BMI indicates body mass index (calculated as weight in kilograms divided by height in meters squared). BMI is age-standardized and estimates are weighted. Error bars represent 95% CIs. Specific numbers for prevalence and CIs are given in eTable 1 in the Supplement and omitted from the Figure for clarity. Data for Chad, Guatemala, and Turkey were restricted to women aged 15 through 49 years with children.
aTurkey had no individuals with a BMI lower than 16 and is not included in this graph.
BMI indicates body mass index (calculated as weight in kilograms divided by height in meters squared). BMI is age-standardized and estimates are weighted. Error bars represents 95% CIs. Specific numbers for prevalence and CIs are given for BMI lower than 16 in eTable 2 in the Supplement.
eAppendix 1. DHS Survey Methods
eAppendix 2. Stata Code
eTable 1. Prevalence of BMI<16.0 kg/m2 in Women in 60 Countries
eTable 2. Average Annualized Change in Prevalence of BMI<16.0 kg/m2 in Women in 40 Countries With Repeated Surveys
eTable 3. Total Female Population Age 20-49 With BMI<16.0 kg/m2 by Country Based on Weighted Estimates
eTable 4. Comparison of Prevalence of BMI<16.0 kg/m2 Between Men and Women in 13 Countries with Male Anthropometry
Razak F, Corsi DJ, Slutsky AS, Kurpad A, Berkman L, Laupacis A, Subramanian SV. Prevalence of Body Mass Index Lower Than 16 Among Women in Low- and Middle-Income Countries. JAMA. 2015;314(20):2164-2171. doi:10.1001/jama.2015.15666
Body mass index (BMI) lower than 16 is the most severe category of adult undernutrition and is associated with substantial morbidity, increased mortality, and poor maternal-fetal outcomes such as low-birth-weight newborns. Little is known about the prevalence and distribution of BMI lower than 16 in low- and middle-income countries (LMIC).
To determine the prevalence and distribution of BMI lower than 16 and its change in prevalence over time in women in LMIC.
Design, Settings, and Participants
Cross-sectional data analysis composed of nationally representative surveys from 1993 through 2012 from the Demographic and Health Surveys Program. Women aged 20 through 49 years from 60 LMIC (N = 500 761) and a subset of 40 countries with repeated surveys (N = 604 144) were examined.
Wealth was measured using a validated asset index, age was categorized in deciles, education by highest completed level (none, primary, secondary, or greater), and place of residence as urban vs rural.
Main Outcomes and Measures
The primary outcome was BMI lower than 16. Analyses assessed the prevalence of BMI lower than 16, its association with sociodemographic factors, and change in prevalence. Logistic regression models were used to calculate odds ratios (ORs), adjusting for survey design and age structure.
Among countries examined, the pooled, weighted, and age-standardized prevalence of BMI lower than 16 was 1.8% (95% CI, 1.7% to 1.8%) with the highest prevalence in India (6.2% [95% CI, 5.9% to 6.5%]), followed by Bangladesh (3.9% [95% CI, 3.4% to 4.3%]), Madagascar (3.4% [95% CI, 2.8% to 4.0%], Timor-Leste (2.9% [95% CI, 2.4% to 3.2%]), Senegal (2.5% [95% CI, 1.9% to 3.2%]), and Sierra Leone (2.2% [95% CI, 1.3% to 3.0%]); and 6 countries had prevalences lower than 0.1% (Albania, Bolivia, Egypt, Peru, Swaziland, and Turkey). The prevalence of BMI lower than 16 in women with a secondary or higher education level was 0.51% (95% CI, 0.47% to 0.55%), and in mutually adjusted models, a less than primary education level was associated with an OR of 1.4 (95% CI, 1.2 to 1.6). The prevalence of BMI lower than 16 was 0.43% (95% CI, 0.37% to 0.48%) in the highest wealth quintile with an OR of 3.0 (95% CI, 2.4 to 3.7) in the lowest wealth quintile. Among the 24 of 39 countries with repeated surveys, there was no decrease in prevalence. In Bangladesh and India, rates were declining with an average absolute change annually of −0.52% (95% CI, −0.58% to −0.46%) in Bangladesh and −0.11% (95% CI, −0.12% to −0.10%) in India.
Conclusions and Relevance
Among women in 60 LMIC, the prevalence of BMI lower than 16 was 1.8%, and was associated with poverty and low education levels. Prevalence of BMI lower than 16 did not decrease over time in most countries studied.
There has been increasing focus globally on obesity, particularly in low- and middle-income countries (LMIC).1,2 Even in countries that have historically faced a primary burden of undernutrition, it has been argued that there is a need to shift focus toward the societal effects of excessive weight gain.3,4 However, the average body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) among the least affluent in many LMIC remains low.5
In 1998, the United Nations sought a method to quantify the population prevalence of severe chronic undernutrition, introducing a definition of severe chronic energy deficiency: a BMI lower than 16.6 Although the definition of BMI lower than 16 may capture individuals with fluctuations of weight due to acute caloric restriction or illness, at a population level, the prevalence of BMI lower than 16 represents a lack of human growth rather than loss of weight.7
The low weight of individuals with BMI lower than 16 is sustained by important physiologic adaptations, and these changes may have important consequences for social functioning, morbidity, and mortality patterns. Very low BMI is associated with decreases in muscle strength and work capacity.8,9 Women with low BMI are more likely to deliver low-birth-weight newborns and the children of women with low BMI have greater rates of wasting and stunting.7 BMI mortality curves, especially in low-income settings, show a marked increase in mortality at low BMI levels.10,11
There have been few cross-national studies to examine the prevalence of BMI lower than 1612,13 and, to our knowledge, no recent estimates. In this study, we examined the prevalence and distribution of BMI lower than 16 using samples of women aged 20 through 49 years from 60 LMIC and examined the change in prevalence of BMI lower than 16 over time.
This study used publicly available deidentified data and was considered exempt by the ethics board at the Harvard T.H. Chan School of Public Health. Data were obtained from the Demographic and Health Surveys (DHS) program,14 which has conducted nationally representative household sample surveys in more than 85 countries since 1984.15 The surveys in the DHS program were designed to provide information on child and maternal health, fertility, and nutrition in LMIC.15 In all surveys, reproductive-aged women (aged 15-49 years) were selected for inclusion using multistage probabilistic sampling in which primary sampling units and households were drawn from geographic-based sampling frames that cover the entire territory of each country (eAppendix 1 in the Supplement).16
Weight and height measurements in the DHS program surveys were obtained by field interview teams, which typically included 2 dedicated and trained health investigators to conduct the anthropometric assessments. Adults were weighed wearing light-fitting clothing and without shoes using digital scales (seca 874 digital floor scale, seca) with precision to 0.01 kg. Standing height was measured without shoes using adjustable measuring boards (ShorrBoard) designed for use in survey settings and recorded to the nearest 0.1 cm.
This study used 2 data sets and analyses. The first was a cross-sectional data analysis across countries where at least 1 survey was completed. This was used to examine prevalence of BMI lower than 16, and the association of BMI lower than 16 with socioeconomic and demographic factors. The second was a repeated cross-sectional analysis of data from countries that had at least 2 surveys. If a country had more than 2 surveys available, the first and last survey year were used to construct this data set. This data set was used to calculate the change in prevalence of BMI lower than 16 over time (eAppendix 1 in the Supplement).
BMI lower than 16 was used to identify individuals in the most severe category of chronic undernutrition.6 Limited evidence suggests that basal metabolic rate among those with severe undernutrition does not vary by ethnicity, and the original classification scheme using BMI alone was determined based on the theory that a single standard was appropriate across ethnic groups.17,18
We examined the association of BMI lower than 16 with socioeconomic and demographic variables: age, area of residence (urban or rural, using country-specific definitions), household wealth, and education. Wealth was defined using an asset index developed for the DHS program and validated elsewhere.19,20 This index was developed within each country using a standardized list of assets and utilities, with creation of a score via factor analyses. Wealth was divided into quintiles and treated as an ordinal variable. Education was divided into 3 ordinal categories: none or incomplete primary, primary, or secondary or higher.
All prevalence estimates were weighted and standard errors adjusted for the complex survey design of the DHS. In the pooled data set, sampling weights were rescaled to account for arbitrary differences in sample sizes across surveys and reduce the influence of larger survey samples, such as those from India or Colombia. Furthermore, prevalence estimates for BMI lower than 16 were age-standardized to the World Health Organization’s standard population for 2000-2025 using the direct method.21
A logistic regression analysis was conducted on the pooled data set to examine factors associated with BMI lower than 16. The model was weighted with rescaled sampling weights and it accounted for the survey-specific primary sampling unit clustering and survey design and stratification characteristics. This was a fixed-effects model to account for between-country differences, including year of survey. Odds ratios (ORs) and 95% CIs were obtained from the logistic regression. A 2-sided P value less than .05 was considered statistically significant.
Average country annualized change in prevalence of BMI lower than 16 was estimated using the difference between final prevalence (pfinal) and initial prevalence (pinitial) multiplied by the inverse of number of years duration between the 2 surveys: Δ = [1 / No. of years] × [pfinal − pinitial], with variance estimated as [Var(pfinal) + Var(pinitial)] / (No. of years)2. Prevalence and variance estimates accounted for survey weights, survey design characteristics, and clustering within primary sampling units. A Pearson product-moment correlation coefficient was calculated between last survey year and change in prevalence of BMI lower than 16 to detect whether there was an association between these variables.
Estimates of the total population of women with a BMI lower than 16 were constructed using United Nations’ country population estimates22 of the total number of women aged 20 through 49 years. The weighted prevalence of BMI lower than 16 was multiplied against the total population of women aged 20 through 49 years for the matching survey year. As a supplementary analysis, for 13 countries where DHS program data also contained men, the Pearson product-moment correlation coefficient between male and female rates of BMI lower than 16 was estimated.
All analyses were performed using Stata statistical software (StataCorp), version 13.1, and Excel for Mac (Microsoft), version 15.15.
The cross-sectional data set of the most recent survey year comprised 60 countries and 500 761 women (60 surveys), and the repeated survey data set comprised 604 144 participants in 40 countries (80 surveys). Participation rates typically exceeded 90% (52 of 60 countries), with a range of 77.0% to 98.9%.15
Of 500 761 women surveyed, there were 7948 unweighted cases of BMI lower than 16, equivalent to a weighted, age-standardized prevalence of 1.8% (95% CI, 1.7% to 1.8%). The Table shows the association of BMI lower than 16 with demographic and socioeconomic variables. The prevalence of BMI lower than 16 was highest in the poorest wealth quintile (1.5% [95% CI, 1.4% to 1.7%]) and lowest in the wealthiest quintile (0.43% [95% CI, 0.37% to 0.48%]). The prevalence of BMI lower than 16 was highest in those with no education (1.8% [95% CI, 1.6% to 1.9%]) vs those with secondary education or higher (0.51% [95% CI, 0.47% to 0.55%]) and higher for those residing in rural areas (1.3% [95% CI, 1.2% to 1.3%]) compared with those residing in urban areas (0.50% [95% CI, 0.45% to 0.55%]). In multivariable models (Table), the poorest quintile was associated with the highest prevalence of BMI lower than 16 (OR, 3.0 [95% CI, 2.4 to 3.7]) compared with those in the highest wealth quintile. The association of education was attenuated but those with no education were more likely to have BMI lower than 16 (OR, 1.4 [95% CI, 1.2 to 1.6]) compared with those with secondary education or higher. The association with living in rural regions was no longer significant.
Figure 1 and eTable 1 in the Supplement provide the prevalence of BMI lower than 16 across the 60 countries. The highest prevalence of BMI lower than 16 was in India (6.2% [95% CI, 6.0% to 6.5%]) and other countries with high prevalence were from South Asia (Bangladesh, Pakistan, and Nepal) and sub-Saharan Africa (Madagascar, Senegal, Sierra Leone, Ethiopia, and Democratic Republic of the Congo). Six countries had prevalence lower than 0.1%: Albania, Bolivia, Egypt, Peru, Swaziland, and Turkey. Rates in rural regions were significantly higher than urban regions in 7 countries (Bangladesh, Cameroon, Haiti, India, Kenya, Maldives, and Nigeria), and the absolute gap was highest in India (3.13%).
Figure 2 shows the annualized change in prevalence of BMI lower than 16 among 39 countries with repeated surveys at final survey cycle vs baseline survey (eTable 2 in the Supplement contains numeric estimates). Turkey was excluded from this analysis as it had no individuals with BMI lower than 16 at either baseline or final survey. The average interval in each country between baseline and final survey was 11 years, with a minimum of 3 years in Guatemala and Nicaragua and a maximum of 18 years in Ghana. Twenty-three countries had their final survey conducted from 2010 through 2014 and 35 countries from 2005 through 2014. In Bangladesh and India, rates have declined, with an average absolute change of −0.52% (95% CI, −0.58% to −0.46%) and −0.11% (95% CI, −0.12% to −0.10%) annually. However, other countries with high prevalence of BMI lower than 16, such as Madagascar and Senegal, had increased absolute rates of 0.18% (95% CI, 0.16% to 0.21%) and 0.22% (95% CI, 0.19% to 0.25%) annually. Overall, 15 of 39 countries (38% [95% CI, 25% to 54%]) showed a significant decline in absolute annualized rates of BMI lower than 16. There was no meaningful correlation between the year the last survey was published and annual change in prevalence of BMI lower than 16 (r = −0.05).
To our knowledge, this article provides the first broad study of the prevalence of BMI lower than 16 and its association with social, economic, and demographic factors. The 60 LMIC studied here represent an estimated 3 billion individuals. There are 2 major findings. First, using the most recently available nationally representative data on a large range of LMIC, BMI lower than 16 remains a critically important public health entity. BMI lower than 16 was associated with poverty and low education. Second, the prevalence of BMI lower than 16 was not decreasing in most countries.
The prevalence and total population burden of individuals with BMI lower than 16 remains high globally, and if prevalence estimates are generalizable, more than 18 million women (18 264 087 [95% CI, 16 699 908 to 19 831 222]) are affected in the countries studied (eTable 3 in the Supplement).The prevalence of BMI lower than 16 has previously only been studied in a few countries or in isolated regions of low-income countries.12,13,23 A limited comparison of a small subset of countries surveyed in the DHS program more than 15 years ago only included studies of women with children, no longitudinal comparisons and no data from Asia.12 To our knowledge, the current study provides the first estimate for more than one-third of the world’s women using standardized methods, representative sampling, and objectively measured nutritional status. The highest prevalence rates of BMI lower than 16 were found across multiple regions of South Asia and Sub-Saharan Africa. The greatest burden was found in India, with prevalence of BMI lower than 16 for more than 6% of women, translating to an estimated 14 million women (14 486 874 [95% CI, 13 901 734 to 15 072 014]; eTable 3 in the Supplement). However, we are limited in our analyses to women in countries included in the DHS program and it is possible that other countries at similar stages of economic development may have a similarly high burden of BMI lower than 16. Among the 10 largest countries that do not have anthropometry data in the DHS program, 5 are LMIC, with a cumulative population of nearly 2 billion: China, Indonesia, Mexico, the Philippines, and Vietnam. For 13 countries where DHS program data also contained men, the Pearson product-moment correlation coefficient between rates of BMI lower than 16 among men and women was r = 0.88 (eTable 4 in the Supplement). This suggests that countries found to have a high burden of BMI lower than 16 among women may also have a high burden in men. However, representative studies in men across a wide range of LMIC are needed.
Findings from studies in low-income settings suggest that the increased risk of mortality among those with a BMI lower than 16 exceeds the increased risk associated with being overweight or obesity.10 For example, in the largest meta-analysis to examine BMI and mortality in Asia, the hazard ratio for death in East Asians with BMI lower than 15 was 2.8 (95% CI, 1.9 to 4.1) compared with 1.5 (95% CI, 1.3 to 1.7) among those with BMI higher than 35. In South Asians with BMI lower than 15 it was 2.1 (95% CI, 1.8 to 2.6) compared with 1.3 (95% CI, 0.7 to 2.3) among those with BMI higher than 35.10 The specific mechanism underlying this increased mortality remains poorly understood, but the association persists despite extensive methodologic steps to eliminate reverse causality related to subclinical illness or confounding related to common risks such as smoking.10,24 The finding of a large and, in some countries, persistent burden of individuals with BMI lower than 16 supports the need for further study of why mortality rates are increased and supports the value of intervention studies to examine whether mortality can be reduced. In contrast to the extensive number of studies that examine methods to reduce morbidity and mortality in those with obesity-related diseases such as diabetes,25 dyslipidemia26 and hypertension,27 to our knowledge no large-scale randomized trials have been attempted to improve the health of those who have BMI lower than 16. Nutritional supplementation over a short term has shown some encouraging effects among those with BMI lower than 16, but many questions remain unanswered about potentially increased long-term cardiovascular and chronic disease risk when chronic nutritional deprivation is reversed in adulthood.7,28,29
The association of BMI lower than 16 with poverty may reflect the persistent material deprivation that keeps individuals at such severely low weight and also may be related to reverse causality as individuals who are severely underweight have reduced work capacity and efficiency, decreased energy expenditure, and increased illness and sick days.8,9,30,31 Therefore, an association may form between severe energy deficiency and low income, with each contributing to and reenforcing the other. A multigenerational effect of very low BMI may exist, with higher incidence of low-birth-weight newborns among women with very low BMI, and subsequent impaired growth in their children.7
Most countries did not show any improvement in the prevalence of BMI lower than 16 over time, and some countries in Sub-Saharan Africa had increasing prevalence of it. However, prior research has demonstrated that more than 90% of these countries have increasing prevalence of overweight during this period.5 The lack of improvement in the prevalence of BMI lower than 16 in most countries despite rapid expansion in overweight suggests the most severely underweight are receiving the least benefit from overall societal gains. The shape of the BMI distribution has changed markedly over the last 2 decades in LMIC, with slow gain in weight among the underweight and rapid gains among the overweight and obese.32 Low-income countries continue to have severe income inequality−with increasing inequality in some circumstances−and this may be one underlying factor that explains the lack of improvement in prevalence of BMI lower than 16 despite weight gain among other segments of society.5,33 One positive finding in this study was that 2 countries with the largest number of women with BMI lower than 16 showed declining rates: Bangladesh and India. Future research should focus on understanding the specific factors that have allowed these 2 countries to reduce the rates of BMI lower than 16.
This study has a number of limitations. First, this cross-sectional analysis is based on survey data spanning 2 decades. For a very few number of countries the latest survey available was pre-2000. However, more than 95% of the data come from 2001 or later (median year, 2008). This is the best available data to estimate prevalence of BMI lower than 16 across LMIC. To address this limitation, a control variable for year of survey was included in the regression model. The analyses of change indicated that the prevalence of BMI lower than 16 has been stable with time and pooling data across this period appears reasonable. Further, the inclusion of year to the logistic regression models did not substantially alter the findings because models already included fixed effects for countries.
Second, low BMI in isolation does not necessarily imply chronic undernutrition, as basal metabolic rate can act as a confounder and weight loss may be due to episodic caloric restriction or underlying disease. This may explain, for example, why even in the wealthiest quintile there are some individuals with BMI lower than 16 (0.43% [95% CI, 0.37 to 0.48]). However, the definition of severe chronic undernutrition based solely on BMI was thought to represent the best mechanism to provide a broadly comparable population level metric.7 In addition, adding measures of energy intake or energy expenditure did not meaningfully contribute to identifying those with severe undernutrition at the population level.34 Third, although 23 of 40 countries with repeated surveys have data available after 2009, rapid economic transitions have occurred over the last 5 years in many countries and this may affect prevalence of BMI lower than 16. Fourth, we are unable to make an estimate for prevalence of BMI lower than 16 among men given the lack of representative data. However, the high correlation in prevalence of BMI lower than 16 in countries where data on both men and women are available suggests that countries with high prevalence of BMI lower than 16 among women will also have high prevalence in men.
Among women in 60 LMIC, the prevalence of BMI lower than 16 was 1.8%, and was associated with poverty and low education levels. Prevalence of BMI lower than 16 did not decrease over time in most countries studied.
Corresponding Author: Fahad Razak, MD, MSc, Li Ka Shing Knowledge Institute, St Michael’s Hospital, Room 364, 30 Bond St, Toronto, ON M5B1W8, Canada (firstname.lastname@example.org).
Correction: This article was corrected on January 12, 2016, to fix errors in the Author Contributions section.
Author Contributions: Drs Razak and Corsi had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Razak, Corsi, Subramanian.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Razak, Subramanian.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Razak, Corsi.
Study supervision: Razak, Subramanian.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Additional Contributions: Aditi Krishna, PhD (Harvard School of Public Health), provided assistance with design of figures and received compensation for her work. Tom Pullum, PhD (Demographic and Health Surveys Program, ICF International), provided statistical advice and received no compensation for his work.