A Prospective Study of the Causes of Febrile Illness Requiring Hospitalization in Children in Cambodia

Background Febrile illnesses are pre-eminent contributors to morbidity and mortality among children in South-East Asia but the causes are poorly understood. We determined the causes of fever in children hospitalised in Siem Reap province, Cambodia. Methods and Findings A one-year prospective study of febrile children admitted to Angkor Hospital for Children, Siem Reap. Demographic, clinical, laboratory and outcome data were comprehensively analysed. Between October 12th 2009 and October 12th 2010 there were 1225 episodes of febrile illness in 1180 children. Median (IQR) age was 2.0 (0.8–6.4) years, with 850 (69%) episodes in children <5 years. Common microbiological diagnoses were dengue virus (16.2%), scrub typhus (7.8%), and Japanese encephalitis virus (5.8%). 76 (6.3%) episodes had culture-proven bloodstream infection, including Salmonella enterica serovar Typhi (22 isolates, 1.8%), Streptococcus pneumoniae (13, 1.1%), Escherichia coli (8, 0.7%), Haemophilus influenzae (7, 0.6%), Staphylococcus aureus (6, 0.5%) and Burkholderia pseudomallei (6, 0.5%). There were 69 deaths (5.6%), including those due to clinically diagnosed pneumonia (19), dengue virus (5), and melioidosis (4). 10 of 69 (14.5%) deaths were associated with culture-proven bloodstream infection in logistic regression analyses (odds ratio for mortality 3.4, 95% CI 1.6–6.9). Antimicrobial resistance was prevalent, particularly in S. enterica Typhi, (where 90% of isolates were resistant to ciprofloxacin, and 86% were multi-drug resistant). Comorbid undernutrition was present in 44% of episodes and a major risk factor for acute mortality (OR 2.1, 95% CI 1.1–4.2), as were HIV infection and cardiac disease. Conclusion We identified a microbiological cause of fever in almost 50% of episodes in this large study of community-acquired febrile illness in hospitalized children in Cambodia. The range of pathogens, antimicrobial susceptibility, and co-morbidities associated with mortality described will be of use in the development of rational guidelines for infectious disease treatment and control in Cambodia and South-East Asia.


Introduction
Febrile illness in children is a common cause of admission to hospital globally, with significant associated morbidity and mortality [1].In developing countries this is frequently compounded by low rates of immunisation, untreated co-morbidities, and late presentations [2].Febrile illnesses are caused by diverse pathogens, presenting with non-specific symptoms to healthcare facilities with limited diagnostic capacity [3,4].Clinical manage-ment guidelines for acute febrile illness are available [5,6], but rarely supported by knowledge of the locally prevalent causative agents.
The Kingdom of Cambodia lies in South-East Asia and has a mortality rate in children aged ,5 years of 54/1000 live births [7].This has halved over the last decade but remains one of the highest in the region.The prevalence of undernutrition in children ,5 years of age (less than 2 SD of weight for age) is 28% [7].There is little published information on the causes of fever in Cambodian children.

Ethics Statement
Parents of all children recruited to the study gave witnessed, written, informed consent before study enrolment.The Oxford Tropical Research Ethics Committee and Angkor Hospital for Children Institutional Review Board approved the study protocol on 24th September 2009 and 2nd October 2009 respectively.

Study Site and Population
This prospective, year-long study of the causes of fever in children was based at Angkor Hospital for Children (AHC), Siem Reap province, Cambodia (Figure 1).AHC is a 50-bed paediatric hospital providing free universal inpatient and outpatient care to children ,16 years of age from urban and rural settings.It has critical care capacity, including mechanical ventilation and inotropic support, and is one of two paediatric hospitals serving Siem Reap city and province.
The national immunisation schedule included Bacillus Calmette-Gue ´rin (BCG) and hepatitis B virus (HBV) at birth, and diphtheria-pertussis-tetanus, oral poliovirus and measles virus vaccines.79% of children aged 12-23 months have received these vaccines [7].Haemophilus influenzae type b immunisation was introduced during the study period.

Patients and Clinical Methods
Patients admitted to AHC between 12 th October 2009 and 12 th October 2010 were considered for enrolment.Eligibility criteria were age ,16 years, documented axillary temperature $38.0uC within 48 h of admission, and caregiver consent.Febrile postsurgical patients were excluded.All children, except emergencies, were assessed using locally-modified Integrated Management of Childhood Illness (IMCI) guidelines [5] prior to a decision to admit.Admission information was recorded on a study-specific clinical record form.Admissions were reviewed twice daily for eligibility.

Sampling and Laboratory Methods
Blood was taken aseptically from all enrolled patients for culture, complete blood count, blood film (including malaria smear), biochemistry, and nucleic acid amplification tests (NAATs).In addition, study patients aged $60 days had admission blood samples taken for serology and Leptospira spp.culture, and a convalescent serology sample taken on discharge, or after 7 days of admission.Whole blood and serum samples were stored at -80uC until analysed.
Nasal and throat swabs were taken for respiratory virus detection in patients with a recent history of cough or sore throat, and increased respiratory effort.Cerebrospinal fluid (CSF) analysis was performed on patients with suspected central nervous system (CNS) infection.Urinalysis was performed on all children; other sampling (e.g.HIV serology, gastric aspirates) and imaging were performed as clinically indicated.

Bacterial Culture
Blood was taken for culture, when possible before in-hospital antimicrobial therapy and within 48 h of admission.Blood was inoculated into a pre-weighed blood culture bottle and the vented bottles were incubated aerobically at 37uC for 7 days [20].Subculture onto sheep blood and chocolate agar was undertaken at 24 h, 48 h, 7 days, or if the culture was turbid.CSF samples were separated immediately upon receipt into aliquots for staining and microscopy, cell count and biochemistry, culture, and storage at -80uC.Cultured organisms were identified using routine methods [20], including API test kits (bioMe ´rieux, France), disc diffusion antimicrobial susceptibility testing [21] and Etests TM (AB Biodisk, Sweden) performed as appropriate.Samples from other sterile sites (e.g.abscesses, pleural fluid) were cultured using routine methods [20] and were reviewed daily for clinical relevance by the infectious diseases team.Whole blood samples were cultured for Leptospira spp.[22].

Serology
The Panbio Japanese encephalitis virus (JEV) and dengue virus (DENV) IgM Combo enzyme-linked immunosorbent assay (ELISA) (Panbio, Australia) was used to detect anti-JEV and anti-DENV specific IgM antibodies in sera.An ELISA (Standard Diagnostics, Korea) was used to detect DENV NS1 antigen [23].A capture IgM ELISA assay (Venture Technologies, Malaysia) was used to detect anti-JEV and anti-DENV specific IgM antibodies in CSF specimens [24].Table S1 describes the result interpretation.
Full details are in the supplementary material.All nucleic acid extractions and assays were done according to the manufacturer's instructions.

Clinical Data
For each episode, the first recorded parameter after the time of admission was used.The clinical syndrome was categorised by a senior paediatrician (VK) at hospital discharge according to the localising focus of infection, e.g.lower respiratory tract infection (LRTI), or non-infectious cause for fever.On final analysis of all data, a microbiological diagnosis was given when results were consistent with the presenting clinical syndrome.When microbiological results were inconsistent with the clinical syndrome, both were given as a final diagnosis.Two clinicians (KE and MC) made these judgements independently with disagreements resolved by discussion.

Data Management and Analysis
Data was managed on a study-specific database and analysed using Stata 12 (Stata Corp., USA), with weight-for-age z-scores calculated using the WHO Anthro 3.2.2Stata macro (World Health Organisation).Multivariate logistic regression was used to analyse the effects of comorbid undernutrition, heart disease, HIV infection, and anaemia; adjusted for each other and age group.

Baseline Characteristics
There were 3225 patient admissions during the study year, of which 1361 (42.2%) met the inclusion criteria.Of these, 136 (10.0%) were not enrolled, leaving 1225 febrile episodes in 1180 children (Figure 2).1144 children had a single episode, 31 children had two episodes, one child had three episodes and four children had four episodes.The median (inter quartile range [IQR]) age was 2.0 (0.8-6.4) years, with 850 (69.4%) episodes in children ,5 years of age.The median (IQR) duration of illness prior to admission was 3 (2-5) days.Medication was given by the caregiver prior to admission in 53% of episodes.In 43% of these the medication was a known antibacterial, anti-malarial or steroid.Other baseline characteristics are in Table 1.

CSF Sampling
A CSF sample was examined by microscopy and culture from children in 174 (14.2%) of disease episodes.52 (30.0%)CSF samples showed white cell pleocytosis, and 11 (6.3%) were culture or CSF stain positive.There was sufficient CSF for NAATs and serology from 107 (62.6%) samples, with 15 (14.0%) positive by NAAT and 7 (6.5%)positive by serology (Table 5).All CSF samples positive by culture for were also positive by NAATs.NAATs also identified an additional three S. pneumoniae and one H. influenzae positive episodes.

Leptospira spp. and Rickettsial Disease
Leptospira spp.culture was performed for 1068 of 1149 (93.0%) disease episodes in children $60 days of age, and two (0.2%) were positive.NAAT for Leptospira spp. was performed in 1179 episodes

Virology
Serological evidence of a flavivirus infection was determined in 1125 (98.0%) disease episodes in children $60 days of age (Table 7); DENV NS1 antigen was assayed in 1105 (96.4%) of episodes in children $60 days of age and 60 (5.4%) samples were positive.There was pronounced seasonality in ''acute serology'' to DENV (NS1 antigen positive, dynamic rise in anti-DENV IgM titres, or CSF anti-DENV IgM positive) in contrast to lack of seasonality in ''acute/recent serology'' to DENV or other flaviviruses (Figure S2 and Figure S3).
There were 389 febrile episodes that met our criteria to have oral and nasopharyngeal swabs sent for the analysis of respiratory viruses (Table 8).

Discussion
With comprehensive laboratory investigation we identified a microbiological cause in almost 50% of febrile illness in this oneyear study of febrile Cambodian children requiring admission to hospital.The acute mortality rate was 5.6% of children enrolled (6.7% of all eligible).Nurse-led triage using IMCI [5] guidelines and paediatric review prior to admission excluded less-severe infections.Children died despite availability of parenteral broadspectrum antimicrobials, critical care, and laboratory facilities, all unavailable to the majority of the Cambodian population.Prevalence of comorbid undernutrition was 43.7%, and doubled the odds of mortality.Comorbid heart disease, and known HIV infection were also associated with large increases in odds of mortality.
LRTIs, diarrhoeal disease or undifferentiated fever were the main presentations, but a microbiological diagnosis was achieved in only 27%, 29%, and 45% of episodes with these syndromes, respectively.The absence of microbiological examination of faeces, limited use of urine culture, and lack of M. tuberculosis culture (or NAAT) facilities were limitations.More inclusive criteria for the analysis of respiratory viruses, HIV testing, and greater emphasis on CSF sampling may have increased our diagnostic yield.In contrast to a recent report [33], we found no enterovirus-71 in 8 episodes of CSF enterovirus-positive meningoencephalitis.
The serological data from our study is consistent with similar studies in Cambodia [8,14], and neighbouring urban Laos [10].There was evidence of DENV infection in 16% of episodes, but clinical differentiation from bacterial septic shock was difficult, with children frequently treated for both.This strategy is supported by the co-incidence of both invasive bacterial disease (particularly S. enterica Typhi) and malaria, with DENV infection in our cohort.Serological evidence of infection by O. tsutsugamushi and R. typhi was also common.Although the interpretation of serological tests on cohorts of unselected febrile children can be difficult [25,26] even with conservative cut-offs for IgM titres against O. tsutsugamushi and R. typhi, we estimated that 10% of children were infected with these microorganisms.NAATs may increase the specificity of diagnosis, but may lack sensitivity due to short periods of rickettsaemia.
Invasive community-acquired bacterial disease was common, frequently resistant to commonly used antimicrobials, and associated with significant mortality.The most common isolate from blood was S. enterica Typhi, with 90% of isolates with intermediate resistance to ciprofloxacin, and 85% multi-drug resistant (MDR) [18], contrasting with the decline in drug-resistant phenotypes seen in adjacent countries [12].B. pseudomallei (inherently resistant to ceftriaxone and penicillins) was confirmed as a pathogen of major local public health significance [16], and E. coli (of which 1 of 7 isolates demonstrated extended-spectrum blactamase activity) was also prevalent.S. aureus was the commonest isolate from all sterile sites, with one meticillin-resistant isolate [19].Invasive disease caused by S. pneumoniae, H. influenzae, and N. meningitidis is present and potentially vaccine-preventable.The small numbers of neonates in this cohort (not due to refusal of parental consent) and their high mortality emphasizes the need for accessible perinatal care in Cambodia [7].
We are investigating whether pre-treatment with incomplete or sub-therapeutic antimicrobial regimens, especially in an unregulated private sector, contributes to low blood culture yield and high-levels of resistance in Cambodia.Resolution of conflicting demands for rational antibiotic prescribing to help prevent further emergence of antimicrobial resistance, and the urgent need for successful treatment of patients, would be aided by a network of sentinel microbiology laboratories throughout South-East Asia [3,4].Without such laboratories, efforts to treat severe infections in low and middle-income countries will flounder in the dark [39].

Figure 2 .
Figure 2. Flow chart showing enrolment to the study.Notes: a including one home palliative care; b including one home palliative care; c excluded from analyses of outcome (e.g. in odds ratios); d included as ''died'' in analyses.doi:10.1371/journal.pone.0060634.g002

Figure
Figure S1 Flowchart summarising methods for the analysis of cerebrospinal fluid (CSF) from children with suspected meningoencephalitis enrolled in the study.(PDF) Figure S2 Number of admissions with ''acute serology'' (see main text for definition) to DENV against period of admission (30-day intervals starting from study start date 12th October 2012).(PDF) Figure S3 Number of admissions with ''acute/recent serology'' (see main text for definition) to DENV against period of admission (30-day intervals starting from study start date 12th October 2012).(PDF) Table S1 Classification of DENV and JEV serology in samples.a Dynamic rise of $2 Panbio units between acute and discharge samples.b Dynamic fall of #2 Panbio units between acute and discharge samples.c Dynamic rise or fall of #2 panbio units between acute and discharge samples.d Considered negative by manufacturer's criteria.(PDF) Material S1. (DOC)

Table 1 .
Baseline, comorbidity and outcome characteristics in 1225 disease episodes among 1180 children admitted to hospital with fever over one year.

Table 2 .
The presenting clinical syndromes diagnosed for 1225 episodes of febrile illness over the one-year study period.Within these 1225 episodes, a total of 1333 clinical syndromes were diagnosed: 1120 episodes had a single clinical syndrome diagnosed, 102 episodes had two separate syndromes diagnosed, and three episodes had three separate syndromes diagnosed (i.e.1333 syndromes in total). doi:10.1371/journal.pone.0060634.t002
a Ziehl-Neelsen stain only, no culture facilities available.b Non-infectious causes of fever included 11 episodes attributed to haematological malignancy, five episodes attributed to burns and one episode attributed to juvenile idiopathic arthritis.c Using n for entire cohort, see text and Tables 4-8 for denominators by each test.doi:10.1371/journal.pone.0060634.t003(96.
b 8 strains were available for testing, all were susceptible to ceftriaxone.c 4 strains were available for testing, one was ESBL producing.d 6 strains available for testing, one was MRSA.e Constitutively resistant to ceftriaxone.f 2 strains available for testing, one was ESBL producing.g 33 coagulase-negative staphylococci, 27 Gram-positive bacilli.doi:10.1371/journal.pone.0060634.t004

Table 7 .
Acute rickettsial serology'' denotes a .4-folddynamicrise in specific IgM between acute and convalescent (7-days) samples; ''acute/recent serology'' denotes either static IgM titres or raised IgM in a single acute sample.Serological results for flaviviruses for episodes in children of 60 days or older (n = 1125).By definition, all NS1 positive samples were denoted ''acute dengue virus serology'', and all CSF IgM positive samples were ''acute serology'' (although we acknowledge that NS1 antigen assay is not a serological test).Numbers in superscript denote samples with insufficient serum for NS1 antigen assay.
a One additional to serological testing.b Two additional to serological testing.c Three additional to serological testing.doi:10.1371/journal.pone.0060634.t006

Table 8 .
Positive results by NAAT for respiratory viruses in children presenting with sore throat or cough and increased respiratory rate or effort (n = 389).

Table 9 .
Primary diagnosis, and contributing diagnoses, for the 69 children who died during the study.These diagnoses are based on all the available results: hence are not identical to ''clinical syndrome diagnosed'' in Table2.Numbers in parentheses indicate the number of patients with the contributing diagnosis in addition to the primary diagnosis.Superscripts give further information about individual morbid patients: a Also with organism of uncertain significance from blood (Burkholderia cepacia) and pulmonary hypertension secondary to congenital heart disease; b also with organism of uncertain significance from blood (Acinetobacter calcoaceticus);