Sepsis: emerging pathogens and antimicrobial resistance in Ethiopian referral hospitals

Background Sepsis due to multidrug resistant (MDR) bacteria is a growing public health problem mainly in low-income countries. Methods A multicenter study was conducted between October 2019 and September 2020 at four hospitals located in central (Tikur Anbessa and Yekatit 12), southern (Hawassa) and northern (Dessie) parts of Ethiopia. A total of 1416 patients clinically investigated for sepsis were enrolled. The number of patients from Tikur Anbessa, Yekatit 12, Dessie and Hawassa hospital was 501, 298, 301 and 316, respectively. At each study site, blood culture was performed from all patients and positive cultures were characterized by their colony characteristics, gram stain and conventional biochemical tests. Each bacterial species was confirmed using Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry (MALDI TOF). Antimicrobial resistance pattern of bacteria was determined by disc diffusion. Logistic regression analysis was used to assess associations of dependent and independent variables. A p-value < 0.05 was considered as statistically significant. The data was analyzed using SPSS version 25. Results Among 1416 blood cultures performed, 40.6% yielded growth. Among these, 27.2%, 0.3% and 13.1%, were positive for pathogenic bacteria, yeast cells and possible contaminants respectively. Klebsiella pneumoniae (26.1%), Klebsiella variicola (18.1%) and E. coli (12.4%) were the most frequent. Most K. variicola were detected at Dessie (61%) and Hawassa (36.4%). Almost all Pantoea dispersa (95.2%) were isolated at Dessie. Rare isolates (0.5% or 0.2% each) included Leclercia adecarboxylata, Raoultella ornithinolytica, Stenotrophomonas maltophilia, Achromobacter xylosoxidans, Burkholderia cepacia, Kosakonia cowanii and Lelliottia amnigena. Enterobacteriaceae most often showed resistance to ampicillin (96.2%), ceftriaxone (78.3%), cefotaxime (78%), cefuroxime (78%) and ceftazidime (76.4%). MDR frequency of Enterobacteriaceae at Hawassa, Tikur Anbessa, Yekatit 12 and Dessie hospital was 95.1%, 93.2%, 87.3% and 67.7%, respectively. Carbapenem resistance was detected in 17.1% of K. pneumoniae (n = 111), 27.7% of E. cloacae (n = 22) and 58.8% of Acinetobacter baumannii (n = 34). Conclusion Diverse and emerging gram-negative bacterial etiologies of sepsis were identified. High multidrug resistance frequency was detected. Both on sepsis etiology types and MDR frequencies, substantial variation between hospitals was determined. Strategies to control MDR should be adapted to specific hospitals. Standard bacteriological services capable of monitoring emerging drug-resistant sepsis etiologies are essential for effective antimicrobial stewardship.


Introduction
Sepsis is a life-threatening condition resulting from a dysregulated immune response to infection that leads to organ dysfunction [1,2]. In May 2017, the World Health Organization (WHO) recognized sepsis as a major public health problem and called all United Nations (UN) member states to improve sepsis prevention, recognition and management [3]. In 2020, WHO noted that approximately 20% of all-cause global deaths are due to sepsis [4], which affects 49 million people and causes 11 million deaths globally every year [5]. Sepsis disproportionately affects neonates, pregnant women, elderly, patients with severe comorbidities and people living in low-resource settings [2,6]. Various species of Klebsiella, Enterobacter, Acinetobacter, and Pseudomonas, E. coli, and S. aureus are major sepsis etiologies with regional variation [7,8].
Though sepsis is treatable [9], antimicrobial resistance (AMR) contributes to worsening the consequences of sepsis from longer hospitalization to death [10,11]. Resistance to cephalosporins, fluoroquinolones, penicillins, aminoglycosides, monobactams, macrolides and carbapenems is spreading globally [7,8,10,12,13]. In 2015, WHO announced that AMR represents a major threat to human health [14] and in 2019 the UN declared AMR a great threat facing the global community [15]. Findings from around the world showed increasing levels of multidrug resistance, which is worrying for the future [9,16].
In sub-Saharan countries, there is scarcity of data on sepsis etiologies [17] though the problem is assumed to be very large because of limited health access and high rates of other infectious diseases [4,6,12,17]. Similarly, data on sepsis etiologies and AMR from Ethiopia are scanty and only studies done at single sites, with limited sample sizes and inaccurate bacteria identification, are available [18][19][20][21][22]. A recent multi-country study published by Sands et al. [23] focused on neonatal sepsis etiologies and AMR in Ethiopia. However, this study had limitations, since it included only a single hospital and neonates from Ethiopia. It is crucial to identify bacterial etiologies and determine their AMR patterns among sepsis patients at a larger scale, in order to guide future sepsis prevention efforts. Hence, this study aimed to determine bacterial sepsis etiologies and their AMR patterns at four hospitals, which serve millions of people in the central, southern and northern parts of Ethiopia.

Study design and study sites
A multicenter study was conducted between October 2019 and September 2020 at four selected hospitals in central, southern and northern Ethiopia. University and referral hospitals which had established microbiology laboratories or a link with a nearby government regional microbiology laboratory were selected purposefully. These were Tikur Anbessa Specialized Hospital (TASH) and Yekatit 12 Specialized Hospital Medical College (Y12HMC) in the central region, Hawassa University Comprehensive Specialized Hospital (HUCSH) in the southern region and Dessie Referral Hospital (DRH) in the northern region.
TASH is the teaching hospital of Addis Ababa University located in Addis Ababa, the capital of Ethiopia. The hospital, with more than 800 beds, is the main referral hospital and the oldest hospital in the country, staffed with the most senior specialists. It provides tertiary level referral diagnoses and treatment for approximately 400,000 patients yearly who referred from all over the country.
DRH is one of the largest public hospitals in the northern part of Ethiopia, located in Dessie town, 400 km from Addis Ababa. It is a referral hospital with 560 beds, providing services for the surrounding areas with an estimated population of 5 million and to residents of the neighboring regions.
Y12HMC is located in Addis Ababa and provides health care services to Addis Ababa residents, referral cases from health centers in Addis Ababa and its bordering regions. The college trains medium-and higher-level health professionals. The hospital has over 300 beds and serves for more than 5 million people in its catchment area.
HUCSH is located in Hawassa city in southern Ethiopia, 280 km from Addis Ababa. HUCSH is one of the largest health facilities in the southern part of the country and provides teaching, public health services and research activities. It serves more than 20 million people locally and in the neighboring regions. Currently, the hospital has over 400 beds and provides patient care to 90,200 outpatients, 18,100 hospitalized patients and 1100 emergency cases annually.
While three hospitals had established microbiology laboratories, DRH had not started performing services capable of monitoring emerging drug-resistant sepsis etiologies are essential for effective antimicrobial stewardship. bacteriological culture and antimicrobial susceptibility testing at the time of this study. Therefore, the nearby located Amhara Public Health Institute-Dessie Branch microbiology laboratory was used for blood culture processing and antimicrobial susceptibility testing.

Patient recruitment and sample size calculation
All patients with suspected cases of sepsis who sought medical service at the study sites were included in the study population. We used the attending physician's decision to identify eligible patients as sepsis cases. All age groups were included, but patients who had been on antibiotic treatment within the preceding ten days were excluded from the study. A total of 1416 clinically diagnosed cases of sepsis from different wards were enrolled in the study. The sample size was calculated based on a single sample size estimation formula (n = Z 2 P (1-P) / d 2 ) using a proportion of 50% (P = 0.5) [24], due to lack of previous similar multicenter studies. As this was a multicenter study, increasing the sample size was necessary; hence, a precision (d) of 0.03 was used to maximize the sample size and the accuracy of the prevalence estimate. Z stands for Z statistic with the level of confidence of 95%, which is conventional where the Z value is 1.96. With a 10% non-response rate, the total sample size came to 1174 and distributed equally across the four study sites. Keeping the minimum sample size allocated to each study sites, we enrolled a total of 1,416 patients clinically investigated for sepsis to identify sepsis etiologies and characterize AMR patterns of bacterial isolates.

Data collection
Professional nurses who had experience of sample collection for blood culture and microbiologists who were working in the bacteriology laboratory were recruited as data collectors. Training and clear instructions were given to all data collectors on sociodemographic and clinical data collection, blood sample collection and transportation to bacteriology laboratories and blood culture processing. A single bottle blood culture system was used because introducing three blood culture bottles for a single patient was found to be difficult at the study sites due to several reasons raised from data collectors including manpower and laboratory settings shortage. Blood cultures, bacterial identification and drug susceptibility testing were performed in accordance with a standardized laboratory protocol that was applied in all study sites. The findings of each blood culture were communicated to attending physicians for patient management. All bacterial strains were stored at − 70 °C and transported to the Armauer Hansen Research Institute (AHRI) and Sweden for further characterization.

Blood sample collection and transportation
Immediately after the patient was identified, an appropriate blood sample volume was collected and a single blood culture bottle processed for each patient. From children who were < 1, 2-5, 6-11 and 12-17 years old, 1 ml, 2 ml, 3-5 and 5-10 ml was collected, respectively, and from adults, 10 ml. A 1:5 and 1:10 blood to broth dilution was made for children and adults, respectively. Blood samples were collected aseptically using 70% alcohol and 2% tincture of iodine and then transferred to blood culture bottles containing brain heart infusion broth (Oxoid Ltd, UK) and transported to a bacteriology laboratory for culturing and identification.

Blood cultures and bacterial identification
All blood culture bottles were incubated aerobically at 37 °C for seven consecutive days and inspected daily for signs of bacterial growth. Blood samples that grew turbid before the seventh day and blood samples that were non-turbid on the seventh day were sub-cultured on blood agar (Oxoid Ltd, UK) and MacConkey agar (Oxoid Ltd, UK) at 37 °C for 24 h. At each study site, all positive cultures were characterized and identified by their colony characteristics, gram staining and conventional biochemical tests. Triple sugar iron, indole, urea, citrate, lysine decarboxylase, motility and malonate biochemical media were used for identification of gram-negative bacteria. Biochemical tests used to identify gram-positive bacteria at species level were catalase, coagulase and fermentation on mannitol salt agar. Isolations of coagulase negative staphylococci and Bacillus species were considered to be possible contaminants, since a single blood culture bottle was used which limits the ability to differentiate between pathogens and contaminants.

Bacterial strain confirmation using matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS)
All bacteria were re-identified and confirmed using MALDI-TOF MS at the Clinical Microbiology Department of Uppsala University Hospital, Uppsala, Sweden, and Karolinska Institute, Stockholm, Sweden. From fresh cultures, a single colony of bacteria was smeared onto a MALDI-TOF plate and the sample was air-dried. Next, 1 µl formic acid was added to each cell, air-dried, and then 1 µl MALDI matrix solution was applied to the cells and air-dried before reading. MALDI-TOF identification was automatically scored by the system software as between 1 and 3 points. All isolates with score 2 and above were accepted and all results below 1.7 and flagged red were rejected. Samples with score 1.7-2 and flagged yellow were re-analyzed.

Drug susceptibility testing
The susceptibility to antibiotics of bacterial isolates was analyzed using disk diffusion after 16-18 h of incubation at 37 °C. Each zone of inhibition was measured to the nearest millimeter and interpreted as sensitive, intermediate or resistant based on the standardized table supplied by the Clinical and Laboratory Standards Institute [25]. Using a sterile wire loop, 3-5 pure colonies were picked and emulsified in nutrient broth (Oxoid). Standard inocula were adjusted to 0.  [25]. All antibiotics discs were OXOID products (Oxoid Ltd, UK). A bacterium that was simultaneously resistant to three or more antibiotics in different classes was considered MDR.

Quality assurance and quality control
At each study site, blood samples were collected and transferred to blood culture broth in accordance with standard operating procedures designed to all sites. Immediately after collection, blood culture bottles were transported to the microbiology laboratory for analysis. Each laboratory test was processed in accordance with established protocols and recorded carefully. Standard operating procedures were followed strictly for each laboratory test. E. coli ATCC 25,922 and/or Staphylococcus aureus ATCC 25,923 were used as quality control strains for culture and drug susceptibility testing throughout the study. Each MALDI-TOF run also included quality control strains using E. coli ATCC 25,922.

Statistical analysis and interpretation
Descriptive statistics (mean, percentages or frequency and standard deviation) were calculated. Association of possible risk factors with sepsis was assessed using univariate and multivariate logistic regression analysis. A p-value < 0.05 was considered as statistically significant. The data were analyzed using SPSS version 25.

Blood culture findings
In the present study, a total of 1,416 patients from four different hospitals were investigated for sepsis. The number of patients from TASH was 501, and the numbers from Y12HMC, DRH and HUCSH were 298, 301 and 316, respectively (

Multivariate analysis of patient demographic characteristics with blood culture finding among patients investigated for sepsis
Possible risk factors for sepsis were identified and their association was assessed using logistic regression.
The multivariate analysis showed a statistically significant association of blood culture positivity among age groups < 29 days (p = 0.001), ≥ 30 days to ≤ 1 year (p = 0.002), > 1 to ≤ 5 years (p = 0.000) and > 5 to < 18 years (p = 0.001). Medical ward (p < 0.019) and pediatric ward (p < 0.000) showed statistically significant association for blood culture positivity that yielded growth of pathogens (Table 2). While the univariate analysis indicated statistically significant association with EOPD (p < 0.008) with blood culture that yielded pathogens growth, the multivariate analysis did not show statistically significant association. Having underlying disease (p < 0.041) was another variable that showed statistically significant association with growth of pathogens. On the other hand, the multivariate analysis did not show any statistically significant association between gender, referral history, previous hospitalization, hospital stay duration, duration of fever, BMI and previous antibiotic treatment with growth of pathogens (Table 2).

Frequency and distribution of bacterial isolates
A total of 426 pathogenic bacteria were isolated from all blood cultures. Gram-negative isolates (89.7%) were the most frequent, while gram-positive isolates accounted for 10.3% (Fig. 1). Double bacterial growth was detected in 41 cultures. Klebsiella pneumoniae was  Figure 1 shows the frequency of bacterial isolates identified at the four hospitals.
At Y12HMC (n = 57), K. pneumoniae and A. baumannii were frequently isolated pathogens with a proportion of 63.2% and 12.3%, respectively. A Salmonella species (n = 1) was isolated as a rare pathogen, with B. cepacia (n = 1) and S. maltophilia (n = 1) identified as rare sepsis pathogens. At DRH, the most frequent pathogenic bacteria were K. variicola with a proportion of 35.1%, followed by P. dispersa (14.9%), E. coli (11.9%) and K. pneumoniae (10.4%). Almost all P. dispersa were detected at this hospital. Salmonella species (n = 1), R. ornithinolytica (n = 1) and K. cowanii (n = 1) were rare sepsis pathogens at DRH. No pathogenic gram-positive isolates were detected at this hospital and all gram-positive isolates were considered to be possible contaminants.
At HUCSH, K. variicola (28.6%) was the most frequent species, followed by K. pneumoniae with a proportion of 26.5%. While Enterobacter species were commonly isolated in other hospitals, only one E. cloacae strain was detected at HUCSH. Most Enterococcus species were detected at HUCSH, with E. faecium identified frequently at 12.2%. Rarely identified Enterococcus species were E. italicus and E. faecalis. S. aureus was another gram-positive coccus identified at this hospital, with a proportion of 8.2%. Figure 2 shows the frequency and distribution of bacterial isolates identified at each hospital.

Discussion
In the current study, the overall pathogenic bacterial growth among patients investigated for sepsis was 27.2%, with only 0.3% due to yeast cells. In total, 13.1% possible contaminants yielded growth, of which some could be real pathogens, especially for neonates, patients who had underlying diseases and immunocompromised patients [8]. There was substantial variability in the frequency of positive blood cultures among patients investigated for sepsis between hospitals: 38.2% in northern Ethiopia, 29.1% in southern Ethiopia and 24.6% and 18.5% at the two hospitals in central Ethiopia. Explanations could be institutional infection prevention strategies and their implementation, professional capacity to identify sepsis cases early, awareness of sepsis occurrence rate and diagnosis within institutions, and availability and utilization of laboratory facilities [2,17,26]. The most frequent sepsis etiologies were K. pneumoniae (26.1%), K. variicola (18.1%) and E. coli 12.4%. Klebsiella species have been found to cause substantial morbidity in sub-Saharan Africa [12]. Our finding was in line with a study from India that reported gram-negative bacteria as the primary sepsis etiologies, with 23% of sepsis due to K. pneumoniae alone [27]. While E. coli was the third most frequent of all blood culture isolates, most were isolated at TASH and DRH, and rarely at the other hospitals. Similar studies have shown that E. coli is a commonly detected pathogen in sepsis patients [16,28]. Other common gram-negative isolates were A. baumannii 8%, E. cloacae 5.2%, P. dispersa 4.9% and P. aeruginosa 4%. S. aureus and E. faecium were commonly detected gram-positive isolates, mirroring results of other studies elsewhere though they were isolated at TASH and HUCSH only [17,28,29].
Notable disparities on the type and frequency of bacteria isolated as sepsis etiologies were seen between hospitals. While K. pneumoniae was the primary sepsis etiology at TASH and Y12HMC, K. variicola was the primary causative agent of sepsis at DRH and HUCSH. Almost all K. variicola were detected at these two hospitals, located in the north and south of the country. This species is currently recognized as an emerging pathogen that can cause severe infections in humans, and can also colonize plants, insects and animals [30]. A study from Stockholm showed K. variicola as a common blood stream pathogen, though it was inaccurately identified as K. pneumoniae initially using classical species identification methods [31]. Almost all P. dispersa 95.2% were isolated in the northern part of the country, at DRH. This was the second most frequent isolate in blood cultures from this site. It was believed that P. dispersa lives in plants, soil and water and rarely caused human infections until a first neonatal sepsis was reported from India [32] and later adult sepsis in Japan [33]. Another case report from India showed that P. dispersa was detected in a septic patient admitted to ICU; this was reported as the next emerging ICU scare [34]. In our study, most P. dispersa were isolated at the neonatal ICU while two strains were isolated from adult patients admitted to the emergency outpatient department at DRH. Only one strain was isolated at TASH (central Ethiopia); this was in an adult patient hospitalized in a medical ward. Uncommon sepsis etiologies identified were A. xylosoxidans, B. cepacia, L. amnigena, K. cowanii, S. maltophilia and R. ornithinolytica. Such emerging pathogens in patients investigated for sepsis could create challenges in the future [35][36][37]. This is the first time these species, together with K. variicola and P. dispersa, were reported in patients investigated for sepsis in Ethiopia. Isolation of such emerging sepsis etiologies highlights the need for institutional-based diagnostic and intervention strategies.
Enterobacteriaceae showed an MDR frequency of 83.3% and a majority of them were frequently resistant to ampicillin, ampicillin-sulbactam, aztreonam, ceftriaxone, cefotaxime, cefuroxime, ceftazidime, cefepime and gentamicin. Carbapenem resistance among Enterobacteriaceae was 9.4%. Available effective treatment options for Enterobacteriaceae were limited to piperacillin-tazobactam, amikacin and meropenem. Enterobacteriaceae isolated at DRH showed low resistance to tested antibiotics compared with Enterobacteriaceae isolated at other hospitals. The occurrence of higher antibiotic resistance among Enterobacteriaceae was in line with studies from other countries [9,10,12]. K. pneumoniae, the most frequent isolate, showed the highest resistance (> 80%) to ampicillin, cefotaxime, ceftriaxone, ceftazidime, cefuroxime, aztreonam, cefepime, SXT and gentamicin. In our study, an alarming level of carbapenem-resistant K. pneumoniae (17.1%) was detected, though it was lower than in another study [13]. A. baumannii showed high resistance (> 80%) to cefotaxime, ceftazidime, cefepime and SXT, with an overall MDR frequency of 91.2%. This high MDR frequency of A. baumannii was in line with a study from India [13. A. baumannii also showed a remarkably high resistance to meropenem (58.8%). Our findings had similarities with studies from Asia [11,13]. The occurrence of meropenem resistance among infrequently detected A. nosocomialis could be a sign of spreading of carbapenem resistance between Acinetobacter species.
The strengths of this study include the selection of hospitals in multiple parts of the country, enrollment of all age groups, a reasonably large sample size and re-characterizing bacteria using an advanced bacterial identification method. On the other hand, the following three limitations should be considered. First, additional sepsis cases might be missed when data collectors were unavailable during blood sample collection. Second, it was impossible to use two or three blood cultures, which could have increased the growth yield and enabled categorization of coagulase negative staphylococci as true pathogens or contaminants. Third, anaerobic incubation of blood culture that could enable the identification of strict anaerobes was not applied.

Conclusion
In this multicenter study, frequent and diverse gramnegative sepsis etiologies were identified, with substantial variation in primary etiologies between hospitals in different parts of the country. Gram-negative isolates were the primary causative agents of sepsis at all hospitals. Isolation of emerging bacterial strains in all sites showed the growing epidemiology and diversity of sepsis etiologies. High antimicrobial resistance was detected with varying frequency between hospitals. These findings could be taken as a call for strategies to control sepsis occurrence and curb the spread of multidrug resistance. Strategies to control antimicrobial resistance should include context-specific measures. High-quality bacteriological services capable of monitoring emerging drug-resistant bacterial agents of sepsis are essential for antimicrobial stewardship programs to be effective.