Epidemiology of Escherichia coli and Klebsiella pneumoniae bloodstream infections in a general hospital in Singapore: a retrospective cohort study

INTRODUCTION

Escherichia coli (E. coli) and Klebsiella pneumoniae (K. pneumoniae) are the predominant pathogens causing Gram-negative bloodstream infections (GNBSIs) worldwide.[1] E. coli and K. pneumoniae BSI cases show an increasing trend worldwide, with resistance to certain key antimicrobials such as ciprofloxacin, third-generation cephalosporins and carbapenems rising to alarming levels.[2-4] This is of great concern owing to the burden it places on patient safety, healthcare systems and the economy. Gram-negative bloodstream infections have become an international concern, with some countries adopting targets and strategies over the past few years.[5]

There is conflicting evidence on the impact of empirical antibiotics and/or multidrug resistance (MDR) status on mortality related to GNBSIs. Some studies demonstrate no or weak associations, while others demonstrate significant associations.[6] Widely observed variations between study designs including statistical methods may contribute to such conflicting evidence.

The epidemiology of GNBSIs has been studied in the local context; however, it has been limited to specific subpopulations.[6,7] Despite the excellent work published over the last two decades, there are no studies investigating the epidemiology of GNBSIs in an unselected population in Singapore. This study aimed to evaluate the epidemiology of two of the most common GNBSIs, namely those caused by E. coli and K. pneumoniae, and identify the associations between 30-day all-cause mortality and inappropriate empirical antibiotic use in an unselected population at a general hospital in Singapore.

METHODS

A retrospective cohort study was conducted at a 700-bed general hospital in Singapore. The hospital provides acute medical and surgical care to patients and predominantly serves the communities in the western region of Singapore.

The study was conducted between 1 November 2015 and 31 October 2017 (two years). All patients (>18 years) with mono-bacterial blood cultures positive for E. coli and K. pneumoniae during the study period were included. Patients with further positive blood cultures taken within 14 days from the first specimen and positive for the same organism were considered to have the same episode of bacteraemia.

The following data were collected from the Department of Medical Informatics and the electronic medical records (Epic Systems Corp, Verona, WI, USA) for all included cases: demographic data, comorbidities, onset of infection, timing of specimen collection, culture results, other investigations, treatments, devices inserted/manipulated (e.g. central venous catheters, urinary catheters) over the preceding 28 days, procedures and surgeries 28 days prior to admission, and mortality data up to 30 days. In addition, total bed days and total number of admissions were obtained for each year. All blood culture isolates that were included in the final data analysis were successfully linked to antibiotic susceptibility data from the laboratory information system (Epic Systems Corp).

Ethical approval was obtained from the Domain Specific Review Board of National Healthcare Group, Singapore (reference no. 2018/00253). Findings are reported in accordance with the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) guidelines.

We adopted surveillance definitions from Public Health England.[8] A healthcare associated BSI was defined as a laboratory-confirmed positive blood culture for a Gram-negative pathogen in patients within 48 hours of hospital admission and who had received healthcare in either the community or the hospital in the previous 28 days (community onset, healthcare associated; HCA-BSI), whereas a BSI developed more than 48 hours after hospital admission was considered hospital-acquired BSI (HA-BSI). Patients with GNBSIs detected within 48 hours of hospital admission and who had not received healthcare in either the community or hospital in the previous 28 days were considered to have community-acquired BSIs (CA-BSIs).

Multidrug resistant Gram-negative bacteria were defined as those with ‘non-susceptibility to at least one agent in three or more antimicrobial classes’.[9] Empirical antibiotic therapy was defined as initiation of antibiotics for the treatment of infection before antibiotic susceptibility data was available. Antibiotic therapy was considered inappropriate if the blood culture isolate did not show in vitro susceptibility to any of the antibiotics administered prior to availability of susceptibility data.

The source of infection was defined as the underlying primary focus of bacteraemia, and it was determined based on the clinical notes of the attending physician or infectious diseases physician. If the source of infection was equivocal, it was determined by the researchers based on clinical notes and investigations; however, in an event of sustained ambiguity, the source of infection was classified as ‘unknown’. The encounters were classified as disseminated in cases where there was objective evidence of multiple sources of infection.

A detailed analysis was undertaken for E. coli and K. pneumoniae. Data analysis was performed using IBM SPSS version 23.0 (SPSS Inc., Chicago, IL, USA). A P value <0.05 was considered significant. Categorical variables were expressed as frequency and percentage, and general associations between categorical variables were examined using the Chi-square test. Logistic regression was used to determine independent predictors of 30-day all-cause mortality and the patient-level predictors of the aetiologies of BSI.

RESULTS

We identified a total of 1,699 GNBSI episodes from the database. Cases with Gram-negative organisms other than E. coli and K. pneumoniae (n = 527), duplicates (n = 162) and incomplete information on most variables (n = 3) were excluded from data analysis, leaving a total of 1,007 cases comprising 700 E. coli and 307 K. pneumoniae BSI [Figure 1].

Figure 1.

Case-selection flowchart shows selection process, with 1,007 encounters included in the analysis after the initial exclusions.

The overall incidences of E. coli and K. pneumoniae were 87 per 10,000 admissions (194.1/100,000 bed days) and 38.1 per 10,000 admissions (85.1/100,000 bed days), respectively. The incidence of E. coli HA-BSI was 11.1 per 100,000 bed days and that of K. pneumoniae was 9.98 per 100,000 bed days. The mean age of the study participants was 71.2 years (standard deviation 14.2). A vast majority of cases were seen in patients aged ≥65 years, with significant male and female preponderance in K. pneumoniae and E. coli, respectively [Figure 2].

Figure 2.

Chart shows distribution of age categories across the two aetiologies, Escherichia coli and Klebsiella pneumonia.

Overall, CA-BSI accounted for 688 (68.3%) cases, and HCA-BSI and HA-BSI accounted for 177 (17.6%) and 76 (7.5%) cases, respectively. We were unable to classify the onset of BSI in 66 (6.5%) cases owing to lack of pre-hospitalisation data from the medical records [Table 1].

Table 1.

Summary of findings for the causative organism for BSIs stratified based on aetiology.

Characteristic	n (%)		P
	E. coli (n=700)	K. pneumoniae (n=307)
Incidence
Overall admissions	87/10,000	38.1/10,000
Hospital-acquired-BSI bed days	11.1/100,000	9.98/100,000
Agea (yr)	71.5±14.1	68.2±14.2	0.68
Gender
Female	431 (61.6)	121 (39.4)	<0.0001
Male	269 (38.4)	186 (60.6)	<0.0001
Onset of infection
Community-acquired BSI	488 (69.7)	200 (65.1)	0.341
Healthcare-associated BSI	127 (18.1)	50 (16.3)	0.611
Hospital-acquired BSI	40 (5.7)	36 (11.7)	0.001
Insufficient data	45 (6.4)	21 (6.8)
Source of infection
Upper urinary tract	448 (64.0)	115 (37.5)	<0.0001
Hepatobiliary	110 (15.7)	39 (12.7)	0.215
Liver abscess	5 (0.7)	61 (19.9)	<0.0001
Gastrointestinal	32 (4.6)	13 (4.2)	0.812
Lower respiratory tract	31 (4.4)	19 (6.2)	0.236
Unknown	59 (8.4)	35 (11.4)	<0.0001
Others	15 (2.1)	25 (8.1)	0.136
Antimicrobial susceptibility
Non-MDR	499 (71.3)	244 (79.5)	0.007
MDR	201 (28.7)	63 (20.5)
MDR community-acquired BSI	116 (57.7)	23 (36.5)	0.009
MDR healthcare-associated BSI	50 (24.9)	21 (33.3)	0.104
MDR hospital-acquired BSI	25 (12.4)	14 (22.2)	0.056
Insufficient data on the onset of infection	10 (5.0)	5 (7.9)
Appropriate empirical therapy
Yes	546 (78.0)	259 (84.4)	0.019
No	154 (22.0)	48 (15.6)
30-day all-cause mortality
Yes	54 (7.7)	29 (9.4)
No	646 (92.3)	278 (90.6)	0.36
MDR	24/201 (11.9)	10/63 (15.9)
Non-MDR	30/499 (6.0)	19/244 (7.8)	0.001
Community-acquired BSI	17/488 (3.5)	10/200 (5.0)
Healthcare-associated BSI	11/127 (8.7)	6/50 (12.0)	0.003
Hospital-acquired BSI	13/40 (32.5)	8/36 (22.2)	0.186
Insufficient data	13/45 (28.9)	5/21 (23.8)	0.056
Inappropriate empirical therapy	16/154 (10.4)	12/48 (25.0)
Appropriate empirical therapy	38/546 (7.0)	17/259 (6.6)	0.001

^aData presented as mean ± standard deviation. BSI: bloodstream infection, E. coli: Escherichia coli, K. pneumoniae: Klebsiella pneumoniae, MDR: multidrug resistance

Upper urinary tract infections (UUTIs), collectively represented cases of pyelonephritis and renal/perinephric abscesses, were reported as the most common focus of infection, irrespective of the onset of BSI in both organisms (n = 563, 55.9%). Overall, hepatobiliary source was the second commonest cause (n = 149, 14.8%). According to the logistic regression analysis, UUTI was significantly associated with E. coli (adjusted odds ratio [OR] 7.12, 95% confidence interval [CI] 2.14–23.68; P = 0.001), whereas pyogenic liver abscess was associated with K. pneumoniae (adjusted OR 0.15, 95% CI 0.03–0.67; P = 0.013). Furthermore, hepatobiliary (excluding liver abscesses) and gastrointestinal sources were significantly associated with E. coli.

Ciprofloxacin non-susceptibility in E. coli demonstrated a significant year-on-year increase from 25.9% to 33.7% (P = 0.014). No significant trends were observed for K. pneumoniae or MDR rates in both organisms. Regression analysis confirmed that the MDR rate was significantly associated with E. coli isolates (adjusted OR 1.545, 95% CI 1.022–2.335; P = 0.039). Highest MDR rates were observed in HA-BSI for both E. coli and K. pneumoniae [Table 1].

The sources of infection and the aetiologies of BSI did not demonstrate significant associations with 30-day all-cause mortality. However, CA-BSI was significantly associated with lower mortality (adjusted OR 0.13, 95% CI 0.05–0.33; P < 0.001) [Table 2]. Logistic regression analysis revealed that inappropriate choice of empirical antibiotic therapy was significantly associated (adjusted OR 3.38, 95% CI 1.47–7.77; P = 0.004) with higher rates of 30-day all-cause mortality [Table 2].

Table 2.

Summary of the factors associated with 30-day all-cause mortality in patients with Escherichia coli and Klebsiella pneumoniae bloodstream infections.

Factor	Adjusted OR	SE	95% CI (upper, lower)	P
Recurrent bacteraemia
No	1
Yes	1.164	0.46	0.472, 2.869	0.742
Study year
Second year	1
First year	1.052	0.313	0.57, 1.942	0.871
Gender
Female	1
Male	0.801	0.366	0.391, 1.64	0.544
Comorbidity
No	1
Diabetes mellitus	1.084	0.365	0.53, 2.217	0.824
Hypertension	1.059	0.396	0.487, 2.304	0.884
Hypercholesterolaemia	1.378	0.366	0.673, 2.824	0.381
Chronic kidney disease	1.505	0.449	0.624, 3.631	0.363
Ischaemic heart disease	0.449	0.366	0.219, 0.919	0.029
Cerebrovascular accident	1.959	0.422	0.856, 4.482	0.111
Hepatitis	1.428	0.874	0.258, 7.916	0.684
Cirrhosis	0.408	0.554	0.138, 1.209	0.106
Chronic obstructive pulmonary disease	1.867	0.736	0.441, 7.897	0.396
History of cancer	0.542	0.38	0.257, 1.141	0.107
Onset of infection
Insufficient data	1
Community-acquired BSI	0.129	0.487	0.05, 0.334	<0.001
Healthcare-associated BSI	0.618	0.521	0.223, 1.715	0.355
Hospital-acquired BSI	2.526	0.6	0.779, 8.193	0.123
Upper urinary tract infection
No	1
Yes	1.812	1.141	0.194, 16.952	0.602
Hepatobiliary
No	1
Yes	0.924	1.188	0.09, 9.474	0.947
Liver abscess
No	1
Yes	1.536	1.606	0.066, 35.778	0.789
Gastrointestinal
No	1
Yes	0.619	1.259	0.052, 7.296	0.703
Unknown source of infection
No	1
Yes	0.814	1.191	0.079, 8.408	0.863
Blood culture results
K. pneumoniae	1
E. coli	0.628	0.387	0.294, 1.341	0.23
Multidrug resistant organisms
No	1
Yes	0.767	0.454	0.315, 1.866	0.559
Inappropriate empirical antibiotic therapy
No	1
Yes	3.381	0.425	1.471, 7.769	0.004
Immunocompromised host
No	1
Yes	0.053	1.407	0.003, 0.84	0.037
Age	1.028	0.017	0.995, 1.061	0.101
Absolute neutrophil count	1.064	0.019	1.026, 1.105	0.001

BSI: bloodstream infection, CI: confidence interval, OR: odds ratio, SE: standard error

DISCUSSION

To our knowledge, this is the first study to investigate the epidemiology of E. coli and K. pneumoniae BSI in an unselected population in Singapore. In keeping with other studies, E. coli was the commonest GNBSI.[1,10,11] Our incidence rate of E. coli and K. pneumoniae — 87 per 10,000 admissions (194.1/100,000 bed days) and 38.1 per 10,000 admissions (85.1/100,000 bed days), respectively — is much higher than that reported in other similar studies.[12-14]

The most common source of BSI in both organisms was UUTIs. Overall, 24% of patients had possible contributory factors for developing UUTIs such as urinary catheter inserted/manipulated over the preceding 28 days, UTI treatment over the preceding 28 days and recurrent UTI. Of all UUTI patients in our study, 13.8% had a urinary catheter inserted/manipulated in the 28 days prior to onset of BSI. However, what proportion of these catheters was appropriate or adequately monitored is unknown. These findings suggest that a proportion of BSI could potentially be reduced by optimal treatment of UTIs and proper management of urinary catheters. Community-based healthcare plays an important role in reducing E. coli BSIs, as the majority of BSIs secondary to UUTIs are CA-BSIs (54.0%). Other sources of BSIs may offer less potential for interventions to reduce the burden of BSIs unless related to procedures or associated with devices.

Antimicrobial resistance rates were high for both organisms, with a higher MDR rate in E. coli (n = 201, 28.7%) compared to K. pneumoniae (n = 63, 20.5%). The MDR rates were significantly higher in HCA-BSIs and HA-BSIs than in CA-BSIs in both organisms [Tables 1 and 3]. This was consistent with the other studies despite varying definitions of MDR used by the researchers. Although our current carbapenem resistance rate remains low, there is a risk of increasing rates owing to rising cephalosporin and piperacillin/tazobactam resistance encouraging the use of carbapenems.

Table 3.

Summary of the factors associated with Escherichia coli bloodstream infections.

Covariate	Adjusted OR	SE	95% CI (upper, lower)	P
Recurrent bacteraemia
No	1
Yes	0.833	0.255	0.506, 1.372	0.473
Study year
Second year	1
First year	1.172	0.164	0.849, 1.617	0.335
Gender
Female	1
Male	0.542	0.19	0.373, 0.787	0.001
Age category (yr)
≥85	1
15–44	1.64	0.459	0.667, 4.035	0.282
45–84	1.399	0.243	0.87, 2.251	0.166
Admitting ward/specialty
Intensive care unit	1
Medical	0.676	0.516	0.246, 1.858	0.448
Surgical	0.813	0.555	0.274, 2.413	0.709
Onset of infection
No sufficient data	1
Community-acquired BSI	0.796	0.328	0.419, 1.512	0.485
Healthcare-associated BSI	0.699	0.367	0.341, 1.435	0.33
Hospital-acquired BSI	1.548	0.399	0.708, 3.385	0.273
MDR
No	1
Yes	1.545	0.211	1.022, 2.335	0.039
Comorbidity
No	1
Diabetes mellitus	0.556	0.189	0.383, 0.806	0.002
Hypertension	0.862	0.215	0.565, 1.314	0.49
Hypercholesterolaemia	1.294	0.187	0.897, 1.868	0.168
Chronic kidney disease	0.617	0.218	0.402, 0.947	0.027
Ischaemic heart disease	1.341	0.203	0.9, 1.998	0.149
Cerebrovascular accident	1.345	0.215	0.883, 2.047	0.168
Hepatitis	0.718	0.352	0.36, 1.43	0.345
Cirrhosis	0.865	0.367	0.421,	0.693
Chronic obstructive pulmonary disease	2.395	0.483	0.929, 6.169	0.071
History of cancer	0.791	0.242	0.492, 1.27	0.331
Upper urinary tract infection
No	1
Yes	7.121	0.613	2.142, 23.677	0.001
Hepatobiliary
No	1
Yes	5.401	0.625	1.586, 18.393	0.007
Liver abscess
No	1
Yes	0.152	0.761	0.034, 0.675	0.013
Gastrointestinal
No	1
Yes	6.272	0.699	1.593, 24.703	0.009
Lower respiratory tract infection
No	1
Yes	3.061	0.694	0.785, 11.932	0.107
Unknown source of infection
No	1
Yes	3.394	0.642	0.964, 11.947	0.057

BSI: bloodstream infection, CI: confidence interval, OR: odds ratio, SE: standard error

Our mortality rates are lower than those reported in other studies, where E. coli mortality ranged from 11% to >30%. This variation is likely due to some studies examining specific patient populations.[2,10,15] Similarly, the mortality in K. pneumoniae demonstrates a wide variation.[10,15-17] Our low mortality rate may be attributable to the higher proportion of patients with UUTIs (n = 563, 55.9%), higher rates of appropriate empirical antibiotic therapy (n = 805, 79.9%), and the lack of severely immunocompromised patients or low severity of sepsis. Mortality rates were higher in HA-BSIs and HCA-BSIs as opposed to CA-BSIs for both organisms, and the latter was significantly associated with a lower mortality, which was consistent with the previous studies.[10,15,17] Logistic regression analysis demonstrated that inappropriate empirical antibiotic therapy was significantly associated with increased 30-day all-cause mortality (adjusted OR 3.38, 95% CI 1.47–7.77; P = 0.004).

Our study has several limitations. Firstly, it was a retrospective study carried out in a single acute hospital. Thus, it is not representative of the overall epidemiology in Singapore. Secondly, we did not evaluate the severity of BSIs/sepsis at the time of specimen collection, and therefore, this was not included in the regression analysis as a confounder. It is possible that a lower proportion of patients who had organ dysfunction or septic shock in our study may have underestimated our mortality rates. Thirdly, we were unable to study long-term trends as our hospital was opened in mid-2015. Finally, we did not have access to all healthcare services received in the community, and therefore, a proportion of patients categorised as CA-BSIs may actually have HCA-BSIs.

We identified relatively high incidence rates and high MDR rates of GNBSI in our healthcare setting. Our study adds to existing evidence that the inappropriate empirical antibiotic therapy is significantly associated with higher 30-day all-cause mortality. Moreover, we recognise optimal treatment of urinary tract infections and proper management of urinary catheters, in particular, in the community setting as a potential intervention to reduce the incidence of GNBSIs. Given the current incidence rates and MDR rates of GNBSIs, we recommend concerted and streamlined efforts to bring various elements of existing surveillance mechanisms and interventions together.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.