The average of 3 measurements was used for analysis. Each subject was asked to empty the bladder and rested at least 20 min before blood pressure was measured.

Ten to twelve milliliters of spot blood samples were taken in the morning after a 12-h fasting from each eligible participant. All biological samples were sent to the National Center for Cardiovascular Disease of Fuwai Hospital for analysis. Fasting plasma glucose, total cholesterol, triglycerides, and high-density lipoprotein cholesterol were determined by routine methods abiding to standards from the National Guide to Clinical Laboratory Procedures issued by the China Ministry of Health in 2006 [13]. Low-density lipoprotein cholesterol was calculated using the Friedwald formula. Serum insulin was measured by the immunoassay method; hsCRP was assessed by high-sensitivity immunonephelometry (German BN ProSpec system). The sample storage and detection methods were described in detail elsewhere [14].

Body mass index (BMI) was calculated as body weight in kilogram divided by the square of height in meter. Subjects were classified according to BMI as normal weight (18.5–23.9 kg/m²), overweight (24.0–27.9 kg/m²), and obese (≥28.0 kg/m²) according to the Chinese standard [15]. Abdominal obesity was considered for a waist ≥85 cm for men and ≥80 cm for women in the Chinese population [15]. The homeostasis model assessment of insulin resistance (HOMA-IR) index was calculated as [fasting serum insulin × (fasting glucose)/22.5]: insulin concentration is reported in mU/L and glucose in mmol/L. Participants were defined as IR if their HOMA-IR level was in the top quartile of the distribution among nondiabetic subjects [16, 17]. As systemic inflammation is one component of cardiometabolic abnormalities presented by Wildman et al. [10], detailed as hsCRP level > 90th percentile, we also adopted this cutoff in our study.

The definition of the MHO and MUO phenotypes was obtained based on the joint combination of obesity markers (BMI or WC) and cardiometabolic abnormalities listed in Table 1. Five sets of cardiometabolic abnormalities were based on the following definitions: the National Cholesterol Education Program-Adult Treatment Panel III (ATPIII), Karelis, Wildman, Chinese Diabetes Society (CDS), and the HOMA index [10, 18-21]. Thus, subjects were classified into MUO if they belonged to BMI-defined or WC-defined obesity, coexisting with a diagnosis of metabolic abnormalities in any of the 5 definitions above. Obese subjects with absence of metabolic abnormalities were classified as MHO.

The prevalence of MHO and MUO was estimated by age and gender based on the 2010 Beijing municipal population census; the age- and gender-specific weight-adjusted sample was acquired to estimate the prevalence of MHO and MUO. For continuous variables, means ± standard deviations were presented if the data fitted normal distribution. For continuous variables with a non-normal distribution, medians (interquartile ranges) were reported, and the differences of medians between groups were assessed by the Kruskal-Wallis test. The χ² test was used to determine the differences in the categorical variables. Test for trend was performed to explore the possible heterogeneities of prevalence of the specific obesity phenotypes over different age groups.

Among the 4,757 subjects, 1,155 (24.3%) subjects were classified as obese, 1,996 (41.9%) as overweight, 1,548 (32.5%) as normal weight, and 58 (1.2%) as underweight based on BMI. Of the 4,757 subjects, 2,994 (62.9%) had abdominal obesity. Compared with men, significantly more women were obese according to BMI (25.9 vs. 22.8%, p = 0.013), but less women had abdominal obesity (61.5 vs. 64.3%, p = 0.043).

As shown in Table 2, there was a great variation in the prevalence of MHO and MUO defined by the 5 definitions listed in Table 1. When obesity was classified using BMI, the highest prevalence of MHO was obtained from HOMA criteria (13.6%, 95% CI 11.0–16.2), followed by CDS (11.4%, 95% CI 8.7–14.1), ATPIII (10.3%, 95% CI 7.6–13.0), and Wildman (5.2%, 95% CI 2.4–8.0), whilst the lowest prevalence of MHO was derived from the criteria of Karelis (4.2%, 95% CI 1.4–7.0). In contrast to MHO, the highest prevalence of MUO was derived from the criteria of Karelis (20.1%, 95% CI 17.6–22.6), which was 2-fold higher than the lowest prevalence from HOMA (10.6%, 95% CI 7.9–13.3). The MUO prevalence for Wildman, ATPIII, and CDS was 19.1% (95% CI 16.5–21.7), 14.0% (95% CI 11.4–16.6), and 12.9% (95% CI 10.2–15.6), respectively. The MHO and MUO prevalence defined by WC was much higher than that defined by BMI; the prevalence order from high to low was similar to that of BMI-defined obesity, but the prevalence rates were much higher both for MHO and MUO in WC-defined obesity than in BMI-defined obesity. The proportion of MHO and MUO or the ratio of MHO to MUO is demonstrated in online supplementary Figure 1 (for all online suppl. material, see www.karger.com/doi/10.1159/000495852).

When using BMI to define obesity, a significantly higher prevalence of MHO in females than in males was observed except for ATPIII criteria, but there was no significant gender difference in MUO prevalence between the other 4 criteria. In WC-defined MHO, similar findings of a gender difference were found as for BMI-defined MHO. Moreover, there existed gender differences in the prevalence of MUO between the 5 definitions of metabolic abnormalities (Table 2). A detailed sex constituent ratio of MHO and MUO for different criteria is shown in online supplementary Figure 1.

In BMI-defined obesity, the prevalence of MHO defined by criteria of ATPIII, Wildman, CDS, and HOMA increased with age (p for trend < 0.05) and reached the peak in those aged 45 years (Fig. 1a). Then, the prevalence declined with an increment at age 80 years, except for Wildman criteria. The Wildman-defined MHO prevalence kept on declining after age 45 years. The trend of age-specific prevalence in Karelis-defined MHO changed differently from the other 4 definitions. The prevalence remained stable before reaching the peak at age 75 years and dropped to a minimum at age 80 years. In addition, the Karelis-defined MHO prevalence was relatively stable across age groups with a narrow variation ranging from 3.1 to 6.0%. The age-specific prevalence of MHO defined by the other 4 criteria fluctuated greatly across age groups. The most fluctuating MHO prevalence was found for the ATPIII definition with a wide variation in prevalence (range 3.2–15.7%), followed by CDS (range 5.3–15.8%), HOMA (range 6.4–16.8%), and Wildman (range 1.6–7.6%).

The age-specific prevalence of BMI-defined MUO increased below age 65 years and then declined regardless of which criteria were used (Fig. 1b). After age 84 years, the prevalence defined by CDS criterion kept on declining, but when the criteria of ATPIII, Karelis, Wildman, and HOMA were used, the prevalence increased slightly. The differences in prevalence between the maximum and the minimum across all age groups varied greatly as defined by the 5 criteria. The magnitude of differences in the prevalence was ranked highest for ATPIII (range 8.6–21.9%), followed by Wildman (range 13.7–25.8%), Karelis (range 12.6–24.6%), CDS (range 8.6–20.0%), and HOMA (range 7.8–16.2%).

In WC-defined obesity, the age-specific MHO prevalence for different metabolic criteria showed almost no consistency. HOMA-defined MHO had the highest prevalence in almost all age groups, which was consistent with BMI-defined MHO. The prevalence kept an ascendant trend with age increment. The variation in prevalence for both CDS-defined and ATPIII-defined MHO was large with a similar trend. The top values appeared at age ≥50 years (43.3% for CDS and 42.5% for ATPIII) and bottom values at age ≥35 years (29.3% for CDS and 31.5% for ATPIII). The age-specific prevalence of MHO defined by either Wildman or Karelis fluctuated with a narrow variation (from 10 to 20%). The MHO prevalence rose obviously after age 80 years regardless of which metabolic criteria were used (Fig. 2a). The tendency of the age-specific prevalence of MUO was similar for the 5 metabolic criteria (Fig. 2b), i.e., it increased after age 40 years, reached the highest level at age 65 years, and then dropped slightly.

The proportion of IR was 43.8% (506/1,155) in BMI-defined obesity, which was higher than in WC-defined abdominal obesity (36.1% [1,080/2,994]). In the case of BMI-defined obesity, the proportion of IR for MHO and MUO individuals differed significantly regardless of which metabolic criteria were used (all p values < 0.001). The most apparent differences were observed when Wildman and Karelis criteria were used. In WC-defined obesity, a similar trend for the presence of IR was found in MHO and MUO. No significant difference was observed in the proportion of elevated hsCRP between BMI-defined obesity and WC-defined obesity (11.8% [136/1,155] vs. 10.2% [305/2,994]). Except for the Wildman criterion, which contained hsCRP as a metabolic component, there appeared no significant difference between MHO and MUO in the proportion of elevated hsCRP in BMI-defined obesity. However, in WC-defined obesity, the proportion of elevated hsCRP was significantly higher in MUO than in MHO when ATPIII or HOMA criterion was adopted (Table 3).

In the present study, we estimated the prevalence of MHO and MUO phenotypes based on the most up-to-date information from a representative survey of the community-based population in Beijing, China. After using the Chinese definition of obesity based on BMI and WC, as well as their combination with 5 criteria of metabolic disorder to define the obesity phenotypes, we found that the difference in the prevalence of obesity and its phenotypes varied greatly. The prevalence of MHO varied from 4.2 to 13.6% when obesity was defined by BMI and from 14.0 to 40.2% when it was defined by WC. MUO prevalence was 10.6–20.1% and 22.7–49.0%, respectively. No matter which obesity criteria were used, the age-specific prevalence of MHO defined by HOMA was the highest almost in each age group, followed by CDS, ATPIII, and Wildman. These findings indicated that the wide variation in the prevalence of MHO and MUO currently is mainly due to the heterogeneity of both definitions of obesity and metabolic disorder, which make the results amongst studies not comparable [7]. Therefore, using a standard criterion to define an obesity phenotype is essential.

BMI and WC are the most common indicators to measure obesity in a population. BMI has been used to define obesity in most obesity phenotype studies, and the cutoffs of BMI for obesity were generally defined by local guidelines or academic organizations [22, 23]. Hence, we adopted the BMI cutoffs of obesity from the Guidelines on Prevention and Control of Overweight and Obesity in China in the current study. They were issued in 2006 by the Ministry of Health and Center for Disease Control in China [15]. WC, which is used to measure abdominal adiposity and whole-body fat, is also an obligatory element in metabolic disorders. For example, both the International Diabetes Federation and the American Heart Association/National Heart, Lung, and Blood Institute regard WC as one of the basic components of metabolic syndrome [24]. However, WC is not a necessary element in the definition of metabolic disorder in the criteria of Karelis, Wildman, and HOMA in the studies of obesity phenotypes [10, 19, 21]. A study by Velho et al. [25] showed that obesity defined by WC would obtain a higher MHO prevalence than when defined by BMI, and this finding was consistent with the present study (62.9 vs. 24.3%). Therefore, we concluded that the prevalence of WC-defined MHO and MUO was much higher than that defined by BMI, accordingly.

Although dozens of criteria for the diagnosis of metabolic disorder were used in studies concerning obesity phenotypes, only few studies focused on the diversity in obesity phenotype prevalence derived from criteria. Phillips et al. [26] adopted BMI ≥30 kg/m² to define obesity in Irish to estimate the prevalence of MHO and MUO, using criteria of Karelis, Wildman, ATPIII, HOMA, and Aguilar-Salinas to define metabolic disorder. The ranking of MHO and MUO prevalence from high to low was similar to that in our study. In the current study, we adopted criteria of ATPIII, Wildman, Karelis, CDS, and HOMA for the following reasons. The ATPIII was the most frequently used criterion in obesity phenotype studies, and the majority of the reports from China also adopted it to define metabolic subtypes in obesity when estimating prevalence [27-29]. Thus, the adoption of ATPIII will make our results comparable to previous studies both from China and other countries. As previous studies in China used the expert consensus to define metabolic abnormality issued by the CDS in 2013 [30], in which the metabolic components included and their cutoff points might be more appropriate for Chinese, we chose CDS criterion in our study.

Considering that the pathological mechanisms of metabolic disorder and obesity are both associated with IR, the criteria of Wildman and Karelis incorporated IR as a component of metabolic abnormality [10, 19, 31]. In an Italian study [32], the estimated IR calculated by HOMA-IR was the only indicator to define metabolic disorder, despite the presence of other metabolic factors, such as hypertension, hyperlipidemia, and hyperglycemia. So, we adopted Wildman, Karelis, and HOMA criteria as well. However, whether IR is an essential indicator to define metabolic disorder is a controversial issue. A study by Bertoni et al. [33] suggested that HOMA-IR is not especially useful in addition to ATPIII criterion for metabolic syndrome in assessing coronary or carotid subclinical disease. Also, insulin does not belong to the clinical routine laboratory indexes. Moreover, the cutoff value of HOMA-IR varied greatly between the Karelis, Wildman, and HOMA criteria. In the current study, the presence of IR was defined as above the upper quartile of the distribution of the calculated HOMA index among nondiabetic subjects. The definition was also widely used in most studies focusing on IR and diabetes [34, 35]. In our study, the proportions of IR in MHO defined by BMI in combination with ATPIII, Karelis, Wildman, or CDS were 30.5, 16.5, 22.5, and 30.6%, respectively. The corresponding proportions of IR in MUO were approximately 2-fold higher than those of MHO (range 53.0–60.1%). Moreover, the presence of IR was much higher in MHO than in MUO according to the definitions of ATPIII and CDS. Similar findings were generated in WC-defined obesity. These findings suggested that IR might not be an essential component to define the MHO and MUO phenotype.

Apart from IR, the Wildman criterion also involved hsCRP, an inflammatory marker, as one of the metabolic components [36]. There was no significant difference in elevated hsCRP between MHO and MUO when using BMI-defined obesity in the current study, which indicated that hsCRP might not be a necessary component in defining metabolic status. However, the elevated hsCRP was significantly higher in MUO than in MHO when we used WC combined with ATPIII and HOMA (p < 0.001). Further study is needed to clarify the role of hsCRP in obesity phenotype classification.

Using BMI to define obesity, several epidemiologic studies from America showed that the prevalence of MHO ranged from 4.0 to 9.7% and that of MUO from 9.0 to 28.5% [5, 10, 21, 37, 38]. The prevalence of MHO in Europe varied from 1.1 to 6.6% and that of MUO from 7.2 to 21.4% [32, 39-41]. Compared to a European population, Americans had a higher prevalence of MHO, and this was consistent with a meta-analysis [42] focusing on the prevalence of MHO in adults. Studies in Korea showed that the prevalence of MHO varied from 5.7 to 25.8% and the prevalence of MUO from 13.6 to 25.9%, and most studies used ATPIII to define metabolic abnormality [22, 29, 43-45]. Our study revealed that the MHO prevalence was 4.2–11.4% and MUO prevalence was 12.9–20.1% when we adopted ATPIII, CDS, Wildman, and Karelis criteria to define metabolic status. These figures showed a good agreement with the previously published data in China (MHO: 1.0–9.2%, MUO: 4.0–30.4%) [27, 28, 46, 47]. However, a broader spectrum of age groups from 20 to 90 years was used in previous studies. As the prevalence of both obesity and metabolic disorder would vary greatly with age, the impact of age on the heterogeneity of the obesity phonotype prevalence between studies cannot be ruled out and makes comparisons between studies difficult.

Few studies have reported on the age-specific prevalence of obesity phenotypes, and thus, the relevant data are limited. The prevalence of obesity phenotypes varied considerably with age in the current study. This was not in agreement with a study focusing on middle-aged men from Yoo et al. [48]. One explanation for this inconsistency might be the age difference. The participants’ age ranges in Yoo et al.’s study were relatively narrow and young, with the majority of the participants being in their 30s to 40s.

Several studies proposed that MHO is a transient state, and it may turn into MUO in a later life stage. A follow-up study from England [49] showed that 44.5% of MHO individuals had transited into an unhealthy metabolic state within 8 years. In the present study, when using WC to define obesity, the age-specific prevalence of the 2 obesity phenotypes supported the hypothesis that MHO is the intermediate state of MUO. For example, after 65 years of age, the prevalence of HOMA-defined MHO increased, but MUO had an opposite trend. The prevalence of BMI-defined MHO decreased with age after 45 years, and the prevalence of MUO increased with age during 45–65 years, which supported that both MHO and MUO were in an unstable state. However, the prevalence of BMI-defined MHO and MUO both obviously decreased after age 65 years, which implied that a number of obese people transformed into a nonobese state. The BMI- and WC-defined age-specific prevalence of total obesity could explain the diversity of the above findings (online suppl. Fig. 2). The prevalence of WC-defined obesity remained increased with aging, whilst the age-specific prevalence of BMI-defined obesity remained stable before age 65 years and then decreased gradually afterward, and this was similar to the findings of Xu et al. [50] in China.

As there is no standard definition, different indicators, inclusion criteria, and/or cutoffs have been used to define metabolic obesity phenotypes, thus making comparisons between studies difficult. Moreover, the obesity and metabolic status varied for population and ethnicity. Using 1 population to estimate the impact of diverse definitions of obesity phenotypes could be a novel approach. To our knowledge, the current study is the first community-based epidemiological study to provide the whole picture comparing the MHO and MUO phenotypes between different obesity and metabolic disorder criteria in a large-scale population. The sample was well representative of the general population. The biomarkers collected in the current study were comprehensive and making it possible to define obesity phenotypes with different criteria. The large sample size allowed us to perform a subgroup analysis to obtain age- and gender-specific prevalence. However, because of the nature of a cross-sectional study, we could not tell which criterion of obesity and metabolic health would be best to predict adverse health outcomes, which poses the major limitation to the present study. In addition, the differences in obesity phenotype prevalence by different criteria may not be directly extrapolated to other populations because of the possible differences in the genetic and environmental background between populations.

In summary, our study demonstrated that the prevalence of MHO and MUO varies considerably according to criteria/definitions of both obesity and metabolic abnormalities. MHO might be an intermediate state of MUO, and further study is needed to clarify this issue.

This study was supported by grants from the Beijing Municipal Science and Technology Commission (D121100004912002) and by the National Science and Technology Support Program (2011BA/11B01).

The study protocol was approved by the research institute’s committee on human research.

The authors have no conflicts of interest to declare.

L.A.T. and X.F. contributed equally to this work.