Estimating cases of severe malaria at the population-level: Analysis of household surveys from 19 malaria endemic countries in Africa

The burden of severe malaria is uncertain at the population level because existing estimates rely exclusively on data from the formal healthcare system. Using data from population-based surveys this analysis examines severe malaria cases at the population level, which captures children whose caregivers 1) have taken the child to a healthcare facility but the child’s illness did not resolve, or 2) have not sought care for the child’s illness. Direct inclusion of these children in severe malaria estimates has been an underlying data gap. This analysis examined data from 37 Demographic and Health Surveys and Malaria Indicator Surveys across 19 countries in sub-Saharan Africa collected between 2011 and 2018. The outcome of interest is severe malaria, dened as children age 6–59 months who were positive for malaria with at least one self-reported symptom for severe malaria, including loss of consciousness, rapid breathing, seizures, or severe anemia. The study includes a weighted descriptive, country-level analysis and a multilevel mixed-effects logistic regression model to assess the determinants of severe malaria.


Results
Among children positive for malaria across all surveys, 4.5% (95% CI 4.1-4.8) had at least one symptom of severe malaria, which was signi cantly associated with age, residence, wealth, and survey timing at a p-value less than 0.05. Children in the higher malaria transmission zone were more likely to have symptoms compared to those in the lowest transmission zone; however, these results were not statistically signi cant.

Conclusion
An accurate estimate of the burden of severe malaria is essential to assessing the impact of malaria interventions and to guiding future malaria investments. This analysis presents a novel approach of estimating the burden of severe malaria in children under age ve in malaria endemic countries. Estimating severe malaria through household-based surveys allows countries to estimate severe malaria across time and to compare with other countries. Having a population level estimate of severe malaria helps further our understanding of the burden and epidemiology of severe malaria.

Background
In 2018, an estimated 405,000 deaths from malaria occurred globally, with an estimated 70% of all malaria deaths occurring in children under age 5 in sub-Saharan Africa (SSA) [1]. Malaria typically begins as an acute febrile illness. If not appropriately treated, Plasmodium falciparum (P. falciparum) malaria can progress to severe illness, and can often lead to death. Children with severe malaria frequently develop one or more of the following complications: severe anemia, respiratory distress, or cerebral malaria [2,3]. The clinical manifestations of these severe malaria complications in children include impaired consciousness, respiratory distress (acidotic breathing), multiple convulsions, prostrations, shock, and jaundice [2,4]. The clinical epidemiology of severe malaria can also present differently according to age and transmission intensity. Studies have shown that as the intensity of malaria transmission increases, the mean age of severe malaria decreases [5][6][7][8]. In high transmission areas, the risk for severe malaria is greatest among young children (the rst few months of life to age 5). Severe malaria becomes less common in older children when acquired immunity provides a protective effect. Conversely, in low transmission areas, severe malaria is more common in older children and adults [7].
Although the clinical features and transmission patterns of severe malaria are well understood, accurately capturing the burden of severe malaria remains a challenge. First, severe malaria symptoms are non-speci c, which makes it di cult to differentiate severe malaria cases from other diseases that are also common in malaria-endemic countries [2]. Second, severe malaria case estimates from routine data collected through formal healthcare services are often affected by variable data quality and non-standard case de nitions, which make comparisons across countries challenging [9][10][11]. Finally, a proportion of severe malaria illnesses and deaths occur outside the formal healthcare system and are therefore undocumented [7,9,[12][13][14].
An accurate estimate of the burden of severe malaria is essential to assess the impact of malaria prevention and control interventions and to guide future malaria investments. To date, the only robust estimates of severe malaria cases include children who present to the formal healthcare system or data from small epidemiological studies across SSA. It has been a challenge to use these data to model the number of severe malaria cases across SSA because of varying age ranges of reporting, different diagnosis techniques, surveillance methods, and health care utilization [5,8]. The Demographic and Health Surveys (DHS) Program, the primary global source of population-based malaria data, tests children for malaria and also measures hemoglobin concentration to test for anemia. For children who test positive for malaria, The DHS Program collects data on severe malaria symptoms for referral purposes during data collection. However, this information has not been used to examine signs and symptoms of severe malaria. To better understand the burden of severe malaria in SSA, this paper presents an analysis of severe malaria symptoms in children age 6-59 months who are positive for P. falciparum malaria, and whose data are captured in population-based surveys.

Data on severe malaria
This study used data from DHS and Malaria Indicator Surveys (MIS), which are both nationally representative, population-based household surveys of The DHS Program [15]. All surveys are independent but use standardized data collection procedures and tools. When requested by the country, the DHS and MIS collect blood samples for anemia and malaria testing among children age 6-59 months. All children are tested regardless of whether they have signs or symptoms of malaria. Specially trained biomarker health technicians (who are usually nurses) take capillary blood obtained with a nger or heel prick. The blood is immediately tested for anemia and malaria in the eld and the results are provided to respondents' parents or guardians within a few minutes.
In the eld, biomarker health technicians use rapid diagnostic tests (RDT) to determine if children have malaria. The RDTs detect the histidine-rich protein II (HRP-II) antigen of P. falciparum in blood. P. falciparum is the primary cause of severe malaria and is the predominant species found in the countries included in this analysis [2,7]. Often blood is also collected for the examination of blood smears by microscopy in the laboratory. DHS and MIS surveys also typically test children age 6-59 months for anemia with capillary blood collected through a nger or heel prick. The tests use the HemoCue 201 + and, on occasion, the HemoCue 301 point-of-care hemoglobin testing system. At that time, additional questions were added to the survey's questionnaire to screen children for signs and symptoms of severe malaria. To assess severe malaria signs and symptoms, the caregiver is asked: "Does (NAME) suffer from any of the following illnesses or symptoms: extreme weakness, heart problems, inability to drink or breastfeed, vomiting everything, loss of consciousness, rapid breathing, seizures, bleeding, jaundice, or dark urine?" If the child is positive for malaria (by RDT) and the caregiver answers "Yes" to any of the signs and symptoms and/or if the child has a hemoglobin concentration of < 8 grams per deciliter (g/dL), the child is given a referral slip for severe malaria to take to the nearest healthcare facility for care [16]. A child who is negative for malaria (by RDT) is not asked about severe malaria symptoms. (See additional le 1 for more information about The DHS Program malaria referral process.)

Data analysis
Case de nition of severe malaria This analysis used a severe malaria case de nition based on signs and symptoms captured in DHS and MIS surveys that most closely aligns with the clinical symptoms of malaria outlined in the World Health Organization (WHO) Management of Severe Malaria Handbook [2]. Of the 11 severe malaria signs and symptoms collected by DHS and MIS surveys, loss of consciousness, rapid breathing, seizures, or severe anemia (hemoglobin levels < 5 g/dL adjusted for altitude) were the four most common clinical manifestations of severe malaria in children that most closely align with the symptoms identi ed in the WHO Management of Severe Malaria Handbook. Hemoglobin levels are adjusted for the altitude in areas that are above 1,000 meters in elevation [17].

Study variables
Outcome variable: severe malaria. The outcome of interest is severe malaria, de ned as children age 6-59 months who were positive for malaria with at least one symptomatic marker for severe malaria, including loss of consciousness, rapid breathing, seizures, or severe anemia. The authors calculated the percentage of children age 6-59 months with severe malaria of all children age 6-59 months who were positive for malaria, according to RDT.
Covariates: all potential confounders. For this analysis, variables found in the literature related to severe malaria were reviewed and included based on data availability. The model included the following covariates: sex, age, place of residence, wealth quintiles, malaria endemicity, survey timing, and country ( Table 1). The child's age was divided into four age categories: 6-23 months, 24-35 months, 36-47 months, and 48-59 months. Place of residence is de ned as whether a household is located in a rural or urban area. Wealth quintiles were derived from the DHS wealth index, which measures the relative socioeconomic status of households based on household assets and amenities at a point in time [18]. Survey timing is divided into two equal categories based on the eldwork dates of 2011-2014 and 2015-2018. To determine malaria endemicity, we assigned each child's household enumeration area into geographical zones based on malaria transmission risk. To link the DHS and MIS geo-coordinates (latitude, longitude) of each survey enumeration area to transmission risk zones, we used geo-coordinated P. falciparum parasite prevalence rates among children age 2-10 (PfPR 2 − 10 ) from the Malaria Atlas Project 2015 [19]. We assigned every child's household in an enumeration area from the DHS or MIS survey dataset to the same malaria transmission risk zone based on corresponding PfPR 2 − 10 data for that enumeration area. For the transmission zone categories, we used risk categories as outlined in the WHO Management of Severe Malaria Handbook [2] with low transmission de ned as PfPR 2 − 10 ≤ 10% and high transmission de ned as PfPR 2 − 10 >50%. Due to a high number of children in the moderate risk category (11%< PfPR 2 − 10 ≤ 50%), we divided this risk group into two equal risk categories of moderate transmission A (11%< PfPR 2 − 10 ≤ 30%) and moderate transmission B (31%< PfPR 2 − 10 ≤ 50%).

Study population
The inclusion criteria for the study were all malaria-endemic countries in SSA that have conducted a DHS or MIS survey in which children were tested for malaria and anemia, and the surveys included questions on signs or symptoms of severe malaria. In total, this analysis examined results from 37 surveys across 19 countries (Fig. 1). The study population for this analysis included children age 6-59 months who stayed in surveyed households the night before the survey and received P. falciparum malaria parasite (RDT) and anemia tests. Children with any missing household enumeration area PfPR 2 − 10 data were excluded from the analysis.

Regression analysis
The study includes a country-level descriptive analysis weighted for complex survey design and a multi-country weighted pooled analysis, both with 95% con dence intervals. The study used a multilevel (individual-level and country-level) unweighted mixed-effects logistic regression model to assess the determinants of severe malaria. The model includes sex, age of the child, residence, household wealth, malaria transmission zones, and survey timing. All analyses were conducted with StataSE16 (StataCorp LP, College Station, USA).

Descriptive Analysis
Country-level analysis  2) had at least one symptom of severe malaria compared with those in higher wealth quintiles and other transmission zones. Slightly more than half (53.9%, 95% CI 49.8-57.9) of children with at least one severe malaria symptom were surveyed between 2015 and 2018 (Table 3). Regression Analysis The regression analysis shows the odds of reporting at least one symptomatic marker of severe malaria in relation to the different background characteristics of the children. Compared to children age 6-23 months, children age 36-47 months (AOR 0.73, 95% CI 0.65-0.82) and 48-59 months (AOR 0.58, 95% CI 0.51-0.66) were signi cantly less likely to report at least one symptomatic marker for severe malaria. Urban children were signi cantly more likely to report at least one symptomatic marker for severe malaria as compared to rural children (AOR 1.28, 95% CI 1.12-1.48). Socioeconomic status of the household is associated with the likelihood of reporting a severe malaria symptom, with children in the highest wealth quintile having lower odds compared to those from households in the lowest quintile (AOR 0.56, 95% CI 0.42-0.76). Malaria endemicity de ned by different transmission zones was not associated with the likelihood of reporting a symptom of severe malaria. Finally, the children surveyed between 2015-2018 were signi cantly less likely to report at least one symptomatic marker for severe malaria compared to children surveyed between 2011-2014 (AOR 0.72, 95% CI 0.62-0.83) ( Table 4). Level of statistical signi cance *** p < 0·001, ** p < 0·01, * p < 0·05

Discussion
Despite improvements in the diagnosis and documentation of severe malaria cases, there remains uncertainty about the burden of severe malaria cases at the population level. Here we present a comprehensive estimate of severe malaria cases in children from 19 malariaendemic countries in SSA. Our estimates are based on data from population-based household surveys that allow severe malaria cases outside of the formal healthcare system to be directly captured in estimates.
The prevalence estimate of severe malaria in this study (4.5% of malaria-infected children) is consistent with other estimates of severe malaria. This acknowledges that other estimates of severe malaria cases account only for children who access the formal healthcare system [1,12]. Household surveys test all children age 6-59 months for malaria, and while some of the children who were showing signs of severe illness would have eventually accessed the formal healthcare system, some of these children would have died, recovered at home, or received care outside of the formal healthcare system [7,9,[12][13][14]. Estimating severe malaria through household surveys provides countries with a standardized estimate of severe malaria that is comparable across time as well with other countries.
Findings from this study con rm previous observations that severe malaria is dependent on age and transmission intensity [5][6][7][8]. Younger children were signi cantly more likely to have severe malaria, and although not signi cant, the risk of severe malaria was greater in high malaria transmission zones. However, unlike previous research, we did not nd an interaction between age and the intensity of malaria transmission in relation to having at least one symptom of severe malaria (data not shown) [5]. One explanation is the limited age range of children (6-59 months) in this analysis. Past studies that have examined the association of variations in age and endemicity on the clinical manifestation of severe malaria included children up to age 10 [5,6,8].
Children surveyed between 2015-2018 were signi cantly less likely to have severe malaria symptoms as compared to children surveyed between 2011-2015. This is controlling for malaria transmission level and is irrespective of variations in malaria prevalence since all children included in the analysis were positive for malaria, according to RDT. This nding aligns with the 2019 World Malaria Report, which reported a decrease in malaria deaths since 2010 [1]. While the role of malaria control interventions in this difference cannot be assumed, since 2015, there has been an increase in the number of malaria interventions in SSA, including the implementation of seasonal malaria chemoprevention and universal coverage of insecticide-treated nets [1]. The impact of these interventions on severe malaria cases needs further exploration.
This analysis also indirectly highlights potential variation in care-seeking patterns for severe malaria cases across SSA. Urban children were signi cantly more likely to have at least one severe malaria symptom as compared to rural children despite a higher prevalence of severe malaria among the rural population. In addition, the country was highly signi cant in the model even when controlling for malaria endemicity. By examining severe malaria cases at the household level, this analysis is more likely to include children whose caregivers have taken them to a healthcare facility but whose illness did not resolve or have not sought care for the child's illness. While the decision to seek care is ultimately decided by the caregiver, it is highly in uenced by factors such as the availability of government-based facilities, country wealth, cost of care, and education [14,[20][21][22][23]. Further exploration is needed into country and urban-rural variations in care-seeking and its association with severe malaria burden estimates.
This study has a several limitations. Children with severe malaria frequently develop one or more complications, including severe anemia, respiratory distress, or cerebral malaria. This analysis examined children who had at least one symptomatic marker for severe malaria. We did not disaggregate this analysis by proxies for respiratory distress or cerebral malaria. Examining severe anemia is possible because this is a discrete diagnosis based on hemoglobin levels. However, to fully disaggregate by respiratory distress or cerebral malaria would require additional questions about symptoms such as prostration and the number and severity of convulsions [2]. The use of household-level data is a noteworthy advantage of this study, but it also introduces a principal limitation. There is a risk of including uncomplicated malaria cases or non-malaria cases in our proxy de nition. Malaria positivity is based on RDT-detectable antigens that continue to circulate in the blood after the infection has cleared, and severe malaria symptoms are non-speci c [24]. Our de nition of severe malaria relies on caregiver self-report rather than a diagnosis by a clinician at a health facility. Although the interviewer for the biomarker questionnaire is a trained biomarker health technician (usually a nurse), which may improve the questionnaire responses, the non-speci c nature of severe malaria remains an issue. We have addressed this limitation by narrowing the proxy de nition of severe malaria to only examine loss of consciousness, rapid breathing, seizures, or severe anemia (hemoglobin < 5 g/dL adjusted for altitude). These symptoms are more distinct than some other symptoms (extreme weakness and heart problems) caregivers are asked and most closely align with the WHO clinical manifestations of severe malaria. We assumed these symptoms would not be confused by caregivers, even those with a limited education. However, we were unable to examine the reliability of reported signs and symptoms because caregivers were not asked questions on severe malaria symptoms for malaria negative children. In addition, by limiting our de nition of severe malaria symptoms, there is the possibility that we may be missing cases.
This analysis only includes children with malaria according to RDT, which further minimizes the possibility that the child is sick with an illness other than severe malaria. However, as noted above, there is still a risk of including non-malaria cases in our proxy de nition because malaria positivity is based on RDT-detectable antigen that circulates in the blood after the infection has cleared. More sensitive measures of malaria diagnosis than standard HRP-II RDTs should be explored, such as microscopy, highly sensitive RDTs, or polymerase chain reaction (PCR).

Conclusions
This analysis investigated severe malaria symptoms in children age 6-59 months who are positive for malaria as identi ed in populationbased surveys. To date, there has been a gap in knowledge about the burden of severe malaria at the population level since previous estimates have relied exclusively on data from formal healthcare services. This analysis presents the most comprehensive estimate of the prevalence of severe malaria in children age 6-59 months from 19 countries across multiple malaria endemicity zones. The data in this analysis were initially collected for severe malaria case referral purposes, but also provide invaluable insights into severe malaria cases at the population level.