Infection with ARO results in delay in starting effective antibiotic therapy, fewer possible treatment options and increased morbidity and mortality, with prolonged hospital stay and greater costs of hospitalisation [1]. Antibiotic stewardship, including appropriate choice and administration of antibiotics, deescalation of therapy, and a multidisciplinary team approach to managing neonatal sepsis, is recommended to limit inappropriate antibiotic use and prevent the development of ARO [1].

Empiric antibiotic therapy is based on surveillance of antimicrobial sensitivity patterns in culture isolates. Pathogens vary considerably between different neonatal units. In many hospitals, gram-positive organisms cause up to 70% of nosocomial infections in neonates [1] with coagulase-negative staphylococci (CoNS) accounting for more than half of these [2]. In developing countries, gram-negative organisms may be far more prevalent as neonatal pathogens [3], with a higher incidence of antimicrobial resistance. Pathogens also vary over time, for example, the first outbreak of multiple-drug-resistant A. baumannii in extremely low birth weight infants in the USA was described in 2004 [4]. In a survey over 10 years in a tertiary care neonatal unit in India, E. aerogenes was the most common cause of late-onset sepsis (LOS) during the first 5 years, whereas S. aureus became the main organism causing LOS in the latter 5 years [5]. Ongoing surveillance of microbiological isolates and their sensitivity patterns is therefore an essential part of antibiotic stewardship, in particular to guide the selection of empiric therapy.

Blood cultures were performed on babies with a history of obstetric risk factors (e.g., chorioamnionitis, prolonged rupture of membranes) or who presented with signs of suspected sepsis, such as apnoea, lethargy, temperature instability, respiratory distress, hypotension, poor feeding, glucose imbalance, and seizures. A sterile 0.5mL blood sample was sent for culture prior to commencing antibiotic therapy, together with a full blood count and C reactive protein (CRP). The CRP was repeated after 24 hours.

During the study period, EOS was treated empirically with penicillin and amikacin. Empiric therapy for LOS was modified in consultation with the microbiology unit in response to the changing antibiograms prevalent in the unit at the time. Positive blood cultures were each reviewed with the microbiology unit and a joint decision was taken on the significance of the organism and appropriate de-escalation of antibiotic therapy was instituted. A BSI was diagnosed if a pathogenic organism was isolated, the baby was clinically ill, and there were nonspecific laboratory signs of sepsis including a raised CRP, abnormal white cell count, or platelet count. Certain organisms, including Micrococcus species, Bacillus species, and Corynebacterium species, were classified as contaminants. CoNS were regarded as contaminants if the baby was clinically well, the CRP was normal, and there were no indwelling central catheters. Organisms classified as contaminants were excluded from the analysis. If the same organism was isolated from the same patient within 7 days, this was considered to be a single episode of BSI.

Limited clinical and demographic information on each patient was obtained from a neonatal computer database. This included obstetric risk factors for infection, antenatal care, use of antenatal corticosteroids, outcome, duration of hospitalization, gender, birth weight, gestational age, place of birth, birth asphyxia, HIV exposure, nasal continuous positive airways pressure (NCPAP), mechanical ventilation, necrotizing enterocolitis (NEC) grade 2 and 3 [6], surgery, and intraventricular haemorrhage (IVH) grade 3 and 4 [7]. Each pathogenic organism isolated was considered as a separate episode of BSI—if 2 pathogenic organisms were isolated from a single patient, that patient's clinical information would be entered for each separate organism. Variables were entered as yes, no, or unknown. Information regarding total patient days and total live births during the study period was obtained from admission and labour ward statistics. The live births included those from Hillbrow Community Health Centre and Alexandra Clinic.

Statistical analysis was done using SPSS version 19 (IBM). Categorical variables were described as frequencies and percentages and compared using Chi-square analysis. Fisher's exact test was used in those instances where the anticipated size of a cell was below 5. If normally distributed, continuous variables were described as mean and 95% confidence intervals and means were compared using the unpaired t-test. If the distribution was not normal, the data was described using median and interquartile range and compared using Mann-Whitney U analysis. All data was compared considering SRD, EOS, and Gram stain as outcome variables. Significant differences are reported.

Multiple gestation was significantly related to SRD—11/32 (34.3%) versus 14/163 (8.5%) (Chi-square 16,025 P < 0.001). Shock occurred more frequently in SRD—4/32 (6.7%) versus 2/163 (1.2%) (Chi square 11.86 P = 0.008), and SRD was significantly associated with NEC—9/163 (5.5%) versus 8/32 (25%) (Chi square 12.75 P = 0.002). The duration of hospital stay was significantly shorter in cases of SRD—11.78 days (SD 10.9) compared to 37.28 days (SD 22.03) (P < 0.001). The median birth weight of babies that died was 1045 grams (IQR 700 grams) as compared to 1200 grams (IQR 565) in survivors. This was not statistically different (P = 0.08). All other clinical parameters were not significantly associated with SRD.

Guidelines for empiric antibiotic treatment in this unit are informed by ongoing surveillance of the spectrum of bacteria causing neonatal sepsis and their antibiotic resistance profiles. Clinicians and microbiologists review sensitivity profiles and adjust empiric antibiotic therapy accordingly. Based on the bacterial profile in the present study, suitable empiric antibiotic therapy for EOS would be penicillin plus an aminoglycoside. Penicillin will provide adequate cover for S. agalactiae. The addition of an aminoglycoside to the penicillin broadens the antimicrobial cover and provides synergy. Both gram-positive and gram-negative empiric cover is required for suspected LOS. Vancomycin will provide cover for CoNS, MRSA, and E. faecium. The clinical significance of CoNS in sepsis is difficult to assess, and the SRDs for CoNS in this audit were 7% (3/43). However, SRDs for MRSA and E. faecium were 18% each (3/16 for MRSA and 2/11 for E. faecium). Due to the differing antibiograms for the K. pneumoniae ESBLs and the A. baumannii's, a single Gram-negative antimicrobial agent will not provide optimal empiric cover. In critically ill neonates meropenem plus either gentamycin or ciprofloxacin is indicated. Alternatively, the combination of ceftazidime and amikacin will provide cover for the majority of the Enterobacteriaceae and non-fermenters. Once the offending pathogen is known, de-escalation of treatment is imperative. Ongoing surveillance of neonatal sepsis is essential in monitoring trends in infection, ensuring up-to-date appropriate antibiotic therapy, early outbreak detection, and institution of preventative measures. Continuous infection control awareness is vital in light of the predominance of LOS due to ARO like K. pneumoniae ESBL and MDR A. baumannii in this unit.

The rate of neonatal BSI in the CMJAH unit is comparable to that reported from other developing countries but is much higher than that in developed countries. A neonatal unit in the Republic of Korea had 15.1 infection in 10.2 neonates per 1000 patient days [10], whereas the rate of neonatal sepsis in developed countries is reported to be between 5 and 6 per 1000 patient days reported from developed countries [11, 12]. Note that there is a great variation in rates of nosocomial infection between different neonatal units, even in developed countries [11].