The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis

Introduction Viral outbreaks present a particular challenge in countries in Africa where there is already a high incidence of other infectious diseases, including malaria which can alter immune responses to secondary infection. Ebola virus disease (EVD) is one such problem; understanding how Plasmodium spp. and Ebolavirus (EBOV) interact is important for future outbreaks. Methods We conducted a systematic review in PubMed and Web of Science to find peer-reviewed papers with primary data literature to determine 1) prevalence of EBOV/Plasmodium spp. coinfection, 2) effect of EBOV/Plasmodium spp. coinfection on EVD pathology and the immune response, 3) impact of EBOV/Plasmodium spp. coinfection on the outcome of EVD-related mortality. Random effects meta-analyses were conducted with the R package meta to produce overall proportion and effect estimates as well as measure between-study heterogeneity. Results From 322 peer-reviewed papers, 17 were included in the qualitative review and nine were included in a meta-analysis. Prevalence of coinfection was between 19% and 72%. One study reported significantly lower coagulatory response biomarkers in coinfected cases but no difference in inflammatory markers. Case fatality rates were similar between EBOV(+)/Pl(+) and EBOV(+)/Pl(-) cases (62.8%, 95% CI 49.3–74.6 and 56.7%, 95% CI 53.2–60.1, respectively), and there was no significant difference in risk of mortality (RR 1.09, 95% CI 0.90–1.31) although heterogeneity between studies was high. One in vivo mouse model laboratory study found no difference in mortality by infection status, but another found prior acute Plasmodium yoeli infection was protective against morbidity and mortality via the IFN-γ signalling pathway. Conclusion The literature was inconclusive; studies varied widely and there was little attempt to adjust for confounding variables. Laboratory studies may present the best option to answer how pathogens interact within the body but improvement in data collection and analysis and in diagnostic methods would aid patient studies in the future.


Introduction
Viral outbreaks present a particular challenge in countries in Africa where there is already a high incidence of other infectious diseases, including malaria which can alter immune responses to secondary infection. Ebola virus disease (EVD) is one such problem; understanding how Plasmodium spp. and Ebolavirus (EBOV) interact is important for future outbreaks.

Methods
We conducted a systematic review in PubMed and Web of Science to find peer-reviewed papers with primary data literature to determine 1) prevalence of EBOV/Plasmodium spp. coinfection, 2) effect of EBOV/Plasmodium spp. coinfection on EVD pathology and the immune response, 3) impact of EBOV/Plasmodium spp. coinfection on the outcome of EVD-related mortality. Random effects meta-analyses were conducted with the R package meta to produce overall proportion and effect estimates as well as measure between-study heterogeneity.

Results
From 322 peer-reviewed papers, 17 were included in the qualitative review and nine were included in a meta-analysis. Prevalence of coinfection was between 19% and 72%. One study reported significantly lower coagulatory response biomarkers in coinfected cases but no difference in inflammatory markers. Case fatality rates were similar between EBOV(+)/Pl (+) and EBOV(+)/Pl(-) cases (62.8%, 95% CI 49.3-74.6 and 56.7%, 95% CI 53.2-60.1, respectively), and there was no significant difference in risk of mortality (RR 1.09, 95% CI 0.90-1.31) although heterogeneity between studies was high. One in vivo mouse model laboratory study found no difference in mortality by infection status, but another found prior acute Plasmodium yoeli infection was protective against morbidity and mortality via the IFNγ signalling pathway. Despite the increase in data available since the West African EVD outbreak, to our knowledge, no one has attempted to synthesize available information on EBOV and Plasmodium coinfection and the overall effect on disease outcomes. There has been increasing threat of Ebola viral outbreaks in recent years yet still little is known about how co-morbidities such as malaria infection affect the Ebola immune response and disease outcomes. This, along with the advent of the Ebola vaccine, rVSV-ZEBOV-GP, means it is important to know how these infections may interact to optimise control and case management in the future [21]. With this in mind we set out to review the literature to answer: How does Plasmodium co-infection affect the immune response, clinical profile and/or clinical outcome to Ebola virus infection? We present a systematic review and meta-analysis of the literature to determine: 1) the prevalence of Plasmodium coinfection among cases of EVD in Africa, 2) the effect of EBOV/Plasmodium coinfection on EVD pathology and the immune response to EBOV infection, and 3) the impact of EBOV/Plasmodium coinfection on the outcome of EVD-related mortality. Finally, we discuss what gaps, if any, remain in the literature and what the implications are for case management and research in future EVD and other viral outbreaks.

Search strategy
The literature search was conducted with a final update on 15 th October 2020 by the lead study author in PubMed and Web of Science to look for published peer-reviewed papers with primary data on i) prevalence or seroprevalence of Plasmodium infection among individuals with current EBOV infection (cases of EVD) or with evidence of past EBOV exposure (indicated by seroprevalence), and/or ii) quantitative or qualitative data on the impact of EVD-related pathology and/or immune response, and/or mortality. To get a complete overview of the evidence we included for review epidemiological studies of infection in humans, clinical case reports and experimental laboratory studies using ex vivo and in vivo infection models. The search was conducted with the following search terms: (Ebola[Title/Abstract]) AND (malaria � [Title/Abstract] OR Plasmodium[Title/Abstract]). While this search strategy was very general, the number of returned articles was deemed feasible to screen without further narrowing of the search terms, thus ensuring all relevant papers could be identified. An initial screen of titles and abstracts excluded irrelevant papers based on being either: review papers, comments/editorials/correspondence, unrelated to EBOV or Plasmodium infection, research on vaccines or drug modes of action, not in the English language. Remaining papers were reviewed in their entirety and papers were included if they met the following criteria: 1) primary peer-reviewed research, 2) either i) epidemiological studies or genomic sequencing studies with explicit measurement of Plasmodium and EBOV co-infection prevalence and/or any measure of impact on EVD pathology/immune response/mortality, or ii) clinical case studies with disease pathology/ immune response measures of co-infected patients, or iii) laboratory studies with ex vivo and/ or in vivo models of EBOV and Plasmodium coinfection and a measure of effect on disease pathology and/or immune system modulation and/or mortality. Study of any Plasmodium or EBOV species and within any country and any date were included. In a subsequent meta-analysis, only epidemiological studies with prevalence of coinfection and CFR by malaria infection status and/or effect estimate data on the outcome of EVD-related mortality from countries in Africa were included. We did not include grey literature.
analysis, data were extracted on the following variables: 1) study first author and publication details, 2) study design; 3) location and study period; 4) details of study population, including population size, mean age and sex breakdown; 5) source of data; 6) EVD-related data including method of diagnosis, EBOV species, number of EBOV positive (EBOV(+)) cases; 7) malariarelated data including method of diagnosis, Plasmodium species, number of sample that received a malaria diagnostic test, number of positive tests; 8) Mortality data including crude EVD case fatality rate (CFR), CFR of coinfected patients; 9) Effect estimates (e.g. risk ratios (RR), odds ratios (OR) and hazard ratios (HR)) for the effect of Plasmodium coinfection on EVD-related mortality; 10) other variables with a significant impact on mortality (potential confounders). Extracted data were independently verified (i.e. double verification) by another member of the study team. Where necessary, study authors were contacted to verify study details. The authors conducted this meta-analysis in concordance with PRISMA standards of quality for reporting meta-analyses and the guidelines for Meta-Analyses and Systematic Reviews of Observational Studies [22,23].

Statistical analysis
We used a random-effects meta-analysis of proportions approach to quantify i) the prevalence of Plasmodium coinfection among EVD cases (EBOV(+)/Pl.(+)), ii) crude CFR of EVD cases, iii) CFR of EBOV(+)/Pl.(+) cases and iv) CFR of EBOV(+)/Pl.(-) cases. This was conducted using the metaprop function from the meta package in R [24,25]. Metaprop allows specific analysis of binomial data and the computation of exact binomial and score test-based confidence intervals [26]. Metaprop was run with a random-effects model to account for heterogeneity between studies in terms of study population, differences in exposure and infection and diagnostic procedures. Random effects models assume the observed studies represent a distribution of possible effects and can incorporate both within-study variance and between-study heterogeneity. To conduct a meta-analysis with individual study weights, we used the inverse variance method of pooling with logit transformation and Clopper-Pearson 95% confidence intervals (CI). The prediction interval was calculated to give a range within which measurements of the same effect size would be expected to fall in future studies. Sub-group analyses were conducted where necessary and as described in the results (i.e. including Zaire EBOV (ZEBOV) species infections only).
For effect estimates related to the impact of Plasmodium infection on EVD mortality, where available, numbers of exposed (EBOV(+)/Pl.(+)) and unexposed (EBOV(+)/Pl.(-)) and with and without the outcome of death were used to calculate unadjusted RR and used in a random effects meta-analysis of RR using the metabin function within the meta package which allows meta-analysis of studies with comparison of two groups with binary outcomes [24]. Finally, any adjusted RRs reported in included studies were included in a meta-analysis using the metagen function which performs fixed and random effects meta-analysis based on estimates and their standard errors using inverse variance weighting for pooling [27].
Heterogeneity between studies was quantified for each proportion and effect estimate measure using the I 2, Chi 2 and Tau-squared (τ 2 ) statistics. I 2 is a measure of the percentage of variability in the effect sizes that is not caused by sampling error, where 25% is considered low heterogeneity, 50% moderate heterogeneity and 75% substantial heterogeneity [28,29]. Although I 2 is not sensitive to the number of studies, it is sensitive to the precision of the included studies. Tau 2 measures the between-study variance within a random effects model and is insensitive to the number and precision of included studies. The square-root of Tau 2 is the estimated standard deviation of the underlying effects across all the studies [30]. In all models between-study variance was estimated using the Hartung-Knapp-Sidik-Jonkman (HKSJ) method, which is considered to be superior to the standard DerSimonion-Laird method in estimating variance of the pooled effect, particularly when the number of studies is small [31].

Assessment of study quality and bias
Due to the small number of studies ultimately included in the effect size meta-analysis, publication bias via assessment of funnel plot asymmetry could not be assessed since the power to distinguish true asymmetry from chance is limited when the number of studies is <10 and is, therefore, not recommended [32]. However, studies included in the meta-analysis were assessed for quality using the Newcastle-Ottawa Scale, developed to assess the quality of nonrandomised studies incorporated into meta-analyses [33]. The scale uses a 'star system' to judge publications on three key areas: i) selection of study cohort(s), including representativeness of the exposed cohort and measure of, ii) comparability of cohorts based on whether important confounding factors were adjusted for in analyses, and iii) ascertainment of the outcome (for cohort studies) of interest including how it was assessed and whether rates of follow-up were sufficient to avoid potential study bias. In line with previous meta-analyses, included studies were assessed to be good, fair or poor based on the number of stars awarded within each category: "good" meant 3 or 4 stars in selection, 1 or 2 stars in comparability, and 2 or 3 stars in outcomes; fair meant 2 stars in selection, 1 or 2 stars in comparability, and 2 or 3 stars in outcomes; and "poor" meant 0 or 1 star(s) in selection, or 0 stars in comparability, or 0 or 1 star(s) in outcomes. Thus, studies had to achieve at least one star in each section to be deemed good or fair.

Search results
Following removal of duplicates, the initial search returned 322 results which had titles and abstracts screened for eligibility (Fig 1). Following the initial screen, 65 results were reviewed in their entirety and a final 17 papers were included for qualitative review as they had a measure of EBOV and Plasmodium coinfection prevalence and/or impact on disease outcome according to our inclusion criteria (Table 1). An additional paper was retained since it presented additional demographic information on the same cohort of patients as another which presented coinfection outcome data [34,35]. We noted potential overlap of the study population between some studies which were confirmed with study authors. Since the studies had different sample sizes, methods and results, all were retained for review and the overlap of populations clearly marked in Table 1. Overlapping studies were excluded in the quantitative meta-analysis so that the study with the largest study population was included in each instance.
The final set of papers qualitatively reviewed thus included 14 observational studies (13 cohort studies and one cross-sectional study), two laboratory studies and one clinical case report. A final set of nine cohort studies were included in the quantitative meta-analysis based on having a unique cohort of patients with similar patient characteristics. Studies were excluded if drawn from the same sampled cohort as another more up-to-date or larger study (n = 3), or if they sampled only children (n = 1).
Sample size of the study cohorts varied from 52 to 2741 and EVD cases from 52 to 1512. Twelve of the studies included all age groups in their cohort, while one included only children aged <18 years. Percentage of males among EVD(+) cohorts was mostly between 44 and 52% (n = 5), although ranged from 13.5% to 63%.
EVD was diagnosed by real-time PCR (RT-PCR) in eleven cohort studies including seven studies [34,36,38,40,42,46,47] which used quantitative RT-PCR (qRT-PCR) through which cycle threshold (C t ) was analysed as a proxy for viral load. Cycle threshold is defined as the cycle number of the assay at which the fluorescence exceeds a certain threshold, with lower Ct values indicative of higher viral load and vice versa [48]. The remaining two studies utilised a metagenomics (mNGS) approach to identify EBOV RNA in blood and plasma samples from EVD patients. One of these used a combination of RT-PCR test data from both DRC and USbased laboratories and mNGS results to determine whether patients were confirmed (at least two positive tests) or probable (one positive test) EVD cases. Although rarely explicitly stated, from the outbreak date and location, most of the observational studies were inferred to be

PLOS ONE
The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis

PLOS ONE
The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis  investigating Zaire ebolavirus (ZEBOV, n = 12); only one investigated Bundibugyo ebolavirus (BDBV). Malaria diagnosis was conducted by rapid diagnostic test (RDT) in most of these studies except two in which malaria diagnosis was made by qRT-PCR, with Ct � 30 indicative of a positive Plasmodium infection considered to provide equivalent sensitivity to microscopy [37,38]. These two studies did, however, have overlap of the study population and only one explored the effect of parasite load on mortality-related outcomes using Ct level as a proxy [38]. Another two studies identified RNA reads mapping to Plasmodium to determine malaria infection prevalence.
Plasmodium falciparum (Pf) was specified as the causal agent of malaria in one cohort study [47] and one other estimated 95% of malaria infections in their sample were Pf although subsequent analyses included all species [37]. The mNGS studies also identified Pf as the causal agent in their samples. The rest did not specify Plasmodium species in their analyses but all used RDTs that detect all four major Plasmodium species (Pf, Plasmodium vivax (Pv), Plasmodium malariae (Pm) and Plasmodium ovale (Po)), except one which used a HRP2 RDT detecting Pf only [34].
Only one study [39] attempted to look at signalling/response pathways between EBOV (+)/Pf(+) and EBOV(+)/Pf(-) patients in an attempt to discern any immune response modulation, and one other compared clinical symptoms between the two infection groups [43].
Cross-sectional studies. One cross-sectional national survey was identified in Gabon which covered a general rural population of 4,272 and measured past exposure to ZEBOV infection using seroprevalence of ZEBOV-specific IgG antibodies, and current Plasmodium spp. infection using in-field blood smear and Plasmodium mitochondrial DNA sequencing. DNA sequencing identified Pf, Pm and Po infections within the sample but main analyses included all-species Plasmodium but was also reported for Pf and Pm infections only.
Clinical case reports. One clinical case report was identified describing the pathology of EVD and malaria coinfection during pregnancy. The clinical case report included one pregnant woman from Gulu, Uganda admitted with Sudan virus in 2000, and a second from Isiro, DRC, infected with BDBV in 2012. EVD was diagnosed using species-specific RT-PCR in each location as well as ELISA in Gulu. Histological methods were employed to analyse EVD and malaria pathology in placental and foetal tissues following delivery.
Laboratory studies. Both laboratory studies used an in vivo murine model of coinfection to analyse the impact of prior Plasmodium infection on outcomes related to EVD. The model in one study was of CD1 mice infected via intraperitoneal inoculation with Plasmodium yoeli (Py) at a dose of 10 4 parasitized erythrocytes, followed by challenge with mouse-attenuated EBOV (MA-EBOV) at a 100 median lethal dose at different time points post Py infection. In each time-group, 6 mice were followed for survival to 28 days post infection (dpi), and four were euthanised on day 4-dpi to analyse blood and liver RNA concentrations for Py and MA-EBOV [49].
The second laboratory study used wild-type, interferon (IFN-αβ) receptor knock-out mice (Ifnar -I-), and IFN-γ receptor null mice (Ifngr -I-) inoculated intraperitoneally with a higher dose of 10 6 parasitized red blood cells of Py and challenged with MA-EBOV or recombinant vesicular stomatitis virus (rVSV)/EBOV glycoprotein (GP) at three different concentrations: 1-, 10-and 100-iu on the sixth day following Py infection. In each group, seven mice were followed for survival up to 18 days and three were euthanised on day 3 to determine viremia and viral load. Knock-out mice were used to study the signalling pathways implicated in initial lab findings. The study also analysed Ifnar -Imice with P. chabaudi (Pcc) infection in the first round of experiments but found no difference in outcomes compared to Py so was discontinued. To test the durability of in vivo effects, parallel experiments challenged Py-infected mice with rVSV/EBOV GP at weeks 1, 3, 5, and 7 post-Py inoculation. In addition, an ex vivo model of infection was analysed using naïve human and mouse macrophages challenged with sera from parasitaemic rodents and from rhesus macaques infected with Plasmodium cynomolgi [50].

Review of key findings
Prevalence and seroprevalence of EBOV/Plasmodium spp. coinfection. Thirteen cohort studies reported data on prevalence of coinfection. Prevalence of Plasmodium coinfection among EVD patients was found to be 33.3% among patients with BDBV in DRC though this was only based on nine malaria RDTs. Among patients infected with ZEBOV, prevalence of Plasmodium coinfection in EVD(+) cohorts ranged from 18.9% to 72.2% for studies utilising malaria RDTs, and 15.4% and 19.4% for the two studies utilising qRT-PCR. In addition to these cohorts, RNA sequencing data from EBOV(+) cases in Guinea, found that 84% (156/ 186) contained reads mapping to Pf, higher than malaria prevalence found by RDT on the same sample of patients (33%, 40/121). However, read depth, a proxy for parasitaemia density, varied widely from 59 to 358820 FPKM (a measure for read depth) and the discrepancy between diagnostic methods reduced as number of reads mapping to Plasmodium increased. RNA sequencing in Boende, DRC, found 24.3% of EBOV(+) cases were coinfected with Pf. One study which focused only on children aged <18years found prevalence in this group to be 40%. Prevalence of Plasmodium infection was reported as higher in children aged and less common in EBOV(+) compared to EBOV(-) individuals [34,46].
The cross-sectional survey found evidence of an epidemiological overlap of the two infections with 10.2% (n = 425) of adults in rural Guinea positive for both current Plasmodium infection and ZEBOV-specific IgG antibodies, a significant overabundance than would be expected by chance (χ 2 = 59.4, df = 1, p<0.0001).
Eight observational cohorts were included in the meta-analysis. The prevalence of Plasmodium coinfection among EVD patients varied from 18.9% to 44.4% between studies, with the overall meta-analysis estimate calculated at 21.7%, 95% CI 18.7-25.1% (Fig 2). Between-study heterogeneity was low-to-moderate with I 2 = 36%, 95% CI 0-72%; chi 2 = 11.01, p<0.14; and Tau 2 = 0.05. The prediction interval was 13.3-33.4%. Sub-group analysis including only studies with ZEBOV as the causal agent had a similar overall estimate and similar heterogeneity. Sub-group analysis by malaria diagnostic showed a similar overall estimate and high heterogeneity from studies utilising only RDTs compared to qRT-PCR or NGS an RT-PCR, though these latter methods only included one study each (S1 and S2 Figs).
Effect on disease pathology and the immune response. Human subjects. Description of the impact of Plasmodium coinfection on EVD pathology and immune response was given in the clinical case study and in both metagenomic studies. The clinical case report presented evidence of a potential pathophysiological interaction of the two infectious agents in the placenta [51]. Although she recovered from infection, one pregnant woman infected with Sudan virus in Uganda had a spontaneous still birth during her illness. Subsequent histological and immunohistochemical analysis showed colocalization of malarial parasite pigment and EBOV antigen within the placenta. Malarial parasite pigment was observed in fibrin and within macrophages embedded within the fibrin. EBOV antigen was similarly primarily localised to areas of fibrin deposition and maternal mononuclear cells, including the malarial parasite-pigment-laden macrophages. This interaction was not observed in a second pregnant woman infected with BDBV in DRC despite testing positive for malaria by RDT. She was, however, kept on a course of artemether-lumefantrine following the positive RDT result. Information on malaria diagnosis and treatment was not reported from the clinical case in Uganda.
Animal models. One experimental lab study measured immune response biomarkers in infected mice and found that acute Plasmodium infection protected against a low but lethal dose (1-iu) of virus (both MA-EBOV and rVSV/EBOV) by eliciting the proinflammatory interferon gamma (IFN-γ) pathway, rendering cells resistant to EBOV infection [50]. Morbidity and mortality were reduced, as well as viremia and viral load in the liver, spleen and kidneys of coinfected mice. The same protection was not conferred by administration of a single large bolus of recombinant IFN-γ, suggesting a persistent, subpatent Plasmodium infection and/or a sustained level of IFN-γ confers protection against EBOV viral replication (effect on mortality discussed below).
Each CFR was similar when looking only at ZEBOV species alone (Fig 3) and when looking at studies that used malaria RDTs only (S1 and S2 Figs). Although heterogeneity between study estimates of CFR among coinfected cases was reduced when analysing only studies that used malaria RDTs (S1 and S2 Figs).
Human subjects-effect estimates. Seven studies found no significant difference in CFR by Plasmodium infection status in unadjusted analyses [36,39,40,42,43,45,47]. Within these, Kratz  reported that mortality appeared to be increased by Plasmodium coinfection using RDT data, but this was not found using prevalence as determined by genomic sequencing data. They did find that higher burden of Plasmodium infection appeared to have higher mortality, but no absolute correlation was observed.
Two studies found no effect of coinfected in adjusted analyses. Hartley et al. found coinfection was associated with significantly higher mortality when adjusted only for age and sex (53.6% v 74.3%, OR = 3.9, p = 0.03), however, these individuals also had higher viral loads and once this was taken into account the detrimental effect was abrogated. Gignoux et al. found no effect after adjusting for multiple confounders including antimalarial prescription, age group, sex, Ct value at admission, days from symptom onset to admission, receipt of IV fluids, and number of inpatients at the Ebola treatment centre on day of admission (aRR = 1.23, 95% CI 0.52-2.86). However, this study did find that Plasmodium coinfection inhibited the apparent beneficial effect of ASAQ treatment; ASAQ treatment appeared to reduce mortality in EBOV (+)/Pl(-) cases compared to those treated with A-L whereas no beneficial effect was seen in EBOV(+)/Pl(+) cases.
Two studies reported increased mortality in either the general study population (Waxman et al.) or in association with age (Kerber et al.) In the former, although no significant effect was found from a smaller study drawn from the same source population (Garbern et al.), with a larger sample Waxman et al., found a significant detrimental effect of Plasmodium coinfection on mortality with an elevated CFR (52% v 66%) and 28-day mortality (8.2 per 1,00 persondays). In a focused subgroup analysis, adjusted for age only, there was greater mortality among EBOV(+)/Pl(+) compared to EBOV(+)/Pl.(-) cases (aHR 1.69, 95% CI 1.14-2.52, p = 0.009). Compared to EBOV(-)/Pl(-) patients, the hazard of dying was highest in EBOV(+)/Pl.(+) cases (aHR 9.4, 95% CI 6.2-14.2, p < .0001) followed by EBOV(+)/Pl.(-) cases (aHR 6.0, 95% CI 4.4-8.0, p < .0001). Kerber et al, meanwhile, found that Plasmodium coinfection only had an impact on mortality among 5-15-year-olds among whom it increased CFR by >20% and increased odds of death to 4.2 times greater than individuals of the same age group without Plasmodium co-infection. No other age groups appeared to be affected, suggesting an interaction between age and infection status.
In contrast, just one study found coinfected patients had significantly lower CFR than singly infected cases (p = 0.007) [38]. Plasmodium infection was associated with an increased survival when adjusted for age, sex and EBOV viral load (aRR = 1.2, 95% CI 1.1-1.4, p = 0.004).

PLOS ONE
The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis However, the effect was more pronounced when stratified by plasmodium parasitaemia level, using Ct as a proxy. High Plasmodium burden was then associated with increased risk of survival to 1.4, 95% CI 1.2-1.7, p<0.0001) whereas low parasite burden infection did not have a significant effect (RR 1.2, 95% CI 0.98-1.4, p = 0.08) P = 0.007). These effect estimates did not include individuals aged <5 years due to low numbers of cases in that age group.
In addition to Plasmodium infection, studies also reported other significant variables affecting CFR as including age, viral load, aspects of clinical treatment (e.g. referral time, length of stay) and certain clinical features (e.g. bleeding and other symptoms).
Only three studies reported final adjusted effect estimates for the whole study population, though with varying confounders adjusted for. These three estimates pooled to a combined

PLOS ONE
The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis aRR estimate of 1.14, 95% CI 0.41-3.17 with high heterogeneity (I 2 = 82%, 95% CI 45-94%, Chi 2 = 11.22, p<0.01)) and an incomprehensibly wide prediction interval (Fig 6A). Since only two of these adjusted for EBOV viral load, which was considered to be a significant factor in predicting outcome in seven studies, a meta-analysis was conducted solely on these two studies, estimating a RR of 0.86, 95% CI 0.12-6.30 (Fig 6B).
Animal models. In addition to these observational studies in the field, the two laboratory studies also report the effect of coinfection on mortality in murine models. Rosenke et al. 2018 found no effect of coinfection on mortality in mice when challenged with a high dose of MA-EBOV at 100-median lethal dose. In contrast, Rogers et al. found that acute Plasmodium infection protected from lethal virus challenge at low (1-iu) MA_EBOV challenge but less so with moderate to high (10-iu and 100-iu) doses. This protective effect waned over time, with Py-infected mice protected from EBOV challenge three weeks after Py infection, but not by 5 weeks. This was IFN-γ dependent as discussed above.
Quality of studies included in meta-analysis. The eight studies included in the metaanalysis were analysed for quality using the Newcastle-Ottawa Scale ( Table 2). Four studies (Kratz, Waxman, Vernet and Li) were deemed to be 'Poor' because, while they scored well under the selection and outcome categories, they did not score under comparability, i.e. they did not attempt to adjust analyses to take into account important confounding factors. Although the other four were deemed 'Good', all studies lost points on the representativeness of the exposed cohort since all only had diagnostic data from hospital-based cohorts, and four studies had follow-up rates that were deemed high enough to introduce potential bias into their final estimates (Kratz, Gignoux, Rosenke and Vernet).

Discussion
We have attempted to consolidate findings on the prevalence and effect of Plasmodium coinfection on mortality from EVD which could have significant implications for future outbreaks of EVD and other viral outbreaks with similar pathology. The number of studies identified, particularly those for the meta-analysis, was limited. This could be because EVD outbreaks have historically been small and, indeed, the majority of studies stem from the 2014 West

PLOS ONE
The impact of malaria coinfection on Ebola virus disease outcomes: A systematic review and meta-analysis Africa EVD outbreak which had a huge affected population in comparison and a high level of support from national and international NGOs and research institutions [6,52]. EVD outbreaks also develop quickly and in countries with limited surveillance and weak health systems, limiting resources and planning for research purposes. Although we identified 13 observational studies, there were overlapping study populations meaning only eight unique studies could be included in the meta-analyses. This is largely because research data came from ETUs which were set-up to specifically manage EVD patients during the 2014 outbreak.
Overall, one-quarter of EVD cases were infected with Plasmodium (26.9%, 95% CI 16.8-40.0%), and, aside from one, appeared relatively consistent across individual studies. Plasmodium spp. prevalence was lower in EVD confirmed cases compared to patients without EBOV infection, most likely because presenting symptoms were due to the viral infection and not malaria in the former group. Indeed, Hartley et al. found that malaria infection could be used as one variable (among others) to help distinguish EVD cases from other diagnoses [35]. Despite this, one-fifth of EVD cases is a high proportion and would have important public health implications if coinfection were to alter morbidity or mortality in a significant way. There was a lack of consensus, however, between studies in relation to coinfection and its effect on EVD outcomes.
From the cohort studies reviewed, the majority found no significant effect of Plasmodium infection on mortality, while two found a significant detrimental effect, and one found a significant protective effect. Overall, the meta-analysis estimated that the CFR of coinfected patients was not significantly different from the CFR of EBOV(+)/Pl(-) patients, and a meta-analysis of effect sizes found no significant effect of coinfection on mortality (RR = 1.09, 95% CI 0.90-1.31). Although tests of heterogeneity have low statistical power with a small number of studies, use of the chi-square p-value and I 2 confidence intervals can provide more confidence in these estimates and in this case suggest high heterogeneity between studies for both the overall effect estimate and the CFR of coinfected patients (EBOV(+)/Pl(+)) [53]. Conversely, between-study heterogeneity appeared to be lower between estimates of the CFRs of individuals infected with EBOV alone (EBOV(+)/Pl(-)). The observed heterogeneity in the estimates In terms of study protocols, only two studies included in the meta-analysis used a malaria diagnostic that was more sensitive than RDT: qRT-PCR and a combination of RT-PCR and mNGS. One of these [38] (using RT-PCR) was also the only one to find a positive effect of coinfection on surviving EVD. When these studies were removed from the meta-analysis, heterogeneity was reduced and the overall estimate, although it remained insignificant, was biased toward a detrimental effect of Plasmodium on EVD mortality. However, the difference in testing procedure doesn't logically explain why Rosenke et al. would observe a protective effect from Plasmodium since RT-PCR is more sensitive than RDT, particularly in detecting low parasite density infection, and so could be expected to identify more malaria cases and would actually bias the effect estimate toward a detrimental effect (a higher RR if considering mortality as the outcome) [54,55]. Conversely, the studies using RDTs could be biased toward a detrimental effect since they used RDTs based on HPR2 for Pf identification which has been shown to be associated with false positive results [56,57].
Additionally, the RT-PCR and mNGS tests could detect different stages of malaria infection to the RDTs, i.e. detecting sub-patent infection compared to more active/acute infection. These stages of malaria are associated with different immune responses which could affect EVD outcomes differently. Younger age groups are more at risk of acute Plasmodium infection compared to adults who gain some level of immunity to severe infection and are more likely to have a chronic, subpatent infection [58,59]. In the cohort studies presented here, younger people were found to be more likely to be associated with both coinfection and greater EVDrelated CFR yet age was not adjusted for in most analyses. If a study had a younger age profile this could result in an inflated CFR among coinfected cases. Furthermore, Rosenke et al.'s analysis did not include cases age <5 years, while other evidence pointed to a potential interaction between age and Plasmodium infection status; Kerber et al. found that coinfection among 5-14-year olds increased their risk of death but the same was not seen in other age groups [46]. This could be suggestive of differences in the intensities of infection and/or immune responses between age groups in either EBOV or Plasmodium infection.
Acute Plasmodium infection is associated with activation of pro-inflammatory pathways and has been implicated in influencing response to other concurrent infectious agents such as respiratory viruses, enteric bacteria and HIV [17][18][19][20]. HIV, like EBOV, is an RNA virus, however, it is a chronic infection as opposed to the acute infection seen in EVD which could lead to alternative interactions. However, prior Plasmodium infection was found to reduce viremia and associated pathologies of chikungunya virus, a more acute RNA virus, via proinflammatory responses through IFN-γ production [60]. Despite this, only one study in our meta-analysis attempted to discern the effect of more acute Plasmodium infection. Rosenke et al. looked at parasite burden (using Ct as a proxy) and found that higher parasite load (Ct � 20), which would be expected from more acute infection, was associated with improved survival (reduced mortality) compared to those with a lower parasite burden for which protection was not significant. A similar protective effect from acute infection was seen in the laboratory study by Rogers et al which found acute Plasmodium infection protected mice and human macrophages from EBOV challenge via a low but sustained IFN-γ response which protected from EBOV challenge up to three weeks post Plasmodium-infection but dwindled thereafter. However, in the only study to monitor similar human biomarkers, Carroll et al found that markers of the innate immune response to acute Plasmodium infection, including IFNγ, were not different between Pf(-) and Pf(+) EVD cases whereas biomarkers associated with the coagulation pathway were reduced in patients Pf(+). This analysis thus suggests the inflammatory pathway found to be important in mice is perhaps not important in human cases, although, the analysis by Carroll et al was based on a very small sample size of individuals (10 and 13 per group).
The protective effect observed from the laboratory study by Rogers et al. was not observed in the second laboratory study identified, however, this second study used only a very high lethal dose of MA-EBOV (100-iu) [49]. Use of similarly high doses by Rogers et al. (10-iu and 100-iu) also showed limited effect on morbidity and outcomes compared to the low EBOV dose (1-iu) which was significantly protective, thus suggesting high viral load overrides any potential protection from coinfection [50]. In human studies, viral load was only quantified in seven studies and, where analysed, all found that higher viral load led to greater mortality. Despite this, only three studies reported analyses that adjusted for viral load. Among these, Hartley et al. found that coinfected cases had a higher viral load than EBOV-only infected cases and when accounted for in their analyses viral load abrogated the apparent detrimental effect of Plasmodium infection on mortality [34]. The two other studies that adjusted for viral load included the positive effect estimate from Rosenke et al. and an insignificant finding from Gignoux et al. [36,38]. Not accounting for this confounding variable is a clear limitation in the other effect estimates reported.
Due to the failure to adjust for important confounders, four of the eight studies included in the meta-analysis were deemed poor quality based on the Newcastle-Ottawa Scale analysis. As well as not adjusting for viral or parasite load, many of the studies failed to account for other potential confounders including age, sex, and time from symptom onset to admission. The majority of studies did not specify Plasmodium species and it could be expected that different species would be associated with a different proportion of chronic versus acute infections and differences in immune response. However, it is unlikely to explain differences in this context since the vast majority of malaria infection is caused by Pf in each of these countries [61]. Similarly, it is conceivable that different EBOV species could have different immune responses that may interact differently. However, all but one observational study examined ZEBOV and thus does not explain differences observed.
Outside of the cohort and laboratory studies, the case report of a pregnant woman in Uganda implicated other potential interactions between the two infectious agents in the placenta of pregnant women [51]. It is unclear whether the co-localization of these two pathogenic agents had an impact on the outcome of still birth experienced in this case. As a one-off observation, this does need to be explored in other future cases to see if there is a legitimate interaction of concern.
So, what does all this mean and what are the implications for future outbreaks of EVD and other viral diseases? From the literature as it stands, it remains unclear whether Plasmodium coinfection has an effect on EVD morbidity and mortality. Although pooled meta-analyses suggest no significant effect, there is a high amount of variability between study effect estimates that have not all had appropriate confounding variables considered, and/or had follow-up rates that may have biased results. With the advent of the new Ebola vaccine (rVSV-ZE-BOV-GP, manufactured by Merck) which has shown high efficacy in trials, the clinical interaction of comorbidities with EBOV may have less significance in future EVD outbreaks [21]. However, it will take time for a vaccine to be delivered widely and is not yet fully assessed in all populations. Furthermore, the learnings described here in terms of research conducted and gaps remaining can be applied to improve our knowledge of malaria coinfections on other viral outbreaks in the future. Since other viral diseases with similar immune response profiles could be affected similarly by Plasmodium infection, improvements in study design and data collection could be made to ensure that all essential clinical information is collected during subsequent outbreaks. In this case, features of clinical presentation, details of clinical acute care, data on viral loads, parasite densities and immune system biomarkers in particular are important and lacking.
Planning for research purposes is difficult in an outbreak setting where the outbreak is fastmoving and resources are limited but being better prepared with a planned set of guidance on what data should be collected prior to the next outbreak would enable faster implementation of data collection for subsequent research. Following analyses, all studies should be encouraged to publish all results related to these important factors, including raw numbers, unadjusted and adjusted estimates, to better enable meta-analytics.
Even despite these improvements, due to the complex nature of the immune system, it may be that experimental lab methods will give the best insights. Findings by Rogers et al are interesting, but more comparison needs to be done to look into other disease models, as well as how well these animal models transfer to the human context. Mouse models of malaria and EVD do not mimic all facets of the diseases in humans, but findings from these models can be used to narrow down the field of investigation in the clinical setting-e.g. explicit measurement of immune system biomarkers such as IFN-γ, stratified by Plasmodium parasite load and viral load, should now be recommended in future clinical research.
Remaining limitations on implementing better-planned research may be sample size restrictions if future outbreaks are not large enough, and diagnostic and clinical capacities to conduct the research. In this case, retrospective genomics analysis on stored blood samples may offer a useful tool since this is increasingly affordable and samples can be stored and analysed in retrospect or in partner institutions with greater research capacity. RNA/DNA read data can provide quantifiable information on prevalence of infection, viral and parasite loads, and biomarkers in one protocol, and thus perhaps be able to answer some of gaps in the data presented here.