Antimicrobial resistance among Enterobacteriaceae, Staphylococcus aureus, and Pseudomonas spp. isolates from clinical specimens from a hospital in Nairobi, Kenya

Background Antimicrobial resistance among pathogens of public health importance is an emerging problem in sub-Saharan Africa. Unfortunately, published information on the burden and patterns of antimicrobial resistance (AMR) in this region is sparse. There is evidence that the burden and patterns of AMR vary by geography and facility. Knowledge of local epidemiology of AMR is thus important for guiding clinical decisions and mitigation strategies. Therefore, the objective of this study was to determine the burden and predictors of AMR and multidrug resistance (MDR) among bacterial pathogens isolated from specimens submitted to the diagnostic laboratory of a hospital in Nairobi, Kenya. Methods This retrospective study used laboratory records of 1,217 clinical specimens submitted for bacterial culture and sensitivity testing at the diagnostic laboratory of The Karen Hospital in Nairobi, Kenya between 2012 and 2016. Records from specimens positive for Enterobacteriaceae, Staphylococcus aureus, or Pseudomonas spp. isolates were included for analysis. Firth logistic models, which minimize small sample bias, were used to investigate determinants of AMR and MDR of the isolates. Results A total of 222 specimens had bacterial growth. Most Enterobacteriaceae isolates were resistant to commonly used drugs such as penicillin/β-lactamase inhibitor combinations (91.2%) and folate pathway inhibitors (83.7%). Resistance to extended-spectrum cephalosporins was also high (52.9%). Levels of AMR and MDR for Enterobacteriaceae were 88.5% and 51%, respectively. Among S. aureus isolates, 57.1% were AMR, while 16.7% were MDR. As many as 42.1% of the Pseudomonas spp. isolates were aminoglycoside-resistant and 15% were fluoroquinolone-resistant, but none exhibited resistance to antipseudomonal carbapenems. Half of Pseudomonas spp. isolates were AMR but none were MDR. Significant predictors of MDR among Enterobacteriaceae were organism species (p = 0.002) and patient gender (p = 0.024). Conclusions The high levels of extended-spectrum cephalosporin resistance and MDR among Enterobacteriaceae isolates are concerning. However, the relatively low levels of MDR S. aureus, and an absence of carbapenem resistance among Pseudomonas isolates, suggests that last-line drugs are still effective against S. aureus and Pseudomonas infections. These findings are relevant for guiding evidence-based treatment decisions as well as surveillance efforts and directions for future research, and contribute to the sparse literature on AMR in sub-Saharan Africa.

The above reports are concerning and hence epidemiologic surveillance, with an emphasis on pathogens of high priority for public health, is warranted to gain a more comprehensive understanding of the burden of AMR and MDR. Encouragingly, Kenya is enrolled in the WHO Global Antimicrobial Surveillance System (GLASS), and is building a national AMR surveillance system (World Health Organization, 2018). However, the findings of locally focused studies are useful to complement these broader surveillance efforts (Kariuki et al., 2018). Information from such local studies are useful for guiding clinical decisions and evidence-based control strategies. Therefore, the objective of this study was to determine the burden and predictors of antimicrobial and multidrug resistance among bacterial pathogens isolated from specimens submitted to the diagnostic laboratory of a hospital in Nairobi, Kenya.

Design, setting and data source
This retrospective epidemiologic study used laboratory records of 1,217 clinical specimens submitted to the diagnostic laboratory of The Karen Hospital for bacterial culture and sensitivity testing between 2012 and 2016. Records of specimens positive for Enterobacteriaceae, Staphylococcus aureus, or Pseudomonas spp. were included in the study. Only records of these organisms were included for further analysis because they were the most frequently isolated organisms and sample sizes of other organisms were not sufficient to enable meaningful analysis. The following data were extracted from the laboratory records of specimens selected for inclusion in the study: medical record number, sample identification, patient gender, age, specimen site, bacterial growth, bacterial species isolated, and results of antimicrobial susceptibility testing. Hospital records for both inpatient and outpatient visits were also obtained. The following data were extracted from the hospital records: medical record number, marital status, visit date, admission date, discharge date, and inpatient/outpatient designation.

Bacterial isolation and antimicrobial susceptibility testing
The hospital laboratory followed the European Committee on Antimicrobial Susceptibility testing (EUCAST) procedures for bacterial isolation and antimicrobial susceptibility testing (

Data management
Data management was performed in SAS 9.4 (SAS Institute, Cary, NC, USA). The data were assessed for duplicate entries but none were found. Patient hospital records were joined to laboratory records based on medical record numbers and if the date of hospital visit, admission, or discharge occurred within 14 days of sample submission or laboratory report. Patient entries were coded as ''inpatient'' or ''outpatient'' based on a visit identification indicator when available. If a visit identification indicator was not available, hospital status (inpatient vs outpatient) was determined based upon the presence or absence of an admission date.
Records for non-duplicate specimens positive for Enterobacteriaceae, Staphylococcus aureus, or Pseudomonas spp. isolates were included for further analysis in the study because these were the most frequently isolated organisms. Specimen site was re-categorized to reflect the most common sites of sample collection. For enteric bacteria and Pseudomonas spp. this included stool, blood, and sputum. Isolates obtained from all other sites were categorized as ''other''. The most common specimen sites for S. aureus isolates were blood, wounds and abscesses; isolates from all other sites were categorized as ''other''.
The results of susceptibility testing were re-coded to two categories, ''resistant'' and ''susceptible''. Isolates with ''intermediate'' susceptibility were re-coded as resistant (Magiorakos et al., 2012). In order to further classify isolates as antimicrobial or multidrug resistant, antimicrobial agents were placed into clinically significant antimicrobial categories listed in guidelines established by the European Center for Disease Prevention and Control (ECDC) and United States Centers for Disease Control and Prevention (CDC) (Magiorakos et al., 2012). Isolates resistant to one or more drugs in at least one category were considered to be antimicrobial resistant (AMR). Isolates resistant to one or more antimicrobial agents in at least three categories were classified as multidrug resistant (MDR) (Magiorakos et al., 2012). If a bacterial species was intrinsically resistant to an antimicrobial agent or category, results of susceptibility testing for isolates of that species were not reported, and the antimicrobial was excluded during classification of the isolate as AMR or MDR (Bouza & Cercenado, 2002;Pfeifer, Cullik & Witte, 2010;Magiorakos et al., 2012;Leclercq et al., 2013;Ruppé, Woerther & Barbier, 2015;Hughes et al., 2016;Goto et al., 2017;CLSI, 2020).

Statistical analysis
All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA). The Shapiro-Wilk test was used to assess for normality of distribution of patient age. Since patient age was non-normally distributed, median and interquartile range were used as measures of central tendency and dispersion. The Cochran-Armitage trend test was used to assess for significant temporal trends in AMR and MDR. Univariable Firth logistic regression models were used to assess whether resistance to one antimicrobial class could be used to predict resistance to other antimicrobial classes among Enterobacteriaceae isolates. The Firth logistic regression models were necessary for these analyses because of the small number of isolates in some of the analyses. Ordinary logistic models, which use maximum likelihood estimation, are not appropriate for these models because they suffer from small-sample bias (Heinze & Schemper, 2002). To mitigate small-sample bias, the Firth models use penalized likelihood functions to obtain parameter estimates rather than maximum likelihood used by ordinary logistic models (Firth, 1993).
The following potential explanatory variables were investigated for associations with AMR and MDR: patient hospitalization status (inpatient vs outpatient), specimen site, patient gender, patient age category, patient marital status, and bacterial species isolated (for Enterobacteriaceae). Firth logistic regression models were used to investigate determinants of: (a) AMR and MDR among enteric bacteria, (b) AMR and MDR among S. aureus isolates, and (c) AMR among Pseudomonas spp. isolates. Model building was performed using a two-step process. First, univariable Firth logistic regression models were used to identify explanatory variables significantly associated with each of the outcomes. In the second step, variables that had significant univariable associations with the outcomes (using a relaxed p-value of <0.15) were considered in multivariable Firth logistic regression models. Model building was performed using manual backwards elimination, with a cutoff p-value of ≤0.05. Explanatory variables were considered to be confounders if their removal resulted in >20% change in the coefficients of the variables in the model. Model fit was assessed using the Hosmer-Lemeshow goodness-of-fit test.

Ethics approval
This study was approved by the Standards and Ethics Committee of The Karen Hospital and the University of Tennessee Institutional Review Board. UTK IRB16-03358-XM. The clinical specimens used in this study were collected as part of the routine diagnostic testing procedures used for patient care in the hospital. Since this was a retrospective study, no contact was made with patients and no attempt was made to link records to specific patients.

Descriptive statistics
Of the 1,217 specimens submitted for culture and sensitivity testing, 222 had bacterial growth of which 166 met the inclusion criteria and were included for further analysis. The most frequently isolated pathogens were enteric bacteria (108 specimens), followed by S. aureus (42), and Pseudomonas spp. (20). Other bacterial species isolated included: Acinetobacter spp., Enterococcus spp., Helicobacter spp., Moraxella catarrhalis, coagulase negative Staphylococcus spp., untyped Staphylococcus spp., Streptococcus pneumoniae, Streptococcus viridans, and untyped Streptococcus spp.
Among the Enterobacteriaceae, Klebsiella spp. were the most frequently isolated (28.7%), followed by Escherichia coli (26.9%), Proteus spp. (15.7%), and Salmonella spp. (14.8%) ( Table 1). Bacterial species included in the ''other'' category were: Citrobacter spp., Enterobacter spp., Raoultella ornithinolytica, and Shigella spp. Most (64.4%) enteric bacterial isolates were from inpatients, and approximately half (51.1%) were obtained from stool specimens (Tables 1 and 2). The majority of specimens positive for Enterobacteriaceae were collected in 2015 (42) and 2016 (48). There was an approximately even gender distribution, with 47.2% of Enterobacteriaceae isolated from female patients and 52.8% from male patients. Age of patients from whom enteric bacteria were isolated ranged from 1 to 93 years, with a median of 46.5 years and an interquartile range of 23-70.
The majority (77.3%) of S. aureus isolates were also from inpatients, and blood was the most common specimen (48.3%) followed by wound or abscess (31.0%) ( Table 1). Specimen sites in the ''other'' category included sputum (four), eye swab (one), and central venous catheter (one). Most (64.3%) S. aureus positive samples were from males. Patient age for S. aureus positive samples ranged from <1 year to 88 years, with a mean of 40.8 years. The majority of patients with S. aureus infections were married (67.7%). As with Enterobacteriaceae, most S. aureus positive specimens were collected in 2015 (17) and 2016 (17).
Almost half (45.5%) of Pseudomonas spp. isolates were obtained from sputum samples, followed by blood (18.2%) and stool (18.2%). The majority of specimens positive for Pseudomonas spp. were from male (80.0%) and married (64.3%) patients. The age of patients who were positive for Pseudomonas spp. ranged from 21 to 85 years, with a median of 71 years and an interquartile range of 43-82.

Patterns of antimicrobial and multidrug resistance
Overall, 75.9% of the isolates included in the current study were AMR, while 36.1% were MDR. There were no significant temporal trends in the overall burden of AMR (p = 0.2489) or MDR (p = 0.6340).

(a) Enterobacteriaceae
Susceptibility test results for one or more antimicrobial agents were available for 104 of the 108 enteric bacterial isolates. Among Enterobacteriaceae, the percentages of resistant isolates were highest for the following antimicrobial categories: penicillin/β-lactamase inhibitor combinations (91.2%), folate pathway inhibitors (83.7%), and penicillins (67.6%) ( Table 3). Resistance to extended-spectrum cephalosporins was observed in about half (52.9%) of Enterobacteriaceae isolates. In addition, about one-third (30%) of the isolates exhibited resistance to at least one quinolone and/or fluoroquinolone, including 25.3% (22/87) that were resistant to one or more fluoroquinolones. Enteric bacteria had the lowest levels of resistance (6.5%) to carbapenems. As many as 88.5% and 51% of the Enterobacteriaceae isolates were AMR and MDR, respectively (Table 4). Moreover, 6.7% of the isolates exhibited MDR involving five antimicrobial categories (Table 4). While AMR exhibited a statistically significant temporal decrease between 2013 and 2016 (p = 0.0016), no significant temporal trends were observed for MDR (p = 0.4258). Among Enterobacteriaceae, resistance to aminoglycosides was a significant predictor of resistance to carbapenems (Odds Ratio [OR] = 10.9; 95% confidence interval [CI] 1.3, 89.3; p = 0.027). In addition, resistance of enteric bacteria to extended-spectrum cephalosporins was a significant predictor (OR = 3.9; 95% CI 1.3, 11.6; p = 0.016) of resistance to quinolones and fluoroquinolones, as was resistance to penicillins (OR = 6.5; 95% CI 1.1, 40.3; p = 0.044). Resistance to extended-spectrum cephalosporins was also a predictor (OR = 17.9; 95% CI 3.7, 87.3; p = 0.0004) of resistance to penicillins.

(c) Pseudomonas spp
Among Pseudomonas spp. isolates, 42.1% were resistant to one or more aminoglycosides, and 15% were resistant to a fluoroquinolone (Table 3). Fifty percent of the Pseudomonas spp. isolates were AMR, and none were MDR (Table 4). Pseudomonas spp. isolates were only observed in 2015 and 2016, and there were no statistically significant differences (Fisher's exact p = 0.6285) in the proportions of AMR isolates between the two years.

Distribution of AMR and MDR isolates
The distribution of patient and specimen characteristics among AMR and MDR isolates, as well as p-values for associations assessed using univariable Firth logistic regression models, are displayed in

Determinants of MDR
Species of the infectious agent and gender of the patient were significant predictors of MDR (

DISCUSSION
This study investigated the burden of antimicrobial and multidrug resistance among the most frequently identified bacterial species isolated from clinical specimens submitted to a diagnostic laboratory in Nairobi, Kenya. Additionally, predictors of multidrug resistance were investigated. The findings are relevant for clinicians and may be useful in guiding evidence-based treatment decisions. Moreover, information from this study adds to the body of knowledge on antimicrobial resistance in sub-Saharan Africa, where there is limited published information on the burden and patterns of AMR and MDR. Continued surveillance to better characterize the burden of AMR, with a focus on pathogens of public health importance, is crucial to address the problem.  TMS (Sang, Oundo & Schnabel, 2012;Swierczewski et al., 2013). High levels of resistance to amoxicillin/clavulanate (71.6%) and TMS (77%) have also been reported among Enterobacteriaceae isolated from clinical specimens in Ethiopia (Teklu et al., 2019). These findings are not surprising, as these medications are inexpensive, orally available, and frequently used, leading to widespread selection for resistance in many countries in Africa (Leopold et al., 2014;Tadesse et al., 2017). The levels of resistance to aminoglycosides (27.6%) among Enterobacteriaceae isolates were comparable to levels of gentamicin resistance among Enterobacteriaceae reported by Kenyan (16-35%) (Bejon et al., 2005) and Ethiopian (31.6%) studies (Bitew & Tsige, 2020). However, susceptibility to aminoglycosides has been reported to vary between individual drugs. Another Ethiopian study reported that while 43.4% of Enterobacteriaceae were resistant to gentamicin, far fewer (13.8%) were resistant to amikacin (Teklu et al., 2019).
Resistance to one or more fluoroquinolones was also observed among a considerable proportion (25.3%) of Enterobacteriaceae isolates. While this is concerning, it is not entirely surprising because the availability of oral formulations of these drugs make them quite popular and hence more likely to be misused (Chattaway et al., 2016). Fluoroquinolone resistance, while previously uncommon, has expanded relatively recently in sub-Saharan Africa compared to other parts of the world (Chattaway et al., 2016). Some recent reports from the region have described high levels of fluoroquinolone resistance among Enterobacteriaceae isolates. For instance, a recent Kenyan study reported that 28% of P. mirabilis, 56% of K. pneumoniae, and 78% of E. coli isolates exhibited resistance to ciprofloxacin (Wangai et al., 2019). Among K. pneumoniae isolated from blood cultures of patients in another hospital in Kenya, 44.4% were resistant to ciprofloxacin (Apondi et al., 2016). Similar proportions of fluoroquinolone-resistant Enterobacteriaceae (45.9%, 46.3%) were documented in reports from Ethiopia (Teklu et al., 2019;Bitew & Tsige, 2020). The current study found comparatively lower (25.3%), albeit substantial, levels of resistance to fluoroquinolones. This difference may reflect numerous factors, including patient characteristics, antimicrobial use patterns, and bacterial species, highlighting the importance of relevant local data for characterizing the burden of AMR.
The level of resistance to extended spectrum cephalosporins (52.9%) among Enterobacteriaceae isolates is of particular public health significance (World Health Organization, 2017). Extended-spectrum β-lactamase (ESBL) production is typically responsible for resistance to third-generation cephalosporins among Enterobacteriaceae, and plasmid-mediated mechanisms may be spread quickly between isolates (Ruppé, Woerther & Barbier, 2015). The findings from the current study were consistent with those of two laboratory-based studies in Ethiopia, where resistance to extended-spectrum cephalosporins ranged from 47.5% to 51.6% in one report, and 60.3% to 62.2% in another (Teklu et al., 2019;Bitew & Tsige, 2020). In contrast, a study conducted in Uganda reported resistance to third generation cephalosporins among Enterobacteriaceae isolates to be considerably lower (27.5%) (Ampaire et al., 2015).
The proportion of ESBL-producing isolates can differ between species of enteric bacteria, as evidenced by a study from Tanzania where ESBL production ranged from 24.4% among E. coli to 63.7% among K. pneumoniae isolates (Mshana et al., 2009). Similarly, levels of resistance to third generation cephalosporins among E. coli isolates (66-75%) were surpassed by K. pneumoniae (82-83%) in a recent report from Kenya (Wangai et al., 2019). Although there is apparent variation by geographic location and bacterial species, the substantial level of cephalosporin resistance observed in the current study, along with findings of other recent reports, suggests that cephalosporin resistance is an emerging problem. Unfortunately, very few studies have investigated the temporal trends of ESBLproducing isolates in sub-Saharan Africa. However, a significant increase in the proportion of ESBL-producing K. pneumoniae isolates was documented in a hospital in Kenya (Henson et al., 2017). A study from Zimbabwe also reported a temporal increase in ESBL production among Enterobacteriaceae, supporting the need to prioritize this issue (Mhondoro et al., 2019). While a temporal decrease in AMR among Enterobacteriaceae was observed in the current study, the ability to draw meaningful conclusions from this finding is limited by small sample size during the earlier years of this study. In addition, sample size precluded the assessment of temporal trends for individual drugs or antimicrobial categories. Clearly, further investigation of such trends is warranted in future, larger scale studies.
While the overall percentage (52.9%) of Enterobacteriaceae that exhibited resistance to extended-spectrum cephalosporins was quite concerning, far fewer carbapenem-resistant Enterobacteriaceae (CRE) were identified (6.5%). The findings of the current study are comparable to those from Ethiopia and Uganda, where carbapenem resistance was reported in 5.2% and 3.5% of enteric isolates, although additional isolates with genetic determinants of carbapenem resistance were identified in the Ugandan study (Ampaire et al., 2015;Teklu et al., 2019). Even lower levels of CRE (<1%) have been documented in other reports from Kenya and Zimbabwe (Maina et al., 2013b;Mhondoro et al., 2019). In contrast, some hospitals within Kenya have reported high levels of carbapenem resistance, particularly for K. pneumoniae, which is a serious public health concern (Apondi et al., 2016;Wangai et al., 2019). Carbapenems are considered ''last-line'' antimicrobials, and a very limited number of treatment options exist for infections with CRE (Morrill et al., 2015). The low level of resistance to carbapenems observed in the current study provides evidence that these antibiotics remain a viable option for the treatment of infections with ESBL-producing Enterobacteriaceae in this patient population. However, careful monitoring and judicious use of these last-line drugs is critical to ensure that they remain effective in treatment of potentially life-threatening multidrug-resistant infections.

Prediction of antimicrobial resistance between drug categories
Resistance to extended-spectrum cephalosporins was a significant predictor of resistance to penicillins, an expected finding given that β-lactamases mediating cephalosporin resistance also act upon penicillins (Ruppé, Woerther & Barbier, 2015). The other observed associations, however, may reflect co-transference of antimicrobial resistance genes via mobile genetic elements (MGEs). For example, extended-spectrum cephalosporin resistance was a significant predictor of resistance to fluoroquinolones and quinolones. Associations between plasmid-borne fluoroquinolone resistance and extended-spectrum β-lactamase production have previously been reported, consistent with these findings (Bouchakour et al., 2010;Crémet et al., 2011;Ruppé, Woerther & Barbier, 2015;Salah et al., 2019). For instance, the proportion of ciprofloxacin-resistant isolates among Gram negative bacteria isolated from hospital specimens in Tanzania was significantly higher among ESBL producers than non-ESBL producers (Mshana et al., 2009).
Aminoglycoside resistance was a significant predictor of resistance to carbapenems. Aminoglycoside resistance is frequently encountered among carbapenemase-producing Enterobacteriaceae, and can be mediated by aminoglycoside modifying enzymes acquired via MGEs (Livermore et al., 2011a;Livermore et al., 2011b;Ruppé, Woerther & Barbier, 2015). Carbapenemase-or ESBL-producing Enterobacteriaceae may carry resistance genes for multiple other drugs, limiting the available options for treatment (Ruppé, Woerther & Barbier, 2015). The observed relationships are consistent with previously reported mechanisms of antimicrobial resistance, and awareness of such associations may help inform evidence-based antimicrobial selection in situations where results of bacterial culture and sensitivity testing are not available.

Burden, distribution and determinants of MDR
The fact that approximately half of the Enterobacteriaceae isolates were MDR is quite concerning but is consistent with reports from recent studies conducted in Ethiopia that also reported high levels of MDR among enteric isolates (42.1% and 68.3%) (Teklu et al., 2019;Bitew & Tsige, 2020). Moreover, one of these studies also identified extensively-drug resistant (XDR) and pan-drug resistant (PDR) Enterobacteriaceae (Bitew & Tsige, 2020). In contrast, MDR was much less common among isolates from clinical specimens collected from hospital patients in Uganda (27.8%) (Ampaire et al., 2015). Among K. pneumoniae isolated from invasive infections in a Kenyan hospital, as many as 79% were MDR, consistent with findings from the current study that 71% of Klebsiella spp. were MDR (Henson et al., 2017).
Compared to E. coli, Proteus spp. and Salmonella spp. had significantly lower odds of MDR. This is consistent with reports of differences in the prevalence of MDR across bacterial species from other studies (Teklu et al., 2019;Bitew & Tsige, 2020). The odds of MDR for Enterobacteriaceae isolated from male patients were significantly greater than that for female patients, which could reflect differences in previous antimicrobial exposure and/or gender differences in healthcare-seeking behavior. Such gender differences differ by both geographical location and sociodemographic factors, including income and educational attainment (Torres et al., 2019). A study conducted in northern Uganda reported that males were significantly more likely than females to self-medicate with antimicrobials, while a study from Sudan had contrasting results (Awad et al., 2005;Ocan et al., 2014). A review of factors influencing this practice in low-and middle-income countries (LMIC) suggests that it is generally more common among females, although this is not a consistent finding (Torres et al., 2019). A Nairobi survey reported that men are more likely to visit public health facilities over self-treatment, a finding that did not persist for private facilities, possibly due to gender disparities in resources (Muriithi, 2013). Thus, the association observed in the current study could reflect gender differences in self-medicating practices, care-seeking behavior, access to healthcare, education, or economic resources, among others. Further research is obviously warranted to explore this association.
A much lower percentage of S. aureus isolates were MDR (16.7%) compared to Enterobacteriaceae (51%) in the current study. This is also much lower than findings from two other studies in the region, which reported that 63% and 64.8% of S. aureus isolates were MDR (Sangeda et al., 2017;Iliya et al., 2020). However, the criteria used to categorize isolates in the above studies as MDR were different from those used in the current study, limiting comparability. This calls for consistent use of a standardized classification scheme for identifying MDR isolates. This study used the framework recommended by the ECDC and CDC, which has the advantage of employing clinically and epidemiologically relevant categories of antimicrobials (Magiorakos et al., 2012).

(c) Pseudomonas spp.
Resistance to aminoglycosides was by far the most common (42.1%) among Pseudomonas spp. isolates in the current study. This is much higher than findings of two Kenyan studies, where lower levels of resistance to amikacin (11%) and gentamicin (9 and 18%) were observed among P. aeruginosa isolates (Bejon et al., 2005;Wangai et al., 2019). However, other studies have reported findings similar to those of the current study. For example, resistance to amikacin and gentamicin were observed in at least 40% of P. aeruginosa isolates from rural and urban hospitals in Kenya (Mukaya et al., 2018;Thuo et al., 2019). Similar findings were reported by a study in Tanzania, where 34.8% of Pseudomonas spp. isolated from hospital patients exhibited resistance to gentamicin (Mshana et al., 2009). Although monotherapy with aminoglycosides is not recommended for treatment of suspected P. aeruginosa infections, these agents play an important role in combination therapy (Bassetti et al., 2018). The substantial proportion of aminoglycoside-resistant isolates observed in our study and others in the region should be taken into consideration by clinicians formulating a plan for empirical treatment.
The percentage of Pseudomonas spp. isolates resistant to antipseudomonal fluoroquinolones (15%) was considerably lower than to aminoglycosides, a finding that is consistent with those of some other studies (Mshana et al., 2009;Thuo et al., 2019). Fluoroquinolones, which have a broad spectrum of activity and the advantage of oral administration, may be recommended for treatment of certain suspected P. aeruginosa infections, either alone or as part of combination therapy (Bassetti et al., 2018). Reports of fluoroquinolone resistance also appear to vary markedly between different facilities. While a 2005 study found that just 2% of P. aeruginosa isolates from inpatients at a Kenya hospital were resistant to ciprofloxacin (Bejon et al., 2005), this was as high as 52.7% in another more recent hospital-based study (Mukaya et al., 2018). Given the high proportion of aminoglycoside-resistant isolates and the limited number of useful antimicrobials due to natural resistance mechanisms of Pseudomonas, it is important to carefully monitor resistance to these agents (Bassetti et al., 2018).

Strengths and limitations
Firth logit models used in this study generate better estimates than ordinary logistic models, which use maximum likelihood estimation which suffer from small-sample bias (Heinze & Schemper, 2002). Parameter estimates of Firth models are generated using penalized likelihood, which reduces the small-sample bias associated with maximum likelihood estimation (Firth, 1993). Additionally, penalized likelihood estimations produce finite, consistent estimates of regression parameters in situations when maximum likelihood estimates do not exist due to complete or quasi-complete separation (Heinze, 2006;Williams, 2019). Therefore, although sample size was a statistical challenge, it was adequately addressed using this modeling approach.
Since this was a retrospective study that used secondary data, the investigators did not determine the antimicrobial agents included in the panel for susceptibility testing. The panels used by the diagnostic laboratory were not consistent for all isolates of the same bacterial species, highlighting the need for standardization of laboratory procedures and reporting in the region. Detailed patient medical history was not available. Therefore, the associations of AMR or MDR with clinical factors (such as current and prior antimicrobial use, underlying health conditions, primary diagnosis, and treatment outcomes) could not be investigated. Information about whether the isolated organisms were considered primary infectious agents was also not available. For instance, it is possible that some isolates from stool specimens were commensal organisms and may not have been responsible for clinical signs. However, commensal bacteria represent a potential reservoir for antimicrobial resistance genes that may be transferred to pathogens (Singh, Verma & Taneja, 2019). Therefore, monitoring patterns of AMR in these organisms provides valuable information.
Despite the above limitations, the findings of this study provide useful information for clinicians and contribute to literature on AMR patterns in Kenya specifically and sub-Saharan Africa in general. Moreover, the study focused on clinically and epidemiologically relevant antimicrobial categories as well as pathogens of public health importance. The study findings are therefore useful to guide future AMR surveillance programs and directions for future research.

CONCLUSIONS
The high levels of extended-spectrum cephalosporin resistance and MDR among Enterobacteriaceae observed in this study are concerning. However, the relatively low levels of MRSA and MDR S. aureus and an absence of carbapenem resistance among Pseudomonas spp. isolates is encouraging. This may suggest that last-line drugs remain available for the treatment of infections with resistant S. aureus and Pseudomonas spp. isolates when necessary. However, continued antimicrobial stewardship, adherence to biosecurity practices, and surveillance are warranted to ensure this does not change.

ADDITIONAL INFORMATION AND DECLARATIONS Funding
The authors received no funding for this work.