Extended-Spectrum β-Lactamase-Producing Escherichia coli and Klebsiella pneumoniae: Risk Factors and Economic Burden Among Patients with Bloodstream Infections

Introduction

Bloodstream infections (BSI) rank among the most prevalent and severe bacterial challenges in clinical medicine, significantly contributing to global disease burden.^1–3 A European study estimated over 160,000 healthcare-associated (HA) primary BSIs in the European Union and the European Economic Area (EU/EEA) per year, leading to more than 24,000 fatalities.^4,5 This burden is surpassed only by hospital-acquired pneumonia (HAP) among HA infections, presenting a formidable public health crisis. Furthermore, in 2019, antimicrobial resistance (AMR) was a complicating factor in approximately 1,500,000 BSI-related deaths worldwide, severely limiting treatment options and escalating the healthcare challenge.²

Klebsiella pneumoniae (KP) and Escherichia coli (EC) stand out as leading pathogens in bloodstream infections (BSI), with their global prevalence escalating notably in the past decade.^3,5 In China, EC and KP are found in 35% of blood samples, with elevated presence of resistance mechanisms, including ESBL carrying rates (surpassing 40% among isolates).⁶ A Guangdong Province study highlighted that over 40% of pediatric EC and KP isolates are ESBL-producers, hinting at possibly higher current rates.⁷ ESBL genes, predominantly housed on mobile genetic elements (MGE) like plasmids, facilitate widespread transfer between EC and KP.^8,9 These genes not only confer resistance to β-lactam antibiotics like cephalosporins but often coexist with multiple drug-resistance genes,^10,11 complicating clinical treatment decisions and potentially escalating patient care costs.¹²

Literature has suggested differing attempts to quantify the economic and burden impact of ESBL on EC/KP-induced bacteremia, including the importance of study designs.¹³ Recent figures from the World Health Organization (WHO) Global Antimicrobial Resistance and Surveillance System (GLASS) showed that the proportion of BSIs caused by third-generation cephalosporins-resistant EC was 3-times higher in low- and middle-income countries (LMICs) compared to high-income countries, (58.3% and 17.53%, respectively).¹⁴ A meta-analysis showed that patients with antibiotic-resistant Gram-negative BSI in China had higher excess mortality, length of hospital stay (LOS) and intensive care unit (ICU) admissions compared with other LMICs.¹ However, the study did not differentiate regions across China, which largely failed to account for differences in epidemiology and clinical impacts across China.¹⁵

Here, we explore the epidemiological status and disease burden, including economic impacts, of local carriers of ESBL bacteria through a single-center retrospective study.

Data Source and Methods

Patients and Sample Selection

We extracted EC and KP blood specimen data from the laboratory’s electronic system, adhering to specific criteria: (1) Positive blood cultures for EC or KP, (2) Clinical signs of BSI, and (3) Comprehensive clinical, microbiological, and cost data. Only the first bacteremia episode was considered for patients with multiple EC BSI cases during readmission or within six months post-discharge. Exclusions applied to individuals under 18, those with incomplete records, or patients discharged, deceased, or under palliative care on the day of blood collection. Patients were categorized into two groups: those with ESBL-positive EC or KP infections (ESBL group) and those with ESBL-negative infections (non-ESBL group)

Microbiological Tests

Clinical isolates were identified using the matrix-assisted laser desorption ionization time-of-flight microbial identification system (MALDI-TOF) (biomacimrieux, France). VITEK-2 COMPACT automatic microbial identification system (biomacimrieux, France) was used to perform drug susceptibility testing. The ESBL production of EC/KP was determined by Gram Negative Susceptibility Card VITEK 2 AST-334 (biomacimrieux, France). Drug susceptibility card implemented CLSI M100.¹⁶

Clinical and Sociodemographic Characteristics

We collated clinical characteristics and sociodemographics. We included hospital-acquired BSI, defined as positive blood cultures 48 hours after admission, whereas community-acquired referred within 48 hours of admission. BSI’s severity was assessed utilizing the APACHEII score;¹⁷ the SOFA score was used to describe organ dysfunction/failure in patients with sepsis;¹⁸ the overall systemic health status of patients was determined by the age adjusted Charlson Comorbidity Index (aCCI).¹⁹ Initial antibiotic therapy, considered as empiric treatment pending susceptibility outcomes, was deemed appropriate if it included at least one in-vitro sensitive antibiotic.²⁰ Conversely, it was classified as inappropriate empirical antibiotic treatment (IEAT) if it failed to deliver any effective in-vitro active antimicrobial within 48 hours of diagnosing the infection. For ESBL-EK BSIs, oxyimino-β-lactams (eg, cefuroxime, cefotaxime, ceftriaxone, ceftazidime, cefepime, and aztreonam) were defined as inappropriate, regardless of in vitro susceptibility.²¹ Records of transfers in and out of the ICU and discharge records containing a summary of the patient’s condition at discharge were included. Patient’s age and sex were also included.

The clinical outcome of this study was in-hospital and 28-day post-discharge mortality in the ESBL and non-ESBL groups calculated as the total number of deaths/total number of cases.

Cost Analysis

Treatment costs were retrieved from the hospital information system. Direct and indirect costs (ie, lost productivity due to hospitalization and death) were analyzed. All costs were converted to USD (on average from 2020–2022, 1 USD = 6.69 RMB). The total direct expenses included medical care (treatment and nursing in the ward), treatment (surgical anesthesia, laboratory tests), rehabilitation, drugs (including antibacterial drugs), traditional Chinese medicine, diagnosis (pathology, microbiology, laboratory, imaging), disposable consumables, blood transfusion, and other expenses.

The indirect economic burden of EC bacteremia was analyzed through DALYs and human capital approaches. These methods impose an indirect economic burden equal to DALYs times gross-domestic product (GDP) per capita times productivity weights.²² Different age groups have different productivity weights, children aged 0–14 were assigned a weight of 0.15; 15–44 and 45–59 years old were assigned a weight of 0.75 and 0.80, respectively; and over 60 years old were assigned a weight of 0.1.²³

BSI-associated DALYs were calculated by summing all years of life lost (YLLs) and years living with disabilities (YLDs).²⁴ Other characteristics, such as life expectancy, age (standardized), time discount rate, and disability weights were included in the calculation of YLLs and YLDs.²⁴ We followed Murray and Lopez et al to calculate DALYs.^25,26 We used Brecht Devleesschauwer’s calculator²⁷ to compute economic costs following WHO’s recommendations.^27–29 Disability weights (D) were assigned in line with the acute infection’s severity, categorizing them into mild, moderate, and severe. These categories corresponded to disability weights of 0.006, 0.051, and 0.133, respectively. To calculate “L” (years lost due to death and years discounted due to disability), we used the Standard Expected Years of Life Lost (SEYLL) as a good approximation for life expectancy.³⁰

Statistical Analyses

We investigated the link between various clinical factors and ESBL EC/KP, employing propensity score matching to balance baseline characteristics associated with ESBL. Subsequently, we analyzed the ESBL’s influence on health outcomes and burdens, adjusted for PSM. First we divided clinical and sociodemographic variables into three categories.①Variables preceding the BSI onset (such as hospitalization time before bacteremia or the aCCI) were analyzed utilizing the chi-square (χ²) test to confirm the risk factors of ESBL infection, and at the same time, the difference between groups was excluded when calculating the economic cost.② Variables following the BSI onset, but its impact on clinical outcome of patient was independent of ESBL (such as SOFA score). These variables were not included in the risk factor analysis of ESBL in the study, but they were used as confounding factors to analyze the impact of ESBL on economic burden; excluding differences between groups due to the use of propensity score matching (PSM). ③Factors influenced by ESBL, like initial antimicrobial susceptibility, are not adjusted for economic assessments using PSM, and their mediating impact was verified.³¹

We computed descriptive statistics across all variables, and reported the mean, standard deviation or median (interquartile range [IQR]) for continuous variables, and proportions (%) for categorical variables. We used t-test (variables with normal distribution) and Mann–Whitney U-test (variables with non-normal distribution) to assess difference in means across ESBL and non-ESBL groups. Chi-square (χ2) test was used for categorical variables. We computed a PSM-corrected logistic regression using a backward stepwise method. We matched individuals from the ESBL and non-ESBL groups utilizing a caliper with a match tolerance equal to 0.02. Mediating effects were confirmed by Baron and Kenny’s steps for mediation analysis between ESBL and hospitalization costs via the mediating impact of LOS and ineffective empirical antimicrobial therapy.^32,33 We employed a holistic and comprehensive approach considering LOS and IEAT as simultaneous mediators of the association between ESBL infections and hospitalization costs. We utilized structural equation modeling (SEM) through the “lavaan” package (in the R software) to account for the correlations and interactions between mediators and ESBL and economic costs using linear regressions. Bootstrapped standard errors with 1000 replications. P-value of less than 0.05 was considered statistically significant. Statistical analyses were performed in R version 4.3.1 and SPSS version 26.0 (SPSS, Chicago, IL, USA).

Results

Analytical Sample

From January 2020 to December 2022, we identified a total of 372 EC and KP isolates from 367 patients with BSIs. After removing duplicates and missing data, we included 318 EC and KP unique isolates. We found 152 non-ESBL-EC/KP BSI patients (47.8%) and 166 (52.2%) ESBL-EC/KP BSI patients.

Descriptive Statistics Before Matching and by ESBL Group

Of the 318 patients, 179 were male, and the mean ages for ESBL-BSI and non-ESBL-BSI were 56.5 and 63 years, respectively (p=0.064). There was no statistically significant difference in strain distribution between the two groups in this study (p=0.407). The ESBL detection rates of EC and KP were 54.0% and 49.1%, respectively. The organ transplant rate (non-ESBL=1.3% versus ESBL=6%, p=0.028) and ICU admission rate (non-ESBL=7.2% versus ESBL=19.9%, p=0.001) were higher in the ESBL group before onset than in the non-ESBL group, but nosocomial infection rates were not significantly different between the two groups (non-ESBL=40.1% versus ESBL=35.5%, p=0.399).

The most common comorbidities in ESBL group and non-ESBL group were malignant tumor (non-ESBL=36.2% versus ESBL=26.5%), hepatobiliary disease (non-ESBL=14.5% versus ESBL=22.9%) and diabetes (non-ESBL=14.5% versus ESBL=16.9%). Among comorbidities, only moderate to severe nephropathy had a statistically significant difference between groups (6.6% for non-ESBL versus 15.7% for ESBL, p=0.011). Notably, while there were partial differences in comorbidity distribution between the two groups of patients, there was no statistically significant differences in quantified aCCI (non-ESBL=4(2,6) versus ESBL=4(3,6), p=0.232). Correspondingly, the APACHEII (non-ESBL=10(7,13) versus ESBL=12(7,19), p=0.001) score and SOFA score (non-ESBL=1(0,4) versus ESBL=3(0,7), p=0.004) reflecting the patient’s basic condition and the severity of sepsis in the ESBL group were higher than those in the non-ESBL group.

Although more carbapenems were used in the ESBL group (non-ESBL=41.4% versus ESBL=53%, p=0.039), the effective rate of empiric antibacterial therapy was lower than that of the non-ESBL group (non-ESBL=80.3% versus ESBL=58.4%, p<0.001), and the total mortality rate (non-ESBL=17.8% versus ESBL=27.1%, p=0.047) was higher than that of the non-ESBL group (Table 1).

Table 1 Characteristics of Patients with EC/KP BSIs, Stratified by ESBL Production

Estimation of the Propensity Score

The raw and PSM-adjusted ESBL+ and non-ESBL proportions are depicted in Table 1. We observed no significant difference in clinical characteristics considered as confounding factors between the two groups (all p>0.05), including gender, age, bacterial species, hospital infection rate, hospitalization time before infection, organ transplantation during hospitalization, empirical drug category, APACHE II score, SOFA score, and aCCI. We found similar distributions across ESBL proportions after correcting the estimates utilizing the PSM (Table 1).

Association between ESBL+ and outcome variables related to clinical burden

We noticed that after removing the interference of confounding factors, the ESBL group still had a lower rate of effective empirical antimicrobial therapy than the non-ESBL group (non-ESBL=79.3% versus ESBL=59.5%, p=0.001), longer total hospitalization time (ESBL=18(11, 28.5) versus non-ESBL=14(10, 22.5), p=0.02) and longer post-infection hospitalization time (non-ESBL=9(6, 13) versus ESBL=12(7.5, 19.5), p<0.001), but there was no significant difference in the overall mortality rate between non-ESBL group and ESBL group (non-ESBL=19.0% versus ESBL=20.7%, p=0.747) (Table 1).

Economic Costs

After PSM, the median total hospitalization cost was $5638 for non-ESBL patients and $7685 for ESBL patients (p=0.014) (Table 2). In the two groups, the cost of antibiotics accounted for 27.3% and 39.0% of the drug, respectively (p=0.002). The median cost of antibiotics during hospitalization was $400 for non-ESBL-EC BSI patients and $834 for ESBL-EC BSI patients, p<0.001) (Table 2 and Figure 1). In the non-ESBL-EC group after PSM matching, patients lost an average of 1.84 DALYs, whereas it was 2.12 among ESBL-EC group (mean test p=0.098). There was no significant difference in mean indirect loss in the ESBL-EC group compared to the non-ESBL-EC group.

Table 2 Costs of Patients with EC/KP BSIs, Stratified by ESBL Production

Figure 1 Economic costs by ESBL group before (A) and after (B) adjusting for PSM results. Abbreviations: ESBL, Extended Spectrum Beta-lactamase producing; PSM, Propensity score matching.

Mediation Analyses

ESBL was positively associated with IEAT (Estimate = 0.215, SE=, p<0.05) and LOS (Estimate = 0.151, SE=0.05, p<0.05) (Table 3). However, ESBL did not show a direct impact on economic costs (−0.03, SE=, p>0.05), whereby IEAT and LOS were significantly and directly associated with economic costs (Estimate= 0.13, SE=0.06, p-value=0.037, Estimate= 0.62, SE=0.14, p-value<0.001, respectively). All estimated variances for the residuals (errors) of the three models (a, b, and c) are significant, indicating variability around the predicted values of these variables (Table 3, variance). The total indirect impact of ESBL on cost was significant (Estimate= 0.12, SE=0.06, p-value<0.027), with marginal indirect impacts from IEAT and more significant impacts via LOS (Table 3 and Figure 2, impacts of indirect effects). The total indirect effect impact was significant suggesting that the relationship between ESBL and cost is largely or fully mediated by IEAT and LOS.

Table 3 Mediation Effect Analysis Results

Figure 2 Diagram of the mediation effects (LOS and IEAT) between ESBL and economic costs. Abbreviations: IEAT, Ineffective empirical antibiotic treatment; LOS, Length of hospital stay; ESBL, Extended Spectrum Beta-Lactamase.

Discussion

Our findings align with the China Antimicrobial Surveillance Network (CHINET)³⁴ regarding ESBL positivity in EC. However, the ESBL rate in Klebsiella pneumoniae (KP) among adults exceeded our previous provincial study (49.1% versus 28.7%).⁷ Compared to the non-ESBL group, patients with ESBL-related BSI faced higher empirical treatment failure, mortality rates, extended hospital stays, and increased medical expenses. Significant disparities were noted in effective empirical therapy, hospitalization duration, and costs between the ESBL and non-ESBL groups if cofounding factors were excluded.

We included DALYs to calculate the overall burden on patients and society, which indicated reduced productivity due to disease.³⁵ Similar to most studies included in a recent systematic literature review written by Zhen X et al,¹³ we found that patients with ESBL BSIs had longer hospital stays and higher financial burdens. We found a 33.3% increase in median post-bacteremia LOS (from 9 days to 12 days), a 28.6% increase in total LOS (from 14 days to 18 days), and a 36.3% increase in hospital costs (an additional $2047 per patient) in the post-PSM ESBL group.

The impact of initial treatment efficacy and antibiotic costs on the total direct burden remains debated, potentially due to regional variations in medical services and drug pricing.¹³ Our study revealed that ESBL significantly affected IEAT and LOS, which in turn significantly impacted hospitalization expenses. This suggests that targeting ESBL could indirectly reduce costs by altering IEAT and LOS. Investigating these mediating factors could further elucidate how ESBL influences hospitalization costs. In our analysis, the median difference in antimicrobial expenditure represented 21.2% (434/2037) of the total median hospitalization cost difference. Notably, the ESBL group incurred higher antimicrobial costs, contributing substantially to overall Western medicine expenses (27.3% versus 39%, p = 0.002), indicating that the elevated drug costs associated with ESBL-BSI also escalated overall hospitalization expenses.

In our study, we noted that the average disability years (1.84–2.12 years) and the average hospitalization duration (14–18 days) contributed to less than 1% of the YLDs within the DALYs after disease weighting. This is expected due to the brief duration of bloodstream infections and their high mortality rate. We propose that for researchers lacking comprehensive data, relying solely on YLDs, in the absence of long-term sequelae associated with bacteremia, can adequately represent the disease burden in adults with acute bloodstream infections.

Our analysis revealed that carbapenem use by clinicians was influenced by factors other than ESBL presence alone (non-ESBL=45.4% versus ESBL=53%, p=0.039). However, this difference became statistically insignificant (p=0.153) after adjusting for SOFA and APACHE II scores. This suggests clinicians might prioritize clinical severity over ESBL-related risk factors when choosing treatments before receiving drug sensitivity reports. A notable finding was the lack of recorded antibacterial drug use information in the six months preceding admission in most electronic medical records. Given that non-β-lactam antibiotic use is also a reported risk factor for ESBL, this omission presents a significant concern.³⁶ Therefore, we advocate for detailed documentation of all antimicrobial drugs during admission consultations to enhance the precision of antimicrobial stewardship.

In previous studies, nosocomial infection was usually a high-risk factor for ESBL-producing bacterial infection.³⁷ In this study, despite the longer LOS in the ESBL group pre-PSM the nosocomial infection rates between the groups showed no significant difference (non-ESBL=40.1% versus ESBL= 35.5%, p=0.399). This indicates a widespread community transmission of blaESBL in the region. Given the substantial local ESBL prevalence, clinicians are advised to exercise caution in prescribing third-generation cephalosporins until definitive drug susceptibility results are obtained.^7,34 Studies in China indicate that using antimicrobials effective against ESBL-producing bacteria as empirical therapy can enhance cure rates and decrease both direct and indirect societal costs.³⁰ Clinics must, therefore, weigh individual patient needs against broader societal considerations. We advocate for a more rigorous implementation of the WHO’s ESBL detection tricycle plan and the AWaRe tool, along with the establishment of multi-dimensional molecular epidemiological surveillance and robust antimicrobial stewardship programs, to address the complex, multidisciplinary challenge posed by antimicrobial resistance.^38–41

Despite controlling for various confounders, potential selection biases in our study cannot be entirely ruled out. To optimize patient inclusion in the PSM, we did not factor in individual comorbidities, given the varying weights assigned to diseases in the Charlson score. While previous research suggests a link between ESBL plasmids and certain virulence factors,⁴² our study focused solely on the impact of drug resistance genes. We mitigated potential biases from virulence factors by integrating metrics like the SOFA score into our PSM criteria.

In conclusion, our findings affirm that BSIs caused by ESBL-producing EC/KP diminish the effectiveness of empirical treatments, extend hospitalization, and heighten financial strain on patients. The study underscores the critical need for accurate ESBL risk assessment and precise antibiotic utilization to alleviate patient burdens. Ultimately, ongoing epidemiological monitoring and risk evaluation of antimicrobial resistance are essential for both clinical management and effective antimicrobial stewardship.