Various definitions of metabolic health have been proposed to explain differences in the risk of type 2 diabetes within BMI categories. The goal of this study was to assess their predictive relevance.
We performed systematic searches of MEDLINE records for prospective cohort studies of type 2 diabetes risk in categories of BMI and metabolic health. In a two-stage meta-analysis, relative risks (RRs) specific to each BMI category were derived by network meta-analysis and the resulting RRs of each study were pooled using random-effects models. Hierarchical summary receiver operating characteristic curves were used to assess predictive performance.
In a meta-analysis of 140,845 participants and 5,963 incident cases of type 2 diabetes from 14 cohort studies, classification as metabolically unhealthy was associated with higher RR of diabetes in all BMI categories (lean RR compared with healthy individuals 4.0 [95% CI 3.0–5.1], overweight 3.4 [2.8–4.3], and obese 2.5 [2.1–3.0]). Metabolically healthy obese individuals had a high absolute risk of type 2 diabetes (10-year cumulative incidence 3.1% [95% CI 2.6–3.5]). Current binary definitions of metabolic health had high specificity (pooled estimate 0.88 [95% CI 0.84–0.91]) but low sensitivity (0.40 [0.31–0.49]) in lean individuals and satisfactory sensitivity (0.81 [0.76–0.86]) but low specificity (0.42 [0.35–0.49]) in obese individuals. However, positive (<3.3 in all BMI categories) and negative (>0.4) likelihood ratios were consistent with insignificant to small improvements in prediction.
Although individuals classified as metabolically unhealthy have a higher RR of type 2 diabetes compared with individuals classified as healthy in all BMI categories, current binary definitions of metabolic health have limited relevance to the prediction of future type 2 diabetes.
Introduction
Obesity and the “metabolic syndrome,” two highly prevalent and often coexisting conditions, are major risk factors for type 2 diabetes and cardiovascular disease (1–3). The observation that some obese individuals have a favorable metabolic profile and appear to be at low risk of obesity-related complications has led to the notion of “metabolically healthy obesity” (4,5). The topic has received much attention in recent times, with an increasing number of studies using definitions of metabolic health in BMI categories either as a risk factor or an outcome (5–9).
In the existing literature, however, there is little consensus on the definitions of metabolic health (4). In addition, the relevance of metabolic health definitions to the prediction of incident type 2 diabetes in BMI categories has not been investigated.
Establishing the predictive value of currently used definitions of metabolic health has been deemed of primary importance, as an accurate risk classification may justify selective preventive action in high-risk individuals (4). In addition, the construct of the “metabolic syndrome,” which is used as a basis for several definitions of metabolic health, has been proposed as a clinically useful predictor of the risk of future type 2 diabetes (10). It is of particular interest to establish whether current definitions of metabolic health help identify lean individuals at high risk of type 2 diabetes (i.e., the “metabolically unhealthy lean”) or obese individuals at low risk of type 2 diabetes (i.e., the “metabolically healthy obese”).
We therefore reviewed the literature on the definitions of metabolic health and assessed their relevance to the prediction of incident type 2 diabetes in lean, overweight, or obese individuals.
Research Design and Methods
Literature Searches
This report adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines where applicable (11). We used three complementary strategies in order to assess the existing literature on the definitions of metabolic health in BMI categories. Firstly, we reviewed titles and abstracts of all references cited by Stefan et al. (4) and Kramer, Zinman, and Retnakaran (5) in recent reviews on the topic. Secondly, we reviewed titles and abstracts retrieved by a MEDLINE search from inception through to the 1 September 2014 with the following terms (search 1): “metabolically-healthy obesity OR metabolically-healthy obese OR metabolically healthy obesity OR metabolically healthy obese.” In order to maximize sensitivity for the detection of studies of incident type 2 diabetes, we conducted a second MEDLINE search from 2000 to 1 September 2014 with the following strategy (search 2): “(Diabetes Mellitus, Type 2[MeSH Terms] OR “type 2 diabetes”[Title/abstract] OR “diabetes”[Title] OR Type II Diabetes Mellitus OR Type 2 Diabetes Mellitus OR Noninsulin-Dependent Diabetes Mellitus OR Diabetes Mellitus, Type II OR Diabetes Mellitus, Type 2 NOT “type 1 diabetes”[Title/abstract]) AND (adiposity OR “body mass index”[MeSH Terms] OR BMI[Text Word] OR “overweight”[MeSH Terms] OR overweight[Text Word] OR “obesity”[MeSH Terms] OR obesity[Text Word]) AND (metabolic health OR metabolically healthy OR metabolic status OR high risk OR risk category OR risk stratification OR cardiometabolic health OR cardiometabolic risk) AND (epidemiologic studies[MeSH Terms] OR epidemiologic study OR observational study OR case-control OR cross-sectional OR case-cohort OR longitudinal study OR cohort OR cohort study OR follow-up study OR cohort analysis OR incidence study) AND (humans[MeSH Terms]) AND (“2000”[Date - Publication]: “3000”[Date - Publication]) AND English[Language].”
Titles, abstracts, and full articles were reviewed by one author (L.A.L.) with the following criteria. For title review, the title had to refer to the definition of metabolic health or cardiometabolic risk stratification in lean, obese, or overweight individuals. For abstract and full-article reviews, the following inclusion criteria were used: 1) the study had a cross-sectional, case-control, cohort, cohort-derived design (nested case-control or case-cohort), and 2) the study provided an explicit definition of metabolic health in lean, overweight, or obese individuals or used one or more variables to stratify cardiometabolic risk in these categories. Exclusion criteria were as follows: 1) the manuscript reported a randomized controlled trial, another intervention study (e.g., lifestyle interventions or studies on individuals who were candidate to bariatric surgery), or a review of the literature; 2) the study used obesity, BMI, or another anthropometric variable as the risk factor or as the outcome, rather than stratifying, variable; and 3) the study was restricted to metabolically healthy or unhealthy individuals only, was not conducted in adult humans, was a genetic association study, or was restricted to patients with diabetes or other cardiometabolic disease.
One author (A.-S.S.) was asked to independently review 10% of the records of each stage of each search. Inconsistencies were resolved by repeated review and discussion. Concordance was high (96% for titles, n = 256 of 269; 97% for abstracts, n = 68 of 70; and 100% for articles, n = 18 of 18). Data on definitions of metabolic health from all screened articles (n = 126) and full information from articles reporting on incident type 2 diabetes (n = 16) were extracted. Studies on incident type 2 diabetes were qualitatively assessed using a modified version of the scoring system proposed by Bell, Kivimaki, and Hamer in a recent review (6).
Definition of BMI and Metabolic Health Categories in Studies of Incident Type 2 Diabetes
For studies in non-East Asians, BMI categories were defined as follows: lean <25, overweight 25–29.9, and obese ≥30 kg/m2 (12). For East Asians, we used categories of BMI associated with “acceptable, increased and high” risk of metabolic disease according to a recent communication of the World Health Organization, i.e., lean <23, overweight 23–27.4, and obese ≥27.5 kg/m2 (13). Where the authors did not use these cutoffs, we contacted them using a standardized electronic mail message and asked for a reclassification of the participants.
Metabolic health definitions, mainly consisting of insulin resistance and metabolic syndrome, were carried over from the original reports.
Two-Stage Meta-analysis
Our meta-analysis was conducted in two stages. In the first stage, we used network meta-analysis to derive, for each study, the relative risk (RR) of type 2 diabetes of metabolically unhealthy compared with healthy individuals in the lean, overweight, and obese categories. In the second stage, we pooled the resulting RRs using random-effects models.
Network Meta-analysis
The primary objectives of this study were to 1) assess the risk of type 2 diabetes associated with current definitions of metabolic health within the lean, overweight, and obese categories and 2) to assess the predictive relevance of these definitions. However, only one study (14) reported RR comparing unhealthy and healthy individuals within each BMI category. In all the other instances, articles reported the risk of all groups relative to the metabolically healthy lean (6,15–28). These comparisons may be of limited value when assessing the predictive relevance of metabolic health definitions. For instance, a comparison of metabolically unhealthy obese versus metabolically healthy lean simultaneously evaluates the contribution of two risk factors (i.e., BMI and level of metabolic health).
In instances where there was a need to contact the authors for clarifications, we asked for additional analyses within BMI categories (Supplementary Table 1). When there was no need to contact authors, we inferred the RR in the overweight and obese categories using network meta-analysis within each study (Supplementary Table 1 and Text). Network meta-analysis is a meta-analysis approach allowing indirect comparison of evidence, which has previously been applied in the context of randomized controlled trials (29,30). Here, we used this method in order to estimate, for each study, the RR of metabolically unhealthy versus healthy individuals in the overweight and obese categories. To do this, we used the comparisons with the metabolically healthy lean category as input evidence (Supplementary Data).
Meta-analysis of RR Estimates
In the second stage of our meta-analysis, adjusted estimates of RR were pooled across studies using random-effects models. Random-effects models were used because of the heterogeneity in the definitions of metabolic health. Analyses were stratified by sex (i.e., studies with proportion of women ≤30% vs. all other studies), age (i.e., mean or median age in the 3rd or 4th decade vs. 5th or 6th decade), country, or population (i.e., non–East Asian vs. East Asian, European vs. other non–East Asian vs. East Asian), definition of metabolic health (i.e., metabolic syndrome vs. insulin resistance), sample size (i.e., fewer than vs. more than 5,000 participants), length of follow-up (fewer than vs. more than 7 years of mean or median follow-up), or extent of adjustment (i.e., crude or minimal vs. extensive). Because three of the 16 selected studies were conducted in the same source population with overlapping but not identical criteria of selection and follow-up periods, only the largest of these studies was included in the main analysis. Analyses with either of the other two studies, instead of the largest, yielded comparable results (Supplementary Table 2). In studies using more than one definition, we arbitrarily chose one for inclusion in our main analysis. In doing so, we chose those analyses using metabolic syndrome as a criterion for the definition of metabolic health (i.e., in the instance of Meigs et al. [18] and Arnlöv et al. [19]) and those with more relaxed criteria for the definition of metabolic health (i.e., in the instance of Bell, Kivimaki, and Hamer [6]). The Egger test and funnel plots were used to assess publication bias. The I2 statistic was used to quantify heterogeneity. The source of heterogeneity was investigated by the aforementioned stratified analyses.
Absolute Type 2 Diabetes Risk Estimation
The cumulative incidence of type 2 diabetes in metabolic health and BMI categories was estimated using a probabilistic analysis (20,000 simulations), which simultaneously incorporates the uncertainty in the estimates of each of the input parameters. Input parameters in the model were 1) the proportion of participants who were metabolically healthy within each of the three BMI categories in our meta-analysis of 140,845 participants, 2) the RR of type 2 diabetes within each BMI category as estimated by our random-effects model meta-analysis, and 3) the cumulative incidence of type 2 diabetes at 5 and 10 years of follow-up from the European Prospective Investigation into Cancer and Nutrition (EPIC)-InterAct case-cohort study (31).
Predictive Relevance
The predictive relevance of metabolic health definitions was studied by hierarchical summary receiver operating characteristic curves and Fagan nomograms. The meta-analysis of predictive test accuracy was performed by fitting a two-level mixed-effects logistic regression model, with independent binomial distributions for the true positives and true negatives conditional on the sensitivity and specificity in each study and a bivariate normal model for the logit transforms of sensitivity and specificity between studies. Sensitivity and specificity were expressed as proportions, and their use in this study pertains to the prediction of future disease in diabetes-free individuals. Fagan nomograms represent the predictive performance of a test with lines intercepting three parallel vertical axes. The leftmost axis indicates the pretest probability of disease, the rightmost axis the posttest probability, and the central axis the positive or negative likelihood ratio. Lines are then drawn from the pretest probability on the left through the likelihood ratio in the center and extended to the posterior probabilities on the right to represent scenarios of a positive or negative test result in each of the three BMI categories. Likelihood ratios were used to evaluate contribution to prediction using cutoffs recommended by the authors of the fagan STATA package (32).
Statistical Analysis
Analyses were carried out using STATA, version 13.1 (StataCorp, College Station, TX). The network package was used for network meta-analysis (30), metan for random-effects model meta-analysis (33), metandi (32) and fagan (http://fmwww.bc.edu/repec/bocode/f/fagan.ado; author, Ben Dwamena, Division of Nuclear Medicine, Department of Radiology, University of Michigan Health System, Ann Arbor, MI) for meta-analysis of predictive test accuracy (32). The probabilistic analysis for estimation of cumulative incidences was carried out using openBUGS (34).
Results
Literature Search
Figure 1 depicts the literature search work flow. A total of 3,122 MEDLINE records were reviewed, and 126 articles matching our search criteria were selected for data extraction. The articles reported a total of 177 analyses using a definition of metabolic health. Definitions of metabolic health were mostly binary (i.e., healthy or unhealthy) and fell into five broad categories: 1) metabolic syndrome, 2) metabolic syndrome combined with insulin resistance or other criteria, 3) insulin resistance, 4) cardiorespiratory fitness, and 5) miscellaneous. Metabolic syndrome defined according to the Adult Treatment Panel III criteria (35), insulin resistance according to HOMA (36), and combined definitions provided by Wildman et al. (7) and Karelis, Brochu, and Rabasa-Lhoret (8) were most frequently used. Supplementary Table 3 reports the breakdown of the 177 definitions into these five broad categories.
Work flow of the review. aSearch 2, as described in Research Design and Methods. bSearch 1, as described in Research Design and Methods. cArticle restricted to unhealthy lean or used BMI as the end point/only risk factor rather than stratifying variable. MHO, metabolically healthy obesity.
Work flow of the review. aSearch 2, as described in Research Design and Methods. bSearch 1, as described in Research Design and Methods. cArticle restricted to unhealthy lean or used BMI as the end point/only risk factor rather than stratifying variable. MHO, metabolically healthy obesity.
We identified a total of sixteen manuscripts reporting on cohort studies of incident type 2 diabetes in BMI and metabolic health categories (6,14–28) (Table 1). Studies were mostly population-based cohort studies in middle-aged individuals with a mean or median follow-up ranging from 4 to 17.5 years. Qualitative assessment revealed generally high quality (Table 1). In the main analysis, we retained only the largest (25) of three studies that investigated the same source population (16,25,28). A total of fourteen studies, comprising 140,845 participants and 5,963 incident cases, were included in the quantitative meta-analysis (6,14,15,17–27).
Cohort studies investigating incident type 2 diabetes in metabolic health and BMI categories
| First author, year; PubMed identification no. | Country/years of recruitment | Definition of metabolic healtha | Definition of BMI categories (kg/m2) | Study population | Sample size and incident cases (Ntotal/NT2D) | Incident T2D ascertainment | Length of follow-up (years) | Adjustment | Quality scoreb |
|---|---|---|---|---|---|---|---|---|---|
| Meigs, 2006; 16735483 | Framingham Offspring Study, U.S./1991–1995 | 1) MetS: 3 of 5 ATPIII criteria (FG <5.6, WC ≤88 102 M or 88 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85); 2) IR: HOMA-IR ≤75th percentile of distribution in subjects without T2D | Lean <25, OW 25–29.9, obese ≥30 | Offspring of community-based study; European ancestry, free from CVD and T2D at baseline, mean age 54 years, women 55% | 1) 2,902/141; 2) 2,803/135 | FG ≥7.0 or new use of hypoglycemic therapy | Mean 6.8 | Age, sex, family history of diabetes, impaired glucose tolerance | 6 |
| Arnlöv, 2011; 20852030 | Uppsala Longitudinal Study of Adult Men, Sweden/1970–1973 | 1) MetS: 3 of 5 ATPIII criteria (FG <6.1,c BMI ≥29.4,d TG <1.7, HDL ≥1.04, BP <130/85); 2) IR: HOMA-IR ≤75th percentile of distribution in subjects without T2D, i.e., 3.43 | Lean <25, OW 25–30, obese >30 | Community-based study of men born in 1920–1924 free from T2D at baseline, mean age 50 years, women 0% | 1) 1,675/160; 2) 1,385/117 | FG ≥7.0 at follow-up or data from national hospital-discharge registry | Up to 20 | Age, smoking status, level of physical activity | 6 |
| Hadaegh, 2011; 21609497 | Tehran Lipid and Glucose Study, Iran/1999–2001 | MetS: 3 of 5 harmonized criteria (FG <5.5, WC <94.5, TG <1.7, HDL ≥1.04 M or 1.3 W, BP <130/85) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study in Tehran; mean age 42 years, women 58% | 5,250/369 | Self-reported or OGTT based at 2 follow-up visits | Median 6.5 | Age, family history of T2D, history of CVD, education, smoking status | 6 |
| Kim, 2012; 22621338 | South Korea/2005 | MetS: 3 of 5 2009 harmonized criteria (FG <5.6, WC <90 M or 80 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5 | Subjects attending baseline and follow-up visits at Health Promotion Centre; mean age 48 years, women 35% | 8,748/308 | FG ≥7.0 or HbA1c ≥6.5% or treatment | 5 | Age, sex, smoking, alcohol consumption, physical activity | 6 |
| Bo, 2012; 23034958 | Italy/2001–2003 | MetS plus IR: 3 of 5 harmonized criteria (FG <5.6, WC <94 M or 80 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85,and HOMA-IR <2.5) | Lean <25, OW 25–30, obese >30 | Caucasian volunteers from local health units; mean age 54 years, women 53% | 1,658/72 | Self-reported, FG, demographic registries | 9 | Nonee | 6 |
| Appleton, 2013; 23491523 | North West Adelaide Health Study, Australia/1999–2003 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean 18.5–24.9, OW 25–29.9, obese ≥30 | Community-based study; adults of European ancestry, free from T2D and CVD at baseline; mean age 42 yearsf, women 57%f | 2,315/112 | Self-reported doctor diagnosis or FG ≥7.0 | Median 8.2 | Age, sex, household income, family history of diabetes | 6 |
| Soriguer, 2013; 23559087 | Prospective Pizarra Study, Spain/1997–1998 | MetS plus IR: 3 of 3 criteria (FG <6.1, TG <1.7, HOMA-IR <90th percentileg) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study; mean age 40 years, women, 62%h | 387/38e | Self-reported or FG at follow-up | 11 | Age, sexe | 5 |
| Aung, 2014; 24257907 | San Antonio Heart Study, U.S./1979–1988 | MetS plus IR: 4 of 5 criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85, HOMA-IR ≤5.13) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study of Mexican and Caucasian Americans; mean age 42 years, women 57% | 2,814/262 | OGTT or medication at follow-up | Median 7.4 | Age, sex, ethnicity, family history of diabetes, FG | 6 |
| Sung, 2012; 22338098 | South Korea/2003 | IR: HOMA-IR <2, i.e., 75th percentile | Lean <23, OW 23–27.49, obese ≥27.5e | Participants of health examination at hospital; mean age 41 years, women 29% | 12,853/223 | Self-reported, medical history, or FG at follow-up | 5 | Age, sex, alcohol, smoking status, exercise, educational status, baseline glucosee | 5 |
| Bell, 2014; 24661566 | English Longitudinal Study of Ageing, U.K./2004–2005 | Definition 1, MetS plus CRP: 3 of 5 customized criteria (HbA1c <6%, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85, CRP <0.3 mg/dL); definition 2, MetS plus CRP: 4 of 5 of the criteria used in definition 1 | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study; mean age 65 years; women 57% | 3,060/138 | Self-reported physician diagnosis | Mean 5.9 | Age, sex, smoking, alcohol intake, physical activity, wealth, depressive symptoms | 5 |
| Jung, 2014; 24706434 | South Korea/2005–2006 | MetS: 4 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean <23, OW 23–27.4, Obese ≥27.5e | Cohort study of employees of large Korean company and their spouses; mean age 37 years; women, 44% | 34,994/889e | FG ≥7, HbA1c ≥6.5%, or medication | 5 | Age, sex, smoking, alcohol intake, physical activitye | 6 |
| Heianza, 2014; 24823457 | Toranomon Hospital Health Management Centre Study, Japan/1997–2002 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1.03 M or 1.29 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5e | Cohort study of Japanese government employees; mean age 48 years, women 27% | 8,090/274e | FG ≥7, HbA1c ≥6.5%, or self-reported | 5 | Age, sex, smoking, physical activity, alcohol intake, family history of diabetese | 6 |
| Rhee, 2014; 24870949 | South Korea/2005 | MetS plus IR: 3 of 4 criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85 and HOMA-IR <90th percentile) | Lean <23, OW 23–27.4, obese ≥27.5e | Participants of medical check-up program; mean age 43 years, women 27% | 6,748/277 | FG, HbA1c, or self-reported history or medication | 4 | Age, sex, ALT, creatinine, total cholesterol, hs-CRPe | 6 |
| Heianza, 2014; 25131796 | Japan/1999–2004 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1.03 M or 1.29 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5e | Cohort study of individuals occupational health examinations; mean age 47 years; women 36% | 27,891/1,668e | FG ≥7, HbA1c ≥6.5%, or self-reported | 8 | Age, sex, smoking, physical activitye | 6 |
| Twig, 2014; 25139886 | Metabolic, Lifestyle and Nutrition Assessment in Young Adults, Israel/1995–2011 | MetS: 3 of 4 ATPIII criteria (FG <5.6, TG <1.7, HDL ≥1, BP <130/85e,i) | Lean <25, OW 25–29.9, obese ≥30 | Cohort study of men from the Israel Defense Forces; mean age 31 years; women 0% | 33,939/734 | FG or physician diagnosis | Median 6.1 | Age, family history of diabetes, country of origin, WBCe | 6 |
| Hinnouho, 2015; 24670711 | Whitehall II Study, U.K./1991–1993 | MetS: 3 of 4 ATPIII criteria (FG <5.6, TG <1.7, HDL ≥1.04 M or 1.29 W, BP <130/85) | Lean 18.5–24.9, OW 25–29.9, obese ≥30 | Cohort study of office workers in central London; mean age 49 years; women 30% | 7,122/798 | OGTT, physician diagnosis, or use of medication at follow-up | Median 17.5 | Sex, socioeconomic status, marital status, ethnicity, physical activity, smoking, alcohol, fruit and vegetable consumption, CVD medications and procedures | 6 |
| First author, year; PubMed identification no. | Country/years of recruitment | Definition of metabolic healtha | Definition of BMI categories (kg/m2) | Study population | Sample size and incident cases (Ntotal/NT2D) | Incident T2D ascertainment | Length of follow-up (years) | Adjustment | Quality scoreb |
|---|---|---|---|---|---|---|---|---|---|
| Meigs, 2006; 16735483 | Framingham Offspring Study, U.S./1991–1995 | 1) MetS: 3 of 5 ATPIII criteria (FG <5.6, WC ≤88 102 M or 88 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85); 2) IR: HOMA-IR ≤75th percentile of distribution in subjects without T2D | Lean <25, OW 25–29.9, obese ≥30 | Offspring of community-based study; European ancestry, free from CVD and T2D at baseline, mean age 54 years, women 55% | 1) 2,902/141; 2) 2,803/135 | FG ≥7.0 or new use of hypoglycemic therapy | Mean 6.8 | Age, sex, family history of diabetes, impaired glucose tolerance | 6 |
| Arnlöv, 2011; 20852030 | Uppsala Longitudinal Study of Adult Men, Sweden/1970–1973 | 1) MetS: 3 of 5 ATPIII criteria (FG <6.1,c BMI ≥29.4,d TG <1.7, HDL ≥1.04, BP <130/85); 2) IR: HOMA-IR ≤75th percentile of distribution in subjects without T2D, i.e., 3.43 | Lean <25, OW 25–30, obese >30 | Community-based study of men born in 1920–1924 free from T2D at baseline, mean age 50 years, women 0% | 1) 1,675/160; 2) 1,385/117 | FG ≥7.0 at follow-up or data from national hospital-discharge registry | Up to 20 | Age, smoking status, level of physical activity | 6 |
| Hadaegh, 2011; 21609497 | Tehran Lipid and Glucose Study, Iran/1999–2001 | MetS: 3 of 5 harmonized criteria (FG <5.5, WC <94.5, TG <1.7, HDL ≥1.04 M or 1.3 W, BP <130/85) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study in Tehran; mean age 42 years, women 58% | 5,250/369 | Self-reported or OGTT based at 2 follow-up visits | Median 6.5 | Age, family history of T2D, history of CVD, education, smoking status | 6 |
| Kim, 2012; 22621338 | South Korea/2005 | MetS: 3 of 5 2009 harmonized criteria (FG <5.6, WC <90 M or 80 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5 | Subjects attending baseline and follow-up visits at Health Promotion Centre; mean age 48 years, women 35% | 8,748/308 | FG ≥7.0 or HbA1c ≥6.5% or treatment | 5 | Age, sex, smoking, alcohol consumption, physical activity | 6 |
| Bo, 2012; 23034958 | Italy/2001–2003 | MetS plus IR: 3 of 5 harmonized criteria (FG <5.6, WC <94 M or 80 W, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85,and HOMA-IR <2.5) | Lean <25, OW 25–30, obese >30 | Caucasian volunteers from local health units; mean age 54 years, women 53% | 1,658/72 | Self-reported, FG, demographic registries | 9 | Nonee | 6 |
| Appleton, 2013; 23491523 | North West Adelaide Health Study, Australia/1999–2003 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean 18.5–24.9, OW 25–29.9, obese ≥30 | Community-based study; adults of European ancestry, free from T2D and CVD at baseline; mean age 42 yearsf, women 57%f | 2,315/112 | Self-reported doctor diagnosis or FG ≥7.0 | Median 8.2 | Age, sex, household income, family history of diabetes | 6 |
| Soriguer, 2013; 23559087 | Prospective Pizarra Study, Spain/1997–1998 | MetS plus IR: 3 of 3 criteria (FG <6.1, TG <1.7, HOMA-IR <90th percentileg) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study; mean age 40 years, women, 62%h | 387/38e | Self-reported or FG at follow-up | 11 | Age, sexe | 5 |
| Aung, 2014; 24257907 | San Antonio Heart Study, U.S./1979–1988 | MetS plus IR: 4 of 5 criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85, HOMA-IR ≤5.13) | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study of Mexican and Caucasian Americans; mean age 42 years, women 57% | 2,814/262 | OGTT or medication at follow-up | Median 7.4 | Age, sex, ethnicity, family history of diabetes, FG | 6 |
| Sung, 2012; 22338098 | South Korea/2003 | IR: HOMA-IR <2, i.e., 75th percentile | Lean <23, OW 23–27.49, obese ≥27.5e | Participants of health examination at hospital; mean age 41 years, women 29% | 12,853/223 | Self-reported, medical history, or FG at follow-up | 5 | Age, sex, alcohol, smoking status, exercise, educational status, baseline glucosee | 5 |
| Bell, 2014; 24661566 | English Longitudinal Study of Ageing, U.K./2004–2005 | Definition 1, MetS plus CRP: 3 of 5 customized criteria (HbA1c <6%, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85, CRP <0.3 mg/dL); definition 2, MetS plus CRP: 4 of 5 of the criteria used in definition 1 | Lean <25, OW 25–29.9, obese ≥30 | Population-based cohort study; mean age 65 years; women 57% | 3,060/138 | Self-reported physician diagnosis | Mean 5.9 | Age, sex, smoking, alcohol intake, physical activity, wealth, depressive symptoms | 5 |
| Jung, 2014; 24706434 | South Korea/2005–2006 | MetS: 4 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85) | Lean <23, OW 23–27.4, Obese ≥27.5e | Cohort study of employees of large Korean company and their spouses; mean age 37 years; women, 44% | 34,994/889e | FG ≥7, HbA1c ≥6.5%, or medication | 5 | Age, sex, smoking, alcohol intake, physical activitye | 6 |
| Heianza, 2014; 24823457 | Toranomon Hospital Health Management Centre Study, Japan/1997–2002 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1.03 M or 1.29 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5e | Cohort study of Japanese government employees; mean age 48 years, women 27% | 8,090/274e | FG ≥7, HbA1c ≥6.5%, or self-reported | 5 | Age, sex, smoking, physical activity, alcohol intake, family history of diabetese | 6 |
| Rhee, 2014; 24870949 | South Korea/2005 | MetS plus IR: 3 of 4 criteria (FG <5.6, TG <1.7, HDL ≥1 M or 1.3 W, BP <130/85 and HOMA-IR <90th percentile) | Lean <23, OW 23–27.4, obese ≥27.5e | Participants of medical check-up program; mean age 43 years, women 27% | 6,748/277 | FG, HbA1c, or self-reported history or medication | 4 | Age, sex, ALT, creatinine, total cholesterol, hs-CRPe | 6 |
| Heianza, 2014; 25131796 | Japan/1999–2004 | MetS: 3 of 4 IDF criteria (FG <5.6, TG <1.7, HDL ≥1.03 M or 1.29 W, BP <130/85) | Lean <23, OW 23–27.4, obese ≥27.5e | Cohort study of individuals occupational health examinations; mean age 47 years; women 36% | 27,891/1,668e | FG ≥7, HbA1c ≥6.5%, or self-reported | 8 | Age, sex, smoking, physical activitye | 6 |
| Twig, 2014; 25139886 | Metabolic, Lifestyle and Nutrition Assessment in Young Adults, Israel/1995–2011 | MetS: 3 of 4 ATPIII criteria (FG <5.6, TG <1.7, HDL ≥1, BP <130/85e,i) | Lean <25, OW 25–29.9, obese ≥30 | Cohort study of men from the Israel Defense Forces; mean age 31 years; women 0% | 33,939/734 | FG or physician diagnosis | Median 6.1 | Age, family history of diabetes, country of origin, WBCe | 6 |
| Hinnouho, 2015; 24670711 | Whitehall II Study, U.K./1991–1993 | MetS: 3 of 4 ATPIII criteria (FG <5.6, TG <1.7, HDL ≥1.04 M or 1.29 W, BP <130/85) | Lean 18.5–24.9, OW 25–29.9, obese ≥30 | Cohort study of office workers in central London; mean age 49 years; women 30% | 7,122/798 | OGTT, physician diagnosis, or use of medication at follow-up | Median 17.5 | Sex, socioeconomic status, marital status, ethnicity, physical activity, smoking, alcohol, fruit and vegetable consumption, CVD medications and procedures | 6 |
ALT, alanine aminotransferase; ATPIII, Adult Treatment Panel III; BP, blood pressure; CRP, C-reactive protein; CVD, cardiovascular disease; FG, fasting glucose; HOMA-IR, HOMA of insulin resistance; IDF, International Diabetes Federation; IR, insulin resistance; MetS, metabolic syndrome; M, men; OGTT, oral glucose tolerance test; OW, overweight; T2D, type 2 diabetes; TG, triglycerides; W, women; WBC, white blood cells; WC, waist circumference.
aAll cutoff values expressed as cm for waist circumference; mmol/L for fasting glucose, HDL cholesterol, or triglycerides; kg/m2 for BMI; and mmHg for blood pressure. In some studies, treatment with medication (e.g., antihypertensive drugs) was used as a complementary factor to adjudicate metabolic risk criteria.
bWe used a quality score similar to the one reported by Bell, Kivimaki, and Hamer (6). Study quality was assessed according to the definition of exposure, outcome, and to the extent of adjustment. Points were assigned as follows: 2 points if the study considered metabolic risk factor clustering as in the metabolic syndrome; 1 point if the study considered insulin resistance only; 2 points if diabetes diagnosis was based on objective clinical measurements (e.g., fasting or 2-h glucose levels); 1 point if diabetes adjudication was based on self-report only; 2 points for extensive adjustment, i.e., age and sex plus at least two of the following: family history of diabetes, ethnicity, alcohol consumption, smoking status, physical activity, dietary habits and socioeconomic status, and impaired glucose tolerance status; 1 point for basic adjustment, i.e., age and sex; and 0 points for crude estimates. Studies were scored out of 6 possible points. Adjustment in the original report was used to adjudicate the extent of adjustment.
cOf fasting blood glucose, corresponding to fasting plasma glucose of 5.6 mmol/L.
dBMI used in lieu of waist circumference criterion.
eInformation as reported by the authors in a personal communication (G. Twig, A. Tirosh, C. Gutiérrez Repiso, G. Rojo Martínez, F. Soriguer, K.-C. Sung, S. Bo, Y. Heianza, H. Sone, Y. Arase, K. Kato, S.R.Y. Chang, E.-J. Rhee, and W.-Y. Lee).
fAverage of metabolically healthy lean, metabolically healthy obese, metabolically unhealthy obese groups.
gThe authors reported 4 different definitions of metabolic health. Here, we report the one we used in the meta-analysis, for which the authors provided detailed results of type 2 diabetes incidence in a personal communication (C. Gutiérrez Repiso, G. Rojo Martínez, and F. Soriguer).
hIn the full baseline study.
iIn the original report, metabolic health was the absence of any metabolic syndrome criteria and the risk of type 2 diabetes was evaluated for individuals with 1, 2, 3, or more criteria separately.
Network Meta-analysis and Random-Effects Meta-analysis of RR Estimates
In a meta-analysis of type 2 diabetes risk within BMI and metabolic health categories, all groups had higher risk compared with the healthy lean group (metabolically unhealthy lean group RR 4.0 [95% CI 3.0–5.1], metabolically healthy overweight group 1.8 [1.5–2.2], metabolically unhealthy overweight group 6.2 [4.8–8.0], metabolically healthy obese group 4.1 [3.3–5.1], metabolically unhealthy obese group 10.9 [8.5–13.9]). (See also Supplementary Table 4 and Supplementary Fig. 1.)
Where analysis results were not available from the manuscript or from direct contact with the authors, network meta-analysis was used to derive within-overweight and within-obese category RRs and 95% CI. In instances where both estimates were available, the central estimate of RR provided by the authors strongly correlated with that obtained by network meta-analysis (average r2 > 0.95; n = 18). (See Supplementary Table 1.) The RRs obtained in this first stage were then pooled using random-effects meta-analysis. Random-effects meta-analysis revealed that—when compared with healthy individuals—metabolically unhealthy individuals are at higher risk of type 2 diabetes in all BMI categories (compared with healthy individuals, lean RR 4.0 [95% CI 3.0–5.1], Nat risk/type 2 diabetes = 67,281/1,393; overweight 3.4 [2.8–4.3], Nat risk/type 2 diabetes = 58,060/2,903; obese, 2.5 [2.1–3.0], Nat risk/type 2 diabetes = 15,504/1,667) (Fig. 2). The RR of type 2 diabetes associated with classification as unhealthy was highest in the lean category and lowest in the obese category. Funnel plots and Egger test (P > 0.1 for all comparisons) indicated no publication bias (Supplementary Fig. 2). There was heterogeneity between studies in the estimated RR within the lean and overweight categories but not the obese category (Fig. 2). Stratified analyses revealed that ethnicity was likely the driver of heterogeneity, with RR higher in East Asian populations in all BMI categories (Fig. 2). Further stratification for geographic region (European vs. East Asian studies) resolved residual heterogeneity in estimates in the lean category (Supplementary Fig. 3). Results were similar in studies using metabolic syndrome or insulin resistance as definitions for metabolic health.
RR of type 2 diabetes in metabolically unhealthy compared with healthy individuals by BMI category and country (non–East Asian or East Asian).
RR of type 2 diabetes in metabolically unhealthy compared with healthy individuals by BMI category and country (non–East Asian or East Asian).
Absolute Type 2 Diabetes Risk Estimation and Predictive Relevance
Using a probabilistic analysis, which simultaneously incorporates uncertainty in all the input parameter estimates, we estimated the cumulative incidence of type 2 diabetes at 5 and 10 years in all BMI and metabolic health categories (Table 2). Metabolically healthy obese individuals had a cumulative incidence of type 2 diabetes over 10 years of 3.1% (95% CI 2.6–3.5). Cumulative incidence estimates from a sensitivity analysis after exclusion of studies in East Asian populations were largely overlapping with those of the main analysis (Supplementary Table 5).
Cumulative incidence of type 2 diabetes in metabolic health and BMI categories
| BMI category | BMI category cumulative incidence at 5 yearsa | BMI category cumulative incidence at 10 yearsa | Risk category | Proportion of healthy or unhealthy individuals in each BMI category | RR within BMI category | Risk category 5-year cumulative incidence (95% CI)* | Risk category 10-year cumulative incidence (95% CI)* |
|---|---|---|---|---|---|---|---|
| Lean | 0.3% | 0.8% | Metabolically healthy lean; metabolically unhealthy lean | 0.82;0.18 | 1;4.0 | 0.2% (0.1–0.2); 0.6% (0.6–0.8) | 0.5% (0.5–0.6); 2.2% (1.9–2.5) |
| Overweight | 0.8% | 2.7% | Metabolically healthy overweight; metabolically unhealthy overweight | 0.59;0.41 | 1;3.4 | 0.4% (0.3–0.5); 1.4% (1.3–1.5) | 1.3% (1.1–1.6); 4.5% (4.2–4.9) |
| Obese | 1.8% | 5.9% | Metabolically healthy obese; metabolically unhealthy obese | 0.38;0.62 | 1;2.5 | 1.0% (0.8–1.1); 2.4% (2.2–2.5) | 3.1% (2.6–3.5); 7.6% (7.3–8.0) |
| BMI category | BMI category cumulative incidence at 5 yearsa | BMI category cumulative incidence at 10 yearsa | Risk category | Proportion of healthy or unhealthy individuals in each BMI category | RR within BMI category | Risk category 5-year cumulative incidence (95% CI)* | Risk category 10-year cumulative incidence (95% CI)* |
|---|---|---|---|---|---|---|---|
| Lean | 0.3% | 0.8% | Metabolically healthy lean; metabolically unhealthy lean | 0.82;0.18 | 1;4.0 | 0.2% (0.1–0.2); 0.6% (0.6–0.8) | 0.5% (0.5–0.6); 2.2% (1.9–2.5) |
| Overweight | 0.8% | 2.7% | Metabolically healthy overweight; metabolically unhealthy overweight | 0.59;0.41 | 1;3.4 | 0.4% (0.3–0.5); 1.4% (1.3–1.5) | 1.3% (1.1–1.6); 4.5% (4.2–4.9) |
| Obese | 1.8% | 5.9% | Metabolically healthy obese; metabolically unhealthy obese | 0.38;0.62 | 1;2.5 | 1.0% (0.8–1.1); 2.4% (2.2–2.5) | 3.1% (2.6–3.5); 7.6% (7.3–8.0) |
aFrom the EPIC-InterAct Study.
*Derived via a probabilistic analysis, which simultaneously incorporates the uncertainty in each of the parameter estimate (openBUGS) (20,000 simulations).
Hierarchical summary receiver operating curve analysis revealed increasing sensitivity and decreasing specificity in higher BMI categories (Fig. 3A–C). In lean individuals, testing metabolic health was not sensitive (pooled estimate 0.40 [95% CI 0.310.49]) but was specific (0.88 [0.84–0.91]). In overweight individuals, both sensitivity (0.65 [0.56–0.74]) and specificity (0.68 [0.61–0.74]) were low. In obese individuals, sensitivity was acceptable (0.81 [0.76–0.86]), but specificity (0.42 [0.35–0.49]) was low. The pooled estimates of the positive likelihood ratio were 3.3 (95% CI 2.7–3.9) in the lean category, 2.0 (1.8–2.3) in the overweight category, and 1.4 (1.3–1.5) in the obese category. The pooled estimates for the negative likelihood ratio were 0.7 (95% CI 0.6–0.8) in the lean category, 0.5 (0.4–0.6) in the overweight category, and 0.5 (0.4–0.5) in the obese category. These likelihood ratios show that current binary definitions of metabolic health make only small or insignificant contributions to the prediction of future type 2 diabetes in BMI categories (Fig. 3D and E).
![Figure 3. Performance of metabolic health definitions in the prediction of future development of type 2 diabetes. The top panels report hierarchical summary receiver operating characteristic (HSROC) curves of the predictive performance in lean (A), overweight (B), and obese (C) individuals. Solid squares, pooled estimates; open circles, individual study estimates with a size that is proportional to the weight of each study; solid lines, HSROC curves; broken lines, uncertainty in the estimates of the summary points (95% confidence region) or of the HSROC curves (95% prediction region). In D and E, Fagan nomograms represent scenarios of positive (i.e., metabolically unhealthy [D]) or negative (i.e., metabolically healthy individual [E]) results of the binary classification of metabolic health in BMI categories.](/view-large/figure/3949432/dc151218f3.jpeg)
Performance of metabolic health definitions in the prediction of future development of type 2 diabetes. The top panels report hierarchical summary receiver operating characteristic (HSROC) curves of the predictive performance in lean (A), overweight (B), and obese (C) individuals. Solid squares, pooled estimates; open circles, individual study estimates with a size that is proportional to the weight of each study; solid lines, HSROC curves; broken lines, uncertainty in the estimates of the summary points (95% confidence region) or of the HSROC curves (95% prediction region). In D and E, Fagan nomograms represent scenarios of positive (i.e., metabolically unhealthy [D]) or negative (i.e., metabolically healthy individual [E]) results of the binary classification of metabolic health in BMI categories.
Performance of metabolic health definitions in the prediction of future development of type 2 diabetes. The top panels report hierarchical summary receiver operating characteristic (HSROC) curves of the predictive performance in lean (A), overweight (B), and obese (C) individuals. Solid squares, pooled estimates; open circles, individual study estimates with a size that is proportional to the weight of each study; solid lines, HSROC curves; broken lines, uncertainty in the estimates of the summary points (95% confidence region) or of the HSROC curves (95% prediction region). In D and E, Fagan nomograms represent scenarios of positive (i.e., metabolically unhealthy [D]) or negative (i.e., metabolically healthy individual [E]) results of the binary classification of metabolic health in BMI categories.
Conclusions
In this study, we reviewed definitions of metabolic health, which have been used to classify the risk of metabolic disease in BMI categories. We also sought to assess the risk of type 2 diabetes associated with being classified as metabolically unhealthy in lean, overweight, and obese individuals. We found that being classified as metabolically unhealthy is associated with higher risk of type 2 diabetes relative to the healthy group in all categories of BMI.
Our study is the largest meta-analysis of the risk of type 2 diabetes associated with metabolic health definitions and the only one to have assessed risk within BMI categories. In a meta-analysis by Bell, Kivimaki, and Hamer (6) of eight studies, with a total of 27,982 participants, RR was calculated using the healthy lean category as a reference. This is not informative about the risk of type 2 diabetes within BMI categories or the predictive relevance of metabolic health definitions. In this study, we obtained within BMI-category risk estimates using network meta-analysis, a method for the pooling of indirect evidence used in meta-analyses of randomized controlled trials. The method accurately estimated the RR in BMI categories, with a loss of precision, due to the uncertainty of indirect estimations, which was offset by the large overall sample size of this meta-analysis. Using this method, we were able to show that the RR of type 2 diabetes is higher in all categories of BMI for metabolically unhealthy individuals compared with those who are classified as metabolically healthy. RRs within the lean and overweight categories were higher in East Asian populations. This probably reflects the higher prevalence of abdominal obesity and insulin resistance in these populations compared with Europeans at a given level of BMI (37,38).
However, RR only partially accounts for the predictive relevance of a given definition. Predictive relevance has to be evaluated also in the context of absolute risk. The absolute risk of individuals deemed to be “metabolically healthy obese” was high, with an estimated cumulative incidence of type 2 diabetes over 10 years exceeding 3%. This raises doubts about the predictive value of currently used binary definitions of metabolic health. An analysis of hierarchical summary receiver operating characteristics curve and Fagan nomograms revealed limited predictive relevance in all three BMI categories.
Metabolic health definitions had a predictive performance opposite of the desirable, with low sensitivity in the lean category and low specificity in the obese category. In the lean category, metabolic health definitions had high specificity and could therefore be considered as a confirmatory test. However, there is presently no screening test to identify at-risk lean individuals in the population. With an absolute risk of 2.2% at 10 years, one would argue that metabolically unhealthy lean individuals would not be candidates for particular preventive measures besides those recommended for the general population. In overweight individuals, current binary definitions of metabolic health had low sensitivity and specificity. The metabolically healthy overweight individuals had an absolute risk of type 2 diabetes greater than that of the lean category and the metabolically unhealthy overweight individuals an absolute risk smaller than that of the obese category. In addition, the metabolically unhealthy overweight group accounted for 40% of the overweight category, which is in many countries the largest BMI category in the general population. Therefore, it is difficult to conceive preventive measures that could efficiently target such a large portion of the population. In obese individuals, using current metabolic health definitions may be sensitive. However, specificity in this group was well below acceptable levels and “metabolically healthy” obese individuals still had an absolute risk greater than that of the overweight category.
In addition to these limitations, defining metabolic health entails invasive biological sampling for the measurement of biomarkers such as glucose, triglycerides, HDL cholesterol, or insulin. In particular, the criteria for some of the definitions of metabolic health include fasting glucose, which is a major predictor of type 2 diabetes (39). Therefore, the value of using any of the other criteria (e.g., HDL cholesterol or triglycerides) included in these definitions in addition to fasting glucose is likely to be limited. Overall, there is little support for use of these definitions for the prediction or classification of type 2 diabetes risk in BMI categories. These considerations apply to currently used binary definitions of metabolic health. It is possible that more comprehensive approaches to the definition of metabolic health may yield better predictive performance. Also, our meta-analytic approach pooled evidence from studies using different definitions of metabolic health. However, our analytical approach accounted for possible differences in the performance of definitions of metabolic health used in the constituent studies of the meta-analysis.
Summary
In conclusion, in a meta-analysis of 140,845 participants, being classified as metabolically unhealthy compared with healthy using current binary definitions of metabolic health was associated with higher RR of type 2 diabetes in all BMI categories. However, when considering predictive performance in the context of absolute risk, we found that current binary definitions of metabolic health have limited predictive relevance. Our study does not support the use of current definitions of metabolic health for the prediction or classification of type 2 diabetes risk.
Article Information
Acknowledgments. The authors acknowledge the role of the following investigators, who provided additional information on their published studies, as contributors of the study: Gilad Twig, Sheba Medical Center and the Israel Defense Forces Medical Corps, Ramat Gan, Israel; Amir Tirosh, Brigham and Women's Hospital and Harvard Medical School, Boston, MA; Carolina Gutiérrez Repiso, Unidad de Gestión Clínica Intercentros (UGCI) de Endocrinología y Nutrición, Hospital Regional Universitario de Málaga, Instituto de Investigación Biomédica de Málaga (IBIMA), Málaga, Spain; Gemma Rojo Martínez, UGCI de Endocrinología y Nutrición, Hospital Regional Universitario de Málaga, CIBERDEM, Málaga, Spain; Federico Soriguer, UGCI de Endocrinología y Nutrición, Hospital Regional Universitario de Málaga, IBIMA, Málaga, Spain; Ki-Chul Sung, Department of Cardiology, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, Korea; Simona Bo, Department of Medical Sciences, University of Turin, Turin, Italy; Yoriko Heianza and Hirohito Sone, Department of Internal Medicine, Niigata University Faculty of Medicine, Niigata, Japan; Yasuji Arase, Health Management Center, Toranomon Hospital, Tokyo, Japan; Kiminori Kato, Niigata Association of Occupational Health, Niigata, Japan; Seungho Ryu Yoosoo Chang, Department of Occupational and Environmental Medicine, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, Korea; Eun-Jung Rhee and Won-Young Lee, Department of Endocrinology and Metabolism, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, Korea; and the EPIC-InterAct Study Consortium.
Funding. The research leading to these results received support from the Innovative Medicines Initiative Joint Undertaking under EMIF grant agreement 115372, resources of which are composed of financial contribution from the European Union's Seventh Framework Programme (FP7/2007-2013) and European Federation of Pharmaceutical Industries and Associations companies’ in kind contribution. This work was supported by the Netherlands Organization for Scientific Research (NWO) and the Medical Research Council (grant no. MC_U106179471). A.A. is supported by a Rubicon grant from the NWO (project no. 825.13.004).
Duality of Interest. D.W. and A.-S.S. are full-time employees of GlaxoSmithKline. J.M.B. is a full-time employee of Pfizer. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. L.A.L., A.A., S.J.S., R.A.S., and N.J.W. conceived and planned the study. L.A.L., S.J.S., and A.-S.S. acquired and analyzed data. L.A.L., A.A., S.J.S., C.L., and N.J.W. drafted the work. All authors interpreted results, revised the draft for important intellectual content, gave final approval of the version to be published, and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
