Background

Smoke arising from solid biomass cooking (wood, dung, charcoal, crop residue) is widely perceived to act as a mosquito repellent among communities [1,2,3], therefore protecting against mosquito-borne disease. However, solid biomass cooking produces health harming levels of household air pollution (HAP), estimated to be responsible for around 450,000 deaths in children aged under 5 years worldwide [4], compared to only 274,000 estimated deaths from malaria in 2019 [1]. This discordance in perceived compared to actual health risks associated with malarial transmission could impact upon uptake of structural interventions (e.g., cleaner fuel transition [LPG, electricity, biogas]) and behavioural changes intended to reduce harmful HAP exposure, notably among those living in endemic malarial regions.

Sub-Saharan Africa (SSA) has the highest malarial prevalence globally with 94% of cases and deaths, caused by predominantly by Plasmodium falciparum [5]. Identified risk factors for malarial infection include poor household construction [6,7,8] (e.g., open eaves), animals sleeping in the house [9] and presence of standing water near the house [10, 11]. The use of mosquito nets, household insecticidal spraying, and larval source management [12] have become common practice advocated in malarial prevention, driven in part by the World Health Organization’s (WHO) coordinated response [5]. Another, common preventive measure is use of mosquito repellent smoke from the burning of certain types of plant materials, such as churai in West Africa [2, 13].

There is little evidence supporting reduced mosquito infiltration [14, 15] or malarial transmission associated with solid biomass fuel cooking [2, 16]; although there is some evidence that solid biomass cooking reduces the risk of arboviruses in Guatemala [17]. Therefore, to better understand this disease risk paradigm, this study investigates the association of malarial acquisition among children aged under 5 years with regard to: (i) cleaner or solid biomass fuels and kerosene cooking; (ii) charcoal or wood fuel usage; and (iii) indoor and outdoor cooking, within households in 17 SSA countries using the population-based Demographic and Health Survey (DHS) data.

Methods

Data sources

This cross-sectional study uses publicly available survey data for 17 malarial-endemic SSA countries with available malarial data (Fig. 1), obtained from the DHS program supported by the United States Agency for International Development (USAID) within the last 10 years (2010–2020). The DHS undertakes full surveys every 5 years, and intermediate Malaria Indicators Surveys (MIS) [18]; only some of the full DHS survey modules undertake malarial testing. For those DHS surveys including malaria modules, malarial testing is undertaken by trained fieldworkers on a sub-sample of eligible children aged 6–59 months using a malarial rapid diagnostic test (RDT) [18]. A two-stage stratified sampling technique was employed to obtain a representative population-based sample, with residential households randomly selected. Eligible households included those with an ever-married (married, widowed or divorced) woman aged between 15 and 49 years and resident the night before the survey. Ethical approval for data collection was gained from the relevant government authority [18], and authorization for data access was provided by the DHS.

Fig. 1
figure 1

Flow diagram for included countries. N Number of countries

Full size image

Malarial endemicity was generated for each cluster by assessment of malarial prevalence obtained from the open source Malaria Atlas Project [19] within eligible countries, and defined as holoendemic (> 75%), hyperendemic (51–75%), mesoendemic (11–50%), hypoendemic (< 10%) [20]. Those data points that fell within hypoendemic areas were excluded from the analysis due to lower rate of malarial infection and testing. Malarial prevalence data were geocoded to the cluster geographic coordinates using the spatial analyst tool in ArcMAP 10.7 [21]; a method that has been previously used for this purpose [22].

As the wealth index provided by DHS contains cooking fuel as an indicator variable, a new modified wealth index was calculated in SPSS [23] using principal component analysis [24] to prevent circularity [8]. The index indicator variables included source of drinking water, house construction material, provision of a toilet facility and household assets, which varied by country (Additional file 1).

Predictor and outcome variables

Proxies for household air pollution (HAP) exposure levels

Three analyses were undertaken (Table 1), undertaking comparisons by the main type of cooking fuel used and cooking location respectively: cleaner vs solid biomass fuels and kerosene fuels; charcoal vs wood fuels; outdoor vs indoor cooking (indoors, in a separate building).

Table 1 Analyses, sub-analyses and exploratory analyses undertaken with detail on categorisation of the exposure of interest
Full size table

Measure of malarial diagnosis

A malarial infection was determined by a positive RDT (n = 17 countries) and in some countries a subsequent blood smear test via microscopy taken at the point of interview (n = 11 countries), both of which were modelled as a binary (negative, positive) outcome variable, in separate analysis within this study. The RDT was undertaken using the SD BIOLINE Malaria Ag test, in all countries, which has estimated sensitivity of 99.7% and specificity of 99.5% [25]. Whereas, only certain countries collected blood samples which were collected with the parasites detected in the blood at time of survey using microscopy [18], with estimated sensitivity of 95.7% and specificity of 97.9% [26].

Explanatory variables

Covariates were included for the relevant contextual, household and individual factors identified as influencing both HAP exposure and malarial risk. Covariates were included in regression models as categorical variables other than household altitude, which was modelled as a continuous variable. Regional level variables were: malarial endemicity (mesoendemic, hyperendemic and holoendemic), season (dry, wet), rural or urban residence (rural, urban), cluster altitude (metres). Household level variables were: number of household members (≤ 6, > 6), household smoking (no, yes), modified wealth index (lowest, low, middle, high, highest), biomass cooking fuel type (where applicable; kerosene, coal/lignite, charcoal, wood, straw/shrubs/grass, agricultural crop, animal dung), household insecticide spraying within the last 12 months (no, yes) and dwelling construction (traditional, modern). Child variables were: age (< 1, 1, 2, 3, 4 years), birth order (first born, not first born), child’s sex (male, female), slept under mosquito net last night (no, yes—treated (ITN) net, yes—untreated net). The season variable is created using regional and country level information from the CIA fact book [27] and the World Bank climate change knowledge portal [28]. The household construction variable is a composite variable derived from the wall, roof and floor material. Firstly, each of the three materials were categorized into natural, rudimentary and finished construction material using the criteria outlined by Tusting et al. [8], followed by the creation of the household construction variable where modern household construction was define as wall, roof and floor being made of finished materials.

Data analysis

Data preparation and analysis was undertaken in R studio [29]. Each variable was described within the combined dataset using number of cases (n), and percentage (%) and median and Interquartile range (IQR) for continuous variables. The level of missing data ranged from 0 to 48% of clinically relevant variables at a country level, which was imputed using the MICE package [30] with 50 iterations [31, 32]; to prevent bias from list-wise deletion [33]. To test the association between cooking practices and malarial infection, multivariable logistic regression using the survey package [34], was used to account for the complex sampling strategy; reporting adjusted odds ratios (AOR) and 95% confidence intervals (95% CI). The MIS survey did not contain information on cooking location and household smoking, therefore a sub-analysis was undertaken using countries where these variables were available for analysis. Sub-analyses were also undertaken for rural, urban, wood cooking fuel houses and mesoendemic areas. In addition, the analysis was repeated to include additional covariates among a sub-set of countries where additional variables of interest were available. This enabled investigation of the influence of (i) household cooking location; (ii) household smoking; and (iii) household insecticidal spraying, as some of the variables are missing from certain countries.

Results

This study identified 85,263 children aged under 5 years children living in 17 participating countries (DHS = 9, MIS = 7) from 2011 to 2019, with a total of 74,461 RDT and 48,491 microscopy test results. Within the pooled full dataset, median child age was 3 years (IQR: 2–4). The proportion of girls ranged from 48.0% in Guinea (2012) to 51.0% in Cote d’Ivoire (2011–2012), with overall 49.4% in the pooled dataset (Table 2).

Table 2 Characteristics of included surveys
Full size table

Malarial infection was positively identified by RDT among 34.6% of children in the combined dataset at the time of testing, with the highest point prevalence in Guinea 2012 (51.8%) and lowest in Tanzania 2017 (7.07%) (Table 3). However, where microscopy was undertaken malarial infection was identified in 28.2% of children, with the highest prevalence in Guinea 2012 (48.7%) and lowest in Uganda 2018–2019 (11.3%). Of the areas surveyed, most were in mesoendemic areas (Fig. 2), with holoendemicity in Cote d’Ivoire 2011–2012, DRC 2013–2014, Guinea 2012 and Liberia 2016. Of those children with a positive malarial RDT result, 1.3% resided in cleaner cooking households. Whereas, 35.2% in outdoor cooking households and 35.7% in a household where cooking was typically undertaken in a separate building (Table 3).

Table 3 Descriptive statistics for the combined dataset (N = 85,263)
Full size table
Fig. 2
figure 2

Malarial endemicity and prevalence among children under five for each country. N number of child observations, PR prevalence rate of positive RDT result

Full size image

Analysis 1—Solid biomass fuel usage and risk of malarial infection

In pooled analyses, cooking with solid biomass fuels and kerosene fuels was observed to be independently associated with a 57% increase in the adjusted odds ratio for malarial infection, compared to cleaner cooking (electricity, LPG) (Fig. 3) (RDT AOR: 1.57 [1.30–1.914]; Microscopy AOR: 1.58 [1.23–2.04]) (Table 3). A 61% increase in adjusted odds ratio was also observed when investigating the effect of cooking location and household smoking with solid biomass fuels and kerosene compared to cleaner cooking fuels (RDT AOR: 1.61 [1.28–2.02]; Microscopy AOR: 1.61 [1.20–2.15]. The increased malarial infection adjusted odds ratio associated with solid biomass fuels and kerosene cooking remained in the stratified sub-analysis among rural locations (RDT AOR: 1.41 [1.02–1.95]; Microscopy AOR: 2.10 [1.34–3.32]), urban locations (RDT AOR: 1.58 [1.24–2.03] only) and mesoendemic regions (RDT AOR: 1.58 [1.28–1.95]; Microscopy AOR: 1.59 [1.21–2.08]) (Table 4).

Fig. 3
figure 3

Adjusted odds ratio of malarial infection with biomass cooking compared to cleaner cooking. AOR adjusted odds ratio, 95% CI 95% confidence interval, N Number of child observations. Table of unadjusted and adjusted results can be found in Additional file 2: Table S2.1

Full size image
Table 4 Odds ratio of malarial infection for each cooking practices for the combined dataset, exploratory and sub-analysis
Full size table

Analysis 2—Biomass fuel type and risk of malarial infection

Among biomass fuel households only, use of wood compared to charcoal fuel was associated with an increased adjusted odds ratio of malarial infection (RDT AOR: 1.77 [1.54–2.04]; Microscopy AOR: 1.21 [1.08–1.37]) (Fig. 4), with a similar effect being observed in the exploratory analysis controlling for cooking location and household smoking (RDT AOR: 1.26 [1.10–1.46] only) and in mesoendemic areas (RDT AOR: 1.77 [1.49–2.09]; Microscopy AOR: 1.26; [1.10–1.44]) (Table 4). In the stratified sub-analysis it was observed that urban areas had a greater adjusted odds ratio of malarial infection associated with wood compared to charcoal cooking (RDT AOR: 2.25 [1.79–2.78]), in comparison to rural areas (RDT AOR: 1.43 [1.21–1.70]).

Fig. 4
figure 4

Adjusted odds ratio of malarial infection with wood cooking compared to charcoal cooking. AOR Adjusted odds ratio, 95% CI 95% confidence interval, N Number of child observations. Table of unadjusted and adjusted results can be found in Additional file 2: Table S2.2

Full size image

Analysis 3—Household cooking location and risk of malarial infection

No significant association was observed between household cooking location and malaria adjusted odds ratio (RDT AOR: 0.94 [0.83–1.05]; Microscopy AOR: 0.97 [95% CI 0.83–1.05]) (Fig. 5). In comparison, cooking in a separate building was associated with a reduced adjusted odds ratio of malarial infection by 74% compared to indoor cooking (Fig. 5) (RDT AOR: 0.74 [0.66–0.83]; Microscopy AOR: 0.75 [0.67–0.84]). The same reduced malarial infection adjusted odds ratio associated with cooking in a separate building was observed in stratified sub-analyses for wood cooking (RDT AOR: 0.75 [0.67–0.85]; Microscopy AOR: 0.77 [0.67–0.87]), rural (RDT AOR: 0.70 [0.62–0.80]; Microscopy AOR: 0.73 [0.64–0.84]) and mesoendemic areas (RDT AOR: 0.73 [0.65–0.82]; Microscopy AOR: 0.74 [0.65–0.83) only (Table 4).

Fig. 5
figure 5

Adjusted odds ratio of malarial infection with cooking location (outdoor, in a separate building) compared to indoors. AOR Adjusted odds ratio, 95% CI 95% confidence interval, N Number of child observations. Table of unadjusted and adjusted results can be found in Additional file 2: Table S2.3

Full size image

Discussion

This large exploratory study of over 85,000 children aged under 5 years living in 17 malaria-endemic SSA found no evidence to suggest that use of cleaner fuels (e.g., LPG, electricity, biogas), charcoal vs wood, or outdoor cooking location are associated with an increased risk of malarial infection. Indeed, the findings suggest that solid biomass fuel usage may be associated with a higher incidence of malarial infection among children in SSA. There are a number of factors that may account for the increase in infections, such as the longer cooking times and thus of carbon dioxide production [35], a major mosquito attractant [36], found with solid biomass fuel cooking [37]. Additionally, the use of solid biomass fuels, particularly wood, crop residue and dung, require women, to typically collect fallen or harvest branches from woods and forests where mosquitoes commonly reside, often taking children under 5 years on their backs, thereby increasing risk of mosquito bites.

It is highly likely that risk of within household acquisition of malaria is also influenced by socioeconomic factors such as household construction characteristics (eaves space, wall type) and living conditions [8, 38,39,40,41] which are not fully captured in the DHS composite wealth index. It is also recognized that use of cleaner domestic energy sources, cooking in a separate building and selection of biomass cooking fuel type may reflect socio-economic determinants, also related to malarial microepidemiology at the household level [42, 43]. The child’s age is also a key factor in malarial infection risk, with an observed increased risk with increasing age, potentially reflecting behavioural, nutritional or exposure differences. In terms of modifiable factors for malarial infection prevention and control, there is strong evidence supporting the sustained use of ITN bed nets, larval source management and household insecticide spraying [12]; of which only ITN bed nets could be controlled for in the main analyses. The importance of household insecticide spraying can be seen in the subsidiary analysis undertaken among countries for which this information was available, identifying that there was no association with type of biomass fuel and malarial infection risk (RDT: AOR 1.23 [0.94–1.61]; Microscopy AOR: 1.07 [0.77–1.47]; Table 4); however, this sub-analysis is likely to be underpowered and should be interpreted with caution.

The analyses presented also did not explore broader contextual factors associated with household or village level clustering of malarial transmission, including position of households in relation to mosquito sites and local attitudes to malarial treatment which are recognized to influence local variations in malarial prevalence [44]. The DHS dataset did not contain information on cooking practices such as timing or duration, both of which influence the amount of smoke produced and therefore HAP exposure, and may also generate higher localized levels of indoor CO2 [35] thereby attracting mosquitoes into the home [36]. In addition, season could only be accounted for at country or broader regional level, which does not take into account microclimates, in addition, the DHS is normally undertaken in the dry season and the MIS in the wet season when the malarial transmission risk is increased [18]. HAP interventions should be developed to include activities which communicate that cooking practices which produce less smoke do not increase risk of malaria transmission to residents. It is also important to reinforce health protection advice regarding evidence-based measures for mosquito control. Further qualitative and quantitative research is merited, for a detailed investigation of the relationships between cooking location, fuel choice and risk of malarial acquisition, considering a wider range of transmission risk factors at a local level.

The rural–urban differences in cooking activity patterns, which can be most clearly noted within the differences observed in distribution between fuel types, is likely to reflect relative economic development, improved access to cleaner fuel sources in urban areas and reduced potential for cohabitation with livestock [45]. However, the rural–urban divide was not as distinct within the cleaner fuel or cooking location sub-analysis, indicating that other contextual and compositional factors exist which may influence malarial infection risk (e.g., nutrition). Although season, malarial endemicity and altitude were captured as confounding factors within our analyses, information was not available for other contextual factors of relevance to malarial infection risk, such as temperature [46].

Additionally, although the cooking practices are reported at the time of interview, this survey question does not take into consideration longer-term trends which may vary on a seasonal basis. Further prospective research is required to better understand environmental influences upon malarial microepidemiology including objective pollutant exposure assessment, capture of household design characteristics, lifestyle and time-activity factors to assess relationships with mosquito breeding conditions, malarial parasitaemia and outcomes among adults and children.

Conclusion

This large-scale observational study suggests that use of cleaner fuels and outdoor cooking practices typically associated with lower levels of household smoke, were not associated with an increased malarial acquisition risk among children living in SSA. Further mixed-methods research is required to better understand the relationships between cooking practices, cooking fuel emissions, mosquito activity and risk of malarial acquisition at household and community levels in this world region.