Climate variability, water supply, sanitation and diarrhea among children under five in Sub-Saharan Africa: a multilevel analysis

Sub-Saharan Africa (SSA) has been identified as the region of the world most vulnerable to the effects of climate change (Ramin 2009; IPCC 2014). According to the Intergovernmental Panel on Climate Change (2014), climate variability in recent decades has already led to profound changes in natural systems, including increased droughts and floods. Also, variations in temperature and precipitation influence the breeding cycle of vectors of arboviruses and waterborne diseases that cause diarrhea in children (Ramin 2009; Mordecai et al. 2020; WHO 2020). Stanke et al. (2013), Nataro & Guerrant (2017) and Kotloff et al. (2019) report that diarrhea has negative consequences on children's growth, cognitive development and metabolism. Diarrhea remains the second driving cause of mortality for children under five in developing countries (Khan & Bhutta 2017). It is responsible for the death of approximately 525,000 children every year with most of them occurring in African countries (Reiner et al. 2018).

Children are more exposed than adults to extreme weather events around the world (Xu et al. 2012; Akresh 2016; Larr & Neidell 2016). Moreover, compared to children in rich countries, those from poor countries are the most affected (Hanna & Oliva 2016). The consequences of extreme weather events on children's health are the critical health issues in African countries (Schulte-Uebbing et al. 2015; Opoku et al. 2021). For example, in Southern Africa following Cyclones Idai and Kenneth in April 2019, Mozambique recorded an increase in cholera cases among children in Cabo Delgado, Pemba, Mecufi and Sofala; and Zimbabwe counted nearly 2.3 million people with signs of severe malnutrition (USAID 2019).

The effects of climate change on a population's health is rarely direct, but depends on a set of contextual factors that reinforce its impact on pre-existing situations of vulnerability (UNECA 2011; IPCC 2014). Among these factors, lack of access to safe water and quality sanitation facilities increases the risk of exposure to waterborne diseases (Alemayehu et al. 2020; Voth-Gaeddert et al. 2020). In 2020, although many developed regions have achieved the UN Millennium Development Goals (MDGs) related to water and sanitation, 48 less developed countries, particularly those in SSA, still fare poorly in this area with almost 113 and 108 million people, respectively, using unimproved water source and facilities (World Health Organization & UNICEF 2021). In places where people use unimproved sanitations such as flush toilets that do not drain into a sewer system, pit latrines without slabs and the practice of open defecation, surface runoff from heavy rainfall carries feces to unprotected water sources that people use (Medina et al. 2007; Bhavnani et al. 2014; Bhandari et al. 2020a, 2020b; Holcomb et al. 2020). Human feces is the primary source of diarrhea pathogens (Khan & Bhutta 2017). Drinking or using water from these sources increases the risk of diarrhea if not treated first (Carlton et al. 2014; Wolf et al. 2018).

The risk of diarrhea increases under certain temperature conditions. Changes in ambient temperature may alter the distribution, survival and virulence of diarrhea pathogens and the patterns of transmission (Carlton et al. 2016). During drought, people in the southern Sahara have inadequate access to drinking water (World Health Organization & UNICEF 2021). Scarcity of water increases the risk that the remaining sources usually contaminated with diarrhea pathogens (Escherichia coli, Campylobacter, Vibrio cholera, Rotavirus, Giardia, Cryptosporidium) will be used by human beings (Stanke et al. 2013; Asmall et al. 2021). Cann et al. (2013) and Hofstra (2011) found that bacterial pathogens that cause diarrhea are in high concentrations in drinking water when evaporation increases as a result of rising temperatures. Similarly, high temperature leads to an extension of the life expectancy of diarrhea pathogens (in particular, bacteria and protozoa) in their zoonotic host, facilitating prolonged pathogen transmission (Lal et al. 2012). Also, drought increases susceptibility to diarrheal diseases, due to negative effects of drought on agricultural crops, livestock and fisheries (Stanke et al. 2013; Levy et al. 2016).

Associations between climatic conditions, sanitation, water sources and diarrhea have been studied in various countries and with different methods. Bandyopadhyay et al. (2012) using pseudo-panel analysis found that a 1 mm increase in rainfall reduced the prevalence of diarrhea by 3% in 14 SSA countries. They also found no relation between unimproved water and diarrhea. In the same way, Mukabutera et al. (2016) using multivariate logistic regression show that increased runoff was protective against diarrhea (Odd Ratio (OR) = 0.54) when household use unimproved type facilities in Rwanda. In contrast to these studies, Dunn et al. (2020) using spatial Poisson regression found a positive association between diarrhea and rainfall on the one hand, and a negative association between diarrhea and toilet type on the other in West Africa. Bhavnani et al. (2014) and Carlton et al. (2014) show that unimproved water source and unimproved sanitation increased the risk of diarrhea (OR = 6.8; OR = 1.38) after maximum rainfall in Ecuador. Thiam et al. (2019) using Bayesian conditional autoregressive models show that the unit increase in rainfall and increase in the proportion of people with untreated stored drinking water was associated with an increase in diarrhea risk by 22% in Senegal. These studies use a standard regression model with assumption that health outcomes are independent or uncorrelated for all of the individuals. Diez-Roux (2000) and Leyland & Groenewegen (2020) point out that the health outcomes of individuals who share the same context (e.g. climate, health) cannot be independent. Also, Merlo et al. (2005) noted that individuals with similar characteristics who live in different contexts may have different health statuses. When the independence assumption is not satisfied, analysis using standard regression methods underestimates the standard errors and gives biased results (Merlo et al. 2005; Leyland & Groenewegen 2020).

The previous studies, by confusing the macro level in which individuals are nested and interact with the micro level, may lead to attributing either the macro relationships to relations at the individual level (ecological fallacy or ecological error) or the individual relationships to relations at the macro level (atomistic fallacy or atomistic error) which bias the results (Diez-Roux 2000; Leyland & Groenewegen 2020). In practice, whenever the macro and micro levels are confused when analyzing a phenomenon, information from one level is aggregated or disaggregated to obtain information from the other level. For example, Bandyopadhyay et al. (2012) aggregates individual DHS data to obtain regional characteristic such as per cent of children who have diarrhea, percentage of households with no toilet facilities, percentage of households with access to safe water. Dunn et al. (2020) and Mukabutera et al. (2016) disaggregate information of the household level (mother's education, mother's occupation, type of toilet) to the individual level.

Relaxing the assumption of independence, multilevel analysis (or hierarchical modeling) distinguishes the different levels of analysis of a phenomenon and leads to a better understanding of the cross-level interactions (Leyland & Groenewegen 2020). Moreover, compared to standard regression, it allows us to measure the climatic heterogeneities that exist between regions in terms of health. The existence of such heterogeneities justifies the need to implement climate mitigation and adaptation strategies adapted to each context (Nyiwul 2021). This study seeks to determine individual and joint effects of climate variation, water supply and sanitation on the occurrence of diarrhea among children under five in SSA using multilevel mixed-effect Poisson regression including cross-level interactions. To do this, we use Demography Health Survey (DHS) data from 25 SSA countries covering the period 2000–2019 and temperature and precipitation data from geolocated DHS databases. It contributes to the existing literature by providing the first attempt to explore the relationship between climate change, water, sanitation and diarrhea using multilevel analysis.

The data used in this research is derived from the fusion of individual country and climate data. The individual data comes from the DHS of 25 SSA countries carried out between 2000 and 2019, i.e. a total of 57 DHS. The criteria for selecting SSA countries in this study were: (i) the country must have climate data available under the DHS program; (ii) it must have a DHS dataset with standardized questions on drinking water sources and toilet type at the household level over the period 2000–2019. DHS are nationally representative household surveys that provide indicators in the areas of population, health and nutrition. This study uses file Kid's Recode (KR) of DHS data type that contains child health indicators such as recent occurrences of diarrhea, fever and cough for young children and treatment of childhood diseases. This file can be downloaded from https://dhsprogram.com/data/available-datasets.cfm. The climate data, in particular on rainfall and temperature, come from the geolocalized databases developed by the DHS program for each of these surveys. After merging these databases, our work covers a sample of 530,625 children under five. Table 1 presents the different DHS for which geolocated data are available per country and per year. The geolocated data are available on http://spatialdata.dhsprogram.com/covariates.

Data analysis was carried out in STATA version 14 (StataCorp; College Station, TX, USA).

The household variables are of two types: household environment variables and household characteristics variables. The variables in the household environment are source of water supply (1. improved, 2. unimproved), type of sanitation (1. improved, 2. unimproved), which are collected at the household level in the DHS. We use the ‘improved’ or ‘unimproved’ classification scheme of WHO & UNICEF (2021) on page 100. In tropical regions, 80% of diseases are transmitted by germs in the water, or by vectors that thrive in it (Ntouda et al. 2013).

The variables that capture the household characteristics are the number of cases of diarrhea among children under five in the household who constitute our dependent variable (this variable is constructed from the variable h11 (diarrhea in the last 2 weeks) in DHS data, the standard of living (1. poor, 2. average, 3. rich), the place of residence of the household (1. urban, 2. rural), the sex of the head of the household (1. male, 2. female) and the average number of children per household.

The climate variables are GPS location data of surveyed DHS clusters. These are mean annual temperature data (units: Degrees Celsius) and mean annual rainfall level (units: millimeters per year) collected at the cluster level in each DHS in each country presented in Table 1.

Through the estimation of these models, multilevel analysis can detect the context effect and the composition effect associated with the variability of a phenomenon through variations in the inter-group variance for each model. If the variance decreases from M0 to M1 ⁠, the differences in diarrhea occurrence between households are due to a compositional effect, i.e. differences are attributable to differences in group composition (i.e. in the characteristics of the household who make up the clusters). In contrast, if the variance decreases from M0 to M2 ⁠, the differences in diarrhea occurrence are due to a context effect, i.e. differences are attributable to differences in group characteristics (climatic characteristics).

The multilevel analysis is also based on the Variance Partition Coefficient (VPC) and the Median Rate Ratio (MRR), both measures of higher level residual heterogeneity. Goldstein et al. (2002) introduced the VPC to describe the percentage of variation in a dataset that is attributed to a particular level. In our study, VPC measures the proportion of the total variance in the occurrence of diarrhea attributable to climatic differences. Generally, this statistic is the ratio of the inter-group variance to the total variance.

MRR measures the change in the occurrence of diarrhea when comparing two identical individuals randomly selected from different groups classified according to the degree of the rate (high or low). Higher MRR indicates larger cross-country differences.

The use of multilevel analysis is relevant because inter-group variance is statistically different from 0 in the M0 model. Table 2 presents the inter-group variance, VPC and MRR of each model. The results show that 77.7% (VPC (M4) = 0.777) of the variation in the occurrence of diarrhea in Sub-Saharan households are due to climatic differences between clusters.¹ It also emerges from the MRR analysis that a household residing in a cluster with a high incidence of diarrhea among children under five is 1,567 times more likely to have diarrhea cases than a household from a cluster with a low incidence of diarrhea. Similarly, the decrease in inter-group variance between model M0 (0.424) and M1 (0.365) reflects the existence of a context effect. In addition, there is a composition effect (decrease in inter-group variance between model M0 and M2, 0.424 < 0.302). Thus, the variation in the occurrence of diarrhea between cluster is due to differences in household characteristics (composition effect) but also to differences in climate (context effect).

The results in Table 3 show that the increase in mean temperature of 1 °C results in the increase of the incidence rate (IRR) of diarrhea of 6.7% in Sub-Saharan households (IRR = 1.067, p < 0.001), while a 1 mm increase in rainfall results in a 0.6% increase in incidence rate (IRR = 1.006, p < 0.001). The results show that in model M2, the type of sanitation influences the occurrence of diarrhea in households where water supply is not significant. However, after the introduction of the climate and inter-action variables in model M4, not only does the water supply variable becomes significant but there is also an increase in the effect of the climate variables between M1 and M4. Household environment variables mediate the effect of climate variations on the occurrence of diarrhea. Table 3 shows that the incidence rate of diarrhea in poor and middle-income households is respectively 1.21 (IRR = 1.21, p < 0.001) and 1.12 (IRR = 1.12, p < 0.001) times greater than that of rich households. Similarly, the incidence rate of diarrhea in rural households is 1.049 (IRR = 1.049, p < 0.001) times higher than in urban households.

The previous table shows that the interaction between the climatic variables (level 2) and the household level variables (level 1) is statistically significant (IRR = 1.013, 1.000, 1.001, p < 0.001). Since it is not easy to interpret the Incidence Rate Ratio (IRR) of these interaction variables, the analysis of the results is based on the analysis of the marginal plot from the estimation of the interaction model (Figure 1). Thus, Figure 1 shows that the combined effect of average rainfall and average temperature on the occurrence of diarrhea in children under five varies according to the type of sanitation and the source of water supply used by the household. Thus, as the average rainfall and temperature increases, households using unimproved sanitation or unimproved water have more cases of diarrhea than those using improved sanitation or improved water. Specifically, when the temperature is between 15 and 25 °C and is associated with increased rainfall, the occurrence of diarrhea in households using unimproved sanitation or unimproved water increases, while it decreases in households using improved sanitation or improved water. In contrast, an average temperature of 30 °C combined with increased rainfall results in an overall decrease in the occurrence of diarrhea in households (regardless of type of sanitation or water source).

The results of the multilevel Poisson regression model obtained in our study show that there are climatic heterogeneities in SSA countries. It is estimated that 77.7% of the variation in occurrence of diarrhea among children under five in households are due to climatic differences. Moreover, the results show that climatic variables have a positive effect on the occurrence of diarrhea in children, which is in line with the findings in the studies of Musengimana et al. (2016), Azage et al. (2017), Abu & Codjoe (2018) and Alemayehu et al. (2020) and contradicts to the work of Bandyopadhyay et al. (2012) and Mukabutera et al. (2016) which found that rainfall has a negative effect on the prevalence of diarrhea. In our study, the positive and significant effect of rainfall on the occurrence of diarrhea in Sub-Saharan African countries can be explained by the fact that, in countries in the region where a large part of the population practice open defecation (22%), use flush toilets that do not drain into a sewer system or pit latrines without slabs and use unimproved water (World Health Organization & UNICEF 2021), heavy rainfall results in surface water runoff that carries feces materials to the unprotected water sources (boreholes and wells) that people use (Wilby et al. 2005; Gray 2014; Bhandari et al. 2020a, 2020b). According to some studies, runoff water contains waterborne diseases including Rotavirus (Koroglu et al. 2011), Cryptosporidiosis and Giardiasis (Lane & Lloyd 2002; Baldursson & Karanis 2011) Campylobacter and Escherichia coli (Strauch 2011); all of which are responsible for episodes of diarrhea in children. Drinking unimproved water is one of the factors responsible for diarrhea in children.

Another finding in this study is the significant and positive impact of temperature (between 15 and 25 °C) on the occurrence of diarrhea in households using unimproved water or unimproved sanitation. This is also in line with the studies of Bandyopadhyay et al. (2012), Chou et al. (2010), Mukabutera et al. (2016) and Azage et al. (2017). This can be explained by the fact that high temperature is associated with rapid multiplication and survival of diarrhea pathogens for longer periods (Lal et al. 2012; Carlton et al. 2016). Hoque et al. (2006) and Levy et al. (2016) note that warm climatic conditions lead to temporal changes in human behavior, such as higher water consumption, use of unimproved drinking water and less hygienic practices due to water scarcity. Moreover in SSA, the most severe cases of diarrhea in children are caused by the Rotavirus, Campylobacter, Cryptosporidium spp., Escherichia coli and Shigella spp. (Galloway & Cohen 2021). Campylobacter in unimproved water proliferates at ambient temperatures between 5 and 28 °C (Oberheim et al. 2020) while Rotavirus is present in high concentrations at 18 °C (Barril et al. 2015) which explains why at these temperature conditions, the prevalence and incidence of diarrhea increases. On an additional note, our findings on the decrease in the occurrence of diarrhea in children in SSA for a temperature of 30 °C is in line with the results obtained by Rice & Johnson (2000), Obiri-Danso et al. (2001), van Gaalen et al. (2017) and Bhandari et al. (2020a, 2020b). The latter shows that bacteria (Campylobacter, E. coli, Shigella), protozoa (Gardia, Cryptosporidium) and viruses (Rotavirus) that cause diarrheal diseases are inactive at higher temperatures (more than 30 °C), resulting in fewer cases of diarrhea. This suggests that higher temperatures play an important role in regulating diarrhea cases through less viral transmission.

People's vulnerability to climate change depends on their standard of living. In this study, household poverty is a factor that increases the incidence ratio of diarrhea in children under five. Fagbamigbe et al. (2021), Novignon & Nonvignon (2012) and Yusuf et al. (2010) reveal that in low- and middle-income countries the prevalence of diarrhea is higher among children from poor households than among those from non-poor households. Moreover, Shiras et al. (2018) and Simiyu et al. (2017) show that in many low-income countries, poor households often share sanitation facilities, circumstances that increase the risk of diarrhea.

Our results suggest that efforts must be made at the macro and micro levels to achieve Sustainable Development Goal (SDG) 6 which specifies that access to safe drinking water and sanitation must be guaranteed for all within 10 years, in order to mitigate the effects of climate change on health in SSA. Indeed, we have shown that fluctuations in temperature and rainfall in an unsanitary environment contribute to a more rapid increase in diarrhea cases in households. This implies that adaptation to climate change in the water sector will be an appropriate response to control the occurrence of diarrheal diseases. In Africa, the water sector is the second most vulnerable sector to the effects of climate change, behind the agricultural sector (Nyiwul 2019). The most common adaptation strategies can be summarized as follows: storage (groundwater recharge dams or reservoirs and subsurface dams, conservation ponds and reservoirs), reutilization and community-based water resource management (IPCC 2014; Nyiwul 2019). In addition, investment in flood protection technologies is another aspect on which SSA must focus on in order to adapt to climate change (Calzadilla et al. 2013).

Certain limitations regarding data analysis should be mentioned. The study reveals a selection bias problem due to the constitution of our country's sample which is conditioned by the availability of DHS data over the period 2000–2019. It is difficult to overcome such a bias in the statistical analysis of survey data collected at the regional or sub-regional level. The main way to reduce this bias would be to have panel data on a large sample of SSA countries, although in practice such data are difficult to obtain. Also, our estimates do not include random effects such as the influence of water quality on diarrhea occurrence which may vary from one context to another. These influences could be caused by the presence of unobservable heterogeneity such as household knowledge of water purification methods. Thus, future studies could opt for an estimation method that captures such effects in order to better appreciate the relationship between climate change and diarrhea occurrence. These studies could also incorporate into their model variables on climate change adaptation strategies in the water sector.

Water supply and sanitation mediate the effect of climatic variations on the prevalence of diarrhea. This study highlights the need for a multilevel analysis of the diarrhea phenomenon in SSA by splitting the micro level (or individual level) and the macro level (or contextual level) to avoid atomistic fallacy, ecological fallacy that biases the results. We found that variations in the occurrence of diarrhea are due to climatic differences (context effect). Furthermore, we found that there was a positive association between temperature, rainfall and occurrence of diarrhea when households used unimproved water and unimproved sanitation. These results highlight the need to adopt climate change adaptation strategies in the water sector both at the macro (national policies) and micro (household) levels in order to reduce the occurrence of diarrheal diseases within households. At the macro level, this means protecting water sources from flooding or surface water, increasing people's access to water and improved sanitation facilities. At the micro (household) level, this involves training individuals in non-potable water treatment methods such as chlorination and filtration and in hygiene practices.

All relevant data are available from an online repository or repositories (https://dhsprogram.com/data/available-datasets.cfm and https://spatialdata.dhsprogram.com/covariates/).