Spatial-temporal trends and risk factors for undernutrition and obesity among children (<5 years) in South Africa, 2008-2017: findings from a nationally representative longitudinal panel survey.

OBJECTIVES
To assess space-time trends in malnutrition and associated risk factors among children (<5 years) in South Africa.


DESIGN
Multiround national panel survey using multistage random sampling.


SETTING
National, community based.


PARTICIPANTS
Community-based sample of children and adults.


SAMPLE SIZE
3254 children in wave 1 (2008) to 4710 children in wave 5 (2017).


PRIMARY OUTCOMES
Stunting, wasting/thinness and obesity among children (<5). Classification was based on anthropometric (height and weight) z-scores using WHO growth standards.


RESULTS
Between 2008 and 2017, a larger decline nationally in stunting among children (<5) was observed from 11.0% to 7.6% (p=0.007), compared with thinness/wasting (5.2% to 3.8%, p=0.131) and obesity (14.5% to 12.9%, p=0.312). A geographic nutritional gradient was observed with obesity more pronounced in the east of the country and thinness/wasting more pronounced in the west. Approximately 73% of districts had an estimated wasting prevalence below the 2025 target threshold of 5% in 2017 while 83% and 88% of districts achieved the necessary relative reduction in stunting and no increase in obesity respectively from 2012 to 2017 in line with 2025 targets. African ethnicity, male gender, low birth weight, lower socioeconomic and maternal/paternal education status and rural residence were significantly associated with stunting. Children in lower income and food-insecure households with young malnourished mothers were significantly more likely to be thin/wasted while African children, with higher birth weights, living in lower income households in KwaZulu-Natal and Eastern Cape were significantly more likely to be obese.


CONCLUSIONS
While improvements in stunting have been observed, thinness/wasting and obesity prevalence remain largely unchanged. The geographic and sociodemographic heterogeneity in childhood malnutrition has implications for equitable attainment of global nutritional targets for 2025, with many districts having dual epidemics of undernutrition and overnutrition. Effective subnational-level public health planning and tailored interventions are required to address this challenge.

I, the Submitting Author has the right to grant and does grant on behalf of all authors of the Work (as defined in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance with the terms applicable for US Federal Government officers or employees acting as part of their official duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd ("BMJ") its licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the Work in this journal and any other BMJ products and to exploit all rights, as set out in our licence.
The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to the Submitting Author unless you are acting as an employee on behalf of your employer or a postgraduate student of an affiliated institution which is paying any applicable article publishing charge ("APC") for Open Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and intends to pay the relevant APC), the terms of reuse of such Open Access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this Work are set out in our licence referred to above.
Other than as permitted in any relevant BMJ Author's Self Archiving Policies, I confirm this Work has not been accepted for publication elsewhere, is not being considered for publication elsewhere and does not duplicate material already published. I confirm all authors consent to publication of this Work and authorise the granting of this licence.
Results: Between 2008 and 2017 there was a significant decline nationally in stunting prevalence among children under 5 years of age from 11.0% to 7.6% (p=0.007), while thinness/wasting (5.2% to 3.8%, p=0.131) and obesity (14.5% to 12.9%, p=0.312) decreased insignificantly. Stunting prevalence appears relatively evenly spread across South Africa, while obesity is more pronounced in the east of the country and thinness/wasting more pronounced in the west. Only 16/ 52 districts had an estimated wasting prevalence below the 2025 target threshold of 5% in 2017.African ethnicity, male gender, low birth weight, lower socio-economic and maternal/paternal education status and residence in a rural area were significantly associated with stunting. Children living in a lower income and food insecure household with young malnourished mothers were significantly more likely to be thin/wasted while African children, with higher birth weights, living in lower income households in KwaZulu-Natal and Eastern Cape were significantly more likely to be obese.
Conclusions: While improvements in stunting have been observed, thinness/wasting and obesity prevalence remain largely unchanged. The geographic and socio-demographic heterogeneity in childhood malnutrition has implications for equitable attainment of global nutritional targets for 2025. Many districts appeared to have dual epidemics of under and over nutrition (high within district heterogeneity and inequality). Effective public health planning and tailored interventions are required at the sub-national level to address this challenge.

Strengths and limitations of this study
 As primary panel study was not designed/powered for provincial and lower geographic level analysis, we cannot discount the resultant impact on precision/random variability when analysing at provincial/district level (administrative tier just below province) and further stratification by socio-demographic correlates.

Background
Despite reductions in malnutrition 150.8 million children (22.2%) under five are stunted and a further 50.5 million children are wasted 1 . Furthermore rapidly rising trend in overweight and obesity in children and adults 2-4 5 has emerged as one of the most serious global public health issues of the 21 st century 6 . Sub-Saharan Africa (SSA) has among the highest levels of child malnutrition 1 globally. This problem is particularly illustrated by South Africa 7 , a middle income country with high levels of wealth/economic inequality that is undergoing rapid socioeconomic and lifestyle changes that have precipitated a nutritional transition, high prevalence of overweight/obesity in children 8 . The dual burdens of undernutrition and overweight/obesity are not distributed in a spatially homogenous manner 9 , and the health risks associated with malnutrition vary by age, gender, ethnicity and geographical location 10 .
Progress to tackle all forms of child malnutrition remain much too slow 1 . In order to support the delivery of public health interventions that will be most effective at reducing malnutrition, an understanding of the geographical distribution of malnutrition is required. Limited data are collected at lower administrative unit level making it difficult to identify specific groups of high-risk individuals and thus, determine the most suitable and 1 Child malnutrition is defined as a pathological state as a result of inadequate nutrition, including undernutrition due to insufficient intake of dietary energy and other key nutrients resulting in stunting (low height for age) or wasting (low weightfor-length) and overweight and obesity due to excessive consumption of dietary energy and reduced levels of physical activity.

Study population
We restricted our analysis to children <5 years of age.

Outcomes
We calculated height for age (HA) and BMI-for-age (BA) z-scores using the WHO 2007 growth standards 15 16 .

Data analysis
Analyses were performed using Stata software version 15 [StataCorp. 2017. Stata Statistical Software: Release 15.
College Station, TX: StataCorp LLC]. Clustering, as well as survey design effects, were accounted for using sample weights to estimate standard error and 95% confidence intervals (CIs) around mean anthropometric zscore point estimates, both overall and stratified by other socio-demographic variables such ethnicity and gender, socio-economic status, and residence location type. Extrapolated population totals of malnourished children (< 5) by yearly age were estimated using the survey weights.
Space-time Bayesian modelling: Furthermore, we employed a Bayesian joint (shared component) space-time binomial model 18 to estimate stable malnutrition prevalence rates at provincial and district levels across the 5 waves. The model splits the risk of malnutrition into three spatio-temporal components: a shared component for all three malnutrition types (stunting, thinness/wasting and obesity) and two additional components that capture that unshared differences between the three types. The model formulation contains an additive decomposition for the shared part, space-time interaction terms common to the three malnutrition types and additional heterogeneity terms. This methodology was employed in an attempt to stabilise estimates at district level given that the primary sampling design was not developed to provide point estimates at this level of geographic disaggregation. Survey weighted prevalence's were applied to sample size totals by district and panel to obtain a survey weighted numerator count by malnutrition type in the binomial distribution. The joint space-time was fitted in WINBUGS  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Risk factors analysis: Two-way tabulations of key socio-demographic covariates, year and child nutritional status were performed using the 'svy: tab' function to produce survey weighted prevalence estimates. Tests of independence for complex survey data survey (weighted Pearson's chi-square test) was utilised to assess the significance of bivariate associations between malnutrition burden and year as well as socio-demographic covariates.
Ethical approval: Approval for the primary study was granted by the Ethics Committee of the University of Cape Town. The current analysis is a secondary data analysis of an open access dataset and does not require further ethical approval. Patient and Public Involvement: As this was a data analysis utilising secondary data from a national community based panel survey, the development of the research question was not informed by the study subjects. Likewise, we could not involve study participants in the design of this study. Study participants were not involved in conduct of the primary study. Results will be disseminated in the form of peer reviewed article as well as through presentation to senior members of our National Department of Health and KwaZulu-Natal Department of Health.

Study population
The sample of children <5 years of age in the 7,301 households included in the SA-NIDS survey increased from

Temporal changes in burden of malnutrition from 2008 to 2017)
Between 2008 and 2017, the prevalence of stunting among children aged under 5 years decreased significantly from 11.0% to 7.6% (p=0.007) ( Table 1). Over the same period, both the prevalence of wasting/thinness (and the prevalence of obesity decreased non-significantly (from 5.2 to 3.8%, p= 0.131 and 14.5% to 12.9%, p= 0.312 respectively). The prevalence of thinness was significantly (p<0.001) higher in children under 2 years of age (8% in 2008; 6% in 2017) compared to 4% in 2008 and 3% in 2017 among children 2 years and older. The prevalence of obesity was also significantly (p<0.001) higher among children under 2 years of age and increased over the study period (18.4% in 2008 vs 21.7 in 2017, p=0.331).

Factors associated with child nutritional status
A bivariate analysis of demographic, maternal, socio-economic and household factors at individual nutritional status level suggests that African ethnicity (p<0.001), male gender (p=0.002), low birth weight (<0.001), residing in lower socio-economic status household (p<0.001), province of residence (p=0.012), lower maternal/paternal education status (p<0.001, 0.020 respectively) and residence in a rural/tribal authority area (p<0.001) were significantly associated with stunting ( Table 2). Children living in lower income households (p=0.053), lower food security (as measured through child hunger in last year) (p<0.001), province of residence (p=0.002), having a younger mother (<20) (p=0.012) and mother having a lower BMI classification (p=0.005) was significantly associated with thinness/wasting status. Children of African ethnicity (p<0.001), higher birth weight (p=0.006), living in lower income households (p=0.001) in KwaZulu-Natal and Eastern Cape (p<0.001) as well as paternal educational attainment (p=0.033) were significantly associated with obesity status ( Table 2).

Discussion
Main findings: The present study illustrates that while stunting has declined among South African children over the last 10 years, wasting and obesity appear largely unchanged, suggesting that development and public health interventions have had a variable impact. Stunting prevalence appears relatively evenly spread across South Africa, but obesity burden is more pronounced in the east of the country, whereas thinness/wasting is more pronounced in the west, with only 16 of 52 districts with estimated wasting prevalence below the 5% (WHO 2025 target threshold) in 2017. A concerning pattern observed was the increase prevalence of obesity in children under the age of two years. Key socio-demographic factors associated with malnutrition status were identified which likely underpins the spatial patterns (and heterogeneity) observed across the country. African children with lower birth weights residing in lower income households in rural areas with less educated mothers and fathers were particular more likely to be stunted. Children in lower income, food insecure households with malnourished young mothers appeared particularly more likely to be thin/wasted while African children, with higher birth weights, living in lower income households in KwaZulu-Natal and Eastern Cape were also more likely to be obese.. Furthermore, low household income appeared to be positively associated with all 3 nutritional types. Declining childhood stunting rates from 2008-2017 may well have resulted from government initiatives to support food security and child health (among other things), but our findings of distinct geographic and socio-demographic variability in undernutrition and obesity rates suggest that tacking malnutrition in South Africa is complex.
Contribution to existing literature: Two previous studies in South Africa among primary school aged children dating back 25+ years (1993and 1994 respectively) utilised cross sectional data 19 20 , thus limiting insight into temporal trends. Furthermore, the study by Jinabhai et al. 19 was restricted to KwaZulu-Natal limiting national representativeness. Another cross sectional study in South African in 2001-2003 among primary school children in five South African Provinces suggested that relative to 1993 prevalence of undernutrition had decreased while obesity had increased 20 21 . Thus these previous data are now outdated, were largely focused on primary school aged children as well as cross sectional in nature and geographically restricted. This is also the first spatial-temporal Bayesian shared component analysis of malnutrition trends among children in South Africa utilising geographically representative repeated panel data over a 10-year period. The current study focusing on children under 5 year of age suggests that there is prominent geographic heterogeneity in malnutrition burden in South Africa in this youngest age group. This is in line with findings from other settings in Africa that have documented similar spatial heterogeneity 22 and persistence of these malnutrition inequalities has been demonstrated in an 80 country study further highlighting this ongoing public health conundrum 23 24 . Our results demonstrate a strong west to east gradient of higher underweight burden on the western side of South Africa and greater obesity on the eastern seaboard (Eastern Cape and KwaZulu-Natal). A map of poverty and inequality in South Africa 2 illustrates the co-existence of high levels of poverty and inequality in many parts of KwaZulu-Natal and the Eastern Cape with high levels of overweight/obesity. This is further confirmed by our individual child level analysis which suggested a significantly higher obesity prevalence in lower income households. Metropolitan areas displayed high levels of nutritional inequality that complement national studies of poverty and inequality 25 .
Under and over nutrition status appeared positively associated with lower household income classification. This finding of stunting and wasting disproportionately affecting the poor has been often demonstrated 26 . Other studies in Africa in particular have documented similar patterns i.e. children living in low SES households, children who live in peripheral areas and whose mothers had little or no schooling were at significantly higher risk of malnutrition 27 . The inconsistent challenges facing health authorities are occurring in the face of rapid urbanization 2 https://southafrica-info.com/people/mapping-poverty-in-south-africa/  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 F o r p e e r r e v i e w o n l y and industrialization that simultaneously attract both the rich and the poor to live in the same geographic districts 28 .The heterogeneous geographic relationship between household income and undernutrition is also affected by the allocation of household income that is a function of maternal education, access to markets, infrastructure and sanitation 29 . Additionally, these data suggest that there is a strong and highly significant association between higher food insecurity (child hunger frequency in the preceding year) and increased thinness/wasting. Community and government based packages of support need to be highly targeted to the poorest and most food insecure households to further reduce inequality in this regard and maximise reductions in malnutrition.
Our findings suggest that children with low birth weight (due to pre-term delivery, fetal/intrauterine growth restriction or a combination of the two) were significantly more likely to be stunted than normal weight babies and this has been demonstrated in many other low and middle income settings (for example 30 ). Socioeconomic status/factors are known risk factors for LBW 31 and may in part explain the significant association found between stunting and lower household income. South Africa has the higher number of incident and prevalent HIV infections globally 32 . A further important contextual risk factor for LBW is maternal HIV status. A systematic review and large observational studies focussing on low and middle incoming countries, suggest a strong and significant association between maternal HIV infection and LBW 33 34 . Evidence from South Africa also suggests the anthropometric z-score of HIV-infected children appear to be consistently lower when compared to HIVexposed but uninfected children 35 . We also observed a significantly higher prevalence of stunting among male children which has been demonstrated previously in a meta-analysis for sub-Saharan Africa 36 , the suggested cause of which might be that male children are more vulnerable to health inequalities relative to female children of the same age. Strengthening community-based packages of care and community health worker (CHW) performance/skills in rural and high burden geographies are key strategies to improve primary health care delivery through better identification of women at higher risk of poor birth outcomes (e.g. HIV positive, history of previous poor birth outcomes and/or currently malnourished), higher referral rates for facility births, and improved linkage to other health as well as social services 37 . Lastly given the high adolescent fertility rates in many parts of South Africa 38 , there is also much scope to improve CHW identification of households with higher risk malnourished adolescent girls prior to pregnancy to ensure more optimal linkage to government and social support to ensure adequate nutrition as well as improved awareness regarding family planning practices e.g. ensuring adequate birth spacing 39 .
Obesity in children has a complex aetiology that includes a wide range of socioeconomic, demographic, environmental and cultural variables 40 such as household composition, mother's education, household income,  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y household size, environmental factors, rural versus urban location, and sanitation 9 41 . The high burden of obesity is likely associated with a progressive increases in the per capita food supply and consumption of high calorific foods (e.g. fat, sugar, fast and/or processed foods) in South Africa 42 . This rapidly changing dietary pattern has, in part, been attributed to urbanisation, growing and expanding supermarkets /formal food retailers, and the availability of fast/processed foods 43 . An interesting finding in these data was the significant positive association between child obesity status and residing in a lower income household. This association has been demonstrated previously [44][45][46] and this evidence base is growing. This conforms with the idea that lower and higher income households/families often have a higher obesity risk than middle income households i.e. so called U-shaped association. Lower income or economically deprived families often replace health fresh food options with cheaper and more calorific processed foods 45 . Multiple studies have demonstrated that the majority of low-income South Africans have a low dietary diversity, and, therefore, consume a limited food range consisting predominantly of a starchy staple such as bread and maize, with low intakes of vegetables and fruit 42 . Future work will characterise food purchasing patterns (and changes over time) among households in South Africa which will be compared with paired longitudinal anthropometric measurements to identify specific dietary patterns associated with child nutritional status.
Lastly and contextually, body mass is culturally influenced in South Africa, and the high level of obesity in KwaZulu-Natal and Eastern Cape may at least in part be a result of cultural beliefs that associate overweight with wealth and good health 47 . Geographic patterns of higher obesity in South Africa appeared to overlap areas of high poverty particular on the eastern side of the country 3 and thus not solely concentrated among higher socioeconomic households.

Strengths:
To our knowledge this is the first spatial-temporal analysis of malnutrition trends among children under five years of age in South Africa. We used standardised anthropometric measurements of children and their mothers from a nationally representative repeated panel data over a 10-year period. The panel nature of the design allows assessment of change in malnutrition burden within the same individuals/households observed at multiple time points. A further strength was the implementation of a fully Bayesian space-time shared component model to produce more stable joint estimates of malnutrition by province, district and year. 3 https://southafrica-info.com/people/mapping-poverty-in-south-africa/  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Weaknesses: The study has several limitations. Firstly, missing or invalid weight/height measurements (especially in wave 2, and among infants -Supplementary Material 2) may have introduced selection bias (if not missing at random), and may thus have affected both the internal validity and the representativeness the findings in the broader South African context. Secondly as the primary panel study was not designed/powered for provincial 48 and lower geographic level analysis, we cannot discount the resultant impact on precision/random variability when analysing at provincial/district level (administrative tier just below province) and further stratification by socio-demographic correlates. Thirdly, we cannot discount the effect of inter-observer variability across different study districts, despite extensive interviewer training and standardization of study protocols. All anthropometric measurements (e.g. weight, height) were taken in duplicate in NIDS 26 which would have ensured better reliability.

Cost of malnutrition, policy and research needs:
Estimating the cost of child malnutrition in South Africa is extremely complicated and no locally-determined cost data exist. Data from the United States, suggest that the incremental lifetime direct medical cost for a 10-year-old obese child relative to a 10-year-old normal weight child ranges from USD 12 660 to USD 19 630 49 . Estimates of the cost of treating wasted children are approximately USD 200 per child 50 while stunting has been consistently linked to worse economic outcomes in adulthood 51 and estimates suggest that, on average, the future per capita income penalty for a stunted individual could be as large as 9-10% in SSA 52 . Urgent investments are needed to accelerate the reduction of all forms of malnutrition, as well as to curb the obesity epidemic among young children in South Africa. There is also considerable evidence indicates that childhood wasting and stunting can be reduced by 60% and 20% respectively using ten nutritionspecific interventions 53 , with an estimated return on investment (ROI) of 18:1, i.e. for USD 1 spent on implementing effective programmes there would be USD 18 return in future economic benefits 54 . Very few obesity prevention interventions targeting children have been effective and a comprehensive multifaceted strategy tackling diet, physical inactivity, coupled with psychosocial support and local food environment change may prove more effective. Nutrition policies tackling child obesity must promote household nutrition security and healthy growth, decrease overconsumption of nutrient-poor foods, better shield children from increasingly pervasive marketing of energy-dense, nutrient-poor foods and sugar sweetened beverages as well as reduction of growing physical inactivity 55 .
Our findings suggest the need to implement evidence-based child health strategies and policy (e.g. further social grant support to vulnerable and impoverished households) that is tailored to specific geographies and socially  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y disadvantaged sub-populations. Integrated nutrition programs in low and middle income countries (LMIC) have had a substantial impact on child nutrition and health via a combination of multisector targeted interventions 56 . Furthermore implementation and/or strengthening of school-based food program can provide a launching pad for preventive programs including education and awareness, provision of healthier/more nutrition food options and micronutrient supplementation, deworming, increased immunization coverage and improved growth monitoring as well as counselling 56 . This may be especially true of obese children where the highest prevalence was observed in higher income households with higher food purchasing power and where local food environments are likely is likely to be an important contextual determinant. A higher prevalence of child thinness/wasting among younger mothers (<25) in poorer, food insecure household, highlights the importance of policies that enable younger mothers to adequately care for their children in all settings.

Conclusions
The heterogeneity of malnutrition is a feature of spatial inequality and rapid urbanization that has manifested in widening levels of inequality in South Africa's districts and a need to reassess where nutrition programmes need to be further decentralised to the highest risk municipalities and local communities to maximise effectiveness.
This work provides the first district level ranking of childhood overweight, thinness/wasting and stunting and allows a differentiated pro-active tailored intervention to be developed for each municipal district. The dual epidemic of undernutrition and overweight/obesity requires differential geographical policy inputs in metropolitan areas and districts across the rural-urban divide. The current and future health cost of malnutrition among South African children cannot be overstated. There is an urgent need to address nutrition problems among preschool aged children in South Africa and other low and middle income countries. Effective public health planning and geographically/contextually tailored interventions are required at sub-national level to address this challenge. The analytical framework employed in this study we believe will have definite utility in other settings.

Funding
This study forms part of the Sustainable and Healthy Food Systems (SHEFS) project supported by the Wellcome Trust's Our Planet, Our Health programme (grant number 205200/Z/16/Z). The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report.

Competing interests
None declared.

Patient consent for publication
Not required.

45
Natal and Eastern Cape were significantly more likely to be obese.

75
Progress to tackle all forms of child malnutrition remain much too slow 1 . In order to support the delivery of 76 public health interventions that will be most effective at reducing malnutrition, an understanding of the 77 geographical distribution of malnutrition is required. Limited data are collected at lower administrative unit level 78 making it difficult to identify specific groups of high-risk individuals and thus, determine the most suitable and 1 Child malnutrition is defined as a pathological state as a result of inadequate nutrition, including undernutrition due to insufficient intake of dietary energy and other key nutrients resulting in stunting (low height for age) or wasting (low weightfor-length) and overweight and obesity due to excessive consumption of dietary energy and reduced levels of physical activity.

98
We restricted our analysis to children <5 years of age.

100
We calculated height for age (HA) and BMI-for-age (BA) z-scores using the WHO 2007 growth standards 15 16 .

101
We generated z-scores by transformation of child anthropometric data using the "lambda mu sigma" method 102 ('zanthro' function in Stata 15). As recommended, weight-for-length was used in children 0 to <2 years of age, 103 and BMI-for-age in children 2 years of age and older 17 . We defined obesity as weight-for-length z-score ≥+2 for 104 children under 2 years of age and BMI for age z-score of >2+ for children age 2 and older 17 . We defined wasting 105 as weight-for-length z-score < -2 for children under 2 years of age and thinness as BMI for age z-score < -2 for 106 children 2 years and older. Stunting was defined as HA z-score of < -2.

127
We employed Bayesian spatial-temporal modelling approach in an attempt to stabilise estimates at district level 128 given that the primary sampling design was not developed to provide point estimates at this level of geographic 129 disaggregation and resultant zero prevalence estimates for particular districts and waves. We choose a Bayesian 130 spatial-temporal formulation to model each of the anthropometric outcomes independently using an autoregressive 131 approach, suggested by a recent methodological comparison 19 , which fuses ideas from autoregressive time series 132 to link information in time and by spatial modelling to link information in space. We also opted for an 133 autoregressive model which only included the spatial term for every period and did not include a heterogeneous 134 term which resulted in a more parsimonious description of risk 20 .

161
The following prior distributions were assumed for the parameters defined above:         Western Cape (at 5.6%) followed by Northern Cape (4.9%) and North West (4.6%) (Figure 2b). The estimates 221 suggest that at provincial level 7 of 9 provinces were above the 5% target threshold for wasting in 2017 (only

222
Eastern Cape and KwaZulu-Natal were the exceptions). There appeared to be a general gradient of higher burden        and industrialization that simultaneously attract both the rich and the poor to live in the same geographic districts 301 33 .The heterogeneous geographic relationship between household income and undernutrition is also affected by the 302 allocation of household income that is a function of maternal education, access to markets, infrastructure and 303 sanitation 34 . Additionally, these data suggest that there is a strong and highly significant association between 304 higher food insecurity (child hunger frequency in the preceding year) and increased thinness/wasting. Community

305
and government based packages of support need to be highly targeted to the poorest and most food insecure 306 households to further reduce inequality in this regard and maximise reductions in malnutrition.

307
Our findings suggest that children with low birth weight (due to pre-term delivery, fetal/intrauterine growth 308 restriction or a combination of the two) were significantly more likely to be stunted than normal weight babies and 2 https://southafrica-info.com/people/mapping-poverty-in-south-africa/

328
Obesity in children has a complex aetiology that includes a wide range of socioeconomic, demographic, 329 environmental and cultural variables 45 such as household composition, mother's education, household income,

345
Lastly and contextually, body mass is culturally influenced in South Africa, and the high level of obesity in

398
The heterogeneity of malnutrition is a feature of spatial inequality and rapid urbanization that has manifested in 399 widening levels of inequality in South Africa's districts and a need to reassess where nutrition programmes need 400 to be further decentralised to the highest risk municipalities and local communities to maximise effectiveness.

401
This work provides the first district level ranking of childhood overweight, thinness/wasting and stunting and

Ethics approval 564
This study utilised open access data and hence ethical approval was not necessary.

Pairwise correlation for anthropometric outcomes and bivariate spatial autocorrelation
We have performed additional supplementary analyses (suing GeoDa: Anselin L, Syabri I, Kho Y. GeoDa: an introduction to spatial data analysis. Geographical analysis. 2006 Jan;38(1):5-22) which assesses pairwise correlation/association between the 3 outcomes as well as bivariate Moran's I to assess if there was significant spatial autocorrelation between the outcomes. This analysis suggests that there is no significant association between stunting and thinness/wasting while there is weak positive but significant spatial autocorrelation between stunting and obesity prevalence as well as weak negative spatial correlation between thinness and obesity (please see detailed analyses below).

Variances and covariances of random effects
Variances and covariances of random effects

Variances and covariances of random effects
Variances and covariances of random effects

Univariate spatial autocorrelation
Based on the univariate Moran's I statistics for each anthropometric outcome there appeared to be significant spatial heterogeneity present for all 3 outcomes.

Supplementary 8: post hoc power analysis
We performed a post hoc power analysis to assess the minimum effect size detectable among infants which has the smallest number of observations. The post hoc power analysis suggests that the sample size in the smallest age group has the power to detect a small effect size (w~0.

76
Progress to tackle all forms of child malnutrition remain much too slow 1 . In order to support the delivery of 77 public health interventions that will be most effective at reducing malnutrition, an understanding of the 78 geographical distribution of malnutrition is required. Limited data are collected at lower administrative unit level 79 making it difficult to identify specific groups of high-risk individuals and thus, determine the most suitable and 1 Child malnutrition is defined as a pathological state as a result of inadequate nutrition, including undernutrition due to insufficient intake of dietary energy and other key nutrients resulting in stunting (low height for age) or wasting (low weightfor-length) and overweight and obesity due to excessive consumption of dietary energy and reduced levels of physical activity.

101
We restricted our analysis to children <5 years of age.

103
We calculated height for age (HA) and BMI-for-age (BA) z-scores using the WHO 2007 growth standards 16 17 .

104
We generated z-scores by transformation of child anthropometric data using the "lambda mu sigma" method 105 ('zanthro' function in Stata 15). As recommended, weight-for-length was used in children 0 to <2 years of age, 106 and BMI-for-age in children 2 years of age and older 18 . We defined obesity as weight-for-length z-score ≥+2 for

131
We employed Bayesian spatial-temporal modelling approach in an attempt to stabilise estimates at district level 132 given that the primary sampling design was not developed to provide point estimates at this level of geographic 133 disaggregation and resultant zero prevalence estimates for particular districts and waves. We choose a Bayesian 134 spatial-temporal formulation to model each of the anthropometric outcomes independently using an autoregressive 135 approach. We employed a Bayesian hierarchical binomial model that simultaneously attempts to estimate the

207
Temporal changes in burden of malnutrition from 2008 to 2017)

323
Our findings suggest that children with low birth weight (due to pre-term delivery, fetal/intrauterine growth 324 restriction or a combination of the two) were significantly more likely to be stunted than normal weight babies and 325 this has been demonstrated in many other low and middle income settings (for example 37 ). Socioeconomic the anthropometric z-score of HIV-infected children appear to be consistently lower when compared to HIV-332 exposed but uninfected children 42 . We also observed a significantly higher prevalence of stunting among male 2 https://southafrica-info.com/people/mapping-poverty-in-south-africa/

414
The heterogeneity of malnutrition is a feature of spatial inequality and rapid urbanization that has manifested in 415 widening levels of inequality in South Africa's districts and a need to reassess where nutrition programmes need 416 to be further decentralised to the highest risk municipalities and local communities to maximise effectiveness.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   16   417 This work provides the first district level ranking of childhood overweight, thinness/wasting and stunting and 418 allows a differentiated pro-active tailored intervention to be developed for each municipal district. The dual 419 epidemic of undernutrition and overweight/obesity requires differential geographical policy inputs in metropolitan 420 areas and districts across the rural-urban divide. The current and future health cost of malnutrition among South

421
African children is likely substantial based on previous costing estimates. There is an urgent need to address 422 nutrition problems among preschool aged children in South Africa and other low and middle income countries.

423
Effective public health planning and geographically/contextually tailored interventions are required at sub-424 national level to address this challenge. The analytical framework employed in this study we believe will have 425 definite utility in other settings.