Antimicrobial Resistance in Invasive Bacterial Infections in Hospitalized Children, Cambodia, 2007–2016

To determine trends, mortality rates, and costs of antimicrobial resistance in invasive bacterial infections in hospitalized children, we analyzed data from Angkor Hospital for Children, Siem Reap, Cambodia, for 2007–2016. A total of 39,050 cultures yielded 1,341 target pathogens. Resistance rates were high; 82% each of Escherichia coli and Klebsiella pneumoniae isolates were multidrug resistant. Hospital-acquired isolates were more often resistant than community-acquired isolates; resistance trends over time were heterogeneous. K. pneumoniae isolates from neonates were more likely than those from nonneonates to be resistant to ampicillin–gentamicin and third-generation cephalosporins. In patients with community-acquired gram-negative bacteremia, third-generation cephalosporin resistance was associated with increased mortality rates, increased intensive care unit admissions, and 2.26-fold increased healthcare costs among survivors. High antimicrobial resistance in this setting is a threat to human life and the economy. In similar low-resource settings, our methods could be reproduced as a robust surveillance model for antimicrobial resistance.

W orldwide, invasive bacterial infections are a leading cause of childhood deaths, mostly in lowand middle-income countries (1). Management of such infections is threatened by the rising prevalence of antimicrobial resistance (AMR), particularly among neonates (2). However, data on AMR in invasive bacterial infections in children from low-and middle-income countries are scarce (3)(4)(5)(6).
To combat the global threat of AMR, improved surveillance to detect emerging and long-term resistance trends is vital (7). Several global initiatives, such as the Fleming Fund, have been recently established to improve laboratory capacity in low-and middle-income countries (7,8), and the World Health Organization (WHO) Global Antimicrobial Resistance Surveillance System (GLASS) (9) has targeted 6 invasive pathogens for routine antimicrobial resistance surveillance: Escherichia coli, Klebsiella pneumoniae, Acinetobacter baumannii, Salmonella spp., Staphylococcus aureus, and Streptococcus pneumoniae. Monitoring resistance in these pathogens is particularly important for invasive bacterial infections in children in low-and middle-income countries, where most treatment is empirically prescribed and must be based on reliable contemporaneous resistance data to be effective.
Recent systematic reviews of AMR in invasive bacterial infections in children highlight the paucity of data available and do not report temporal resistance trends (5,6). In addition, although recent studies indicate excess deaths caused by AMR in low-and middle-income countries (10), there is limited evidence describing the economic and mortality burden of resistance at the patient level, particularly among children.
We analyzed 10 years of continuous AMR surveillance data for invasive bacterial infections in children from a sentinel surveillance site in Cambodia and describe resistance trends over time, by age group, and by site of acquisition (community or hospital). To evaluate the excess deaths and cost burden associated with third-generation cephalosporin resistance in community-acquired gramnegative bacteremia in hospitalized children, we analyzed patient-level data.

Procedures and AMR Reporting
We processed blood and CSF culture specimens as described elsewhere (11,13). Antimicrobial susceptibility testing was undertaken by disk diffusion and Etest MIC, according to Clinical and Laboratory Standards Institute guidelines (14) (online Technical Appendix Methods and Table 1, https://wwwnc.cdc.gov/EID/article/24/5/17-1830-Techapp1.pdf). Resistance proportions are reported as number of resistant isolates/number of isolates tested.

Outcome Analyses
We included in patient outcome analyses community-acquired monomicrobial Enterobacteriaceae (excluding Salmonellae) and A. baumannii bacteremia. These pathogens represent common causes of sepsis in children worldwide where third-generation cephalosporins would be a firstline/empiric treatment. We obtained clinical and costing data from hospital records and calculated cost per patient as admission cost plus antimicrobial costs.

Statistical Analyses
We treated isolates from specimens taken within 48 hours of admission as community-acquired infections and after 48 hours as hospital-acquired infections. However, Salmonella enterica serotypes Typhi and Paratyphi and Burkholderia pseudomallei isolates were always considered community-acquired infections. To ensure sufficient data per period, we grouped isolates into 2-year blocks. We assessed associations between resistance and year of isolation, patient age group, and site of acquisition (community vs. hospital) by univariable and multivariable logistic regression. Multivariable models included all variables. According to assessment of model fit by calculation of Akaike information criterion and plotting of observed versus predicted data, we considered time (year of isolation) a factor unless otherwise stated. For the outcome analyses, we used univariable analysis to compare variables by third-generation cephalosporin resistance status and patient outcome by using the Mann-Whitney-Wilcoxon rank-sum test for continuous variables and the χ 2 test with Yates correction for categorical variables. For multivariable logistic regression, outcome variables were hospital deaths and intensive care unit (ICU) admissions, and covariates were resistance, age group, age <10 years, malnutrition, sex, and organism type (Enterobacteriaceae vs. A. baumannii). We conducted multivariable linear regression by using admission duration and cost for survivors as outcome variables and using the same covariates. The linear model variables were log transformed, and results are presented with log-and back-transformed coefficients, which is interpreted as a multiplicative rather than an additive model. Analyses were undertaken by using the R statistical package (15).

Results
During the 10-year study period, 39,050 sterile site samples were collected for culture: 36,358 (93.1%) blood and 2,692 (6.9%) CSF (online Technical Appendix Figure 1). The sampling rate, indicated by the blood culture:hospital admission ratio, rose throughout the study period as utility of the clinical microbiology service increased. Approximately 1 blood culture was sent for every 3 admissions in 2007 (1,293 blood cultures:3,829 admissions), rising to 1 blood culture per admission in 2013 (5,294 blood cultures:5,208 admissions) and subsequently remaining stable (online Technical Appendix Figure 2). From 2012 through 2016, the proportion of blood cultures from neonates rose from 9.1% to 21.2%, and the proportion from children >5 years of age dropped from 35.4% to 22.4% (online Technical Appendix Table 2).

Overall AMR Rates
Overall AMR rates were high, especially among gramnegative GLASS organisms ( The most frequently isolated non-GLASS pathogen was Burkholderia pseudomallei, the causative agent of melioidosis (26.1%, 66), which was universally sensitive to the first-line drugs ceftazidime and co-trimoxazole ( Table 2). Next was Haemophilus influenzae, for which approximately half of isolates were ampicillin resistant (53.6%, 30/56) and one third multidrug resistant (37.1%, 13/35), followed by Enterobacter cloacae, which had a similar resistance profile to K. pneumoniae and E. coli. The remaining non-GLASS pathogens (group A Streptococcus, Pseudomonas aeruginosa, and Neisseria meningitides) exhibited low-level resistance to the key antimicrobials reported.

AMR Time Trends
The most frequently isolated organisms were K. pneumoniae, E. coli, Salmonella Typhi, S. aureus, and S. pneumoniae; AMR time trends were heterogeneous (Tables 1, 3-5; Figure 1). S. pneumoniae penicillin resistance fluctuated over time; 55.6% (5/9) of isolates were resistant in 2007-2008, dropping to 41.0% (16/39) in 2011-2012 before rising to 68.8% (22/32) in 2015-2016. During 2011-2016, when we tested S. pneumoniae for multidrug resistance, the proportion of multidrug-resistant S. pneumoniae isolates increased from 50.0% to 84.4%. Salmonella Typhi fluoroquinolone resistance also increased over the study period, from 88.6% to 100%. Multivariable logistic regression analysis in which time was a continuous variable showed an increased probability of Salmonella Typhi fluoroquinolone resistance over time (adjusted odds ratio [aOR] 2.14, 95% CI 1.29-3.74; p = 0.005), although not statistically significant in the model when time was a factor (online Technical Appendix Table 3 and Figure 3). Conversely, during 2015-2016, the proportion of resistant K. pneumoniae isolates fell dramatically for most antimicrobials tested, a phenomenon not seen for E. coli (online Technical Appendix Table 4). For E. coli, ampicillin-gentamicin and third-generation cephalosporin resistance remained stable at ≈50% with no evidence of significant change over time, as did rates of methicillinresistant S. aureus, which remained low throughout the study period. The proportion of K. pneumoniae isolates from neonates peaked in 2011-2014 at 46%-50% before dropping in 2015-2016 to 35% (online Technical Appendix Table 5), paralleling the change in the proportion of resistant isolates seen. To determine any subtle shifts in susceptibility, we examined changes in zone diameter distribution over time for E. coli and K. pneumoniae and found no clear trends (online Technical Appendix

AMR by Patient Age Group
Isolates from younger children were more often resistant to clinically important antimicrobials (Tables 3-5; Figure  2; online Technical Appendix Tables 7-9). Multivariable logistic regression controlling for year of isolation and site of acquisition indicated that K. pneumoniae isolates from neonates were >7 times more likely to be resistant to the first-line treatment agents ampicillin-gentamicin

AMR by Site of Infection Acquisition
Approximately four fifths of included isolates were from community-acquired infections (1,089, 81.2%) and one fifth from hospital-acquired infections (252, 18.8%). In almost all instances, the proportion of hospital-acquired isolates resistant to a given antimicrobial was higher than that of community-acquired isolates (online Technical Appendix Tables 10-12). K. pneumoniae, the main cause of hospital-acquired infections, was >3 times more likely to be resistant to ampicillin-gentamicin (aOR 3.62, 95% CI 1.42-9.58) and third-generation cephalosporins (aOR 3.51, 95% CI 1.27-10.12) in hospital-acquired isolates (Tables  3, 4). Increased likelihood of resistance among hospitalacquired isolates was also found for E. coli ampicillin-gentamicin and third-generation cephalosporin resistance and S. aureus methicillin resistance.

Outcomes
We analyzed patient outcomes for 129 admission episodes for community-acquired monomicrobial gram-negative bacteremia (online Technical Appendix Figure 6). Children from whom third-generation cephalosporinresistant bacteria were isolated were less likely than other patients to have received appropriate antimicrobial therapy (57% vs. 94%; p<0.001). If appropriate therapy was received, it was initiated later for children infected with third-generation-resistant than third-generation-sensitive organisms (2 days vs. 0 days after admission for those who survived [p<0.001]; 0.5 days vs. 0 days for those who died [p = 0.004]). Patients who died were younger (median age 1.4 vs. 10.2 months; p = 0.002), were more likely to have been admitted to an ICU (88% vs. 27%; p<0.001), stayed for a shorter time in hospital (3 vs. 8 days; p<0.001), and were more likely to have been infected with Enterobacteriaceae than A. baumannii (97% vs. 71%; p = 0.003) (online Technical Appendix Table  13). A. baumannii infections were associated with high levels of third-generation cephalosporin resistance (90%) but a low mortality rate (3%) despite only 48% of patients having received appropriate antimicrobials. Conversely, Enterobacteriaceae infections were associated with a high mortality rate (33%) despite 84% of patients having received appropriate antimicrobials (online Technical Appendix Table 14).

Discussion
In this hospitalized population of children in Cambodia, AMR levels were high, particularly among the gram-negative GLASS pathogens K. pneumoniae, E. coli, and A. baumannii. These organisms exhibited concerning resistance to WHO-recommended first-line sepsis treatment, emphasizing the urgent need for revised treatment guidelines (4). Few studies inform prevalence estimates of antimicrobial resistance in low-and middle-income countries in Asia, but compared with what is known, the high levels of gramnegative resistance reported here are not uncommon (6,16).
For the gram-positive GLASS pathogens, S. pneumoniae resistance was broadly similar to that of the wider region (67.7% vs. 59.3% multidrug resistance, respectively) (17).
Rates of methicillin-resistant S. aureus were comparatively lower; only 40.0% of hospital-acquired isolates were methicillin resistant compared with a regional average of 67.4% (18). A major strength of this study is the observation of resistance trends over an extended period, something rarely possible in low-and middle-income countries because of lack of longstanding microbiology services. We found heterogeneous trends in resistance over time; resistance increased in some organisms (Salmonella Typhi) and decreased in others (K. pneumoniae). The most surprising temporal trend observed was a drop in the proportion of resistant K. pneumoniae isolates for most antimicrobials tested, in contrast to largely stable resistance levels in E. coli. For K. pneumoniae resistance by site of acquisition, in community-acquired isolates, resistance sharply declined in 2015-2016, perhaps suggesting loss of a plasmid coding for multiple resistance determinants. Confirming this trend will require a larger dataset from multiple sites in Cambodia and further analysis of the underlying resistance mechanisms at work using a method such as whole-genome sequencing. The genetic determinants of resistance in colonizing K. pneumoniae and E. coli isolates from this population have been reported elsewhere (CE Moore, CM Parry, P Turner, NPJ Day, N Stoesser, unpub data; N Stoesser, C Turner, P Turner, BS Cooper, unpub data), whereas whole-genome sequencing of invasive isolates is ongoing. Loss of antimicrobial selective pressure leading to declining resistance may result from changes in national/ regional antimicrobial supply or lack of active drug in antimicrobials used (19).
The number of hospital-acquired K. pneumoniae isolates peaked during 2011-2012. This peak may be the result of a genuine rise in the rate of hospital-acquired infections or the increased rate of blood culture sampling compared with previous years. From 2011-2012 onward, the proportion of resistant hospital-acquired K. pneumoniae isolates declined. This drop may be linked to maturation of a hospitalwide infection-control program implemented in 2010 (20) and enforced by prospective hospital-acquired infection surveillance from 2015 onward (21) or to the clinical microbiology service operating since 2012 with a strong focus on antimicrobial drug stewardship. Indeed, a recent study of prescribing practices at this hospital found 84%-89% of antimicrobial drug prescriptions were appropriate (22). The apparent success of these interventions suggests that they could be useful for combating AMR in similar settings. The perceived temporal drop in K. pneumoniae resistance could also be attributable to changing proportions of isolates from neonates over time; 46%-50% of isolates were from neonates in 2011-2014, dropping to 35% in 2015-2016. 848 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 24, No. 5, May 2018 This study is unusual in that it directly compares different age groups of children, revealing AMR trends associated with age. Of note, the dominant pathogen in neonates, K. pneumoniae, was also more often resistant in neonates. For hospital-acquired isolates, this resistance may result from horizontal acquisition of resistant gramnegative organisms from hospital surfaces, as suggested by a recent multicenter study of sepsis in neonates (2). Indeed, colonization of neonates by resistant gram-negative organisms has been shown to be common at Angkor Hospital for Children and associated with subsequent invasive infection (23). For community-acquired isolates, vertical maternal transfer of resistant organisms may have a substantial role and is currently under investigation at this center.
Similarly, Salmonella Typhi from children <5 years of age was more often multidrug resistant than that from those >5 years of age. Isolates from younger children have greater genetic diversity (24), although how this diversity relates to increased AMR requires further investigation. In Cambodia, the most common indication for antimicrobial drug use is infections in children <5 years of age; thus, children in this age group may be exposed to more antimicrobial drugs, leading to greater resistance in organisms causing infection. That S. pneumoniae isolates were more often penicillin resistant in children <5 years of age is consistent with findings of previous work showing greater colonization of this age group by multidrug-resistant pneumococci (25). Vaccination may have a collateral benefit of reducing AMR (26), which suggests that it could be useful for combating Salmonella Typhi and S. pneumoniae resistance in low-and middle-income countries. Because 85% of Salmonella Typhi isolates are simultaneously fluoroquinolone resistant and multidrug resistant, few agents remain for treating typhoid in this population, placing even greater value on preventive measures such as vaccination. In January 2015, a 13-valent pneumococcal conjugate vaccine was introduced in Cambodia (27) with no catch-up campaign, meaning that only S. pneumoniae isolates from children born in or after December 2014 could have been affected, equating to 5 isolates in this dataset. Pneumococcal vaccination is thus unlikely to have had an appreciable effect on the AMR trends reported here.
The WHO Global Report on Surveillance identified a major gap in research comparing resource use in resistant versus nonresistant pathogens (28), an area that we addressed by demonstrating that resistance is associated with worse healthcare outcomes, including increased deaths and ICU admissions, delayed effective treatment, and more than doubled admission costs. Use of patient records allows these estimates to more closely reflect reality than modeled or ecologic analyses, although it is unclear whether this increased risk for adverse outcomes represents greater virulence, delayed treatment, or confounding. The observed outcome differences between Enterobacteriaceae and A. baumannii infections suggest either a true difference in virulence or that a proportion of A. baumannii isolates were contaminants, an uncertainty that highlights the difficulty of establishing the clinical significance of skin-colonizing organisms.
This study has several limitations. The data derive from a single nongovernmental hospital for children with limited numbers of isolates for some bacterial species; thus, trends and outcomes may not be representative of the wider region. The study was retrospective, and classification of community-acquired and hospital-acquired infections was limited by hospital database and clinical case note accuracy, meaning that some community-acquired infections may have actually been hospital-acquired infections. Widespread prehospitalization use of antimicrobials may have selected for resistant organisms (29). There were no restrictions to blood culture submission over time, but from early 2016 onward, clinicians were asked to focus on children requiring admission, which may have affected certain organisms (e.g., Salmonella Typhi). Microbiology practice variations over time meant that antimicrobial susceptibility testing was not consistent; however, we believe that the value of examining the evidence over a long period outweighed the effect that these variations may have had on results. For example, in 2009, the Clinical and Laboratory Standards Institute sensitivity zone size cutoffs for carbapenems and cephalosporins in Enterobacteriaceae increased, which could have resulted in a small number of isolates previously classed as sensitive being reclassified as resistant. The reported pre-2009 resistance levels are thus conservative and would not negate the downward resistance trends observed. In the outcome analysis, we did not consider prehospitalization factors, clinical diagnosis, and non-HIV/malnutrition co-occurring conditions because quantifying those could have introduced substantial reporting bias. Furthermore, our cost estimates may be higher than actual costs because we did not account for partial/ shared doses and price fluctuations, suggesting that these cost estimates are most useful as a relative indication of cost burden.
In conclusion, the high rate of AMR in this setting of hospitalized children in Cambodia was associated with increased deaths and healthcare costs and threatens the effectiveness of first-line sepsis treatment. AMR represents a major threat to children's health globally (5,6), yet there is a dearth of data for children in low-resource settings (3)(4)(5)(6). By reporting a decade of continuous AMR surveillance data, this study fills a gap in the understanding of antimicrobial drug resistance in children in Cambodia. In the context of the current global drive to combat AMR and the goal of the Fleming Fund to improve surveillance in low-and middle-income countries, our study demonstrates the feasibility and utility of undertaking accurate long-term antimicrobial drug resistance surveillance in these countries. The methods used here are reproducible in similar low-resource settings.