Etiology of Severe Febrile Illness in Low- and Middle-Income Countries: A Systematic Review

Background With apparent declines in malaria worldwide during the last decade and more widespread use of malaria rapid diagnostic tests, healthcare workers in low-resource areas face a growing proportion of febrile patients without malaria. We sought to describe current knowledge and identify information gaps of the etiology severe febrile illness in low-and middle-income countries. Methods and Findings We conducted a systematic review of studies conducted in low-and-middle income countries 1980–2013 that prospectively assessed consecutive febrile patients admitted to hospital using rigorous laboratory-based case definitions. We found 45 eligible studies describing 54,578 patients; 9,771 (17.9%) had a positive result for ≥1 pathogen meeting diagnostic criteria. There were no eligible studies identified from Southern and Middle Africa, Eastern Asia, Oceania, Latin American and Caribbean regions, and the European region. The median (range) number of diagnostic tests meeting our confirmed laboratory case definitions was 2 (1 to 11) per study. Of diagnostic tests, 5,052 (10.3%) of 49,143 had confirmed bacterial or fungal bloodstream infection; 709 (3.8%) of 18,142 had bacterial zoonosis; 3,488 (28.5%) of 12,245 had malaria; and 1,804 (17.4%) of 10,389 had a viral infection. Conclusions We demonstrate a wide range of pathogens associated with severe febrile illness and highlight the substantial information gaps regarding the geographic distribution and role of common pathogens. High quality severe febrile illness etiology research that is comprehensive with respect to pathogens and geographically representative is needed.


Introduction
Fever is a common reason for seeking healthcare in low-and middle-income countries (LMICs) [1]. Among patients with febrile illness requiring admission case fatality ratios are high, sometimes exceeding 20% [2][3][4][5][6]. Fever etiology research [4,7,8] and the more widespread use of malaria diagnostic tests following changes to malaria treatment guidelines [9,10] have highlighted the problem of malaria over-diagnosis among patients with severe febrile illness. Apparent declines in malaria illnesses and deaths associated with malaria control efforts mean that the proportion of febrile patients with malaria has declined over the past decade [11,12].
While the global burden of disease due to diarrhea and pneumonia has been estimated at the syndrome level [13][14][15], such an approach has not been taken for fever without localizing features. Instead, illness and death due to some febrile diseases (e.g., dengue, malaria) are estimated [11,16], while others have been neglected (e.g., leptospirosis, Q fever). Comprehensive, standardized, and high quality, multi-center etiology research is being undertaken to understand the causes of severe childhood diarrhea and pneumonia [13,14] but such an approach has not been taken for fever. The many causes of fever are difficult to distinguish clinically [4,7,8] and laboratory services may be limited or absent in low-resource areas [17]. Consequently, clinicians frequently lack information about the local epidemiology of causes of severe febrile illness needed to adapt international management guidelines. Similarly disease control programs lack data to set priorities for prevention.
A robust contemporary picture of treatable and preventable infectious causes of severe febrile illness is urgently needed to improve patient outcomes and to inform disease control efforts in LMICs. Systematic reviews of studies of community-acquired bloodstream infections in Africa [18] and Asia [19] have demonstrated the importance invasive infections among febrile inpatients. A study mapping studies of the aetiology of non-malarial febrile illness in South East Asia [20] highlighted the diversity and geographical variation in a range of causes of fever. It also revealed the substantial information gaps that remain for a range of relevant pathogens.
To describe epidemiologic patterns and to identify data gaps in our understanding of severe febrile illness in low resource areas, we sought to systematically review prospective hospitalbased studies of the etiology of febrile illness in LMICs.   Table 1. The search string combined the geography terms 'country' and etiology terms 'pathogen' or respective 'disease' (Table 1). For blood stream infections and rickettsial infections, only disease terms were searched without pathogen terms. Adjustments to the search strategy were made in accordance with the requirements of each database. Online translation tools were used to evaluate non-English titles, abstracts, and full text articles.

Title and abstract review
One investigator (NP) reviewed titles and abstracts of articles identified by the search strategy.
Those that appeared to be prospective studies of consecutive febrile patients enrolled in the emergency department or inpatient service of hospitals in an LMIC during the time period

Full-text review
Two investigators (DRM, JAC) reviewed full-text articles identified by the title and abstract review. When required, the third investigator (NP) served as tiebreaker, independently reviewing articles to resolve disagreement between the other two investigators. To be eligible for data extraction, full-text articles were confirmed to be prospective studies of consecutive febrile patients enrolled in the emergency department or inpatient service of hospitals in a low-or middle-income country during the time period 1980 through 2013. For the purposes of this review, febrile patients were defined as a person with a history of fever in the past 48 hours; an axillary temperature 37.5°C; or a rectal temperature 38.0°C. In addition, participants in such studies needed to be evaluated for at least one of the febrile diseases of interest using laboratory-confirmed case definitions (Table 2). We excluded studies of syndromes other than fever; studies of specific subgroups of febrile patients, such as HIV-infected persons; studies of health-care associated infections or studies where such infections could not be distinguished; and studies of outpatients or where outpatients and inpatients could not be distinguished.

Validity assessment
We ensured the validity of the review by adhering to the predefined selection criteria to allow comparison across individual studies. By creating pre-determined case definitions we sought to capture only confirmed cases of infection. However, some variation in microbiological techniques and interpretation of results was unavoidable. We did not exclude studies on the basis of incomplete description of laboratory techniques, blood culture contaminants isolated, or failure to report all pathogens that may have been isolated or identified.

Data extraction
The following data were extracted from each eligible study by one investigator (NP): geographical location of the healthcare facility; healthcare facility rurality; study time dates and duration; study inclusion and exclusion criteria including age range; diagnostic techniques for each infection; number of patients tested for each infection; number tested meeting case definition for each infection; use of additional tests (e.g., HIV serology). When available, we also recorded clinical diagnosis of patients; in-hospital fatality ratio; seasonal variation of pathogens; and pre-admission use of antimicrobials. For the purpose of this review, pediatric studies were defined as those that included patients aged from 28 days to <15 years. Studies with mixed populations of adults and children were analyzed as adult studies. Queries regarding data

Statistical analysis
Following data extraction, infections were organized into four groups: blood parasites; bacterial and fungal bloodstream infections; bacterial zoonoses; and viral infections, as shown in Table 2. Data from all individuals in all studies were aggregated to compare prevalence of febrile diseases across studies and regions. Summary statistics were calculated for key variables. Analyses of associations between patient factors or clinical conditions (e.g., HIV infection) and specific febrile diseases were done for studies with data for both the pathogens and factors being assessed. Chi-squared test was used to establish significance of associations and values were expressed as odds ratios (ORs) calculated with STATA software version 13.0 (College Station, TX, USA).

Role of the funding source
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Search results
The online search completed on 28 September 2013 yielded 135,141 records of which 2,729 articles that appeared to be about febrile illness among humans in LMIC were selected. Of these 863 met criteria for full text review, 823 (95.4%) were available for full-text review of which, 45 (5.5%) were eligible (Fig 1).

Infections searched meeting laboratory case definitions
Of the 25 febrile illnesses searched for in this review (

Bacterial and fungal bloodstream infections
Of the 45 eligible studies and 54,578 patients included in this review, blood cultures and antigen testing was conducted in 28 (62.2%) studies among 49,143 (90.0%) patients. All studies described the microbiological techniques used for blood cultures. However, the media used and methods of identification of organisms varied between studies. Minimum acceptable blood culture volumes were reported by 18 (64.3%) of 28 studies using blood cultures and ranged from 1 mL to 3mL in pediatric studies and from 5 mL to 10mL in adult studies. Of 23 studies reporting results of antimicrobial susceptibility testing, all used disc diffusion, or Epsilometer test (E-test) methods [2,3,5,26,27,[35][36][37][38][46][47][48][49][50][51][52][53][54]. Organisms thought to be contaminants were reported as being excluded from analysis in 16 (57.1%) of the 28 studies. In six studies providing data from all positive blood cultures, contaminants were isolated from 36 (3.9%) of 920 adult blood cultures [2,50] and 107 (4.2%) of 2,550 pediatric blood cultures [3,49,51,55]. Of patients evaluated with blood culture or antigen testing, 4,852 (10.6%) were reported to have positive result. Table 4 provides a summary of the most common bloodstream isolates according to region and age group in eligible studies.
Brucella spp was tested by both blood culture and serological methods,  (Table 4).

Malaria and other blood parasites
Of all studies included in this review, 24 (51.1%) reported the prevalence of malaria parasites identified by thick or thin smear, nucleic acid amplification test (NAAT), or rapid diagnostic test. No study reported the detection of blood parasites other than malaria.
Of studies testing for malaria, 13 (54.2%) were conducted in Eastern Africa. Of the 12,245 patients tested for malaria, 10,535 (86.0%) were tested using thick or thin malaria blood smears only and one study tested 462 (3.8%) patients for malaria using rapid diagnostic tests only [30]. Three studies used two or more tests to diagnose malaria [41,44,53]. Among the 3,106 Plasmodium spp. that were identified to the species level, 2,928 (94.3%) were Plasmodium falciparum, of which 2,907 (99.3%) were found in the African regions. Plasmodium vivax accounted for 173 (5.6%) of speciated Plasmodium spp. Of the 92 Plasmodium spp. identified among patients in South Central and South East Asia 75 (81.5%) were Plasmodium vivax.

Viral infections
Of the 45 eligible studies with 54,578 patients in this review, viral infections were investigated according to our laboratory case definitions in eight (17.0%) studies including 7,939 (14.4%) patients using 10,389 tests for the six infections; chikungunya, dengue, influenza, Japanese encephalitis, West Nile virus infection, and yellow fever virus (Table 4).
Of viral infections, dengue fever was the most commonly assessed and was investigated using an eligible test in six (12.8%) studies; one (20.0%) each in Eastern Africa [4], North Africa [62], South Central Asia [63] and three (50.0%) studies in South East Asia [3]. In total 2,513 (4.6%) patients were tested for dengue fever using virus isolation, NAAT, or serology according to our case definitions.

HIV co-infection
Of the 15 studies with 9,365 patients that included HIV testing, 1,988 (21.2%) patients were found to be HIV seropositive. There were insufficient data in included studies to investigate the association between HIV and infections other than bacterial and fungal bloodstream infection. In nine studies with adequate data for analysis [2,6,27,35,47,49,52,64]

Clinical diagnosis
Five studies, four from Eastern Africa [4,44,47,51,53,56] and one from South Central Asia [46], provided sufficient data regarding both clinical and laboratory confirmed diagnoses for the cause of fever and enabled assessment the accuracy of clinical diagnosis. In Eastern Africa a clinical diagnosis of malaria was recorded in 800 (51.1%) of 1,566 patients of whom 85 (5.4%) had malaria parasites identified through laboratory diagnostic testing. In the study from South Central Asia 52 (50.5%) of 103 patients presenting with fever had a clinical diagnosis of enteric fever. Of the 52 patients with a clinical diagnosis of enteric fever 20 (38.5%) were found to have a positive blood culture for typhoidal Salmonella [46].

Concurrent infections
Seven (15.6%) out of the 45 eligible studies provided information regarding apparent concurrent infections. Of 5,719 patients enrolled in studies reporting such information, 198 (3.5%) were found to have both a positive blood culture for a pathogen and a positive malaria smear [2,48,52,53,55,60,61].
Moreover, one study in South Central Asia showed evidence for mixed infections of S. enterica serotype Typhi with scrub typhus or typhus group rickettsiosis [46].
Data from two Malawian cohorts showed a shift from a predominance of non-typhoidal Salmonella in blood cultures during the wet season to a predominance of S. pneumoniae during the dry season [5,47]. A Mozambique study found no association between season and bloodstream infections [40]. A study from Egypt suggested an association between the onset of the rainy season and a predominance of S. enterica serotype Typhi isolates. In the same study it was shown that brucellosis was reported in all months of the year with peaks in the spring and early summer, coinciding with the parturient seasons of domestic animals [26].
In Nepal S. enterica serotype Typhi and Paratyphi A were the most common bloodstream isolates during both monsoon and winter seasons. However, there was a substantial increase in the proportion of Salmonella Paratyphi A isolates during the monsoon season. In another study done in Nepal, murine typhus was more common during the winter season than the summer [43].
Vector-borne and zoonotic diseases such as dengue and leptospirosis were found to be more common during the rainy seasons in Mauritania and Thailand [33,59]. In Cambodia, influenza virus cases peaked during the rainy season [28].

Discussion
To our knowledge, this is the first systematic review of severe febrile illness etiology for a broad range of pathogens in all LMICs. We show that bacterial and fungal bloodstream infections, bacterial zoonoses, malaria, and viral infections are leading causes of severe febrile illness, and that their relative importance appears to vary by geographic region (Table 4). Our findings confirm that some infectious causes of fever are closely linked to HIV co-infection, that severe febrile illness is associated with high in-hospital case fatality ratios, that some pathogens show seasonal patterns, and that clinical diagnosis is unreliable among febrile patients, especially for pathogens causing systemic infections. Most notably, we demonstrate that there are major gaps in our current understanding of the causes of severe febrile illness in LMICs. Some potentially important pathogens have not been rigorously studied in any country, many studies examined only one or a few pathogens, many countries and some geographic regions had no eligible research on severe febrile illness etiology, and representation of age groups was inconsistent.
Bacterial and fungal bloodstream infections were the most sought febrile disease with a total of 30 (63.8%) out of 47 studies included in this review conducting blood cultures, 22 (47%) of which were conducted in the African continent. Overall, the proportion of severe febrile illness attributed to invasive bacterial and fungal infections was 10.7%, 8.5% among children and 13.9% among adults. Although our ability to examine geographic and age-related patterns of bloodstream infections was limited by incomplete representativeness of studies, some observations are possible. Salmonella enterica was the most common bloodstream isolate. Non-typhoidal Salmonella (NTS) predominated in all African regions, except for Northern Africa where Salmonella Typhi was more common. In Asian regions Salmonella Typhi and Salmonella Paratyphi A predominated. S. pneumoniae was the most common Gram-positive invasive infection in both African and Asian regions. S. pneumoniae was particularly common in paediatric cohorts in both areas accounting for 19.2% of bacteraemia in the African regions and 16.5% in the Asian regions. Fastidious organisms, such as S. pneumoniae, may be less often isolated than those without special growth requirements. This may have affected the relative prevalence of different species in our review.
Plasmodium spp. was the most commonly identified organism among patients with febrile illness overall. As expected, Plasmodium falciparum predominated in the African regions while Plasmodium vivax predominated in Asian regions. Malaria parasite and bacterial bloodstream co-infections were common among patients with positive malaria diagnostic tests especially in areas with year-round intense malaria transmission [2,40,48,51,53,55,60,61]. It is apparent that Plasmodium spp. may act as the prime pathogen; as a co-pathogen, increasing risk for other infections such as NTS bacteremia in some circumstances [65]; or as an bystander in others. In the latter, despite having a positive blood film the patient is suffering from another illness and the parasitemia, that should nonetheless also be treated, is incidental [66].
Studies that used an afebrile control group to calculate the fraction of febrile illness attributable to malaria indicate that incidental parasitemia is likely to be particularly common in malaria-endemic areas [2,47,64]. Such studies could be improved by including routine measurement of malaria parasite density. Furthermore, studies comparing clinical diagnosis with laboratory diagnosis of malaria confirmed that malaria over-diagnosis is common among patients with severe febrile illness. Incidental Plasmodium spp. infection and malaria overdiagnosis increase the risk of the patient not being treated for the actual cause of the current illness [4,8,44,47,51,56,64].
With respect to bacterial zoonoses and viral infections, large geographical areas had no or few studies, and few patients were evaluated for these infections. Those studies that did evaluate for bacterial zoonoses and viral infections varied widely in pathogens sought and diagnostic tests used. Many studies did not collect convalescent serum, precluding conventional standard diagnostic testing and therefore inclusion in our review. Among eligible studies, case fractions were found to be highly variable across regions and age groups and the small number of both studies and participants suggest that prevalence data should be interpreted with caution. Among eligible studies, spotted fever rickettsiosis predominated in African regions, with brucellosis being common in Northern Africa, while typhus group rickettsiosis, scrub typhus, and leptospirosis were particularly common in Asia. Among viral infections, dengue fever was found to be an important cause of febrile illness in Asia and was associated with a high case ratio.
No eligible studies were found from Latin America and the Caribbean, Oceania, some regions of Africa. There were a small number of studies from populous regions of Asia. Furthermore, the majority of LMICs did not have a single eligible study. In addition, we found that there was a limited amount of research in rural settings, despite the majority of countries searched in this review having predominantly rural populations [67]. Future research studies should improve geographic representativeness and include rural study sites.
This systemic review had a number of limitations. We included only studies of severe febrile illness that required admission to hospital emergency or inpatient departments. It is likely that patterns of infection could be quite different in outpatient, primary care, and community settings [68]. We included studies from 1980, just prior to the onset of the global HIV pandemic. We may have missed potentially relevant studies done in earlier years. It is also possible that patterns of infection in the 1980s and 1990s do not reflect the contemporary picture. Because we used conventional standard laboratory-based case definitions, some infections causing the most severe illness resulting in death before acquisition of a convalescent serum sample could not be ascertained. We did not collect data systematically on localised infections among febrile patients. However, such data was rarely reported in the studied included in this review.
Many studies did not enroll all age groups. This meant that age-related differences in severe febrile illness etiology could not be assessed. HIV infection appears to increase risk for a number of infections that may present with severe febrile illness, such as cryptococcal disease, bacteremic disseminated tuberculosis, and NTS bacteremia [2,6,27,35,47,49,52,64]. However, there were insufficient data to assess the role of HIV co-infection for a number of other pathogens evaluated in this study. Several studies found seasonal patterns with some pathogens. However, not all studies ran for a full year and others that did include at least a year of enrollment did not explore seasonality. Ideally, fever etiology research should include all age groups, at least one annual cycle, and routinely assess HIV infection status of participants. Furthermore, while we were restrictive with laboratory case definitions, we were unable to account for variability in some aspects of clinical and laboratory practice at study sites. Finally, data were aggregated by region by combining individual patient results across studies, resulting in a greater influence of larger studies.
We suggest that the current understanding of the etiology of severe febrile illness in LMICs is incomplete. High quality severe febrile illness etiology research that is comprehensive with respect to pathogens and geographically representative could improve patient outcomes by informing patient management guidelines and disease control priorities [69,70]. We recommend that multi-center severe febrile illness research should investigate a broad range of treatable or preventable infections; use standardized and quality assured diagnostic tests with rigorous case definitions; include healthy community controls to allow accurate estimations of attributable case fractions [71,72]; be geographically and demographically representative; have standardized reporting of fever associated with localized infections and should cover at least a full calendar year to incorporate seasonal variation. Such information is an essential component of an effective health system but the gaps in evidence identified by this study are likely to require coordinated resources and expertise to fill in LMICs.