Invasive bacterial co-infection in African children with Plasmodium falciparum malaria: a systematic review

Malaria remains a leading cause of childhood morbidity and mortality worldwide, accounting for 7% of deaths in children younger than five years of age [1]. In the past few years two of the largest clinical trials ever conducted in African children with severe malaria (SM) have concluded, both holding major implications for treatment guidelines [2, 3]. With the introduction of artesunate in sub-Saharan Africa (sSA), baseline mortality in children with SM will at best be between 6% to 8.5% but likely to be substantially higher outside the framework of Good Clinical Practice-run trials. Understanding the key correlates of poor outcome may identify future targets for additional definitive or adjunctive treatments.

Over time there has been a piecemeal accumulation of data indicating that children with Plasmodium falciparum malaria are at risk of invasive bacterial infection (IBI). Blood stream infection, largely secondary to enteric Gram-negative organisms (EGNOs), with a predominance of non-typhoidal salmonella (NTS) species, has been widely reported as a complication of severe malaria. However, it remains uncertain whether malaria infection is a risk factor for invasive bacterial disease since the majority of children in malaria-endemic Africa are frequently infected by P. falciparum throughout childhood and only a minority will develop severe disease. A sub-analysis within a comprehensive systematic review of blood stream infections in Africa indicated that 6.5% of 11,814 malaria infections had concomitant bacteremia [4]. Which children are at greatest risk of developing dual infection and whether this extends across the clinical spectrum (asymptomatic, mild and severe) remains unclear.

This is important for two reasons. Firstly, concomitant IBI in children hospitalized with evidence of recent or intercurrent malaria, with or without severe manifestations, results in a significantly worse outcome. In one epidemiological study an estimated one third of all SM deaths were attributable to bacteremia [5]. Secondly, it has implications when considering revisions of current management guidelines to include routine provision of antibiotics to all children with malaria. Routine antibiotics, along with anti-malarials, are currently recommended for children with SM [6]. However, malaria remains a very common cause of pediatric admission and the indiscriminate use of antibiotics would be both financially costly and could perpetuate the rise of antibiotic resistance. Thus, there remains a need for clarity on whether children with malaria are more susceptible to a bacterial infection and its precise nature in order to tailor antimicrobial management. In this systematic review we bring together the current breadth of published data on the prevalence of IBI among African children with malaria.

We reviewed published observational and epidemiological data in order to establish whether there was evidence of an association between malaria and IBI. We sought to determine the prevalence of co-infection in children with malaria infection, the risk factors for co-infection and the mortality effect in African children, drawing on this knowledge to consider the implications for future treatment guidance.

Studies were eligible for inclusion in the review if they were conducted among children with malaria, who had blood cultures taken and were admitted to hospitals or outpatient clinics in sSA. Although we restricted our review to studies reporting bacteremia, we did not exclude studies describing other bacterial co-infection, such as urinary tract infection or meningitis.

We performed a systematic search for articles describing bacterial infection among children with P. falciparum malaria. Search terms used were ‘(malaria OR plasmodium)’ AND ‘(bacteria OR bacterial OR bacteremia OR bacteraemia OR sepsis OR septicaemia OR septicemia)’ in three major scientific databases, PubMed, Embase and Africa Wide Information (AWI). The Internet search was carried out on 13 and 14 August 2012. Searches of PubMed and Embase were limited to humans while AWI was limited to humans and scholarly (peer reviewed) journals. Abstracts and titles from all years were compiled in Endnote (Thomson Reuters) and screened, after the removal of duplicates. Prospective studies were required to recruit children systematically or consecutively and to evaluate all cases. Our review was restricted to articles either written in English or translated into English. Unpublished data were not included. The protocol and review are not registered.

Eligible publications were retrieved in full text if available online or from academic libraries. Both prospective and retrospective case series of children with malaria undergoing blood culture were subdivided according to principal inclusion criteria (SM, all malaria or non-severe malaria). Studies principally reporting epidemiological associations were tabled separately. The definitions of SM varied widely. In order to facilitate agreement between the reviewers for study inclusion we used the following criteria for assessment of the quality of studies reporting on severe malaria (SM): 1) criteria for SM well-defined; 2) bacterial species defined or quality assurance verified that contaminants were excluded; 3) mortality data for ‘high risk’ SM cohort at least 7.4% (the lowest 95% confidence for artesunate arm of the AQUAMAT multicountry trial [2]; the overall mortality in this arm was 8.5% (7.4% to 9.5%)) [2]. These were ranked as low, moderate and high (1, 2 and 3) according to the number of criteria met. We reviewed the reference lists from articles rated ‘high quality’ to identify additional secondary articles not found during the initial online search.

Both authors examined the manuscripts and excluded studies during abstract review if they did not include a cohort of patients with a confirmed diagnosis of malaria, case reports, review articles or behavioral studies. During full text review, we excluded studies if they were conducted outside of sSA, did not report any microbiological data or featured predominantly adults. Studies reporting a mixed adult child population were only included if the data for children could be readily separated. Studies in which relevant data could not be extracted or where inclusion criteria failed to represent a population of children with malaria, were also excluded. For study sites generating several papers from the same study cohort, we included only the most appropriate study unless time periods varied. Data on neonates were excluded since neonates represent a group largely unaffected by malaria but with an unrelated, distinct vulnerability for IBI. We did not actively consult with specialists in the field to discuss or review.

For each study we worked out case fractions for malaria IBI co-infection as ‘number of children with malaria AND IBI’ (numerator) divided by ‘number of children with malaria’ (denominator). We used the Q test, as a measure of heterogeneity among studies, calculated as the weighted sum of squared differences between individual study effects and the pooled effect across studies. I² was calculated to quantify heterogeneity (between-studies variability) [7]. Since the definitions for SM vary from study to study, we also compared the study inclusion criteria for SM cases with the WHO definition of SM [8].

Our search of the published literature yielded 10,200 articles and 7,897 unique articles following removal of duplicates. After reviewing the abstracts 72 full text reports were identified for closer evaluation, which were reduced to 21 studies after applying the exclusion criteria (Figure 1).

We excluded 15 studies since they described specific populations not primarily defined by malaria status. Studies alluding to an association between malaria and IBI but with incomplete information or unsuitable study design were tabulated separately (Additional file 1: Table S2). Only four studies were excluded due to location outside of sSSA, highlighting the dearth of data from other regions. Three of these studies reported predominantly data from adults [9‐11]. A study of 340 children with severe malaria in Papua New Guinea recorded two cases of malaria with concomitant infection (invasive candida and Klebsiella Pneumoniae); however, not all children had blood cultures taken [12]. Other reasons for exclusion are detailed in Additional file 1: Table S1.

A further four papers were found from scanning reference lists of selected articles, leaving 25 studies from a total of 11 sSA countries. They comprised twenty cohort analyses (including inpatient and outpatients studies), two randomized controlled trials and three epidemiological studies, summarized in Tables 1 and 2 according to study type and clinical severity.

Collated data examining the potential biological relationship between malaria and invasive bacterial infection across a range of studies conducted in sSA is by no means conclusive. For SM the case fraction of children with IBI co-infection was 6.4% (95% CI 5.81 to 6.98%) but with substantial between-study variability, mostly due to differences in study methodology including the case definition of SM. In prospective studies of febrile children admitted to health facilities with all-severity malaria, the case fraction with IBI was lower in children with malaria (5.6%) compared to non-malaria febrile illness (7.8%). For children with all-severity malaria (including recent infection) NTS was more commonly isolated than in non-malarial febrile illness. Added to this, several large studies [5, 24] provide further support that EGNO (including NTS species) bacterial isolates are more common in children with malaria compared to non-malaria illness, but this finding was not universal. In children with SM neither those with cerebral malaria nor children with neurological complications appeared at increased risk of IBI; however, a higher proportion of children with SMA had IBI, principally NTS. Children with malaria and IBI infection had a substantially higher mortality (22%), compared to those without (12%). However, the most compelling data on the predisposition of malaria-exposed children to IBI comes from longitudinal studies capturing declining risk of IBI, spectrum of IBI organisms and risk of mortality in parallel to an epidemiological transition from high to low malaria endemicity. The largest reduction was in the incidence of NTS bacteremia, which was mirrored by a similar reduction in the protective effect of sickle cell trait against this organism. The data we have collated from hospital cohort studies corresponds with the findings reported in epidemiological studies but is less conclusive.

Teasing apart the association between malaria and IBI has been challenging for several reasons. Foremost, both malarial parasitemia and IBI are common in children presenting with fever in malaria endemic areas. Rainy seasons in the tropics predispose to blood borne infection with enteric pathogens as well as malaria outbreaks [50, 51]. Hospitalized children often have overlapping etiological syndromes [52], for example, 27% of Mozambican children with radiologically confirmed pneumonias also had confirmed asexual parasitemia [30]. The limitations imposed by the lack of routine high quality microbiology services in sSA have resulted in a fragmented and patchy understanding of this association and its importance. Hard to culture pathogens, such as NTS [53], may be under-reported, conversely contamination remains a possibility in those studies reporting high incidences of staphylococcal organisms [27, 39, 40, 42] and requires substantiating by further studies.

The findings of this review are subject to methodological limitations. Heterogeneity was high between studies. Only a few papers recorded sufficient data to allow standardization of populations. Inclusion of recent malaria infection, using antigen-based rapid diagnostic tests, has shed more light on the prevalence of co-infection and needs further investigation. We were also unable to control for differences in study design and quality of microbiological data. For many studies, the prevalence of IBI co-infecting malaria was not the primary question, making a formal meta-analysis potentially misleading. Unpublished data were not included and, therefore, our findings may have a publication bias towards studies that report high level of concordance. In children with SM the strength of a biological association with IBI is even more difficult to establish due to the paucity of data from an appropriate control population with severe life threatening illness. Our findings do, however, concur with the conclusions of the systematic review and meta-analysis of community acquired bacterial infection indicating co-infection in 6.5% of children with malaria (any type) [4].

A putative clinical association between malaria and IBI was first suggested in the 1920s in Guyana by Giglioli who observed both increased prevalence and virulence of paratyphoid fever during malaria seasons with a co-infection rate of 29% [54]. A similar rise in typhoid fever in Nigerian children during the rainy season was speculated to be linked to the increased incidence of malaria [55]. These observations of seasonal concurrence of NTS infection and malaria were further explored in Gambian children [26] showing that while stool carriage of NTS remained constant throughout the year, rises in invasive NTS infections occurred in parallel to the annual rise in malaria cases.

The propensity for co-infection with EGNOs has led to speculation that the source of the microorganisms is most likely to be from the intestine due to impaired gut barrier function [56]. A recent mouse experiment may help our understanding of the reasons why children with malaria maybe more susceptible to concomitant NTS at the biological level [57]. The production of hemoxygenase (HO-1), which is induced in response to malaria hemolysis, resulted in impaired defense by mice towards NTS infection. While not yet replicated in humans the underlying mechanisms and portal of entry of pathogens into the bloodstream include either, or a combination of, mechanical and immune-suppressive factors. Histopathological studies have demonstrated intense sequestration of P. falciparum-infected erythrocytes within the endothelial bed of the gut, particularly at the tip of intestinal villi [58, 59]. The presence of small bowel intussusceptions in children with SM [60] has also been reported. The intense sequestration of late stage infected erythrocytes in venules and capillaries as well as the increased rigidity (or non-deformability) of non-parasitized red cells [61] both compromise microcirculatory flow to vital organs and alter endothelial cell function [62]. This may precipitate injury and impaired gut barrier function either directly or indirectly through local cytokine production, with the subsequent transfer of endotoxin and/or pathogenic bacteria into the blood stream. We have recently shown that endotoxemia is common in SM (28%) and results in temporary immune paralysis similar to that observed in patients with sepsis and experimentally-induced endotoxemia [63]. We hypothesized that the most likely origin of endotoxin was from the gut.

Although current recommendations for managing putative sepsis in children with SM are unclear [46, 64], the ideal approach is not obvious [65, 66]. For example, in one recent study almost 50% of bacterial isolates were resistant to the antibiotics most commonly recommended for empirical use by the WHO [5]. Moreover, in the specific case of NTS, the efficacy of gentamicin is doubtful and susceptibility testing unreliable due to the intracellular nature of this infection [67, 68]. For newer antimicrobials, apart from financial considerations, there are concerns that over-use could lead to resistance. Judicious use of antibiotics may prove to be critical to those at highest risk of poor outcome.

Establishing the relationship, both proof and effect, between malaria and IBI has been challenging. The accumulated evidence suggests that children with recent or acute malaria are at risk of bacterial infection, which results in an increased risk of mortality. Establishing which children with SM are at greatest risk of bacteremia would inform a policy for targeted antibiotic therapy that could substantially reduce malaria-associated mortality while minimizing the risks of excess antibiotic prescribing.

JC is a Paediatric specialty trainee and Honorary Research Fellow at Queen Mary, University of London and was a visiting research fellow at the KEMRI-Wellcome research programme last year. KM is a Professor in Tropical Paediatric Infectious Diseases at Imperial College, London, based fulltime in East Africa at KEMRI Wellcome Trust Programme.