Estimating subnational prevalence trends of cardiovascular disease risk factors, tobacco use, overweight, and obesity in East Africa using adaptive kernel density estimation for survey data

doi:10.21203/rs.3.rs-2748445/v1

Download PDF

Research Article

Estimating subnational prevalence trends of cardiovascular disease risk factors, tobacco use, overweight, and obesity in East Africa using adaptive kernel density estimation for survey data

https://doi.org/10.21203/rs.3.rs-2748445/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 24 Aug, 2023

Read the published version in International Journal of Health Geographics →

You are reading this latest preprint version

Background: Cardiovascular disease (CVD) is increasing in Sub-Saharan Africa (SSA). Overweight/obesity and tobacco use are modifiable CVD risk factors, however literature about the spatiotemporal dynamics of these risk factors in the region at subnational or local scales is lacking. We describe the spatiotemporal trends of overweight/obesity and tobacco use at subnational and subpopulation levels over a 13-year period (2003 to 2016) in five East African nations.

Methods: Cross-sectional, nationally representative Demographic and Health Surveys (DHS) were used to explore the subnational spatiotemporal patterns of tobacco use and overweight/obesity in Burundi, Kenya, Rwanda, Tanzania, and Uganda, five East African Community (EAC) nations with unique cultural landscapes influencing CVD risk factors. Adaptive kernel density estimation and logistic regression were used to determine the spatial distribution and change over time of CVD risk factors on a subnational and subpopulation (rural/urban) scale.

Results: Subnational and subpopulation analysis shows that regional and national level analysis masks important trends in CVD risk factor prevalence. Overweight/obesity and tobacco use trends were not similar: overweight/obesity prevalence increased across most nations included in the study and the inverse was true for tobacco use prevalence.

Urban populations in each nation were more likely to be overweight/obese than rural populations, although the magnitude of difference varied widely between nations. Although the prevalence of overweight/obesity increased over time in both urban and rural populations, spatial analysis revealed that the rate of change was not equal across all urban and rural areas.

Rural populations were more likely to use tobacco than urban populations, though the likelihood of use varied substantially between nations. Additionally, spatial analysis showed that tobacco use was not distributed evenly over space: tobacco use increased in and around major cities and urban centers but declined in rural areas.

Conclusions: We highlight the importance of de-homogenizing research about CVD risk factors in SSA. Studies of national prevalence trends for CVD risk factors mask important information about subpopulation and place-specific behavior and drivers of risk factor prevalence. Spatially explicit studies are vital to understand local drivers of risk factor trends which can in turn inform policy and public health intervention.

obesity

tobacco use

adaptive kernel density estimation

Demographic Health Survey

Burundi

Kenya

Rwanda

Tanzania

Uganda

Overweight/obesity and tobacco use are significant and modifiable cardiovascular disease (CVD) risk factors in SSA (1–3). Understanding the spatial and temporal patterns of these risk factors is important to design early interventions in the CVD continuum, especially in resource poor areas. In Sub-Saharan Africa (SSA), literature about the prevalence and spatiotemporal patterns of CVD risk factors is limited. Existing studies are either locally specific ecological studies that do not allow for comparison across space and time or studies that pool data from multiple SSA countries to make inferences about all of SSA (4, 5). However, SSA is a diverse region with varying population structures, cultures, politics, and economic development levels that influence health and disease outcomes. Health, disease, and associated risk factor dynamics differ across populations and time and are not static. Although literature exists about the temporal trends of CVD risk factors in SSA, there is a lack of literature about the spatial dynamics of those trends within and across SSA countries. For instance, trend studies often report overweight/obesity or tobacco use prevalence at the national level and do not delve into subnational or subpopulation differences, such as differences between urban and rural populations (5, 6), despite evidence from SSA suggesting a strong association between residence type and CVD risk factor prevalence (7–11).

In this study, cross-sectional nationally representative data from five East African countries over a 13-year period were used to explore the spatial and temporal patterns of overweight/obesity and tobacco use. Statistical and geospatial methods were employed to examine the subnational spatiotemporal trends of overweight/obesity and tobacco use in an East African context. Administrative boundaries within countries, such as provinces or counties, are not always appropriate as a unit of analysis since variations in spatial phenomena, such as disease, rarely conform to administrative boundaries and variation within boundaries can be lost (12). Additionally, in a region where subnational administrative boundaries are not static, temporal analysis of disease rates linked to administrative boundaries is not feasible. In the East African Community (EAC) first level administration boundaries have been partially or entirely redrawn since 2000; for instance, in 2010 Kenya overhauled level one administrative boundaries from eight provinces to 47 counties (13). This paper employs adaptive kernel density estimation to create maps with continuously varying surfaces, not limited by subnational administrative boundaries, to analyze spatiotemporal trends of CVD risk factors within nations.

Five EAC nations, Burundi, Kenya, Rwanda, Tanzania, and Uganda, with shared borders and varying political, economic, and cultural environments were included in the study to illustrate the variability in subnational CVD risk factor trends in a SSA region. The objective of this study was to describe the spatiotemporal trends of CVD risk factors, overweight/obesity and tobacco use, at subnational and subpopulation levels over a 13-year period (2003 to 2016). Differences in CVD risk factor prevalence between rural and urban areas were analyzed and geospatial methods were used to visualize spatial and temporal trends within nations. This study provides insight into geographic variations of CVD risk factors observed within SSA countries, which ought to be considered in the creation of context-specific and locally appropriate CVD risk factor intervention strategies in the region (14, 15).

Study Area

Burundi, Rwanda, Kenya, Tanzania, and Uganda are member states of the East African Community (EAC) organization in the African Great Lakes region of eastern Africa (Fig. 1). The EAC is a regional organization that links member states in economic partnership and aims to become a political federation. Member nations have unique political, economic, and disease challenges that shape the current sociocultural and health landscapes.

The nations selected for this study fall into two World Bank income categories: low-income countries (Burundi, Rwanda, and Uganda) and lower-middle income countries (Tanzania and Kenya). Gross domestic product per capita (GDP) 2020 estimates in the study area were well below the world GDP (US$10,910) but varied significantly (16). GDP ranged from US$238.99 in Burundi to US$1878.58 in Kenya. GDP 2020 estimates in Rwanda, Uganda, and Tanzania were US$797.86, US$822.03, and US$1076.47, respectively (16). Demographic and result tables are presented in order of increasing GDP (Burundi, Rwanda, Uganda, Tanzania, and Kenya) for ease of comparison and comprehension.

Data Source

Data was obtained from Demographic and Health Surveys from 2003 to 2016 in Kenya, Tanzania, Rwanda, and Uganda and from 2010 to 2016 in Burundi (17). The Demographic and Health Survey (DHS) is a cross sectional and nationally representative survey conducted approximately every five years in low- and middle-income countries. Sampling, questionnaire, and interview techniques are spatially and temporally standardized across all surveys, thus allowing for comparative assessment. A stratified two-stage sample design is designed for each country using census information from national statistical boards. In the first stage, primary sampling units or sample clusters, are randomly selected within strata that represent urban and rural areas in administrative regions. Urban and rural areas are defined by national statistical boards and may vary from country to country. As there is no standard, universally accepted metric to define urban-rural areas, they may be defined by population density or infrastructure or both (18). Second, households within the sampling clusters are randomly sampled without replacement. A weighting variable is provided with each survey to make samples nationally representative.

In most surveys since the early 2000s, DHS provides geographical coordinates (latitude and longitude) for each sample cluster. Households sampled within a cluster are assigned the same coordinates which are offset by two kilometers in urban areas and five kilometers in rural areas to maintain respondent confidentiality. Data from two survey periods (2010 and 2016) in Burundi were included in this study and surveys from three survey periods (around 2005, 2010, and 2015) were included for Kenya, Rwanda, Tanzania, and Uganda. Except for the 2005 Tanzania survey, all other surveys included geographical coordinates for sample clusters.

Variables

Overweight/obesity and tobacco use, CVD risk factors, were derived from the DHS. Interviewers collected anthropometric measures, height and weight, for women using a SECA 878 U digital scale and measuring board. BMI data was not collected for men. Therefore, the prevalence of overweight/obesity in this study refers only to women between 15 and 49 years old who were not pregnant. Overweight and obesity categories were assigned using the World Health Organization (WHO) definition, where BMI ≥ 25 kg/m² is overweight and BMI ≥ 30 kg/m² is obese. To ensure enough statistical power for the analysis, overweight and obesity were combined to form one category, overweight/obese.

The DHS collects self-reported data on tobacco use habits of both sexes, therefore women between 15 and 49 years old and men between 15 and 59 years old were included in this portion of the study. Smoking and smokeless tobacco were combined into one category for this study, to ensure enough statistical power. The prevalence of tobacco use, in any form, was determined using the following questions:

Do you currently smoke cigarettes? (Yes or No) in some surveys this question was asked as: Do you currently smoke cigarettes every day, somedays, or not at all? (not at all, every day, some days)

Do you currently smoke or use any other type of tobacco? (Yes or No) in some surveys this question was asked as: Do you currently smoke or use any other type of tobacco every day, somedays, or not at all? (not at all, every day, some days)

What other type of tobacco do you currently smoke or use? (pipes full of tobacco, cigars or cigarillos, water pipes/shisha, snuff by mouth, snuff by nose, chewing tobacco, other)

All affirmative answers to these questions were coded as one and summed to create an index that ranged from 0 to 3. A tobacco use variable was then created by coding respondents with a 0 score as “0” for no tobacco use and respondents with a 1 to 3 score as “1” for yes.

Analysis

Statistical Analysis

Sampling clusters in the DHS are selected from strata that represent sub-national regions and urban/rural places within those regions. Design weights are provided with each survey to compensate for over or under sampling, non-response, and other issues that may arise to ensure data is unbiased and representative of the national population. The DHS recommends the use of survey design and weights for valid estimates of statistical measures (19). Surveys in each country were merged into one (i.e. 2010 and 2016 Burundi surveys into a pooled ‘Burundi’ file) and design weights were re-calculated to give each survey equal weight and avoid biased estimates due to original sample size (20). Weights for the pooled survey were calculated as follows, where s = survey and w = weights:

w _{s1_new} = (w_s1/∑w_s1) / (∑w_s1 + ∑w_s2 + ∑w_s3 / n_s)

Logistic regression was used to analyze the sub-national dynamics of CVD risk factor prevalence over time and place (urban or rural residence). Statistical analysis was performed at the individual level, so the unit of analysis was each survey respondent, and survey design and weights were considered. Analysis was divided into two parts: 1) analysis of CVD risk factor prevalence over time within each nation and 2) subnational analysis of CVD risk factor prevalence over place (rural/urban). The INDEPTH (International Network for the Demographic Evaluation of Populations and Their Health) 2013 population standard for SSA, based on just SSA populations, was used to age-standardize prevalence rates for descriptive purposes in this study (21). All statistical analysis was performed in R using the survey package for complex survey analysis (22, 23).

Spatial Analysis

Kernel density estimation (KDE) is an interpolation method that uses features within a neighborhood (i.e., kernel) to estimate the value of raster cells (24). A curved surface is fitted over each point and the surface value is highest at the point location and declines with increasing distance from the point until reaching zero at a search radius determined by the user. At each output raster cell, kernel surface values at the raster center are added to determine the value of the cell (24). Selecting the appropriate size for the search radius (i.e., bandwidth) of the kernel can have a strong influence on the resulting surface estimate. Two KDE methods are employed by geo-statisticians; fixed-bandwidth KDE or adaptive KDE. In the fixed-bandwidth method, the same bandwidth is applied across the entire study area, which implies that one bandwidth size will capture a similar number of points across the study area. However, if one is estimating health data in rural and urban areas, the same size bandwidth would capture more points in densely populated urban areas and less points in sparsely populated rural areas (25, 26). Adaptive KDE uses bandwidth sizes that vary depending on the location of the sample points. Therefore, smaller bandwidths are used in heavily populated areas (i.e., urban), moderately sized bandwidths in peri-urban/suburban areas, and larger bandwidths in sparsely populated areas that are more to capture a similar number of points for surface estimation. Selecting the appropriate size for the search radius (i.e., bandwidth) of the kernel can have strong influence on the resulting surface estimate, therefore KDE with adaptive bandwidths is most appropriate for health data that is often unevenly distributed (12, 25, 27).

The GIS data provided by the DHS links multiple individual observations to one spatial location, the sample cluster. The number of observations per sample cluster varies by location, where clusters in densely populated urban areas often have more observations than clusters in rural areas. Larmarange et al. (2011) developed a spatial interpolation package (prevR) which uses Gaussian kernel estimators with an adaptive bandwidth of equal number of observations to interpolate prevalence surfaces from DHS data (28). This method achieves high geographic detail in densely populated/sampled areas and maintains stability in sparsely populated areas with low observations. The adaptive bandwidth depends on a minimum number of observations as defined by the user. In this study, the minimum number of observations (N) was selected through the ‘Noptim()’ function in prevR, where the optimal N value is a function of the observed national prevalence, the number of people sampled, and the number of survey clusters (29). The prevR KDE method with adaptive bandwidths has been used to map health and disease outcomes, model disease risk, equity, and social vulnerability (30–35).

The Larmarange adaptive estimation method occurs in three steps. First, a kernel is drawn around sample cluster points that contain a minimum number of observations (N), as defined by the user, or can be generated within the program using the ‘Noptim ()’ function. The clusters are summed to create an intensity surface of sampled individuals. Second, the method described above is used to create an intensity surface for individuals with the variable of interest, i.e., positive for overweight/obesity or tobacco use. Lastly, the ratio of the intensity of individuals with the variable of interest to the surface of the number sampled is used to calculate the prevalence surface. The use of adaptive bandwidths of a minimum number of sampled people creates interpolated surfaces that adapt to the irregular spatial distribution of sample clusters in urban and rural areas (29). Spatial analysis was conducted in R using prevR, sp, and raster packages (23, 28, 36, 37). Maps were created in ESRI ArcGIS Desktop (38).

Only women between 15 and 49 years old, who were not pregnant during the survey were included in the overweight/obesity portion of the study. A total of 182,310 women were included in the 2003 to 2016 DHS surveys from the EAC. Pregnant women (n = 15,378, 8.43%) were excluded to reduce potential bias from pregnancy related weight gain, resulting in an analytical dataset of 166,932 women. The number of participants varied by country. Uganda had the lowest number of participants in this study (8,755) followed by Burundi (24,382), Rwanda (35,695), Kenya (44,383), and Tanzania (30,575). A total of 250,180 men and women, between age 15 and 59 years old were included in the tobacco use portion of the study. However, the number of participants varied by country. Burundi had the lowest number of participants (n = 38,490), followed by Tanzania (n = 42,410), Uganda (n = 45,845), Rwanda (n = 55,855), and Kenya (n = 67,580).

Overweight/obesity

The national prevalence of overweight/obesity in nonpregnant women between 15 and 49 years old increased from the start of the study, around 2005, to the last year of study, around 2015, in Rwanda, Uganda, Tanzania, and Kenya. Although three years of study were included in all other nations, there were only two years of DHS data available for Burundi, 2010 and 2016. The national prevalence of overweight/obesity in Burundi during that period was not significantly different, remaining at slightly less than 8% prevalence. In the same time frame 2009 to 2016, overweight/obesity prevalence increased by 29% in Rwanda, 26% in Uganda, 33% in Tanzania, and 31% in Kenya (Table 2.1). Over the 10-year period, 2003 to 2016, Rwanda experienced the highest change in overweight/obesity prevalence, from 11.54% (CI: 10.49%, 12.58%) in 2005 to 20.94% (CI: 19.77%, 22.12%) in 2015, an 82% relative difference (Table 2.1).

Women residing in urban areas were significantly more likely to be overweight/obese compared to rural populations in all the nations across all time periods (Table 2.2). Burundi, the lowest income nation in the study, displayed the greatest difference between urban and rural populations across all years of study. In 2010 Burundi, women residing in urban areas were 6.88 (95% CI: 5.35, 8.84) times more likely to be overweight/obesity compared to women in rural areas, by 2016 the odds declined slightly to 6.19 (95% CI: 4.89, 7.83) (Table 2.2). Kenya, the highest income nation in the study, had the lowest overweight/obese prevalence gap between urban and rural populations, in 2003 urban populations were 2.78 (95% CI: 2.35, 3.29) times more likely to be overweight/obese compared to rural populations and by 2014 the odds declined to 2.14 (95% CI: 1.94, 2.47) (Table 2.2). Like Burundi and Kenya, the overweight/obesity prevalence gap between urban and rural areas in Uganda and Tanzania also narrowed over time. In 2006 Uganda, urban populations were 3.48 (95% CI: 2.53, 4.79) times more likely to be overweight/obesity compared to rural populations and by 2016 the odds declined to 2.11 (95% CI: 1.75, 2.54) (Table 2.2). In 2005 Tanzania, urban populations were 3.66 (95% CI: 3.08, 4.36) times more likely to be overweight/obesity compared to rural populations and by 2016 the odds declined to 2.72 (95% CI: 2.38, 3.12) (Table 2.2). Rwanda was the only nation where the difference in prevalence between urban and rural areas increased over time, where urban populations at the beginning of the study were 2.17 (95% CI: 1.73, 2.71) times more likely to be overweight/obese compared to rural populations and at the end of the study those odds increased to 2.89 (95% CI: 2.45, 3.40) (Table 2.2).

The uneven distribution of overweight/obesity between urban and rural communities was also evident on a spatial scale (Fig. 2 to Fig. 6). Spatial interpolation of overweight/obesity prevalence was conducted on two DHS survey waves in each of the countries, around 2010 and around 2015. In earliest year of study across all nations, the highest rates of overweight/obesity prevalence were found in or around capital cities. This trend was still prevalent in the last year of study, however, more urban locations reported high overweight/obesity prevalence rates as well. A ratio of difference between the overweight/obesity prevalence surfaces in the last year of study over the first year of study, showed that the rate of obesity increased across most of the country in all the nations and was more evident in rural areas.

In 2010, the high prevalence of overweight/obesity was isolated in and around Bujumbura, the capital city (Fig. 2A). By 2016, Bujumbura still reported the highest rates of overweight/obesity, although more urban locations in the country reported high rates of overweight/obesity (Fig. 2B). However, the greatest change in overweight/obesity prevalence, ranging up to 5.29 times higher than 2010, occurred in rural areas north of Bujumbura and in the east near Rutana (Fig. 2C). The highest overweight/obesity prevalence estimates in 2010 Rwanda were found in the capital city, Kigali, and in the northeastern and northwestern border of the country, ranging up to 39% (Fig. 3A). In 2015 the highest overweight/obesity prevalence estimates were still located in Kigali, ranging up to 51%, but higher rates were also found around other cities in the country, such as Butare in the south and Kibungo in the east (Fig. 3B). Although the overweight/obesity prevalence increased across most of the nation, in some areas the rates were over 8 times higher in 2015 compared to 2010 (Fig. 3C).

In 2011 Uganda, higher rates of overweight/obesity were isolated to the national capital, Kampala, and several rural and urban areas in the south (Fig. 4A). By 2016, the highest rates of obesity were found predominantly in central and south Uganda, especially around Kampala, although high prevalence values were distributed throughout the southern region (Fig. 4B). The prevalence of overweight/obesity increased across most of the nation, but the change in prevalence was most pronounced in areas around Moroto in the northeast, Arua in the northwest, and in areas around Kampala (Fig. 4C). However, the rate of change between 2011 to 2016 ranged from 0.37 to 9.59 times higher in 2016 (Fig. 4C).

The highest overweight/obesity prevalence in 2010 Tanzania were in and around the capital city, Dar es Salaam, and major border cities Arusha and Moshi in the north (Fig. 5A). In 2016, overweight/obesity prevalence rates remained high near the capital, Arusha, and Moshi but were also high in other urban centers such as Tanga in the northeast, Mbeya in the south, and Morogoro, a town west of Dar es Salaam (Fig. 5B). Although overweight/obesity increased across most of the nation, the highest rates of change, up to 8.23 time higher, were found near Arusha in the north, near the southeastern border, and in areas around the border town of Kigoma in the northeast (Fig. 5C).

In 2009 the highest rates of overweight/obesity in Kenya were in central Kenya, in and around the capital city, Nairobi (Fig. 6A). By 2014, the highest rates of overweight/obesity were still located in central Kenya but also around urban centers on the southern border and along the coastline to the east (Fig. 6B). The rate of change between 2009 and 2014 was most pronounced in the sparsely populated region of northern Kenya, near Wajir in the northeast and close to the Ugandan border in the northwest (Fig. 6C).

Table 1

Subnational trends of overweight/obesity prevalence in urban and rural areas. Rural residence used as reference level for logistic regression.
Country	Period	n	Age-adjusted Prevalence (%, 95% CI)	Weighted Prevalence (%, 95% CI)	Odds Ratio (95% CI)
Burundi	2010	4104	7.63 (6.80, 8.53)	7.57 (6.74, 8.41)	Reference
Burundi	2016	7908	7.84 (7.24, 8.48)	7.90 (7.07, 8.73)	1.05 (0.89, 1.24)
Rwanda	2005	5211	11.57 (10.66, 12.54)	11.54 (10.49, 12.58)	0.67 (0.59, 0.76)^***
	2010	6474	16.22 (15.25, 17.24)	16.26 (15.32, 17.21)	Reference
	2015	6217	20.60 (19.49, 21.76)	20.94 (19.77, 22.12)	1.36 (1.24, 1.51)^***
Uganda	2006	2519	16.23 (14.70, 17.89)	16.35 (14.26, 18.44)	0.85 (0.69, 1.04)
	2011	2420	18.79 (17.08, 20.63)	18.76 (16.73, 20.79)	Reference
	2016	5415	23.54 (22.27, 24.87)	23.70 (22.03, 25.37)	1.35 (1.14, 1.58)^***
Tanzania	2005	9159	17.31 (16.47, 18.18)	17.61 (16.42, 18.80)	0.79 (0.71, 0.89)^***
	2010	9097	20.61 (19.69, 21.56)	21.27 (19.92, 22.63)	Reference
	2016	12027	27.34 (26.43, 28.28)	28.31 (26.92, 29.70)	1.46 (1.31, 1.63)^***
Kenya	2003	7184	22.94 (21.85, 24.07)	23.17 (21.58, 24.75)	0.91 (0.78, 1.06)
	2009	7692	24.30 (23.23, 25.42)	24.95 (22.64, 27.26)	Reference
	2014	13455	31.00 (30.08, 31.94)	32.76 (31.47, 34.05)	1.47 (1.28, 1.68)^***

*p < 0.05 **p < 0.01 ***p < 0.001

Table 2

Subnational trends of overweight/obesity prevalence in urban and rural areas. Rural residence used as reference level for logistic regression.
		Rural (reference)			Urban
Country	Period	n	%	95% CI	n	%	95% CI	Odds Ratio (95% CI)
Burundi	2010	3195	5.22	4.41, 6.02	909	27.45	23.67, 31.24	6.88 (5.35, 8.84)^***
Burundi	2016	6236	5.23	4.48, 5.98	1672	25.46	22.06, 28.86	6.19 (4.89, 7.83)^***
Rwanda	2005	4022	9.96	8.88, 11.04	1189	19.34	16.41, 22.27	2.17 (1.73, 2.71)^***
	2010	5332	14.65	13.63, 15.67	1142	25.26	22.93, 27.59	1.97 (1.70, 2.28)^***
	2015	4599	16.91	15.73, 18.09	1618	37.00	33.71, 40.29	2.89 (2.45, 3.40)^***
Uganda	2006	2095	12.75	10.66, 14.84	424	33.72	28.09, 39.35	3.48 (2.53, 4.79)^***
	2011	1650	14.29	12.16, 16.43	770	34.95	30.76, 39.15	3.22 (2.50, 4.16)^***
	2016	4147	19.89	17.89, 21.89	1268	34.33	31.24, 37.41	2.11 (1.75, 2.54)^***
Tanzania	2005	6868	11.52	10.37, 12.68	2291	32.30	29.45, 35.14	3.66 (3.08, 4.36)^***
	2010	6699	15.12	13.56, 16.68	2398	36.01	33.53, 38.50	3.16 (2.68, 3.72)^***
	2016	8225	20.68	19.22, 22.13	3802	41.51	39.06, 43.96	2.72 (2.38, 3.12)^***
Kenya	2003	4768	18.18	16.44, 19.92	2416	38.20	35.40, 41.00	2.78 (2.35, 3.29)^***
	2009	5282	19.94	17.66, 22.22	2410	39.63	36.92, 42.34	2.64 (2.20, 3.17)^***
	2014	8481	25.81	24.27, 27.35	4974	43.23	40.99, 45.46	2.19 (1.94, 2.47)^***

*p < 0.05 **p < 0.01 ***p < 0.001

Tobacco Use

Tobacco use prevalence by adults of both sexes between 15 and 59 years old declined over time across all the nations included in the study (Table 2.3). Although three years of study were included in all other nations, there were only two years of DHS data available for Burundi, 2010 and 2016. However, in that 6-year period, there was a 41% decline of tobacco use prevalence, from 13.63% (CI: 12.79%, 14.47%) in 2010 to 8.03% (CI: 7.55%, 8.51%) in 2016. Tobacco use prevalence declined by 29% in Rwanda, 32% in Uganda, 28% in Tanzania, and 16% in Kenya in the same period. Uganda experienced the highest relative change of tobacco use, from 8.08% (CI: 7.19%, 8.97%) in 2006 to 3.71% (CI: 3.36%, 4.06%) in 2016, an almost 55% decline in tobacco use over 10-years (Table 2.3).

In all the nations and across all survey years, adults residing in rural areas were significantly more likely to use tobacco compared to urban populations (Table 2.4). Burundi, the lowest income nation in the study, displayed the greatest difference between rural and urban populations across all years of study. Rural populations in 2010 Burundi were 6.07 (95% CI: 4.76, 7.75) times more likely to use tobacco that urban populations and by 2016 the odds declined to 5.76 (95% CI: 4.63, 7.18) (Table 2.4). Kenya, the highest income nation in the study, had the lowest tobacco use prevalence gap between rural and urban populations, in 2003 rural populations were 2.84 (95% CI: 2.40, 3.35) times more likely to use tobacco compared to urban populations and by 2014 the odds declined to 2.18 (95% CI: 1.94, 2.45). Like Burundi and Kenya, the tobacco use prevalence gap between urban and rural areas in Uganda and Tanzania also narrowed over time. In 2006 Uganda, rural populations were 3.16 (95% CI: 2.34, 4.27) times more likely to use tobacco compared to urban populations and by 2016 the odds declined to 2.16 (95% CI: 1.83, 2.55). In 2005 Tanzania, rural populations were 3.35 (95% CI: 2.84, 3.95) times more likely to use tobacco compared to urban populations and by 2016 the odds declined to 2.59 (95% CI: 2.27, 2.95) (Table 2.4). Rwanda was the only nation where the difference in tobacco use prevalence between rural and urban areas increased over time, where rural populations at the beginning of the study were 2.05 (95% CI: 1.65, 2.55) times more likely to be overweight/obese compared to urban populations and at the end of the study those odds increased to 2.74 (95% CI: 2.34, 3.22) (Table 2.4).

The uneven distribution of tobacco use prevalence between rural and urban communities was also evident on prevalence surface maps (Fig. 7 to Fig. 11). Unlike the spatial distribution of overweight/obesity prevalence, the highest prevalence of tobacco use was not located in large urban cities and was mainly located in rural areas in the first years of study. In the last year of study, overall tobacco use prevalence estimates declined over the entire landscape in all the nations (Fig. 7 to 11). A ratio of difference between tobacco use prevalence surface estimates in the last year of study over the first year of study, showed that although the overall use of tobacco use declined in the last year of study, this decline was not evenly distributed across space. In fact, tobacco use prevalence appeared to increase mainly in and around small urban centers.

In the first year of study in Burundi, tobacco use prevalence estimates ranged up to 32% in northern rural areas between Bubanza and Muyinga (Fig. 7A). By 2016, the highest rates were less than 21% and were in approximately the same areas in the north (Fig. 7B). Although overall tobacco use prevalence estimates declined across the country, in some rural areas to the west and south the use of tobacco doubled over the 6-year period (Fig. 7C). In Rwanda, the highest tobacco use prevalence estimates in 2010 were on the southern border with Burundi, ranging up to 24% prevalence (Fig. 8A) and by 2015, prevalence declined to less than 17% across the entire nation (Fig. 8B). However, like Burundi, there were pockets of increased tobacco use in rural areas around the nation and on the outskirts of the capital city, Kigali (Fig. 8C).

In 2011 Uganda, the highest rates of tobacco use, which ranged up to 66%, were located along the Kenyan border in the northeast (Fig. 9A). Although tobacco use estimates declined across the nation in the last year of study, 2016, tobacco use increased up to 7 times on the shores of Lake Victoria, along the Kenyan border to the east, and in and around the capital city of Kampala (Fig. 9B and 9C). The tobacco use prevalence estimates for Tanzania were the lowest across all nations included in the study, ranging up to 16% in rural pockets along the northern Kenyan border and along the southern border in 2010 (Fig. 10A). Although prevalence estimates declined to less than 11% in the last year of study, 2016, there were areas of increased tobacco use in rural areas and around urban centers like Moshi in the north, Kigoma to the west, the island of Zanzibar, and Dar es Salaam (Fig. 10B and 10C). The highest tobacco use estimates in 2009 Kenya were in the north and along the Ugandan border, ranging up to 42% prevalence (Fig. 11A). By 2014 prevalence estimates across the nation declined to less than 25% (Fig. 11B). However, there were pockets of tobacco use increase in urban centers and rural areas from the outskirts of Nairobi, the capital city, to the shores of Lake Victoria to the west (Fig. 11C). Increased tobacco use was also evident in rural areas between Nairobi and the coastal city of Mombasa to the east (Fig. 11C).

Table 3

Prevalence of tobacco use in East African Community (EAC) nations selected for study.
Country	Period	n	Age-adjusted Prevalence (%, 95% CI)	Weighted Prevalence (%, 95% CI))	Odds Ratio (95% CI)
Burundi	2010	13669	16.25 (15.42, 17.14)	13.63 (12.79, 14.47)	Reference
Burundi	2016	24821	9.93 (9.42, 10.46)	8.03 (7.55, 8.51)	0.55 (0.50, 0.61)^***
Rwanda	2005	16141	12.67 (11.86, 13.55)	9.70 (9.13, 10.27)	1.32 (1.21, 1.45)^***
	2010	20000	9.79 (9.19, 10.42)	7.51 (7.08, 7.95)	Reference
	2015	19714	7.18 (6.66, 7.74)	5.36 (4.99, 5.74)	0.70 (0.63, 0.77)^***
Uganda	2006	11034	9.70 (8.87, 10.63)	8.08 (7.19, 8.97)	1.54 (1.29, 1.83)^***
	2011	10969	6.84 (6.12, 7.66)	5.42 (4.78, 6.05)	Reference
	2016	23842	4.72 (4.33, 5.16)	3.71 (3.36, 4.06)	0.67 (0.57, 0.79)^***
Tanzania	2005	12964	5.64 (5.23, 6.06)	5.66 (5.06, 6.27)	1.09 (0.94, 1.27)
	2010	12666	5.03 (4.65, 5.43)	5.22 (4.73, 5.72)	Reference
	2016	16780	3.66 (3.38, 3.96)	3.77 (3.38, 4.16)	0.71 (0.62, 0.82)^***
Kenya	2003	11773	10.25 (9.57, 10.98)	9.25 (8.45, 10.05)	1.36 (1.14, 1.63)^***
	2009	11909	8.16 (7.51, 8.86)	6.95 (5.99, 7.91)	Reference
	2014	43898	6.75 (6.44, 7.08)	5.84 (5.50, 6.18)	0.83 (0.71, 0.98)^*

*p < 0.05 **p < 0.01 ***p < 0.001

Table 4

Subnational trends of tobacco use prevalence in urban and rural areas. Urban residence used as reference for logistic regression.
		Urban (reference)			Rural
Country	Period	n	%	95% CI	n	%	95% CI	Odds Ratio (95% CI)
Burundi	2010	3374	6.75	5.47, 8.03	10295	14.57	13.62, 15.51	6.07 (4.76, 7.75)^***
Burundi	2016	5435	4.40	3.49, 5.32	19386	8.59	8.06, 9.12	5.76 (4.63, 7.18)^***
Rwanda	2005	3746	7.57	6.45, 8.70	12395	10.14	9.49, 10.78	2.05 (1.65, 2.55)^***
	2010	3523	6.35	5.34, 7.35	16477	7.72	7.25, 8.20	1.95 (1.68, 2.25)^***
	2015	5034	3.73	3.13, 4.33	14680	5.76	5.32, 6.20	2.74 (2.34, 3.22)^***
Uganda	2006	1841	5.63	2.55, 8.71	9193	8.58	7.67, 9.48	3.16 (2.34, 4.27)^***
	2011	3193	2.76	1.98, 3.55	7776	6.07	5.31, 6.84	3.21 (2.53, 4.07)^***
	2016	5529	2.64	2.04, 3.24	18313	4.09	3.67, 4.51	2.16 (1.83, 2.55)^***
Tanzania	2005	3114	4.30	3.12, 5.48	9850	6.20	5.48, 6.91	3.35 (2.84, 3.95)^***
	2010	3215	4.05	3.23, 4.87	9451	5.68	5.08, 6.29	3.07 (2.63, 3.60)^***
	2016	5202	3.36	2.65, 4.08	11578	4.00	3.55, 4.46	2.59 (2.27, 2.95)^***
Kenya	2003	3901	8.55	7.37, 9.72	7872	9.49	8.50, 10.48	2.84 (2.40, 3.35)^***
	2009	3699	6.02	4.30, 7.74	8210	7.28	6.15, 8.41	2.61 (2.19, 3.11)^***
	2014	16529	5.73	5.14, 6.31	27369	5.92	5.51, 6.33	2.18 (1.94, 2.45)^***

*p < 0.05 **p < 0.01 ***p < 0.001

The objective of this study was to describe the spatiotemporal trends of CVD risk factors, overweight/obesity and tobacco use, at subnational and subpopulation levels. Overweight/obesity and tobacco use trends were not similar. The national prevalence of overweight/obesity increased across most nations included in the study, except Burundi, and the inverse was true for tobacco use prevalence in the region, which declined over time. Findings are in line with studies that show overweight/obesity prevalence in low-to-middle income countries (LMICs) is increasing as countries progress through stages of epidemiological and obesity transitions (39–41). In this study, the national prevalence rate of overweight/obesity varied between nations and was associated with indicators of economic development, higher per capita gross domestic product (GDP) and urbanicity, which are also major driving forces of globalization and nutritional transition in LMICs (40, 42). Major lifestyle and technological shifts have led to a decrease in physical activity coupled with modifications of global food systems that have increased access to cheaper, less nutritious food (43, 44).

Urban populations in each nation were more likely to be overweight/obese than rural populations, although the magnitude of overweight/obesity difference varied widely between nations. Urban populations in the lowest income nation of the group, Burundi, were almost 6 to 7 times more likely to be overweight/obese compared to rural populations. Yet in Kenya, the nation with the highest GDP in the group, urban populations were 2 to 3 times more likely to be overweight/obese compared to rural populations. Although the prevalence of overweight/obesity increased over time in both urban and rural populations, spatial analysis revealed that the rate of change was not equal across all urban and rural areas. While urban areas had higher prevalence rates at the start and end of the study, the greatest magnitude of change over time occurred in rural areas. These findings not only illustrate how analysis at broader scales, i.e., regional or national, mask variation in space and obscure the place specific indicators and drivers of overweight/obesity prevalence but may also signify obesity transition in East Africa.

Jaacks et al. (2019), propose four stages of obesity transition as a model for the global obesity epidemic (41). In the first stage, obesity prevalence is higher in women, people with higher socioeconomic status, and in adults more than children. Countries in south Asia and SSA, which mainly consist of LMICs, are currently in this stage. In the second stage, obesity prevalence increases in adults, the gap narrows between sexes, and socioeconomic differences among women also narrow. In the third stage obesity prevalence in lower socioeconomic groups surpasses higher socioeconomic groups and plateaus at high rates for women. Most European countries are in this stage of transition. In the final stage obesity prevalence declines in all groups, although no countries have entered this stage. Although Jaacks et al. (2019) do not include urbanicity in their conceptual model, urbanicity is linked to economic development and higher socioeconomic status in LMICs (42, 45). Therefore, the narrowing overweight/obesity gap between urban and rural areas may indicate a transition to the second stage of obesity transition in the higher income East African nations of Kenya and Tanzania.

The urban-rural overweight/obesity gap is likely linked to higher economic development in urban areas which lead to changes in employment, income, food access, physical activity, and transportation. However, as illustrated in this study and others, the rate of overweight/obesity is increasing at a higher rate in rural compared to urban areas (46). As LMICs economies grow, rural areas adopt historically urban characteristics such as changes in infrastructure, transportation, and food access. Infrastructure changes include improved roads, access to electricity, and in some cases, improved water sources. Better roads lead to transportation shifts, such as more public transport and therefore less walking, and may also lead to changes in local food markets. With improved roads and electricity, access to cheap, highly processed obesogenic food increases coupled with more marketing efforts on radio, T.V., and print media (42, 46).

In contrast to the urban/rural patterns of overweight/obesity, rural populations were more likely to use tobacco compared to urban populations and the likelihood of use varied substantially between nations. Adults in rural Burundi were 5 to 6 times more likely to use tobacco than their urban counterparts, while rural adults in Kenya were 2 to 3 times more likely to use tobacco than urban adults. The rate of tobacco use declined in both subpopulations, however, spatial analysis showed that tobacco use was not decreasing across the entire landscape. Prevalence estimate maps reveal that tobacco use increased in and around major cities and urban centers in all the nations, except Burundi, but declined in rural areas.

Although the national prevalence of tobacco use declined over time in all the nations, which aligns with studies about global and national tobacco use trends, spatial analysis showed that the decline in tobacco use was not true for all places with the nations (47–49). Current literature indicates that Sub-Saharan Africa has the lowest average rate of tobacco use compared to other WHO regions, at around 10% compared to a 22% global average in 2019 (49). However, the region has the youngest population and most countries in SSA are in the early stages of a tobacco epidemic and tobacco use prevalence is projected to increase (50–52). According to the WHO the number of tobacco users in SSA from 2000 to 2022 increased and is expected to continue growing (49). All the nations included in this study are signatories of the WHO Framework Convention on Tobacco Control (WHO FCTC). The FCTC aims to reduce tobacco use through several key measures, such as: 100% smoke-free areas, health warnings labels on tobacco products, taxation, and bans on tobacco advertising, promotion, and sponsorship. However, the level of compliance varies between nations, for instance, Burundi and Uganda have implemented smoke-free environments, where all public spaces are completely smoke free, while Rwanda, Kenya, and Tanzania lack absence bans or have up to two smoke-free public spaces (48). Egbe et al. (2022) warn that despite the recent gains in tobacco control across SSA, policy and intervention strategies are hindered transnational tobacco companies interference and are also limited by a lack of locally generated evidence (53). This current study provides evidence at local scales that the growth of tobacco users is not evenly distributed across nations in this region and is likely happening in and around urban spaces.

Although this study was able to illustrate spatial and temporal variations in overweight/obesity and tobacco use prevalence using secondary data, there were several limitations. DHS surveys provide standardized data and large sample sizes; however, the observed trends were not longitudinal and were derived from a series of cross-section analyses of multiple surveys, which represent different sample years and sizes. Second, there are issues inherent with the definition of urbanicity in the DHS, due to a lack of a universally accepted definition of urban and rural areas, there urban and rural areas determined by national statistical offices and may differ across nations and time (18, 54). Third, although BMI is an inexpensive and reliable anthropometric measure for large-scale surveys, it does not accurately reflect body fat composition and may result in misclassification of overweight/obesity (55). Inherent with secondary analysis, these surveys were not designed for this study, therefore measures that are better indicators of overweight/obesity, such as waist circumference were not available. Fourth, smoking and smokeless tobacco were grouped together to create one tobacco use variable due to the questionnaire design and low prevalence rates. However, different tobacco types have users with distinct characteristics which may influence the spatial distribution and prevalence trends of each tobacco type. Fifth, self-reported tobacco use may have been underreported due to the stigma associated with tobacco use, especially among the young and women. Also, the DHS was initially designed to assess maternal, child, and nutrition health metrics and therefore women are overrepresented in the data and women past reproductive age, over 49 years old were not sampled while men up to 59 years old were sampled. As such, prevalence estimates may underestimate the true prevalence of tobacco use in older generations, particularly among older women. Lastly, the use of sample cluster data for spatial interpolation is not ideal since individual observations were linked to one spatial point and displaced by 2 to 10 km; therefore, prevalence surface estimates likely less accurate than the true prevalence distribution and variability below the cluster is masked in aggregation.

To the best of the authors’ knowledge, this is the first study to explore the spatiotemporal trends of overweight/obesity and tobacco use prevalence at a subnational level in this region of SSA. Findings highlight the importance of de-homogenizing studies about CVD risk factors, such as tobacco use and overweight/obesity, in not only SSA but in other parts of the world in epidemiological transition. One question, one method, and one conclusion at regional or national levels does not fit all. Studies of national prevalence trends for overweight/obesity or tobacco use, which constitute the majority of the CVD risk factor literature in SSA, mask important information about subpopulation and place-specific behavior and drivers of risk factor prevalence. As shown in this study, although the national prevalence of tobacco use declined in all the nations, place-specific trends differed and, in some cases, increased. This type of understanding is not only vital to increase knowledge at local scales but can be a starting point for studies to understand local drivers of risk factor trends which can in turn inform policy and public health intervention.

It is imperative that as nations transition to higher levels of cardiovascular and chronic conditions, which are strongly impacted by place and culture, epidemiological analysis should occur at subnational levels and include spatial analysis, especially in regions where administrative boundaries are not static. As illustrated, spatial analysis also provides another means to understand subpopulation behavior, for instance, CVD risk factor trends in urban and rural spaces were place specific and therefore did not experience similar growth or decline patterns. Findings from this study offer insight on 1) the use of existing data, such as DHS or other national surveys, to analyze spatial trends in areas with low data resources and 2) potential future directions for research on how to utilize this type of de-homogenized analysis to build policy and create public health interventions that tailored to and appropriate for specific areas of needs.

CVD: Cardiovascular disease; SSA: Sub-Saharan Africa; DHS: Demographic and Health Surveys; EAC: East African Community; GDP: Gross Domestic Product; WHO: World Health Organization; BMI: Body Mass Index; INDEPTH: International Network for the Demographic Evaluation of Populations and Their Health; KDE: Kernel Density Estimation; LMIC: Low- and Middle-Income Countries

Ethics approval and consent to participate: Procedures and questionnaires for standard DHS surveys have been reviewed and approved by ICF Institutional Review Board (IRB). Additionally, country-specific DHS survey protocols are reviewed by the ICF IRB and typically by an IRB in the host country. ICF IRB ensures that the survey complies with the U.S. Department of Health and Human Services regulations for the protection of human subjects (45 CFR 46), while the host country IRB ensures that the survey complies with laws and norms of the nation.

Consent for publication: Not applicable.

Availability of data and materials: The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests: The authors declare that they have no competing interests.

Funding: None.

Authors’ Contributions: BCK and MC conceived of this study. BCK performed the data analysis, and all authors interpreted the results. BCK wrote the manuscript and MC reviewed the manuscript. All authors read and approved the final manuscript.

Acknowledgements: We are grateful to Caglar Koylu for valuable input about spatial interpolation methods.

Akpalu A, Gebregziabher M, Ovbiagele B, Sarfo F, Iheonye H, Akinyemi R, et al. Differential Impact of Risk Factors on Stroke Occurrence Among Men Versus Women in West Africa: The SIREN Study. Stroke. 2019 Apr 1;50(4):820–7.
Owolabi MO, Sarfo F, Akinyemi R, Gebregziabher M, Akpa O, Akpalu A, et al. Dominant modifiable risk factors for stroke in Ghana and Nigeria (SIREN): a case-control study. Lancet Glob Health. 2018;6(4):e436–46.
Mensah GA. Descriptive epidemiology of cardiovascular risk factors and diabetes in sub-saharan Africa. Prog Cardiovasc Dis. 2013 Nov 1;56(3):240–50.
Sreeramareddy CT, Pradhan PM, Sin S, Pradhan MM, Sin S. Prevalence, distribution, and social determinants of tobacco use in 30 sub-Saharan African countries. BMC Med. 2014 Dec 18;12(1):243.
Sreeramareddy CT, Acharya K. Trends in Prevalence of Tobacco Use by Sex and Socioeconomic Status in 22 Sub-Saharan African Countries, 2003-2019. JAMA Netw Open. 2021 Dec 8;4(12):e2137820.
Kengne AP, Bentham J, Zhou B, Peer N, Matsha TE. Trends in obesity and diabetes across Africa from 1980 to 2014: An analysis of pooled population-based studies. Int J Epidemiol. 2017;46(5):1421–32.
Njelekela M, Sato T, Nara Y, Miki T, Kuga S, Noguchi T, et al. Nutritional variation and cardiovascular risk factors in Tanzania - Rural-urban difference. South African Medical Journal. 2003;93(4):295–9.
Sliwa K, Acquah L, Gersh BJ, Mocumbi AO. Impact of socioeconomic status, ethnicity, and urbanization on risk factor profiles of cardiovascular disease in Africa. Circulation. 2016 Mar 22;133(12):1199–208.
Hendriks ME, Wit FWNM, Roos MTL, Brewster LM, Akande TM, de Beer IH, et al. Hypertension in Sub-Saharan Africa: Cross-Sectional Surveys in Four Rural and Urban Communities. PLoS One. 2012 Mar 12;7(3):e32638.
Ajayi IOO, Adebamowo C, Adami HO, Dalal S, Diamond MB, Bajunirwe F, et al. Urban-rural and geographic differences in overweight and obesity in four sub-Saharan African adult populations: a multi-country cross-sectional study. BMC Public Health. 2016;16(1):1–13.
Pinchoff J, Mills CW, Balk D. Urbanization and health: The effects of the built environment on chronic disease risk factors among women in Tanzania. PLoS One. 2020 Nov 1;15(11 November).
Beyer KMM, Rushton G. Mapping cancer for community engagement. Prev Chronic Dis. 2009;6(1).
Hope KR. Bringing in the future in Kenya: Beyond the 2010 constitution. Insight on Africa. 2015;7(2).
Dummer TJB. Health geography: supporting public health policy and planning. CMAJ. 2008 Apr 22;178(9):1177–80.
Curtis S, Riva M, Rosenberg M, Brown T. Health geography and public health. A companion to health and medical geography. 2010;323–45.
The World Bank. GDP per capita [Internet]. 2022. Available from: https://data.worldbank.org/indicator/NY.GDP.PCAP.CD
ICF. Demographic and Health Surveys (various) [Datasets]. Funded by USAID. Rockville, Maryland: ICF [Distributor].; 2016.
Dorélien A, Balk D, Todd M. What is Urban? Comparing a satellite view with the demographic and health surveys. Popul Dev Rev. 2013;39(3):413–39.
Croft TN, Marshall AMJ, Allen CK, Arnold F, Assaf S, Balian S. Guide to DHS Statistics. Rockville, Maryland, USA; 2018.
ICF International. Demographic and Health Survey Sampling and Household Listing Manual. Calverton, Maryland, USA; 2012.
Sankoh O, Sharrow D, Herbst K, Kabudula CW, Alam N, Kant S, et al. The INDEPTH standard population for low- and middle-income countries, 2013. Global Health Action. 2014;7(1).
Lumley T. Analysis of complex survey samples. Vol. 9, Journal of Statistical Software. American Statistical Association; 2004. p. 1–19.
R Core Development Team. A Language and Environment for Statistical Computing. Vol. 2, R Foundation for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2020. p. https://www.R-project.org.
Silverman BW. Density estimation: For statistics and data analysis. Density Estimation: For Statistics and Data Analysis. 2018;37(1):1–175.
Rushton G. Public health, GIS, and spatial analytic tools. Annu Rev Public Health. 2003;24(1):43–56.
Abramson IS. On Bandwidth Variation in Kernel Estimates-A Square Root Law. The Annals of Statistics. 2007;10(4):1217–23.
Lemke D, Mattauch V, Heidinger O, Pebesma E, Hense HW. Comparing adaptive and fixed bandwidth-based kernel density estimates in spatial cancer epidemiology. Int J Health Geogr. 2015;14(1):15.
Larmarange J, Vallo R, Yaro S, Msellati P, Méda N. Methods for mapping regional trends of HIV prevalence from Demographic and Health Surveys (DHS). Cybergeo. 2011.
Larmarange J, Bendaud V. HIV estimates at second subnational level from national population-based surveys. Vol. 28, AIDS. 2014.
Kienberger S, Hagenlocher M. Spatial-explicit modeling of social vulnerability to malaria in East Africa. 2014.
Duah HO, Amankwa CE, Adomako I, Owusu B, Agbadi P. Comorbid patterns of anaemia and diarrhoea among children aged under 5 years in Ghana: a multivariate complex sample logistic regression analysis and spatial mapping visualisation. Int Health. 2021 Dec 1;13(6):562–72.
Nutor JJ, Duah HO, Agbadi P, Duodu PA, Gondwe KW. Spatial analysis of factors associated with HIV infection in Malawi: Indicators for effective prevention. BMC Public Health. 2020 Jul 25;20(1):1–14.
He WJ, Lai YS, Karmacharya BM, Dai BF, Hao YT, Xu DR. Geographical heterogeneity and inequality of access to improved drinking water supply and sanitation in Nepal. Int J Equity Health. 2018 Apr 2;17(1):1–14.
Fürst MA, McMahon DP, Osborne JL, Paxton RJ, Brown MJF. Disease associations between honeybees and bumblebees as a threat to wild pollinators. Nature 2014 506:7488. 2014 Feb 19;506(7488):364–6.
Hagenlocher M, Castro MC. Mapping malaria risk and vulnerability in the United Republic of Tanzania: A spatial explicit model. Popul Health Metr. 2015 Feb 3;13(1):1–14.
Pebesma E, Bivand RS. S classes and methods for spatial data: the sp package. R news. 2005;5(2):9–13.
Hijmans RJ. raster: Geographic Data Analysis and Modeling. 2021.
ESRI. ArcGIS Desktop. Redlands, CA: Environmental Systems Research Institute; 2019.
Bixby H, Bentham J, Zhou B, Di Cesare M, Paciorek CJ. Rising rural body-mass index is the main driver of the global obesity epidemic in adults. Nature. 2019 May 9;569(7755):260–4.
Popkin BM, Adair LS, Ng SW. Global nutrition transition and the pandemic of obesity in developing countries. Nutr Rev. 2012 Jan 1;70(1):3–21.
Jaacks LM, Vandevijvere S, Pan A, McGowan CJ, Wallace C, Imamura F, et al. The obesity transition: stages of the global epidemic. Lancet Diabetes Endocrinol. 2019 Mar 1;7(3):231–40.
Ford ND, Patel SA, Narayan KMV. Obesity in Low- and Middle-Income Countries: Burden, Drivers, and Emerging Challenges. Annu Rev Public Health. 2017 Mar 20;38:145–64.
Popkin BM, Corvalan C, Grummer-Strawn LM. Dynamics of the double burden of malnutrition and the changing nutrition reality. The Lancet. 2020;395(10217):65–74.
Popkin BM, Slining MM. New dynamics in global obesity facing low- and middle-income countries. Obesity Reviews. 2013 Nov;14(S2):11–20.
Popkin BM. Urbanization, lifestyle changes and the nutrition transition. World Dev. 1999;27(11):1905–16.
Jaacks LM, Slining MM, Popkin BM. Recent Underweight and Overweight Trends by Rural–Urban Residence among Women in Low- and Middle-Income Countries. J Nutr. 2015 Feb 1;145(2):352–7.
Reitsma MB, Kendrick PJ, Ababneh E, Abbafati C, Abbasi-Kangevari M, Abdoli A, et al. Spatial, temporal, and demographic patterns in prevalence of smoking tobacco use and attributable disease burden in 204 countries and territories, 1990–2019: a systematic analysis from the Global Burden of Disease Study 2019. The Lancet. 2021 Jun 19;397(10292):2337–60.
World Health Organization. WHO Report on the Global Tobacco Epidemic, 2021: Addressing new and emerging products. 2021.
World Health Organization. WHO global report on trends in prevalence of tobacco use 2000-2025, fourth edition. Geneva: World Health Organization; 2021.
Eriksen MP, Mackay Judith, Ross Hana, American Cancer Society, World Lung Foundation. The tobacco atlas. 4th ed., c. Atlanta, GA: American Cancer Society; 2012. 132 p.
Thun M, Peto R, Boreham J, Lopez AD. Stages of the cigarette epidemic on entering its second century. Tob Control. 2012 Mar 1;21(2):96 LP – 101.
Méndez D, Alshanqeety O, Warner KE. The potential impact of smoking control policies on future global smoking trends. Tob Control. 2013;22(1):46–51.
Egbe CO, Magati P, Wanyonyi E, Sessou L, Owusu-Dabo E, Ayo-Yusuf OA, et al. Landscape of tobacco control in sub-Saharan Africa. Tob Control. 2022 Mar 1;31(2):153–9.
Hugo G, Champion A, Lattes A. Toward a new conceptualization of settlements for demography. Popul Dev Rev. 2003 Jun 1;29(2):277–97.
Rothman KJ. BMI-related errors in the measurement of obesity. Int J Obes. 2008 Aug 11;32(3):S56–9.

No competing interests reported.

Download PDF

Journal Publication

published 24 Aug, 2023

Read the published version in International Journal of Health Geographics →

Editorial decision: Major revision
12 Jul, 2023
Reviews received at journal
17 May, 2023
Reviewers agreed at journal
01 May, 2023
Reviewers invited by journal
05 Apr, 2023
Submission checks completed at journal
30 Mar, 2023
Editor assigned by journal
30 Mar, 2023
First submitted to journal
28 Mar, 2023

You are reading this latest preprint version

Estimating subnational prevalence trends of cardiovascular disease risk factors, tobacco use, overweight, and obesity in East Africa using adaptive kernel density estimation for survey data

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Study Area

Data Source

Variables

Analysis

Statistical Analysis

Spatial Analysis

Results

Overweight/obesity

Tobacco Use

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1