The study took place from August 2001 to August 2002 at Muhimbili National Hospital, Dar es Salaam, Tanzania. A total of 1787 children (aged 0–7 years) were consecutively enrolled in a prospective cohort study of 1828 admissions. The inclusion criterion was clinical presentation suspect of systemic infection based on the presence of fever (> = 38'C), hypothermia (< 36'C) and other signs and symptoms as detailed in the WHO's IMCI Integrated Management of Childhood Illness guidelines 9 including general danger signs such as convulsions, lethargy, inability to drink or breastfeed, vomiting, and other signs of infection, such as neck stiffness, bulging fontanelles, cough, tachypnea, difficult breathing, chest in-drawings, nasal flaring, grunting, diarrhea, dehydration, ear or eye discharge, oral thrush, jaundice, enlargement of liver or spleen, lymphadenopathy, and signs of infection in the skin and umbilicus (in neonates). The attending clinician decided on inclusion of the patient and subsequently recorded clinical data using a standardized questionnaire and obtained blood for culture, malaria microscopy and HIV testing. Additionally, patients' medical records and departmental registries for admissions, discharges and deaths were reviewed.

Due to the young age of the study subjects (0 – 7 years), the parents or other accompanying, responsible family members were asked for written consent on behalf of the patient. Information was given in writing and verbally in the national language, Kiswahili. Written informed consent was obtained before taking blood for microbiological investigations, if feasible. However, in some circumstances, in the case of critically ill patients, blood specimens were taken based on verbal consent, since these investigations are strongly recommended as routine investigations in severely ill, febrile children, and since it would be inappropriate to delay management of such patients due to paperwork. The responsible family member was then approached in retrospect for written consent to use the specimen and information in the study. The responsible family member was allowed to opt out from the HIV-testing and only consent to participation in the blood culture part of the study. As far as possible, the treatment was guided by the test results. In the following, the term "suspected systemic infection" refers to all included patients in the study, and the term "laboratory-confirmed bloodstream infection" refers to growth of one or more clinically relevant bacterial or fungal isolates from blood-culture from a patient who also confirms to the inclusion criterion of suspected systemic infection. Viral and parasitic infections were not included in the definition of bloodstream infection.

Community-acquired infection was defined as bloodstream-infection with growth of pathogenic bacteria in a blood-culture obtained within the first 48 hours after admission. For classification purposes, the time to blood-culture was calculated as the time from admission to the time of receipt of blood culture in the laboratory, allowing for overnight delay in transport for cultures obtained in the evening or at night. A neonate who was born in hospital within the last 10 days was considered as having hospital-acquired infection. If a patient was discharged and re-admitted within ten days, the episode was considered a single admission. The Muhimbili University College of Health Research Ethics Committee approved the study protocol.

One and five ml blood from neonates and older children, respectively, were inoculated in BACTEC Myco/F lytic blood-culturing vials (Becton Dickinson, Franklin Lakes, NJ), which supports the growth of fungi and bacteria, including mycobacteria 10. The blood-cultures were incubated for six weeks. Positive blood-cultures were subcultured on Columbia II agar base (Oxoid Ltd, Basingstoke, UK) with five percent human blood, chocolate agar and MacConckey agar (Difco/BD Diagnostic Systems, Sparks, MI, USA). The isolates were identified by standard methods 11, including the use of API20E, API20NE and API 20 AUX systems (bioMérieux SA, Marcy l'Etoile, France). Due to economic limitations, we did not perform anaerobic culture, and, in general, only a single blood culture was taken from each patient. Thus, the study was not designed to evaluate the clinical significance of coagulase-negative staphylococci (CoNS) and bacteria of doubtful or limited pathogenicity. Consequently, CoNS and probable contaminant such as diphteroids, Bacillus species and micrococci were not considered pathogens in this study. Enterococcal isolates were included in the study as pathogens. Cultures with polymicrobial growth were considered clinically relevant if known pathogens were among the constituent isolates.

Susceptibilities against antimicrobial agents were tested by the disk diffusion method according to the Clinical and Laboratory Standards Institute guidelines 12. Testing for minimum inhibitory concentration (MIC) of antimicrobials was not routinely performed on all isolates. However, MIC determination by E-test (AB Biodisk, Solna, Sweden) was performed for more detailed characterization of the susceptibilities for antimicrobials in Gram-negative bacteria and enterococci. Gram-negative bacteria were investigated for extended-spectrum beta-lactamases (ESBL) with E-test, PCR and DNA sequencing as described previously 13. Enterococcal isolates were investigated by PCR to affirm identity (E. faecalis, E. faecium) and to detect vancomycin resistance (vanA, vanB). We verified the identity of isolates of S. aureus and detected resistance to methicillin/oxacillin with a multiplex PCR targeting the nuc gene and the mecA gene 14.

Malaria testing was performed by microscopy of Giemsa-stained thick and thin drop blood smears. HIV testing was performed anonymously using a rapid test for HIV1/HIV2 antibodies (ACON HIV 1/2, ACON laboratories, Inc. San Diego, California, USA). HIV antibody-positive sera from children younger than 18 months were analyzed for HIV-1 RNA by reverse transcriptase polymerase chain reaction (RT-PCR) using primers targeting the pol gene, JA17 through JA20 15, and the vif region OG 462/502 16.

Data were entered in a database based on the free-of-charge WHONET software for surveillance of antimicrobial resistance available as a download from the World Health Organization 1718, and further data management was done in Filemaker Database software. Statistical analysis was performed in Stata 8 (Stata Corporation, College Station, Texas, US). Univariate assessment of risk factors for intra-hospital death was done by Fisher's exact test with a two-sided P-value and odds ratios and 95% confidence intervals were obtained by the 'logistic' function in Stata. Significant factors to the level of P < 0.2 from the univariate analysis and a priori important factors such as sex, age and underlying diseases were included in the multivariate analysis. Multivariate analysis was performed by automated and manual backwards step-wise logistic regression where factors with P > 0.2 were removed from the model. We present four logistic regression models using different subsets of the study population analyzing cases of laboratory-confirmed bloodstream infection (n = 216) as well as cases of clinically suspected systemic infection (n = 1527), and for each category we re-analyzed the data for those who had known HIV status (n = 128 and n = 790, respectively). Comparisons of medians of time variables were done by Wilcoxon rank-sum (Mann-Whitney) test.

A total of 1787 patients with a median age of 8.5 months were admitted 1828 times. Forty-four percent (795/1787) were female. Neonates (median age 3 days, range 0–30) accounted for almost a third of the admissions (29.3%, 535/1828), and stayed a median of 6 days (range 1–53) in the hospital. Older children (median age 1 year, range 1 month – 7 years) stayed a median of 7 days (range 1–78) in the hospital.

The clinical outcome was known for 89.3% (1632/1828) of the admissions. Among the 17.0% (277/1632) who died in hospital, 22.4% (n = 62) had laboratory-confirmed bloodstream infection, 17.7% (n = 49) had malaria, 6.5% (n = 18) had concomitant laboratory-confirmed bloodstream infection and malaria-parasitemia and 52.0% (n = 144) had neither. Readmissions comprised 3.2% (n = 59) of all admissions (39 and 1 patient were readmitted once and twice, respectively, 18 patients had been admitted prior to the study).

Information on antimicrobial use was available for 85.2% (1557/1828) of the admissions. The majority of patients (93.8%) received antimicrobial therapy (Table 1) and at least two-thirds (67.2%, 1046/1557) of the patients did so before blood-culture was taken. There was no formal empirical regimen for the treatment of sepsis at the hospital, partly because of the scarcity of local studies on antimicrobial resistance of relevant bacterial isolates. However, most neonates received a regimen of ampicillin + cloxacillin + gentamicin. One-fifth of the neonates received ceftriaxone, in most cases as a second-line regimen in case of unsatisfactory response to the first regimen. In older children there was a more diverse pattern of regimens used, in most cases including one or more penicillins (ampicillin, penicillin G or cloxacillin) and either chloramphenicol or gentamicin. To some extent, accompanying localizing signs did impact on the choice of antimicrobial regimen, most notably in the case of suspected meningitis in children older than 1 month when a regimen based on chloramphenicol and a penicillin would often be used, and in the case of bloody diarrhea when erythromycin was often added. Severe malaria was treated with quinine, while less severe cases were treated with chloroquine in neonates and pyrimethamine-sulfadoxin in older children. Oral thrush was treated with nystatin mixture, but systemic antifungal drugs were rarely used. Ciprofloxacin was not used in children at the hospital.

The incidence of laboratory-confirmed bloodstream infection was 13.9% (255/1828) of all admissions, 15.9% (85/535) among neonates and 13.1% (170/1293) among older children. A single pathogen was recovered from 224 children (12.3%), while 31 (1.7%) had polymicrobial infection with two (n = 26), three (n = 2) or four isolates (n = 3). In total, 294 pathogenic bacterial and fungal isolates were recovered (Table 2). Among all laboratory-confirmed bloodstream infections, half (128/255) were defined as potentially hospital-acquired. Salmonella and Escherichia coli were the most common isolates in community-acquired infections, and klebsiella and Staphylococcus aureus were the most common in hospital-acquired infections (Table 2). Klebsiella was, by far, the most common cause of neonatal bloodstream infection, particularly in early-onset infection (0–6 days of age) where it responsible for approximately one-third of the cases (Table 3). In children older than 1 month of age, salmonellae were the most frequently isolated pathogens (Table 3).

Among all children with suspected systemic infection, those with laboratory-confirmed bloodstream infection had three times increased risk of dying (Table 4, Figure 1), and among those who survived, the duration of hospital stay was significantly longer than for children without laboratory-verified bloodstream infection (median 8 versus 6 days, P < 0.001). Univariate and multivariate analysis of risk factors for death are shown in Table 4 and 5, respectively. The mortality rate from Gram-negative bloodstream infection (45.6%) was more than twice that of malaria (20.2%) and Gram-positive bloodstream infection (16.7%). Positive blood-culture with Enterobacteriaceae, other Gram-negative bacteria and candida were independent risk factors for fatal outcome (Table 5, Figure 2). In a separate logistic regression model examining only the pathogen involved, the significant risk factors for death were, in descending order of odds ratio, growth of Enterobacter spp. (odds ratio 10.5, P = 0.007), Pseudomonas aeruginosa (9.7, P < 0.001), Salmonella Typhimurium (6.4, P = 0.001), E. coli (5.0, P < 0.001), klebsiella (4.5, P < 0.001), and candida (2.5, P = 0.049) (data not shown in table). Growth of Salmonella Enteritidis, Acinetobacter spp., E. faecalis, E. faecium and S. aureus were not significant risk factors for death.

Malaria slides were positive in 21.6% (354/1638). Malaria parasitemia was neither associated with HIV-infection (P = 0.227), nor bloodstream infection caused by any pathogen (P = 0.730), Gram-negative bacteria (P = 0.296), Enterobacteriaceae (P = 0.351), E. coli (P = 0.516), klebsiella (P = 1.000), non-typhoid salmonella (P = 0.664). The case-fatality rate was not significantly higher in patients with malaria parasitemia (20.2%) compared to those without (16.4%, P = 0.102). However, in patients with positive malaria slides, mortality was almost three times higher among those with concomitant Gram-negative bloodstream infection (45.7% versus 17.2%, P < 0.001).

Enterobacteriaceae displayed high rates of resistance to commonly used antimicrobials (Table 6). Only 20% of community-acquired Enterobacteriaceae isolates was sensitive to ampicillin, while two-thirds were sensitive to gentamicin. Two-thirds of community-acquired salmonellae were sensitive to chloramphenicol, while half are resistant to ampicillin and to co-trimoxazole. Isolates of Enterobacteriaceae were almost uniformly sensitive to ciprofloxacin. P. aeruginosa isolates were commonly sensitive to anti-pseudomonas drugs such as ceftazidime, ciprofloxacin and tobramicin (Table 7). ESBL phenotype was found in 18% of the Enterobacteriaceae isolates (E. coli 9/37, klebsiella 9/53, Enterobacter spp. 5/9, salmonella 1/39 and Pantoea spp 2/2) involving TEM-63, SHV-2a, SHV-12 and CTX-M-15 genotypes 13, and in 3 isolates of non-Enterobacteriaceae (one Acinetobacter spp. and the 2 Chryseobacterium spp.). ESBL-producing isolates were resistant to almost all tested antimicrobials except for ciprofloxacin and meropenem.

The majority of S. aureus isolates were sensitive to commonly used anti-staphylococcal agents, including cloxacillin and gentamicin (Table 8). Among the three isolates phenotypically resistant to oxacillin, only one was available for confirmatory detection of the mecA gene by PCR. The child with confirmed MRSA genotype died, the two other survived. Ten of 21 E. faecium isolates showed combined resistance to ampicillin and gentamicin (high-level), but only one patient with bloodstream infection caused by these organisms died. This combined resistance trait occurred in both community-acquired (4/12) and hospital-acquired infections (6/9), and the difference was not statistically different (P = 0.198). Six of 15 E. faecalis isolates were high-level gentamicin-resistant (but not ampicillin-resistant) and were involved in the death of two patients, among whom one had community-acquired infection. The majority of candida isolates were susceptible to fluconazole (96%) and amphotericin B (87%).

Antimicrobial treatment prior to blood culture was significantly associated with resistance to co-trimoxazole (83% versus 69%, P = 0.036) and chloramphenicol (59% versus 42%, P = 0.035) in Gram-negative isolates as a group, with resistance to erythromycin (36% versus 0%, P = 0.014) and chloramphenicol (46% versus 0%, P = 0.005) in S. aureus and chloramphenicol in E. faecalis (89% versus 20%, P = 0.023). In addition, all three MRSA isolates were from patients who had received previous antimicrobial treatment, but this was not statistically significant.

Hospital-acquisition of infection was significantly associated with resistance to amoxicillin-clavulanate and cephalosporins in E. coli and with co-trimoxazole-resistance in klebsiella. The percentage of bloodstream infection episodes caused by an ESBL-producing organism was not significantly different among community-acquired (7%, 9/127) and hospital-acquired infections 13% (16/128, P = 0.206). Patients with hospital-acquired laboratory-verified bloodstream infection did not receive inappropriate antimicrobial treatment more often (41%) than those with community-acquired infection (42%, P≅0.892).

Inappropriate antimicrobial treatment was a risk factors for death irrespectively of whether controls were patients with clinically suspected infection or laboratory-confirmed bloodstream infection (Table 5, Figure 3). In those who survived, inappropriate antimicrobial treatment for the bloodstream-infection was associated with increased duration of hospital stay (median 8 days versus 6 days, P = 0.043).

Hospital-acquired infection was associated with increased duration of hospitalization (median 10 days versus 6 days, P < 0.001), but was not a significant risk factor for fatal outcome in the multivariate analysis (Table 5, Figure 4).

Fifty-one percent (911/1787) of the study subjects were examined for HIV antibodies and 24.0% (219/911) were positive. The HIV-1 PCR was positive in 55.4% (82/148) of children younger than 18 months and 8.9% (5/56) of neonates. Thus, the HIV-1 prevalence was 1.7% (5/302) for neonates, 24.4% (147/603) for older children, and the combined overall prevalence was 16.8% (152/905).

Laboratory-confirmed bloodstream infection was not significantly more frequent in HIV-positive children (17.5, 28/160) than in HIV-negative ones (15.5%, 119/766, P = 0.552). However, bloodstream infection was more frequently caused by non-typhoid salmonella in HIV-positive children (25%, 7/28) than in HIV-negative children (8.4%, 10/119, P = 0.022). There was no such association between HIV-infection and bloodstream infection caused by E. coli (17.9% versus 10.9%, P = 0.339), ESBL-producing organisms (17.9% versus 10.1%, P = 0.321), candida (10.7% versus 10.9%, P = 1.000) or other pathogens, except for bloodstream infection caused by klebsiella, which was negatively associated with HIV infection (7.1% versus 33.6%, P = 0.005). HIV-positive children as a group did not receive more antimicrobial therapy (97.2% versus 93.4%, P = 0.082). However, the given antimicrobial therapy for those with bloodstream infection was inappropriate more often than in HIV-negative children (64.0% versus 40.2%, P = 0.044). HIV-infection was a risk factor for death (Table 5, Figure 5), and associated with significantly longer hospital stay in those who survived (median 9 versus 6 days, P < 0.001).

One sixth (15.5%, 248/1603) of the patients were malnourished. Malnutrition was a risk factor for death (Table 5, Figure 6) and, in those who survived, it was associated with prolonged hospital stay (median 10 versus 6 days, P < 0.001). Other underlying diseases were grouped into infectious diseases, including tuberculosis (n = 97), hepatitis (n = 9) and congenital syphilis (n = 2) and non-infectious conditions, including sickle cell disease (n = 36), cancer (n = 5), congenital heart disease (n = 38), asthma (n = 5), Down's syndrome (n = 11), cerebral palsy (n = 25), congenital malformations (n = 21), epilepsy (n = 10) and hemophilia (n = 2). Other underlying infectious conditions than HIV were a significant risk factor for fatal outcome among all patients with clinically suspected systemic infection.