Interpretation and implications
The analysis presented here indicates that a significant proportion of confirmed P. falciparum malaria cases reported by the HMIS in malaria-endemic countries are likely to be non-malarial fevers coincident with an asymptomatic P. falciparum infection. Across all surveyed country-years, the majority of febrile children for whom treatment was sought did not have a malaria-positive fever, and amongst the children who did have a malaria-positive fever, the malaria infection was not the underlying cause for nearly two-thirds of these fevers. It is important to note that in this analysis, a symptomatic malaria infection is defined solely by febrile illness, and as such, other symptoms of chronic malaria infections (such as anaemia) are not considered.
In this analysis, prevalence of malaria-attributable fever amongst malaria-positive children who sought treatment during the 2 weeks preceding each survey was used to generate an incidence of malaria-attributable and malaria-positive fever at clinics, applicable only to the 2 weeks preceding each survey. Although conventional burden estimates quantify the number of malaria cases over the course of a year, these findings signify that the adjusted case-count approach for burden estimation may significantly overestimate the clinical malaria burden by classing all fevers with a patent malaria infection as a “malaria case”. The reasons for differential proportions of MAF and NMFI amongst malaria-positive fevers between countries are complex and depend on both the national prevalence of both
P. falciparum and NMFI. Many of the surveys from countries with the greatest predicted overestimates for cases using the adjusted case-count approach were completed in the first half of the study decade, coinciding with higher malaria burden (and therefore higher levels of exposure-related immunity [
6]) and worse sanitation and housing conditions relative to later years [
17]. These countries also typically experience higher rates of non-malarial febrile illness. The comparison of MAF and NMFI fractions of clinic-based confirmed cases to contemporaneous malaria cases reported to three Ministries of Health in sub-Saharan Africa (Fig.
5) shows the potential for malaria-positive fevers that are not causally attributable to malaria to be systematically reported as “malaria cases”, which may lead to significant overestimations of the clinical burden of malaria.
Methodological discussion and limitations
Many of the 21 countries included in this study were surveyed multiple times between 2006 and 2016, with three surveys in Angola, Liberia, Madagascar, Tanzania and Uganda, and four surveys in Senegal. In most of these countries, a larger fraction of malaria-positive fevers were causally attributable to malaria in later surveys compared to earlier surveys within the same country (with the exception of 2016 surveys in Tanzania and Uganda). These findings may be explained by a loss of exposure-related immunity amongst children under 5 years of age as malaria prevalence decreases (all six countries showed decline in
PfPR
0–5 over the study period) [
6,
18] resulting in a higher fraction of symptomatic illness amongst infected children (although this pattern was not observed in some other countries where
PfPR
0–5 decreased over the study period). This increase in MAF in the six countries could also be explained by a decrease in NMFI prevalence relative to malaria prevalence over the time period, but this was not observed in the current study [
13]; indeed in most countries, a higher fraction of NMFI compared to MAF was observed amongst febrile children presenting at public health clinics in more recent country-years. A possible limitation of this repetitive surveying through time and space is that the results may not be easily applied to other countries with similar levels of malaria transmission intensity; indeed, this study shows that transmission intensity is not the only factor which affects the proportion of malaria-positive fevers that are attributable to malaria, and that non-malarial fever prevalence and household characteristics also play a role. The model included a survey-level random effect to account for biases introduced by the blocking of surveys through time and space.
Averaging over the 41 surveyed country-years, approximately two-thirds of malaria-positive fevers amongst children under 5 years of age would not be resolved with effective anti-malarial treatment. If clinicians do not investigate other causes of fever given a positive RDT, non-malarial fevers may be systematically missed and remain untreated, potentially leading to missed opportunities to reduce the burden of other important non-malarial febrile illnesses. It is, however, also important to consider that many of the underlying causes of these non-malarial fevers may not warrant treatment, for example if the underlying cause is a self-limiting illness. As such, although this analysis suggests that clinicians should be aware that co-infections with
P. falciparum and other illnesses are common, the results presented here do not directly indicate any requirement for upscaling of antimicrobial medications, especially in areas where bacterial infections are less common. Evidence from literature suggests that clinicians show good adherence to integrated management of childhood illness (IMCI) guidelines when prescribing ACT after a positive RDT [
19,
20], but information is sparse on clinicians’ propensity to investigate alternative causes of illness after either negative or positive RDT outcomes. Evidence suggests that in malaria-endemic areas, a shift from malaria-centric test-and-treat methods towards a holistic IMCI approach in clinics would both improve health outcomes and reduce unnecessary antibiotic use [
21]. Studies have shown that clinicians are more likely to prescribe antibiotics to patients with a negative RDT than malaria-positive patients in a study in Zanzibar [
20], and that after IMCI training clinicians do record diagnoses of malaria co-infections with NMFI [
22]. Diagnosis of common NMFIs is hampered by a lack of routine diagnostics that are cost-effective at the clinic level, and further challenged by the diverse and heterogenous fever aetiology across the continent [
23]. This problem may also be exacerbated by health systems that often require a single cause for febrile illness to be reported, which may lead to an over-reporting of malaria in areas of high prevalence, and an artificial boost of NMFIs in areas where malaria prevalence is lower. The applicability of these results to age groups older than 5 years of age is ambiguous, owing to the lack of information on treatment-seeking rates for older ages.
In this analysis, a number of additional factors hypothesized to affect the relationship between MAF and
PfPR were investigated. No effects of recent
PfPR declines or differential underlying treatment rates were observed on the relationship between MAF and
PfPR (Figs.
2,
3). It is unclear whether these effects were not observed because they do not affect the relationship, were due to insensitivity of the modelling framework, or were due to the calibration dataset being ill-adapted for measuring the effect of recent declines in
PfPR. Multiple sets of longitudinal cohort studies at differing levels of baseline
P. falciparum exposure would be better-suited to measure the effect of recent
PfPR declines on the relationship between MAF and
PfPR as they capture the changing immunity profile of individuals over time within the same cohort. Mechanistic models suggest a loss of immunity following up-scaling of malaria interventions, and a shift towards increased levels of severe malaria-attributable fever in older children following declines in
P. falciparum exposure [
24]. Any lag between reduction in
PfPR and this loss of exposure-related immunity would imply that the profile of the relationship between MAF and
PfPR might temporarily shift until exposure-related immunity reaches a new equilibrium.
The effect that high or low levels of community anti-malarial treatment may have on the relationship between MAF and PfPR is unclear. High levels of prompt and effective treatment may reduce the exposure-related immunity in a community, altering the relationship between MAF and PfPR. Here, the treatment-seeking rate for fever was used as a proxy for the level of prompt and effective treatment, and no effect of treatment was found on the relationship between MAF and PfPR. Case–control studies with prompt treatment of fevers may increase the sensitivity of the approach used here.
The household survey datasets used in this analysis report RDT result from the time of interview, but caregiver recall of fever episodes within the past 2 weeks (rather than fever measurement at the time of interview). In addition to biases in recall of recent fever [
25,
26] the possibility exists that an individual with a patent malaria-attributable fever may seek and receive effective anti-malarial treatment, and return to RDT-negative by the time of interview, despite having experienced a malaria infection in the 2 weeks preceding the survey. If left unadjusted, an overestimation of the proportion of NMFI amongst the individuals who sought and received treatment would be expected. The modelling study presented here is the first of its kind to adjust for the possibility of reversion from RDT-positive to RDT-negative within the 2 weeks preceding the survey, as determined using a previously presented model [
27]. Datasets with RDT outcome at the point of care would increase the sensitivity of this approach, although diagnostic testing for all fever cases presenting at clinics is not universal; amongst febrile individuals who sought care at a public clinic in the household survey dataset only 44.3% had blood drawn at the point of care (used in this study as a proxy for parasite-based malaria diagnosis). The possibility exists that, under this approach, an individual may have been treated for a malaria-positive fever over 14 days prior to the interview (and, therefore, not report any fever, or treatment-seeking) yet still present a positive RDT at the time of interview, due to the persistent positivity of RDTs after treatment [
27]. This potential limitation is not addressed in the current study, although the effect is estimated to be small, given the low treatment-seeking rate for malaria-positive fevers, and the lower-still proportion who received an effective anti-malarial [
3].
Fever caused by
Plasmodium vivax malaria is also, by definition, grouped within NMFI in this analysis, confounding the definition of NMFI.
P. vivax is not prevalent in Africa (in part due to widespread Duffy negativity [
28]), except in Mauritania and East Africa [
29,
30]. Due to location of the surveyed country-years represented here (which mostly fall outside the zones of stable
P. vivax transmission [
31]), the effect of
P. vivax-attributable fevers in the country-years represented in this analysis is presumed to be limited. Caution should, however, be taken when extending the approach presented here to areas of higher
P. vivax burden.
The results presented here suggest a high prevalence of co-infections between malaria and NMFI amongst febrile children presenting at clinics. Here, the effects of co-morbidity (or modulating effects of co-infection on febrile illness) are not accounted for, owing to the lack of diagnosis for NMFI. Future studies may aim to investigate the effects of co-morbidity if differential diagnosis of NMFI is available, or to include spatial distributions of known NMFI-causing pathogens, which were precluded from the current analysis given the lack of availability of cartographic NMFI estimates at the same scale as the PfPR model inputs.
The Ministry of Health reports used to generate the number of paediatric malaria-positive fevers that would and would not be resolved by anti-malarial treatment alone in three sub-Saharan African countries (Fig.
5) are subject to biases not adjusted for in this study. Although the reports document the total number of cases reported from health facilities, not all health facilities report to the HMIS. Biases in which facilities report to the HMIS may mean the estimates presented here are not representative on a national scale. If certain facilities (e.g. those in urban areas, typically associated with a lower
P. falciparum burden [
32], or those receiving more severe malaria cases, such as hospitals) are not equally likely to report to the HMIS than rural facilities, the estimates may not be nationally representative. The estimates for the proportion of the annual cases that are estimated to have been malaria-attributable may also be biased by the short data collection period for each survey. Each survey is collected over a period of usually 2–4 months, with fever and care-seeking information for each child only being valid for the 2 weeks preceding each child’s interview. For this reason, the estimates presented here for the proportion of children with MAF seeking care is only valid for the 2 weeks preceding each survey. Because treatment-seeking rates for MAF may vary over the course of a year, the 2-week period of the survey may not necessarily be representative of the treatment-seeking rate across the year. A more accurate comparison would be to calculate a country-specific incidence rate for each of the three countries to compare with the actual reported malaria cases from the three case study countries, but this was not possible with the datasets and modelling methods used in this analysis.