1.1. Malaria

In 2023, 263 million malaria cases were reported globally [1]. Cases of malaria were declining globally up to 2015, then increased during the 2019−2020 year, coinciding with the beginning of the COVID-19 pandemic [1]. Notably, there were 236 million cases recorded in 2019 and 247 million in 2020 [1]. The observed declines occurring until 2015 may be attributed in part to interventions such as long-lasting insecticidal nets (LLINs) and indoor residual spraying (IRS), and to the efficacy of antimalarials [2]. Malaria-attributed mortality has been decreasing globally since 2000; however, there was a marked increase recorded in 2020, with 567,000 deaths in 2019 compared to the 622,000 in 2020 [1]. In 2023, 597,000 mortalities were recorded, a decrease from the previous year [1].

The life cycle of malaria parasites takes place in the female Anopheles mosquito as well as in infected humans [3]. Malaria parasites are transmitted to the human host from an infected mosquito during a blood meal [3]. The Plasmodium parasites first reproduce in hepatocytes then progress to the blood stages, where symptoms occur primarily due to schizont rupture and the destruction of erythrocytes [3,4]. The life cycle is completed in the Anopheles mosquito after it feeds on an infected human [3]. Climate and weather variables, such as precipitation and temperature, can have an important impact on malaria transmission by affecting the parasite development and vector density [5].

1.2. Malaria, weather, and climate

Precipitation has the ability to both benefit and hinder malaria transmission [6]. Precipitation can create more breeding sites for the vector, which in turn promotes malaria transmission and incidence [5]. However, an overabundance of rain may flood breeding sites which can reduce vector density [6]. Yamba et al. (2023) observed that the range of acceptability for rainfall differed between four African regions, but depending on the region, the maximum monthly rainfall value that allows for malaria transmission was around 400 mm or 600 mm [6]. The minimum was about 80 mm [6].

The impacts of temperature on malaria transmission are also range-dependent [7]. In the optimum range of temperatures, parasite development may be promoted and this can benefit malaria transmission [5,6]. Temperature can also affect factors such as mosquito fitness and life cycle which can in turn affect transmission [8]. The optimal range for high malaria transmission may be between 20°C to 28°C(9). Outside the optimal range, temperature may reduce malaria transmission, with below 15°C-16°C and above 40°C limiting parasite development [6]. These ranges are dependent on the mortality model [7]. With both precipitation and temperature, the exposure indirectly affects prevalence via effects on the vector/parasite. Many factors can impact malaria transmission contemporaneously, including climate and weather variables and the development of insecticide and antimalarial resistance, pandemics, financial and socioeconomic barriers, behavior, and access to healthcare [2,5,9,10].

1.3. Malaria, weather, and climate in Mali

There is sufficient evidence to suggest that meteorological variables, such as precipitation and temperature, affect malaria transmission in Mali and Africa in the general population [5,10]. In 2023, 94% of malaria cases (246,000/263,000) and 95% of malaria-attributable deaths (569,000/597,000) occurred within the WHO African Region [1]. Mali had 3.1% of all malaria cases globally (~8,153/263,000) and 2.4% of malaria-attributable deaths (~14,328/597,000) [1]. There are nuances, however, in the effect that climate and weather variables have in different regions, including Mali, such as geographical location, environment, and human behaviors [10]. For example, in Kolia, Mali, malaria transmission is sustained outside the rainy season by irrigation that keeps water from flowing downstream in the area [10]. According to a previous scoping review, the impact of temperature and precipitation on the prevalence of malaria in children under 5 has not been studied solely in Mali [11].

1.4. At-risk population

Of the deaths in the WHO African Region in 2023, 76% were among children under the age of five (~432,440/569,000) [1]. This population is particularly vulnerable as they lack acquired immunity [12]. As a result, in comparison to other populations, children under five have an increased likelihood of developing severe malaria which may include severe anemia, hypoglycemia, and cerebral malaria [12]. Furthermore, children are more vulnerable to the effects of climate change due to a number of factors, including physiology and behavior [13]. Compared to adults, children have less effective heat adaptation capacity and they spend more time outside [13].

1.5. The problem and objectives

Malaria is a prevalent disease in Mali that affects individuals of all ages in impactful ways [1,2,4,12]. The link between weather and malaria plays an important role in both the transmission and incidence of the disease [5]. Both temperature and precipitation have the capacity to promote and demote the spread of malaria [6]. It is important to develop an understanding of how climate and weather may impact malaria over time to inform policy and adaptation strategies in every country affected by malaria [14]. As such, the objective of this exploratory study was to investigate potential associations between precipitation and temperature and prevalence of malaria infection in children under 5 in Mali, a country with a lack of evidence on this question [11].

2.1. Study design, population, and location

A cross-sectional study was conducted using historical data from 2012/13, 2015, 2018, and 2021 in Mali. Data sources included the Mali Demographic and Health Surveys (DHS) and Malaria Indicators Surveys (MIS) conducted in 2012/13 (Standard DHS), 2015 (MIS), 2018 (Standard DHS), and 2021 (MIS) [15–18]. DHS surveys are conducted by the DHS Program to collect primary data on a variety of public health topics, including malaria [19,20]. The DHS uses different types of questionnaires, including a household questionnaire and individual questionnaires [20]. These standardized surveys are conducted approximately every five years [19]. The MIS is an additional survey that collects data on malaria indicators, largely concerning children under five and pregnant women, and is also conducted every five years [21,22]. The DHS Program is implemented by ICF via the United States Agency for International Development (USAID) [23]. All DHS procedures and questionnaires are reviewed and approved by ICF Institutional Review Board (IRB) [24]. The data was initially accessed for research purposes on November 12, 2023. There is no information that can identify individual participants during or after data collection. Exact geographical coordinates for the surveys are not provided as the DHS program randomly displaces GPS latitude and longitude positions in all surveys for confidentiality purposes [25]. This research received ethics approval from the University of Alberta (ID: Pro00131799).

The population of Mali in 2023 was 23,769,127 [26]. The region was skewed towards younger ages with 46.8% of the population between the ages of 0–14 years and the median age being 16.4 years [27,28]. The population was 46.2% urban, with a 4.57% annual rate of urbanization [28]. Mali had an average household size of 5.8 people [26]. Around 7 out of 10 households had improved sources of drinking water access, 55% had improved sanitation, and 49% had electricity [26]. Concerning education and literacy, 66% of women and 53% of men had no education, and 2% of women and 6% of men had above a secondary education [26]. For literacy, 28% of women were literate compared to 47% of men [26].

In the DHS datasets, Mali is divided into 9 regions, including the capital of Bamako as its own region (Fig 1) [26]. The country is located in Western Africa, south of Algeria, and according to 2018 estimates, 34.1% of land is agricultural and 10.2% is forest area [28]. The vast majority of Malians live in the southern part of the country [28]. Mali’s climate is seasonal [29]. The rainy season is from May to October and peaks in August (average total precipitation of 109.75 mm across 1991–2020) [29,30]. The annual average mean surface air temperature in Mali was 29°C in 2012, 29.3°C in 2015, 29.36°C in 2018, and 29.81°C in 2021 [29]. Annual precipitation In Mali was 365.08 mm in 2012, 339.41 mm in 2015, 353.37 mm in 2018, and 321.85 mm in 2021 [29]. Temperature and precipitation also vary across the regions of Mali [29]. Malaria transmission typically peaks in October, at the end of the rainy season [31].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Regions of Mali.

https://doi.org/10.1371/journal.pone.0342127.g001

2.2. Outcome

The outcome of interest was malaria prevalence diagnosed by rapid diagnostic tests (RDT) in children ages 6–59 months. Malaria prevalence data were sourced from the Mali DHS and MIS. Participants were selected based on a stratified two-stage cluster design in which enumeration areas were based on census files and then a fixed number of households were selected by equal probability systematic sampling in each enumeration area [32,33]. The RDT sampling was done at the time of the interview with consent from parents [21]. Blood samples were immediately tested for anemia and malaria and results were given to participants during the interview [21]. The 2021, 2018, and 2015 surveys were conducted between August/September and November, overlapping with the rainy season [15–17,31]. The 2012/13 survey was conducted between November 2012 and February 2013 [18].

2.3. Exposure data

Exposures examined include the meteorological variables precipitation and temperature. Daily total precipitation (mm) data (2012/13, 2015, 2018, and 2021) were obtained from the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS), which has a spatial resolution of 0.05° (~5.3 km) [34]. Monthly totals were calculated. Daily maximum and minimum temperature (°C) data were obtained from the Climate Prediction Center (CPC) Global Unified Temperature dataset from National Oceanic and Atmospheric Administration (NOAA) Physical Sciences Laboratory. Average temperature (°C) was generated by taking the average of the daily maximum and minimum temperatures. Monthly averages were used. Precipitation and temperature were spatially linked to DHS clusters using ArcGIS Pro. DHS surveys were conducted in different months, with some overlap between the years. Three lags were created for each exposure based on the month that any one individual was surveyed [35]. For example, a child surveyed in September would include temperature and precipitation exposures from June, July, August, and September. CHIRPS was created by the University of California, Santa Barbara Climate Hazards Center (CHC) in collaboration with the United States Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center to provide complete data in regions where rain-gauge stations are not consistently placed [34]. The National Oceanic and Atmospheric Administration (NOAA) is a bureau within the United States Department of Commerce [36].

2.4. Covariates

DHS variables selected as relevant covariates are listed and defined in Table 1. Household-level variables included mother’s highest education level, type of place of residence, household electricity, main floor material, main roof material, main wall material, wealth index combined, mean cluster altitude (m), and the variable indicating whether the dwelling had been sprayed with insecticide in the last 12 months. The rest of the variables were treated as individual-level variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of the health outcome and characteristics for 2021, 2018, 2015, and 2012/13.

https://doi.org/10.1371/journal.pone.0342127.t001

2.5. Statistical analysis

Individual-level (covariates and outcome), household-level (covariates), and cluster-level (exposures) data were used to analyze the associations between temperature and precipitation exposures and malaria diagnosis in children under 5 in Mali. Different DHS datasets (Household Member Recode and Children’s Recode datasets) were merged by using multiple variables in the dataset to uniquely identify each child (using the child’s line number in the household, the cluster number, the household number, and the respondent’s (mother’s) line number). These data, including the covariate and outcome data, were subsequently merged with the exposure data using the cluster number and date measured (month). Missing data were checked individually in the Household Member Recode and Children’s Recode datasets, as well as in the combined dataset, to ensure that variables did not contain a high degree of missingness (>15% missingness). Observations missing either the outcome or exposure were excluded. The 2018 dataset was the only one to have high degrees of missingness, but missingness was less than 15% after dropping observations which did not have a response for malaria RDT. Transformations were undertaken for continuous variables using skewness and kurtosis values. Body mass index was logarithmically transformed, age and altitude were categorized, all other continuous variables were assessed as normal. Model variables were selected a priori. Multicollinearity was checked; anemia level and hemoglobin level adjusted for altitude were highly multicollinear, as was slept under a mosquito bed net the previous night and person slept under a long-lasting insecticidal net (LLIN). Hemoglobin level adjusted for altitude and slept under a mosquito bed net the previous night were kept in the analysis as they included more detail. Anemia level and person slept under a long-lasting insecticidal net (LLIN) were excluded (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Data management flowchart.

https://doi.org/10.1371/journal.pone.0342127.g002

Variables were grouped into four categories for entry into the models: demographics (sex, type of place of residence, mother’s highest education level, age, and cluster altitude in meters), economics (household electricity, main floor material, main roof material, main wall material, and wealth index combined), biological (hemoglobin level adjusted for altitude and BMI), and interventions (slept under a mosquito bed net the previous night, medication taken for fever, received drugs to prevent malaria current month or last, given medication to prevent malaria in applicable year, and dwelling sprayed in last 12 months). This was done so that covariates could be entered into the model sequentially to avoid the potential for overfitting.

Logistic regression was conducted to assess univariate and multivariable relationships between exposures (minimum temperature, maximum temperature, average temperature, precipitation) and malaria for each lag (0, 1, 2, 3) in a pooled analysis of all of the years (2021, 2018, 2015, 2012/13). Two-level multilevel modeling was used where the first level was the individual and the second level was the cluster. Only covariates which were available in all years were included in the analysis. Grouped covariates were sequentially added to the models (final model adjusted for all covariates). A sensitivity analysis was performed to ensure that the exclusion of selected covariates (medication taken for fever, given medication to prevent malaria in applicable year, body mass index (BMI), has dwelling been sprayed in last 12 months, received drugs to prevent malaria this month or last) did not statistically significantly impact the results of the model (S3 Tables). A pooled meta-analysis was attempted, though not successful due to high heterogeneity between the years. Year was included as a covariate in the pooled analyses to adjust for this heterogeneity. Additionally, multilevel logistic regression models were developed for all years individually (S4 Tables).

Variable categories were added sequentially: univariate (model 1), adding demographics (model 2), adding economics (model 3), adding biological (model 4), and adding interventions (model 5/final model). Effect modification by age and sex was assessed in the pooled analysis and individual year models (S5 Tables). All models included multilevel weights [40]. All analyses were completed in STATA MP/18.

3.1. Descriptive summary

3.1.3. Exposures.

The median pooled precipitation level across Mali varied depending on the lag (Table 2). The smallest median is with a zero month lag (43.7 mm); this is opposed to the largest median which is with a two month lag (169.6 mm). The highest median temperature was 33.9°C with a zero month lag and the lowest median temperature was 22.9°C with the same lag. The highest median for average temperature was 29.1°C with a three month lag and the lowest median for average temperature 28.0°C with a one month lag. Exposure descriptives have also been summarized for the yearly populations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of the exposures for 2021, 2018, 2015, and 2012/13.

https://doi.org/10.1371/journal.pone.0342127.t002

3.2. Associations with exposures

3.2.1. Precipitation.

Precipitation was statistically significantly associated with the odds of malaria with a three month lag, where a statistically significant risk factor result was observed for precipitation in every model (Table 3). The adjusted odds of being positive for malaria RDT in children ages 6–59 months was 1.005 times greater for every millimeter increase in total precipitation for the three month lag in the pooled models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Adjusted multilevel logistic regression model associations for every lag, pooled across 2021, 2018, 2015, and 2012/13, for every exposure variable with malaria prevalence diagnosed by RDT in children ages 6 to 59 months.

https://doi.org/10.1371/journal.pone.0342127.t003

3.3. Sensitivity analysis

A sensitivity analysis was completed comparing the main pooled analysis with models including the covariates (medication taken for fever, given medication to prevent malaria in applicable year, body mass index (BMI), has dwelling been sprayed in last 12 months, received drugs to prevent malaria this month or last) excluded in the main analysis due to selected missingness across survey years. The models including these excluded variables were constructed by selectively pooling the years in which these covariates were available. These models followed similar trends to the main analysis suggesting that they did not significantly impact the main findings (S3 Tables). Data for each of the years were also analysed separately (S4 Tables) and did not generate the same results, though this was likely due to sample size limitations.

4.1. Limitations

This study has limitations which must be acknowledged. The sample size of any one survey was limited. While pooling of the years increased statistical power, the different times that the field work was conducted between these surveys may have increased heterogeneity [54]. The analysis was conducted on data from Mali as a whole without accounting for how this relationship may vary geographically across the country. Selected variables were omitted due to collinearity, which may have impacted the association between the outcome and the exposure, though sensitivity analyses were undertaken to test the consistency of the results. Potential confounders, such as mother’s occupation, time children spent outdoors, and the health-seeking attitude of the mother, were not included in the study. Due to the limitations of RDT, it is possible that lower level infections could have been missed [55]. Undetected cases of malaria are still affected by environmental exposures. RDT is also incapable of differentiating between untreated current infections and recently treated current infections [55]. Variables required assumptions, for example the use of the variable “slept under a mosquito bed net the previous night” assumes that previous night is representative of most/all nights. Generalizability is limited; the effect of these exposures varies across regions, as such the results of this study may not extend outside of Mali [11]. The DHS may suffer from volunteer and nonresponse bias wherein ill individuals may be more likely to participate than healthy individuals and respond to different questions. Exclusion bias, which is bias that occurs from the exclusion of a part of the population, occurred in the DHS surveys of Mali due to the purposeful exclusion of regions for security reasons.