The epidemiology of paediatric bloodstream infection (BSI) in Africa is poorly documented. A meta-analysis of prospective studies of community-acquired BSI [1] identified 22 eligible studies (four in Southern Africa) [2-5], where non-typhoidal Salmonella, E. coli, S. aureus and S. pneumoniae infection predominated. Despite published descriptions of community-acquired sepsis in African children [1-9] data on hospital-acquired BSI are extremely limited [10,11]. It is estimated that healthcare-associated BSI may be responsible for 25000 deaths in African children annually [12]. Overall, incidence rates of healthcare-associated infection in developing countries are thought to be at least double that of high-income settings [13]. Further research on the epidemiology of hospital-acquired BSI in African children is needed to quantify the burden and better understand contributory factors.

Significant changes in BSI epidemiology among African children are expected, given increasing access to HIV prevention programmes, paediatric antiretroviral therapy and inclusion of pneumococcal conjugate vaccine (PCV) in the immunisation schedule [14,15]. In light of globally increasing antimicrobial resistance [12], regional antimicrobial resistance prevalence and the efficacy of empiric antibiotic therapy for paediatric BSI in Sub-Saharan Africa also require evaluation. Descriptions of increasing prevalence of Enterobacteriaceae BSI isolates producing extended-spectrum beta-lactamases (ESBL) are concerning given the limited availability of appropriate antibiotics in many African countries [16].

The difficulties in obtaining paediatric blood culture specimens and the increased yield with higher volume blood inoculum are well described [8]. Of concern in our region are two recent audits [17,18] reporting low pathogen yields and high blood culture contamination rates (exceeding accepted rates of < 3%) [19]. We examined trends in paediatric BSI epidemiology over six years at a single academic institution, determining rates of bacteraemia, blood culture yield and contamination, BSI-associated mortality and prevalence of antimicrobial resistance. We also investigated the association between bacteraemia, antimicrobial resistance and HIV infection.

For 63209 children hospitalized over the study period, 17001 blood culture specimens were submitted; 1 blood culture per 3.7 admissions or 68.6 specimens/1000 patient days. From 935 culture–positive specimens, 979 BSI pathogens were isolated (5.5% yield; 95% CI 5.3-5.7%). Blood culture contamination rates were high (1123 contaminated blood cultures from 17001 blood culture specimens submitted [6.6%]; 95% CI 6.4-6.8%), increasing over time (p = 0.003) (Figure 1). Coagulase-negative staphylococci (CoNS) were the most commonly isolated contaminant (650/1123; 57.9%), followed by non-pathogenic streptococci (75/1123; 6.7%), Bacillus species (74/1123; 6.6%), Micrococcus species (63/1123; 5.6%) and diphtheroids (61/1123; 5.4%).

Of 864 discrete BSI episodes, 818 (94.7%) were monomicrobial infections and 46 (5.3%) were polymicrobial (42 with 2 pathogens; 4 with 3 pathogens). The overall BSI rate was 3.5 per 1000 patient days (95% CI 3.3–3.7). A significant decline in the BSI rate from 4.6 in 2008 to 3.1 in 2013 (Chi square for trend p = 0.02; Figure 1) was noted despite substantial increases in hospitalizations (7537 to 11201 [49%]) and number of blood cultures (7816 vs 9185 [17%]) submitted between 2008 and 2013.

While H. influenzae BSI episodes increased between 2008–2010 and 2011–2013 (7/465 [1.5%] to 15/399 [3.8%]; p = 0.05), the proportion of H. influenzae serotype B isolates remained similar (3/7 [42.9%] vs 5/15 [33.3%]; p =0.65). However, S. pneumoniae BSI declined significantly (58/465 [12.5%] to 33/399 [8.3%]) (p = 0.04). Pneumococci as a percentage of all Gram positive pathogens declined from a pre-vaccine high of 43.5% (27/62) in 2008 to 13.1% (8/61) in 2013 [p < 0.001]. The proportion of pneumococcal BSI due to vaccine-serotypes (accounting for PCV7 and PCV13) did not decrease significantly over the two time periods, despite PCV-13 giving broader coverage (20/68 [29.4%] vs 6/23 [26%], p = 0.99). The proportion of pneumococcal BSI exhibiting penicillin-resistance remained unchanged between 2008–2010 and 2011–2013, (18/58 [31%] vs 10/33 [30.3%]; p = 1.0).

Prevalence of antimicrobial resistance was assessed among a subset of pathogens, focussing on fluconazole resistant Candida species and four bacterial pathogens: MRSA, multi-drug resistant Acinetobacter baumannii and ESBL-producing E coli and K. pneumoniae (Figure 2). No carbapenem resistant Enterobacteriaceae (CRE) or vancomycin resistant Enterococci (VRE) were isolated. For the four selected bacterial pathogens, antimicrobial resistance prevalence among community isolates was 25% compared with 70% among hospital-acquired isolates (p < 0.001). For A. baumannii and MRSA, resistance rates were significantly higher among hospital-acquired isolates, whereas E. coli and K. pneumoniae isolates had similar prevalence of ESBLs among hospital- and community-acquired isolates. Among the 65 Candida species, all 21 C. albicans isolates were fluconazole susceptible, while 22 of the 44 (50%) non-albicans Candida species were fluconazole resistant. The prevalence of antimicrobial resistance did not differ significantly between 2008–2010 and 2011–2013 for the selected pathogens [Chi square for trend p = 0.24 and p = 0.14 respectively]. Factors associated with antimicrobial resistance on multivariate analysis included hospital-acquired infection, infancy, HIV-infection and Gram negative sepsis (Table 4).

Variable assessed	Univariate analysis (p-value)	Multivariate analysis (p-value)	Odds ratio	95% CI
Length of stay prior to BSI onset	<0.001	0.53	-	-
Age category (infants)	<0.001	0.003	1.92	1.2 – 3.1
Gender	0.8	0.92	-	-
HIV status (positive)	<0.001	<0.001	2.64	1.7 – 4.2
Year of BSI	0.4	0.19	-	-
Place of BSI onset (hospital-acquired)	<0.001	<0.001	3.68	2.7 – 5.1
Type of BSI pathogen
- Gram negative	<0.001	<0.001	1.99	1.4 – 2.9
Mono- vs poly-microbial BSI	0.18	0.84	-	-
ICU vs general ward at BSI onset	<0.001	0.06	-	-

BSI = bloodstream infection; ICU = intensive care unit; Hospital-acquired BSI = BC submitted > 72 hours after admission. To determine factors associated with antimicrobial resistance, binary logistic regression analyses were performed. A p-value below 0.05 was considered statistically significant.

In keeping with previous studies from Africa, gram negatives predominated in our study. E. coli and Klebsiella spp. were the most prevalent Enterobacteriaceae. Non-typhoidal salmonellae, a prominent BSI pathogen in malaria-endemic regions [27], was uncommon in our study. S. pneumoniae (the most common isolate in community-acquired BSI in African children [1]), was also prominent in our cohort. In addition, the profile of gram positive isolates changed significantly over time owing to reduced frequency of S. pneumoniae detection. The proportion of vaccine-serotype pneumococcal BSI isolates remained stable over time suggesting that the dramatic decline in pneumococcal sepsis rates is not solely attributable to vaccine. However, a study of invasive pneumococcal disease in South Africa did demonstrate a substantial reduction of 89% in disease caused by PCV7 serotypes in children <2 years old [28]. Other factors such as increasing antiretroviral uptake and decreasing HIV prevalence probably contributed. Fungi (especially hospital-acquired) and gram negatives were significantly associated with BSI mortality, as in other African studies [6,10,11].

The burden of paediatric BSI was concentrated among infants (58.6% of the cohort). Young age was not associated with mortality from BSI but was significantly associated with antimicrobial resistant pathogens. Overall BSI mortality in our cohort (20.4%) was higher than that reported for high income settings (11 - 14%) [29,30], but lower than other African settings (27 – 38%) [6,10,11]. Access to intensive care was not reported for the other African BSI studies and likely contributed to the lower case fatality in our cohort, although PICU facilities at our institution are very limited (only 1 in 5 patients was admitted in PICU at BSI diagnosis).

As previously reported [10], HIV-infected children were at increased risk for BSI-associated mortality and more likely to have antimicrobial resistant pathogens. Bacterial colonisation is a risk factor for later invasive infection with high rates of colonisation with antimicrobial resistant pathogens described among HIV-infected children in Cape Town [31]. It was not possible (given the study design) to compare the relative risk for BSI among HIV-infected versus HIV-uninfected children or the effect of antiretroviral therapy. A prospective study is underway at our institution to determine the relative risk for hospital-acquired infection (including BSI) among HIV-infected children.

Hospital-acquired BSI was common (nearly half of all BSI episodes), more prevalent among infants and significantly associated with mortality. The profile of hospital-acquired pathogens was distinct from that of community-acquired BSI with Klebsiella, Acinetobacter and Candida species predominating. In keeping with the only prospective study of hospital-acquired bacteraemia in African children, Acinetobacter sepsis had the highest case fatality rates in our cohort. The rate of hospital-acquired BSI in our study exceeded the rate of nosocomial bacteraemia in rural Kenya (1.63 versus 1.0 episodes per 1000 patient days), however our cohort had a substantially higher HIV prevalence (13.4% vs 2%) [11]. Unsurprisingly, hospital isolates exhibited significantly greater antimicrobial resistance. Unlike the Tanzanian cohort [10], antimicrobial resistance was not associated with BSI mortality, possibly due to our use of carbapenems for empiric treatment of hospital-acquired sepsis.

A worrying observation is the high rate of antimicrobial resistance among community-acquired pathogens, especially. E.coli and Klebsiella spp. Inappropriate empiric antimicrobial therapy (due to ESBL-producing and multi-resistant pathogens) predicted death in the Tanzanian cohort (OR 12.9) [10]. It is possible that pre-hospital antibiotic administration in our cohort may have falsely elevated antimicrobial resistance rates for community-acquired pathogens, by decreasing the frequency of isolation of susceptible pathogens. However our data are in keeping with pooled laboratory data (2010–2012) from public sector hospitals in South Africa demonstrating ESBL-carriage in 68% of 2774 K. pneumoniae BSI isolates [32]. Ongoing surveillance of antimicrobial resistance in community BSI (and monitoring of clinical outcomes among children given ineffective antibiotic therapy) is needed to determine if antibiotic guidelines need revision.

Among hospital-acquired BSI in the last two years of the study period, isolates exhibited highest susceptibility (83.6%) to meropenem plus amikacin, not currently our recommended empiric regimen. A prospective review, which includes clinical data on response to therapy is urgently needed to inform our guidelines for treatment of hospital-acquired sepsis. However, given the need for antimicrobial stewardship (and restriction of carbapenem use), it is also important to assess the efficacy of narrower-spectrum regimens (such as piperacillin-tazobactam and amikacin).

Our review of temporal trends in paediatric BSI epidemiology identified targets for quality improvement. Although not unique to our setting, blood culture contamination rates were high (double the international norm [19]) and increased over time (exceeding the rate of pathogen isolation). The annual pathogen yield declined significantly over time and in 2013 (4.9%; 95% CI 3.9 – 5.2) was substantially lower than blood culture yields from a systematic review in Africa (8.2%; 95% CI 7.9 – 8.4) [1]. There are several explanations for these findings including poor aseptic technique during specimen collection with failure to isolate pathogens because of overgrowth by c ontaminants, low sensitivity of paediatric blood cultures and sub-optimal blood volumes from children [8]. Prior administration of antibiotics also contributes to low blood culture yield: local management guidelines [33] recommend that critically-ill children referred in from primary care receive a single dose of intramuscular ceftriaxone prior to transfer.

Between the two study time periods, BSI rates declined significantly despite a substantial increase in hospitalization (measured by increased inpatient days) and despite an increase in actual numbers of blood culture specimens submitted. We suspect that improvements in PMTCT programmes, paediatric antiretroviral coverage and PCV introduction between 2008 and 2013 are partly responsible. However, declining pathogen yields (as described above) may also have artificially reduced BSI rates.

This study has several limitations, most importantly the possibility that some healthcare-associated BSI (re-admission within 30 days of hospital discharge) may have been misclassified as community-acquired, owing to the retrospective study design. We chose 72 hours as a more conservative cut-off for hospital-acquired bacteraemia (many authorities use 48 hours) to avoid possibly including some community bacteraemias as hospital-acquired. The time of blood culture collection and the time of developing symptoms/signs of infection (as opposed to blood culture collection) were not routinely available. The use of 72 hours as a cut-off, may have slightly underestimated the nosocomial bacteraemia rate. We were unable to evaluate the appropriateness of empiric therapy for community-acquired BSI, as the locally recommended treatment regimen depends on the child’s clinical presentation. Lack of standardized patient selection or technique for blood culture, lack of comprehensive clinical data or information on recent antibiotic use and/or hospitalisation are also limitations. Many patients had unknown HIV status, but were likely tested at a different facility.

Although BSI rates declined over time, we could not determine which factors and practices contributed to this trend. In addition, the BSI rates may have been significantly underestimated owing to pre-hospital antibiotic administration and inadequate blood volumes submitted for culture. We believe that change in the laboratory culturing system and average volume of blood inoculum are less likely explanations for the decline in BSI rate, since the blood culture contamination rate increased significantly over time. However, differences in pathogen yield between the two systems have been described, and the impact of a different blood culture system on the BSI rate cannot be completely discounted [34-36].

Given the relatively good resources and care at our institution (including an infection prevention and control service and PICU facilities), these findings may be more generalizable to better-resourced African settings. Recommendations for local practice arising from this study include urgent review of paediatric blood culture practice (emphasizing aseptic technique and adequate blood inoculum) and review of empiric antibiotic therapy for both community and hospital-acquired BSI.

Children with BSI experienced high mortality, particularly for hospital-acquired infection. S. pneumoniae BSI declined after introduction of PCV and increasing antiretroviral coverage. Pathogens (both community- and hospital-acquired) exhibited substantial antimicrobial resistance. Although BSI rates declined, blood culture contamination rates increased; blood culture sampling technique and local options for empiric antimicrobial therapy require re-evaluation.