Epidemiology of Obesity in Adults: Latest Trends

Search strategy

We used the following cut-off values to define overweight/obesity: overweight (BMI: 25–29.9), class I obesity (BMI: 30–34.9), class II obesity (BMI: 35–39.9) and class III obesity (BMI: ≥ 40) [18]. We defined abdominal obesity (i.e., cut-off values of waist circumference (WC) ≥ 102 cm (40 in) for men and WC ≥ 88 cm (35 in) for women) [19]. In addition, we considered the non-Hispanic Asian cut-points of WC ≥ 90 cm for men and WC ≥ 80 cm for women [20].

We systematically reviewed the literature using a protocol informed by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. We searched for articles published in English, peer-reviewed journals in PubMed. For the global study, we further restricted studies to systematic reviews or meta-analyses. We developed the search syntax in collaboration with a reference librarian (Supplementary Tables 1 and 2). Our query included the following terms, their cognates, and synonyms: obesity AND trend AND United States AND adults AND nationally representative (US studies); obesity AND trend AND global AND adults (global studies).

Inclusion/exclusion criteria

For the search of domestic studies, articles had to fit the following inclusion criteria: 1) estimates of obesity following our criteria above; 2) sampling to be nationally representative or sub-nationally representative of the US; 3) include adults over the age of 18 years; 4) measured (rather than self-report) anthropometry; 5) repeated cross-sectional or longitudinal study design; 6) peer-reviewed; 7) available in English; and 8) published between January 1, 2012 and July 1, 2018 (Supplementary table 1).

For the search of global studies, we restricted studies to systematic reviews and began with the seminal GBD paper [2] as a model and searched for studies that evaluated the prevalence or incidence of obesity using measured height and weight in countries outside the United States. To be included, articles had to fit the following inclusion criteria: 1) estimates of obesity following our criteria above; 2) estimates for areas outside the US; 3) adults over the age of 18 years; 4) measured (rather than self-report) anthropometry; 5) systematic review or meta-analysis; 6) peer-reviewed; 7) available in English; and 8) published between January 1, 2014 and July 1, 2018 (Supplementary table 2).

Study selection and data extraction

We used Covidence, an online platform, to manage screening and selection of studies. A single reviewer completed an initial independent screen of all titles and abstracts retrieved from the database searches. A second reviewer checked a random sampling of titles to guarantee no articles were falsely excluded at this stage, and the second reviewer found no such discrepancies. Three separate reviewers independently reviewed the full texts of studies, whereby each study was dually screened, to determine final study inclusion. All conflicts in the full text review were resolved via discussion with the authorship team.

In our electronic search for the domestic studies, we found 695 references, two of which were duplicates, resulting in a total of 693 studies. In the initial title and abstract screen, the research team deemed 620 studies irrelevant, leaving 73 full-texts to review. We retained and extracted data from 18 studies that met inclusion criteria (Figure 1). Our search for the global studies returned 141 references, of which we deemed 120 studies irrelevant on the basis of initial title and abstract screen, leaving 21 full-texts to review of which one fit inclusion criteria. In addition, we included the GBD study published in July 2017 [2] and used the reference lists of the papers captured in our search to add one NCD-RisC paper [17] (Figure 2).

Overall obesity trends in the US

Flegal et al. [30] estimated obesity prevalence to be 34.6% (men: 33.5%; women: 35.7%) in 2005–2006, which decreased in 2007–2008 to 33.9% (men: 32.2%; women: 35.5%) but then increased to 37.9% (men: 35.2%; women: 40.5%) in 2013–2014 (Table 2). Hales et al. [38], including the most recent cycle of the NHANES, reported that obesity prevalence in 2013–2016 to be 36.5% for men and 40.8% for women.

Ljungvall et al. [21] reported an obesity prevalence of 10% for men and 16% for women in 1959–1962, which increased to 18% for men and 23% for women in 1988–91, 27% for men and 34% for women in 1999–2000, and again to 32% for men and 36% for women in NHANES 2007 – 2008. The increasing trend in obesity prevalence was featured in the other studies in our search as well [22–24, 33, 35].

Several studies that used BMI cut-offs higher than 30 suggest a shift in population distribution for BMI towards the upper end of the BMI distribution. [21, 27, 30, 32]. For example, Ljungvall et al. [21] reported that prevalence of BMI ≥ 35, which was 1% and 5% for men and women, respectively in 1959, increased to 5% and 9% in 1988–1991 and finally to 11% and 19% in 2007–2008. Kranjac et al. [32] reported that the prevalence of BMI ≥ 40 was 1% in 1971, rising to 6% in 2012, with higher prevalence reported in women than in men. The prevalence of BMI ≥ 40 for men and women was 5.5% and 9.9% in 2013–2014 [30] and 5.5% and 9.8% in 2013–2016 [38].

Between 1988–1994 and 2009–2010, abdominal obesity prevalence increased from 29.1% to 42.0% among men and from 46.0% to 61.5% among women [28]. Robinson et al. [26] reported that in 1986–1990 abdominal obesity prevalence was 36.0% (27.5% for men and 44.3% for women) increasing to 52.5% (43.1% for men and 61.5% for women) in 2006–2010. Liu et al. [36] reported that central obesity among US non-Hispanic Asian adults in 2011–2014 was 55.3% and 60.9% for men and women, respectively.

Sociodemographic disparities

Several of the US studies examined differences in obesity prevalence by race/ethnicity, educational attainment, income and urbanization level. For example, Ljungvall et al. [21] reported that probability of obesity in 1960 was higher in non-Hispanic black women than non-Hispanic white women by 10 percentage points and that total increase during the study period (1960 – 2008) was also larger in non-Hispanic black women than non-Hispanic white women (by 5 – 10 percentage points). Romero et al. [22] studied a shorter period of time (1988 – 2004) and showed that a baseline difference observed between non-Hispanic white and non-Hispanic black did not change over time, while a difference between non-Hispanic black and Hispanic increased over time.

Yu et al [33] concluded that educational inequalities in relation to obesity prevalence from 1971 to 2012 were generally larger in women than men and larger in non-Hispanic white than non-Hispanic black and that obesity prevalence did not differ by educational attainment among non-Hispanic black men. In addition, individuals with some college education (i.e., 13 – 15 years of school or associate’s degree) experienced the most rapid increase in obesity prevalence among the four educational attainment categories in non-Hispanic white men, non-Hispanic white women and non-Hispanic black women. Ogden et al. [37], which also examined the trends by educational attainment (high school graduates or less; some college; and college graduates) between 1999–2002 and 2011–2014, showed that obesity prevalence among men with some college tended to increase at a faster pace than other two groups. On the other hand, such difference was not observed among women.

Ljungvall et al. [21] found an initial disparity by income, with higher prevalence observed among women in the lower (versus higher) income groups in 1959; this difference did not diverge over the full period captured (1959–2008). The initial income disparity for men in 1960 disappeared by the 1970s. When Ogden et al. [37] examined trends in obesity prevalence by three categories of household income from 1999–2002 to 2011–2014, obesity prevalence increased among women in the bottom two income groups, but it did not among women in the highest income group. Among men, obesity prevalence in the three income groups increased during the same period.

Hales et al. [38] examined if obesity prevalence differed by urbanization level (large metropolitan statistical areas [MSAs]; medium or small MSAs; and non-MSAs), showing that in 2013–2016, participants living in medium or small MSAs had a higher obesity prevalence compared to those living in large MSAs in both sexes. In addition, women living in non-MSAs also had a higher prevalence compared to women living in large MSAs.

Given less change in obesity prevalence in the first decade of the 2000s compared to previous years, several authors suggested that the increase in obesity prevalence among US adults may be leveling off [39, 40]. Robinson et al. examined differences in obesity prevalence [25] and abdominal obesity prevalence [26] by birth cohort using an age-period-cohort analysis. Robinson et al. [25] found that cohorts born in the 1980s had increased propensity to obesity compared to previous generations, suggesting obesity prevalence may continue to increase as this younger generation reaches the ages of peak prevalence of obesity. Robinson et al. [26] reported that the baby boomers (those born in 1946–1964) seemed to have low cohort effects on abdominal obesity.

Differences in obesity and abdominal obesity patterns

The authors of two studies reported trends in obesity defined using BMI and abdominal obesity (per waist circumference). Robinson et al. [26] reported that in 1986–1990 abdominal obesity prevalence was 27.5% for men and 44.3% for women and increased up to 43.1% for men and 61.5% for women in 2006–2010. During the same period, obesity defined with BMI increased from 18.5% to 32.6% and from 23.6% to 36.0% among men and women, respectively. Ladabaum et al. [28] reported that between 1988–1994 and 2009–2010, obesity prevalence increased from 19.9% to 34.6% among men and 24.9% to 35.4% among women while abdominal obesity prevalence increased from 29.1% to 42.0% among men and 46.0% to 61.5% among women.

Global trends in adult obesity prevalence

The GBD 2015 Obesity Collaborators based their study on a systematic literature search in Medline for studies providing nationally or sub-nationally representative estimates of BMI, overweight, or obesity among children or adults. Information came from 1514 data sources from 174 countries (713 measured and 801 self-report data) between 1980 and 2015. On the other hand, the NCD-RisC estimated global trends of BMI and obesity prevalence between 1975 and 2014 using data collected from 1698 population-based studies (e.g., nationally or sub-nationally representative studies and community-based studies) that used measured height and weight [16]. Following this initial paper, they expanded their study period to 2016 using data collected from 1820 population-based studies [17].

Both research groups described how obesity prevalence has increased in the last few decades. The GBD study showed that between 1980 and 2015 obesity prevalence doubled in 73 countries and showed an increase in most of the other countries as well. The NCD-RisC [16, 17] found that between 1975 and 2014, age-standardized prevalence of obesity increased from 3.2% to 10.8% in men and from 6.4% to 14.9% in women. In 2014, 2.3% and 5.0% of men and women from these 200 countries had BMI ≥35 and 0.64% and 1.6% had BMI ≥ 40.

The GBD study also showed trends in adult obesity prevalence by country’s sociodemographic development quintiles (categorized into quintiles: low, low-middle, middle, high-middle, and high). The GBD study showed that between 1980 and 2015, men aged 25 to 29 and living in countries with a low-middle degree of development experienced the largest relative increase in obesity prevalence (1.1% in 1980 to 3.8% in 2015) among population subgroups stratified by sex, age and country’s level of sociodemographic level.

The NCD RisC [16, 17] emphasized large regional differences in obesity prevalence. Areas with obesity prevalence of ≥25% or higher in 2016 included High-income Western countries (men: 29.6%; women: 29.6%), Central and Eastern Europe (women: 26.1%), Central Asia, Middle East and North Africa (women: 35.2%), Latin America and Caribbean (women: 29.2%), and Oceania (women: 30.0%), while several areas had obesity prevalence of <10%, i.e., East and South East Asia (men: 5.9%; women: 7.4%), High-income Asia Pacific (men: 4.9%; women: 4.3%), South Asia (men: 3.2%; women: 6.0%); and Sub-Saharan Africa (men: 4.8%).

Synthesis of findings

We extracted 18 US studies from the NHANES in our search, which indicate an increase in obesity prevalence over the past 40 years, with the latest prevalence estimates from the NHANES 2013–2016 at 36.6% (men) and 41.0% (women) [38]. In addition, Hales et al. [41], which was published as a letter and thus not included in our search, reported obesity prevalence in 2015–2016 to be 37.9% (men) and 41.1% (women).

Several papers by Flegal and her colleagues published outside of our inclusion window, and thus not included in our search, reported obesity prevalence during the periods between 1960 and 1988–1994 [42], 1988–1994 to 1999–2000 [43], and 1999–2000 and 2007–2008 [39, 44–47]. The earlier two studies showed that obesity prevalence increased significantly from 14.5% to 22.5% to 30.5% between NHANES II (1976–1980), NHANES III (1988–1994), and the first cycle of the continuous NHANES (1999–2000), while the more recent studies suggested a stabilized trend in obesity prevalence in the early 2000s. Whereas some authors report that the increase in adult obesity prevalence may be slowing over time [39, 40], the 18 more recent studies included in our review do not support such trend. In particular, Flegal et al. [30] found a significant increasing linear trend in the prevalence between 2005–2006 and 2013–2014 among women. Significant increases were reported between 2001–2004 and 2013–2016 [38] and between 2007–2008 and 2015–2016 [41]. In addition, US-specific estimates provided by the GBD [2] and the NCD-RisC [17] also suggest an increase in US obesity prevalence (Figure 3).

The studies that examined obesity trends suggest particularly high risk among non-Hispanic black women [21, 22], individuals with some college education versus the other educational attainment categories [33], those whose communities were classified as the intermediate category of urbanization [38] and cohort differences in obesity prevalence. [25, 26].

Our findings for global obesity trends indicate that obesity prevalence has increased in the last few decades across the world. The GBD study showed that between 1980 and 2015 obesity prevalence doubled in 73 countries and the NCD-RisC [16, 17] found that age-standardized prevalence of obesity increased from 3.2% to 10.8% in men and from 6.4% to 14.9% in women across 200 countries between 1975 and 2014 [16].

Differences in surveillance

The NHANES is a repeated cross-sectional survey, in which different participants were randomly sampled from population at several time points. Thus, unlike in longitudinal studies, it is not possible to track within-person changes in weight over time. While there are several representative longitudinal cohort studies in the US that collect measured height and weight (e.g., Add Health [13], the Longitudinal Health and Retirement Study [48], the National Social Life, Health, and Aging Project [49] and the Wisconsin Longitudinal Study [50]), none had data published in the date range that fit our inclusion criteria.

The GBD study and the NCD-RisC study based their estimates on many population-based studies, some of which were cross-sectional and the others were longitudinal follow-up. Rather than looking at the longitudinal association, the GBD and NCD-RisC studies estimated obesity prevalence for each country and year using data collected in each given country and the year of data collection, but allowed for inclusion of data from other years in the same country or from data in other countries across similar time periods within regions.

Analytical issues

Several aspects of the analysis of obesity prevalence trends can impact the magnitude and direction of estimated effects, potentially resulting in inconsistent conclusions across studies. Conclusions in relation to the trends in obesity prevalence differed by time periods covered by each study. Studies that covered a longer period (e.g., 1959–2008 [21]; 1988–2008 [28]) documented increases in obesity prevalence, while Flegal et al. [43] suggested that obesity prevalence from 1999 to 2008 did not continue to increase at the same rate as that observed in the prior 10 years. Another example is a difference between Ljungvall et al. [21] that covered the period between 1959–2008 and Romero et al. [22] that used information collected in 1988–2004. While the former described that the increases in obesity prevalence during the study periods (1960–2008) were similar across the racial/ethnic subgroups (or, at least, smaller than the increase experienced by the whole population), the latter emphasized that non-Hispanic black were at higher risk of obesity compared to non-Hispanic white.

The time point used as baseline for evaluating obesity prevalence trends can also have an important impact on findings [47]. A notable example comes from studies of recent childhood obesity trends; evaluations using 2003–2004 as the baseline time point reported decreases in the prevalence of obesity among 2–5 year old children through 2011–2014 [47, 51], while studies using 1999–2000 as baseline reported no evidence of a decline in obesity prevalence in this or any age group through 2011–2014 [52–54]. Because prevalence estimates can fluctuate substantially between study waves, inclusion of data from several prior years and subsequent years can aid in determining whether prevalence changes at any given time point reflect a transient anomalous dip or a true downward trend [53]. The need to place data in context and see the bigger picture underscores the need for ongoing, consistent monitoring of obesity prevalence and trends in the US and worldwide.

One must consider exclusion criteria for each specific study, even when based on the same study source. For example, in the 18 studies we reviewed, three studies explicitly excluded BMI < 18.5 [23, 27, 34]. Studies also varied greatly in the exclusion of older adults, with some studies excluding adults in their late 60s or 70s while others making no exclusions of older adults, and also varied in exclusion of younger adults aged 18–24 years old. Because obesity prevalence varies across the lifespan and some evidence suggests that obesity trends differ by age [30, 47], the age range of included participants could potentially impact estimated trends or limit comparability across studies. The analytical approaches used to evaluate obesity trends also varied, with methods including pairwise difference testing, linear trend tests, regression modeling to evaluate linear and quadratic trends, and age-period-cohort analysis. Studies using multivariable regression modeling to evaluate changes over time also differed in the selection of covariates used for adjustment in regression models. However, regardless of these differences, studies of US trends included in our review consistently found significant long-term increases in the prevalence of obesity.

There are also analytical issues related to future obesity projections. Flegal et al. [30] cautioned that several previous attempts to use past data to extrapolate to future trends in obesity prevalence may not have provided valid estimates [55–57]. Mehta et al. [31] conducted simulations using NHANES I to simulate mortality for NHANES III participants to test whether a decline in association between BMI and mortality related to statistical nuisance issues, finding that these nuisance contributors, such as the usage of categorical BMI variable (vs. continuous variable) and changes in population distribution altered findings.