What is a “high” prevalence of obesity? Two rapid reviews and a proposed set of thresholds for classifying prevalence levels

Summary Categories such as “low” and “high” have been used for several decades to describe the prevalence of stunting and wasting in populations of children aged under 5 years. They provide support for public health risk assessment and policy‐making, including alerting health departments and aid agencies to national trends and local needs. In the light of the need for monitoring progress to meet globally agreed targets for overweight and obesity, the classification of their prevalence will be a valuable to aid in policy development, to target resources, and to promote public health interventions. This paper reviews the current use of categories to describe obesity prevalence in policy, advocacy, and research literature. Where prevalence categories have been formally proposed, this paper compares their application on large‐scale datasets. The paper then develops a set of recommended threshold values to classify prevalence levels for overweight and obesity among children under age 5 years, children aged 5–19 years, and adults.


| Current use of obesity prevalence categories
In order to understand better the descriptions of obesity and overweight prevalence levels used in the literature, a rapid review was undertaken of published anthropometric surveys to assess how the descriptor "high" may be used to describe obesity prevalence.

| Prevalence surveys
The National Library of Science database (PubMed) was searched in early October 2020 for published papers using the word "high" associated with a report of obesity prevalence, published in the previous 25 years (September 1995 to September 2020). Of more than 30,000 titles returned, the first 200 "best match" were examined. Papers were excluded if they used comparators such as "higher than" and "not as high as" or did not report surveys of children's or adults' adiposity prevalence. Full text papers were examined by one author.

| Public health strategies
In a second survey, public health obesity strategy papers were examined for references to threshold values for obesity prevalence levels. A literature search was undertaken using Google and Google Scholar for documents describing obesity prevalence and sourced from government departments, intergovernmental agencies, and obesity professional societies (members of the World Obesity

| Prevalence surveys
Of the 200 "best match" records returned, 31 were excluded for using the term relatively (e.g., not as high, higher, and highest), 64 were excluded for using the term for comorbid conditions (e.g., high blood pressure and high risk of obstetric problems), and 14 were excluded for using the term to describe subjects or locations (e.g., high school, high income, and high mountain). The remaining 91 records were inspected by the first author and 50 of these were deliberatively selected to demonstrate the range of prevalence levels and range of population samples where the term "high" was being used to describe obesity prevalence. The results are shown in Table 1, where it can be seen that the term is applied to a wide variety of prevalence levels, and based on several different criteria for defining overweight or obesity. The lowest levels of prevalence described as "high" were below 10% for adults and below 5% for children and adolescents.
None of the papers referred to a published definition of "high," and the use of the term "high" appeared to be based on the authors' own judgments. These judgments may have been based on authors' expectations: For example, in surveys of adults in LMICs, the word "high" was used for adult prevalence at levels as low as 6.8% in Nigeria and 5.2% in Malaysia. For children, "high" was used to describe levels of obesity prevalence below 10% in both low-and high-income countries, for example, 4.6% in India, 5% in Norway, 5.4% in Switzerland, and 3.9%-5.1% in Sweden, and the phrase "too high" was used to describe prevalence levels of 3.1%-4.4% among children in France.

| Public health strategies
Of 200 records returned by the searches, 130 records were excluded as not relating to obesity or related conditions (58 of the records), as social media links (24 records), advertising (43 records), and foreign language (5 records). Of the remaining 70 records, documents were downloaded and examined in detail. Of these, a further nine were found to be duplicates held on different sites, leaving 46 governmental and intergovernmental documents and 15 professional society documents included for examination. Documents in which the classification of prevalence levels for overweight or obesity were formally defined were examined and the source reference extracted: These source references were associated with four organizations, namely, the WHO together with UNICEF (cited 53 times), 56,57 the World Bank 58 linked to a paper by Popkin et al. 59 (together cited 6 times), and the World Obesity Federation (cited 2 times). 60,61 The details of the prevalence classifications are shown in Table 2. All papers use the same definitions for overweight and obesity, based on the WHO criteria. 62 A fifth source, The 2020 Global Nutrition Report, 63  that a similar equality of distribution has been maintained in the most recent years and applies across a greater number of countries. Within each classification, the proportion of countries has also remained consistent in the most recent years, as shown in Table 3.
It can also be argued that the definition of a "very low" prevalence of overweight at a threshold of 2.5% of the population has some external validation, given that a population of healthy children used by WHO to represent optimum growth defines overweight at a threshold of weight-for-height Z score above +2.0, equivalent to 2.3% of the population. 65 Given these arguments, the de Onis et al. 56 classification scheme was adopted for the joint WHO-UNICEF 2019 publication Recommendations for data collection, analysis and reporting on anthropometric indicators in children under 5 years old. 57 As Table 2 shows, the WHO-UNICEF classification scheme is not identical to that suggested by either the World Bank or the paper by Popkin et al. Under the World Bank categorization, 81% of countries are classified with a "low" prevalence of overweight, and less than 5% of countries with a "high" or "very high" prevalence, and this is similar to the Popkin et al. scheme which would classify 99% of countries as having less than a "medium" prevalence of overweight. This suggests that the discriminatory power of either of these two classification schemes may be relatively weak for distinguishing countries, and from a public health perspective, they may lead to many countries assuming that obesity prevention in younger children is unnecessary. For that reason, the present paper recommends using the WHO-UNICEF proposals for children under age 5 years (see the recommendations summarized in Table 7

| Criteria for obesity prevalence among adults
The next step moves from child to adult prevalence. A very high correlation (r = 0.91, p < 0.001) is found between child and adolescent (ages 5-19 years) obesity prevalence and age-adjusted adult obesity prevalence, across 192 countries in the WHO Global Health Observatory's 2016 estimates. 67 The regression equation (y = 0.51x À 1.20) indicates that adult obesity prevalence is found at approximately twice the levels found in children and adolescents (see Figure S1). On this basis, it is reasonable to suggest categories for adults based on double the prevalence thresholds for children aged 5-19 years.
However, based on the regression equation, the equivalent figure for 10% prevalence in children is around 23% for adults. This could be rounded to 20% prevalence or 25% prevalence. There is no obvious method for externally validating one alternative over the other, so it is suggested here that the threshold of 20% rather than 25% is used as the recommended threshold to define a "high" prevalence. It should be recalled that very few countries had obesity prevalence levels as high as 20% only a generation ago. A comparison of the two options showing the distribution of countries across the classifications is shown in Table S2. The proposed criteria for adult obesity are shown in Table 7. 3.3.4 | Criteria for prevalence of "at risk of overweight" in children under age 5 years Children under age 5 years with a weight-for-length/height Z score between +1.0 and +2.0 are classified as "at risk of overweight" ("at risk") by the WHO, equivalent to a prevalence between 2.3% at Z = +2.0 and 16% at Z = +1.0 in the population of children with optimum health. In order to propose prevalence categories for "at-risk" children, the data for the prevalence of children "at risk" were regressed against the prevalence of children overweight, based on the dataset provided were provided (including repeated surveys over several years). 68 The surveys were generally in low-and lower-middle-income countries, so the number of countries with higher prevalence levels may have been limited. The results show a good correlation between "at-risk-of-overweight" and overweight prevalence levels (r = 0.71, p < 0.001) and a regression line close to 0.5 (y = 0.52x À 0.73) (see Figure S2), indicating that it would be reasonable to use criteria for "at-risk-of-overweight" children at around double those for overweight children in this age group.
Using latest available data from each of the 123 different countries gave a distribution in which 23% of countries had a low or very low prevalence of children "at risk of overweight" and 29% of countries had a high or very high prevalence of children "at risk" (see Table S3). Based on the "at risk of overweight" and overweight in combination (i.e., the prevalence of children with weightfor-height Z scores > +1.0), 27% of countries had a low or very low prevalence, and 31% of the countries had a high or very high prevalence (see Table S3).

| Criteria for prevalence of overweight in children 5-19 years
Using the same approach for overweight (without obesity) in older children (defined as having a BMI Z score between +1.0 and +2.0 above the median reference population), the prevalence for overweight regressed against that for obesity in this age group, based on 191 countries reported in the WHO Global Health Observatory, shows a strong correlation (r = 0.88, p < 0.001) and a linear gradient of over 0.9 and an offset of 5 percentage points (y = 0.93x À 5.20) (see Figure S3).
This indicates that the corresponding classifications of prevalence, rounded to convenient levels, can be proposed at 5 percentage points above those for obesity and are shown in Table 7. Of 191 countries, 6% have a "very low" prevalence of overweight children and adolescents, and 16% a "very high" prevalence (see Table S4).
Combining overweight with obesity, the distribution of prevalence levels for all children above a BMI Z score of +1.0 shows 7% of countries to have a very low prevalence of overweight including obesity and 14% to have a very high prevalence of overweight including obesity.
The present proposals were compared with those of the World Bank 58 and the Popkin et al. 59 paper, which also suggest criteria for prevalence levels for this group of children. The results are shown in Table 5. The World Bank classification scheme appears skewed towards identifying nearly two thirds of countries as having a "very high" prevalence of overweight and obesity. Setting a target for improvement (e.g., bringing the prevalence down to Medium) may be hard to achieve for many of these countries. The Popkin et al. 59 classification scheme looks less distorted, although nearly 40% of countries are classified as having prevalence levels below "Medium," which may deter those countries from taking action to prevent child and adolescent overweight from increasing.

| Criteria for overweight prevalence in adults
Lastly, the categorization of the prevalence of overweight in adults can be proposed. A scatterplot of adult overweight, nonobese prevalence against adult obesity prevalence shows some nonlinearity, which is primarily explained by outliers (mostly in the Pacific Islands) where the mean BMI is above 30 kg/m 2 (see Figure S4). A linear regression shows a strong correlation (r = 0.67, p < 0.0001) and a gradient of close to 1 with an offset close to 10 percentage points (y = 1.0864x À 11.584; see Figure S4). This indicates the rounded thresholds shown in Table 7.
[Note that excluding the high BMI outliers and regressing overweight on obesity prevalence only for those countries where obesity prevalence is below 45% gives a correlation of r = 0.87 (n = 181; p < 0.001) and a gradient just over 1 and an offset of 12.4, which would make little difference to the proposed thresholds.]  Table 6 shows the distribution for countries using the proposed thresholds for overweight prevalence (including obesity) and compares these with the distributions under the proposed schemes from the World Bank 58 and Popkin et al. 59 The latter two classifications schemes are identical, and both would classify two thirds of the world's countries as having a very high prevalence of overweight and obesity combined.

| DISCUSSION
The prevalence of undernutrition has been classified since the early 1990s for global monitoring of malnutrition in children under age 5 years. The original threshold level for wasting-the "severity index for malnutrition in emergency situations"-is based on the association of prevalence levels with mortality risk. 56 Table 7. The thresholds are grounded on the proposals of de Onis et al. 56 for children under age 5 years used by the WHO and UNICEF and the approach extended by regression analyses to suggest thresholds for categorizing prevalence levels for young children "at risk of overweight" and also to older children and adults, both for obesity prevalence and for overweight prevalence.
The proposed classification thresholds are shown in Table 7.
The proposed categorizations show many countries in the "high" and "very high" categories: For example, 21% of countries have "high" or "very high" overweight prevalence in children under 5 years, rising to 39% of countries with "high" or "very high" prevalence of obesity in children 5-19 years, and rising further to 55% of countries classified with "high" or "very high" adult obesity prevalence levels.

| Limitations
Some care may need to be taken in the use of these thresholds for External validity will depend on the functional value of the thresholds and categories as they are used in practice.

| CONCLUSION
A formal set of criteria for describing and classifying prevalence levels can be of value for policy development and public communication.
T A B L E 7 Proposed prevalence classifications for children under age 5 years, aged 5-19 years, and adults, for categories of obesity As the paper by de Onis et al. 56 concluded, "Harmonized terminology will help avoid confusion and promote appropriate interventions" (p. 175). A review of the use of the descriptor "high" for the prevalence of overweight or obesity found it has been used by researchers somewhat indiscriminately. The present paper proposes a set of threshold values for defining overweight and obesity prevalence in a population as "low," "medium," "high," and so forth based on the approach used by the WHO for children under age 5 years and extended to children aged 5-19 years, and to adults.
As this paper goes to press, these thresholds are being used in a pilot version of the UNICEF-WHO Landscape Analysis tool being applied in several countries in 2020-2021. 5 The recommended threshold values are also expected to be used in future editions of the World Obesity Federation's Obesity Atlas series.

ACKNOWLEDGMENT
The first author is grateful for the support of a consultancy with UNICEF to enable the preparation of this paper.