Introduction
Escherichia coli causes a significant burden of disease in humans since it is the most common cause of bacterial bloodstream infections (BSIs) [Reference Skogberg1–Reference Bonten3]. The importance of using population-based study designs for BSI research has been established [Reference Laupland4–Reference Søgaard6]. By including all incident BSIs in a defined and isolated geographical area, selection biases are minimised and the area population can be used as the population-at-risk [Reference Laupland5]. This allows BSI epidemiology to be studied at a population-level and gain insight into the burden of disease. Population-based studies are essential when researching E. coli BSIs because the vast majority of cases are community-onset [Reference Laupland7–Reference Abernethy12], and not all require hospitalisation [Reference Laupland7]. Therefore, a solely hospital-based data collection approach for E. coli BSI research does not achieve complete case ascertainment and defining the population-at-risk will not be possible.
Mortality and length of hospital stay (LOS) are important measures of burden of disease from the patient and healthcare system perspectives [Reference Naylor13]. Two previous population-based E. coli BSI studies used logistic regression analyses to explore factors associated with all-cause mortality while adjusting for confounding (in-hospital mortality [Reference Laupland7] and 30-day mortality [Reference Abernethy12]). Median total LOS has been reported by two previous population-based studies [Reference Laupland7, Reference Williamson8]. Both used non-parametric tests to assess whether median LOS differed between location of onset types [Reference Laupland7, Reference Williamson8]. We are not aware of a previously published population-based E. coli BSI study using multivariable regression analysis to explore post-infection LOS. Multivariable regression analyses facilitate more accurate estimation of the measures of associations by controlling for confounding variables and exploration of interaction effects. Detailed analyses of E. coli BSI characteristics and outcomes in the western interior area of British Columbia (BC), Canada have not been previously reported.
Using population-based data from a geographically isolated Canadian area, we aimed to: (1) describe characteristics, antimicrobial resistance (AMR), and outcomes of E. coli BSIs; and (2) evaluate factors associated with 30-day mortality and long post-infection LOS.
Methods
Study protocol
Our population-based cohort study included all incident E. coli BSIs in western interior area residents (BC, Canada; 2018 population 191 000) [14] from 1 April 2010 to 31 March 2020, which were the first 10 years of data from the surveillance database. A subsequent E. coli BSI in the same patient was only included as a separate incident case if it was diagnosed at least 1 month from the preceding BSI that had clinically resolved after treatment [Reference Williamson8, Reference Sawatwong11, Reference Blandy15–Reference Madsen17]. Details regarding the area's healthcare infrastructure, population, surveillance database and microbiological methodology have been previously published [Reference Laupland18, Reference Laupland19]. The Interior Health and University of Guelph Research Ethics Boards approved this study and granted waivers of individual informed consent (2013-14-052-I and 2018-10-050, respectively).
Laboratory procedures
A single draw was used to collect blood into a pair of aerobic and anaerobic bottles (blood culture set). During the study, it was standard protocol to draw two blood culture sets from different sites on each patient. The BacT/Alert 3D systems was used for blood culturing (bioMérieux, France). Standard methodology and established protocols were used for bacterial isolation and speciation. Antimicrobial susceptibility testing (AST) was performed using broth microdilution (Vitek 2XL, bioMérieux, France), ETEST®, or Kirby Bauer disk diffusion depending on the antimicrobial. Antimicrobials assessed for susceptibility in BSI E. coli isolates included, but were not limited to: ampicillin, cefazolin, ceftriaxone, ciprofloxacin, gentamicin and trimethoprim sulfamethoxazole (TMS). The results of AST were interpreted as susceptible, intermediate and resistant using Clinical Laboratory Standards Institute (CLSI) clinical breakpoints at the time of AST [20]. Escherichia coli isolates were screened and confirmed for extended-spectrum β-lactamase (ESBL)-production using ESBL detection disc sets (MASTDISC® Combi, Mast Group Ltd, United Kingdom).
Data and definitions
Escherichia coli BSIs were identified by the regional microbiology laboratory and an experienced infectious disease specialist (Kevin B. Laupland (KBL)) confirmed incident cases. Kevin B. Laupland reviewed electronic medical records from the regional clinical information system to obtain and determine case data including: bacterial species isolated from blood culture, blood culture date, patient's age and sex, hospital admission date, location of onset, infection focus, Charlson comorbidity index (CCI), antimicrobial susceptibility data, LOS, and 30-day all-cause mortality. The electronic medical record is directly linked to the regional enterprise master patient index and provincial vital statistics database, which provided the date of death for deceased patients. The BSI location of onset was characterised as nosocomial if the first positive blood culture was obtained at least 48 h after hospital admission or within 48 h of hospital discharge; otherwise, it was characterised as community-onset [Reference Friedman21]. Using the definitions proposed by Friedman et al. [Reference Friedman21], community-onset BSIs were categorised into healthcare-associated or community-acquired [Reference Friedman21]. Detailed medical record review by KBL facilitated determination of the infection foci category: no focus (primary), soft tissue, pneumonia, cardiovascular, intra-abdominal and biliary, and urogenital. The CCI was constructed from medical record data using established methodology [Reference Charlson22]. Intermediate susceptibility was considered resistant for analyses. Bloodstream infections were classified as multidrug resistant (MDR) if resistant to three or more antimicrobial categories [Reference Magiorakos23]. Polymicrobial infections were E. coli BSIs that had at least one more bacterial species isolated within 2 days of the E. coli culture [Reference Nielsen2, Reference Weinstein24]. Thirty-day all-cause mortality was defined as death from any cause within 30-days after the E. coli blood culture. In-hospital all-cause mortality was defined as death from any cause during hospitalisation. Thirty-day in-hospital all-cause mortality was death from any cause during the first 30-days of hospitalisation after the E. coli blood culture. Total LOS was the difference in days between hospital admission and discharge. Pre-infection LOS was the difference in days between hospital admission and the day the blood culture was drawn. Post-infection LOS was the difference between the day the blood culture was drawn and hospital discharge.
Statistical analysis
We performed data analyses in Stata 16.1 [25]. To summarise categorical variables, we calculated proportions (sex, location of onset, focus of infection (urogenital and non-urogenital), AMR data) and determined median and interquartile range (IQR) to summarise continuous variables (age, CCI, pre-infection LOS). We calculated incidence and mortality rates by dividing the number of incident E. coli BSIs or 30-day deaths by the surveillance population during the study period [14, Reference Dohoo, Martin and Stryhn26]. We calculated 30-day case fatality risk by dividing the number of 30-day deaths by the number of incident E. coli BSIs [Reference Dohoo, Martin and Stryhn26]. To summarise LOS measures, median and IQR were calculated.
To identify factors significantly associated with 30-day all-cause mortality in E. coli BSIs, we used a logistic regression model [25, Reference Dohoo, Martin and Stryhn26], which was also used to estimate factors significantly associated with long post-infection LOS (defined as post-infection LOS greater than the 75th percentile) [Reference Blandy15]. The latter model was restricted to the subset of E. coli BSI patients that were hospitalised and survived to hospital discharge. The models were fitted with generalised estimating equations using exchangeable correlation structures to account for the lack of independence from some patients having more than one incident BSI during the study [25, Reference Dohoo, Martin and Stryhn26]. For inclusion in each logistic regression model, we considered the following 12 explanatory variables: age (years); sex (female, male); polymicrobial (monomicrobial, polymicrobial); ESBL (non-ESBL, ESBL); ciprofloxacin resistance (susceptible, resistant); MDR (non-MDR, MDR); infection focus (urogenital, non-urogenital); location of onset (community-acquired, healthcare-associated, nosocomial); pre-infection LOS (days); CCI; season (spring (March–May), summer (June−August), fall (September–November), winter (December–February)); and study year (1 through 10, e.g. study year 1–1 April 2010 to 31 March 2011). Escherichia coli BSI patients that did not require hospitalisation were assigned a pre-infection LOS of zero days to facilitate regression modelling. First, we assessed the linearity assumption for continuous variables (age, pre-infection LOS, and CCI) by examining a locally weighted regression (lowess) of the log odds of the outcome and continuous variable. If it was not appropriate to model them as a continuous variable, then they were modelled as a categorical variable. Next, we completed univariable analysis, and to avoid collinearity in subsequent multivariable model fitting we checked for high correlation between explanatory variables using a Phi coefficient or Spearman's rank correlation coefficient, as appropriate (ρ ⩾ |0.8| was the cut-off value). Multivariable logistic regression models were built using a backwards stepwise model building process starting with all of the explanatory variables that met the liberal P-value threshold of P < 0.2 in univariable analysis, and were not highly correlated [Reference Dohoo, Martin and Stryhn26]. If the explanatory variables, MDR and either ESBL or ciprofloxacin resistance, met the criteria to be considered in the multivariable model, separate multivariable models were built for inclusion of MDR, and ESBL and ciprofloxacin resistance since the latter variables could be a component of MDR. We considered the following two-way interaction effects for inclusion in the multivariable regression models: age and sex, ESBL and ciprofloxacin resistance, age and explanatory variables in the final main effects multivariable model, and sex and explanatory variables in the final main effects multivariable model. Variables remained in the final multivariable model if statistically significant (α = 0.05), part of a significant interaction term, and/or a confounding variable (based on >20% change in another variable's coefficient and meeting causal criteria) [Reference Dohoo, Martin and Stryhn26]. Odds ratios (OR) were reported with 95% confidence intervals (CI).
Missing data
Data were complete for the main descriptive and regression modelling analyses. Incident E. coli BSIs were removed from specific analyses if data were incomplete. The only missing data were for susceptibility to gentamicin, ampicillin and TMS where data for four, two, and 43 incident E. coli BSIs were missing, respectively.
Results
Incidence and characteristics
From 1 April 2010 to 31 March 2020, we identified 1080 incident E. coli BSIs from 1009 patients. Three patients had four incident E. coli BSIs, six patients had three, 50 patients had two, and the remaining 950 patients had a single E. coli BSI. Our population-based study had 1.8 million person-years of follow-up. The overall crude incidence rate was 59.1 E. coli BSIs/100 000 person-years. The age distribution of incident E. coli BSIs was left skewed with a median of 70.8 years (IQR 58.6−81.3) and 57.8% of BSIs were in females (624/1080) (Figure S1, Table 1).
Characteristics of cases with incident E. coli bloodstream infections that did and did not survive 30 days after blood culture collection based on data from a population-based cohort study in the western interior area of British Columbia, Canada (April 2010−March 2020)
IQR, interquartile range; BSI, bloodstream infection; LOS, length of hospital stay; CCI, Charlson comorbidity index.
a Antimicrobial resistance variables used in regression analyses. Details of antimicrobial susceptibility for additional antimicrobials are available in the results section.
b When those not hospitalised were removed (n = 154), the only change was in the 75th percentile for those that died within 30 days; median (IQR) – 0 (0–4).
Polymicrobial infections occurred in 7.8% of incident E. coli BSIs (84/1080) with one E. coli BSI case involving a total of four bacterial species, 12 having three bacterial species, and 71 having two bacterial species (Table 1, Table S1). The five most common bacterial species that were present in polymicrobial infections with E. coli were: Klebsiella spp. (28 BSIs), Enterococcus spp. (16), Pseudomonas spp. (12), Staphylococcus spp. (7), and Streptococcus spp. (7) (Table S1).
Almost half of incident E. coli BSIs were pansusceptible (48.4%, 523/1080). The proportion of E. coli BSIs that were resistant to ciprofloxacin, ceftriaxone, gentamicin, cefazolin, ampicillin, and TMS was 24.7% (267/1080), 10.6% (114/1080), 8.6% (92/1076), 18.6% (201/1080), 44.3% (478/1078), and 22.8% (236/1037), respectively. No meropenem resistance was reported and 8.4% of E. coli BSIs were ESBL-producing (91/1080) (Table 1). Overall, there were a small number of E. coli BSIs that were intermediate, and therefore considered resistant to ciprofloxacin (three isolates), gentamicin (1), cefazolin (3), and ampicillin (10), and there were no intermediate isolates for ceftriaxone or TMS. Multidrug resistance was present in 22.3% of E. coli BSIs (241/1080) (Table 1), with resistance in three (39.0%, 94/241) to seven (5.4%, 13/241) antimicrobial categories in these infections.
Escherichia coli BSIs were 90.1% (973/1080) community-onset with healthcare-associated (48.8%, 527/1080) occurring more frequently than community-acquired (41.3%, 446/1080) (Table 1). The underlying E. coli BSI foci from most to least common were urogenital (65.4%, 706/1080), intra-abdominal or biliary (23.5%, 254/1080), no focus (7.0%, 76/1080), respiratory (2.5%, 27/1080), soft tissue (0.9%, 10/1080), and orthopaedic (0.6%, 7/1080).
The distribution of CCI for incident E. coli BSIs was right skewed with a median of 2 (IQR 0–3) (Table 1). The CCI ranged from zero (35.1%, 379/1080) to 11 (0.2%, 2/1080). The three most common comorbidities were diabetes mellitus (23.5%, 254/1080), cancer (20.8%, 225/1080), and chronic pulmonary disease (15.1%, 163/1080) (Table 1).
Mortality
The 30-day all-cause case fatality risk was 11.5% (124/1080) and the crude 30-day mortality rate was 6.8 deaths/100 000 person-years. The descriptive statistics for characteristics of E. coli BSI cases that did and did not survive to 30 days after blood culture are in Table 1. Most cases with E. coli BSIs required hospitalisation (85.7%, 926/1080) and the in-hospital case fatality risk was 10.7% (99/926). The 30-day in-hospital case fatality risk was 9.9% (92/926). The continuous variables age, pre-infection LOS, and CCI did not meet the linearity assumption and they were categorised as follows: age category (<65-years-old, ⩾65-years-old); pre-infection LOS category (not hospitalised, pre-infection LOS < 2 days, pre-infection LOS ⩾ 2 days); and CCI category (CCI < 3, CCI ⩾ 3). We identified high correlation (ρ ⩾ |0.8|) between the variables for location of onset and pre-infection LOS category and we decided to proceed with the location of onset variable for multivariable modelling since this variable provided data on community-acquired and healthcare-associated E. coli BSIs. Age category, sex, polymicrobial, location of onset, infection focus, and CCI category were all considered for inclusion in our multivariable logistic regression model for 30-day mortality in E. coli BSIs (Table S2). When individual comorbidities were evaluated for their association with 30-day mortality using univariable logistic regression models, E. coli BSI cases with cancer, chronic pulmonary disease, congestive heart failure, myocardial infarction, dementia, mild liver disease, or cerebrovascular disease had significantly higher odds of dying within 30 days compared to those without the specific comorbidity (Table S2). Our final multivariable logistic regression model for 30-day mortality included age category, location of onset, infection focus and CCI category (Table 2). None of the interactions tested were statistically significant. Older E. coli BSI cases had significantly higher odds of 30-day mortality compared to younger cases (Table 2). Compared to community-acquired E. coli BSIs, cases with either healthcare-associated or nosocomial had significantly higher odds of 30-day mortality, and the odds were significantly higher for nosocomial E. coli BSIs compared to healthcare-associated (P = 0.004) (Table 2). Cases with a BSI focus that was anything other than urogenital had higher odds of dying within 30 days compared to those with a urogenital focus (Table 2). Cases with a CCI ⩾ 3 had higher odds of 30-day mortality compared to those with a CCI < 3 (Table 2).
The adjusted ORs for the multivariable logistic regression model estimating the associations between the explanatory variables and 30-day mortality in E. coli bloodstream infections based on data from a population-based cohort study in the western interior area of British Columbia, Canada (April 2010 to March 2020)a
aOR, Adjusted odds ratio; CI, Confidence interval.
a Logistic regression model fitted with generalised estimating equation using exchangeable correlation structure to account for lack of independence from some patients having more than one incident bloodstream infection.
b Overall P-value for variable estimated using a Wald test.
c Contrast comparing nosocomial to healthcare-associated aOR 2.16 (95% CI 1.29–3.65) P = 0.004.
Length of hospital stay
The distributions for all LOS measures were right skewed. For cases with incident E. coli BSIs that required hospitalisation and survived to discharge (n = 827), the median LOS prior to infection was 0 days (IQR 0–0), the median total LOS was 7 days (IQR 4–12), and the median post-infection LOS was 6 days (IQR 4–11). For cases that died in hospital (n = 99), the median LOS prior to infection was 0 days (IQR 0–7), the median total LOS was 7 days (IQR 2–20), and the median post-infection LOS prior to in-hospital death was 3 days (IQR 1–11). The median post-infection LOS in E. coli BSIs that survived to discharge stratified by demographics and characteristics are presented in Table S3. The continuous variables age, pre-infection LOS, and CCI did not meet the linearity assumption and they were categorised as follows: age category (<65-years-old, ⩾65-years-old); pre-infection LOS category (pre-infection LOS<2 days, pre-infection LOS ⩾2 days); and CCI category (CCI < 3, CCI ⩾ 3). We identified high correlation (ρ ⩾ |0.8|) between the variables for location of onset and pre-infection LOS category and we decided to proceed with the location of onset variable for multivariable modelling since this variable provided data on community-acquired and healthcare-associated E. coli BSIs. Age category, polymicrobial, ESBL, ciprofloxacin resistance, MDR, location of onset, infection focus and CCI category were all considered for inclusion in our multivariable logistic regression model for long post-infection LOS in patients that survived to discharge (Table S4). When individual comorbidities were evaluated for their association with long post-infection LOS in patients that survived to discharge using univariable logistic regression models, E. coli BSI cases with diabetes mellitus, chronic pulmonary disease, congestive heart failure, renal disease, dementia, cerebrovascular disease, or diabetes mellitus with end organ damage had significantly higher odds of long post-infection LOS compared to those without the specific comorbidity (Table S4). Our final multivariable logistic regression model for long post-infection LOS in patients that survived to discharge included ESBL, location of onset, and CCI category (Table 3). None of the interactions tested were significant. In patients that survived to discharge, those with ESBL-producing E. coli BSIs had significantly higher odds of long post-infection LOS compared to those with non-ESBL-producing E. coli BSIs (Table 3). Compared to community-acquired and healthcare-associated E. coli BSIs, nosocomial cases had significantly higher odds of long post-infection LOS in patients that survived to discharge (Table 3). In patients that survived to discharge, those with a CCI ⩾ 3 had significantly higher odds of long post-infection LOS compared to those with a CCI < 3 (Table 3). A separate multivariable model was constructed for MDR, but MDR was not statistically significant (results not shown).
The adjusted ORs for the multivariable logistic regression model estimating the associations between the explanatory variables and long post-infection LOS in patients with E. coli bloodstream infection that survived to discharge based on data from a population-based cohort study in the western interior area of British Columbia, Canada (April 2010 to March 2020)a
aOR, Adjusted odds ratio; CI, Confidence interval.
a Logistic regression model fitted with generalised estimating equation using exchangeable correlation structure to account for lack of independence from some patients having more than one incident bloodstream infection.
b Overall P-value for variable estimated using a Wald test.
c Contrast comparing nosocomial to healthcare-associated aOR 5.02 (95% CI 2.97–8.47) P < 0.001.
Discussion
Our population-based study contributes important information regarding factors associated with mortality and LOS to the E. coli BSI literature. There were several key strengths of our study, including 10 years of population-based surveillance data, the geographically isolated nature of the area, and the availability of detailed clinical data. Notably, we have results on the burden of disease from E. coli BSIs at the population-level for the western interior of BC, since we have all incident E. coli BSIs, and associated 30-day deaths and hospitalisations. When we examine burden from the absolute perspective, the number of E. coli BSIs in the population, we report a considerable burden of disease with a crude incidence rate of 59.1 E. coli BSIs/100 000 person-years and a crude 30-day mortality rate of 6.8 deaths/100 000 person-years.
The epidemiological characteristics of the incident E. coli BSIs in our study are mostly similar to those from previous population-based studies. In previous studies, 72.8–91.1% of E. coli BSIs were community-onset, which consisted of 31.7–40.0% healthcare-associated and 34.0–59.4% community-acquired [Reference Laupland7–Reference Al-Hasan10, Reference Abernethy12]. Our study had a similar proportion of community-onset E. coli BSIs (90.1%), but we had more healthcare-associated (48.8%) than previous studies. When previous studies reported E. coli BSI foci, urogenital foci were the most common and this is consistent with our results [Reference Bonten3, Reference Bou-Antoun9, Reference Al-Hasan10, Reference Abernethy12]. Resistance to ceftriaxone and ciprofloxacin, ESBL-production, and MDR are of particular importance for E. coli infections due to their potential implications for treatment and their prevalence can vary widely between different studies. Previous studies reported third-generation cephalosporin resistance of 1.0–10.4% [Reference Laupland7–Reference Al-Hasan10, Reference Abernethy12], ciprofloxacin resistance of 7.0–18.4% [Reference Laupland7–Reference Al-Hasan10, Reference Abernethy12], and ESBL-producing E. coli in 2.0–26.8% [Reference Laupland7, Reference Williamson8, Reference Sawatwong11]. We had a higher prevalence of ciprofloxacin resistance (24.7%), however, differences in population and time periods make interpreting this difference difficult. We did not identify a previous population-based E. coli BSI study that reported MDR. Future studies could easily include the MDR metric generated from AMR data [Reference Magiorakos23]. Previous population-based E. coli BSI studies did not indicate if E. coli BSIs were polymicrobial or monomicrobial. In our study, 7.8% of incident E. coli BSIs were polymicrobial and we contend that this is an important detail to include, especially if performing risk factor analyses of health outcomes.
Mortality is an important measure of burden of disease [Reference Naylor13]. Population-based studies have reported E. coli 30-day all-cause case fatality risks of 8.0–18.2% [Reference Skogberg1, Reference Abernethy12, Reference Mehl16] and our 30-day all-cause case fatality risk of 11.5% is within the lower half of the range. Another previously used definition of mortality is in-hospital all-cause mortality. If we compare our 30-day case fatality risk (11.5%) to our 30-day in-hospital case fatality risk (9.9%), we note that the in-hospital measure is lower. If we take the in-hospital deaths that occurred within 30 days (n = 92) and used the total number of incident E. coli BSIs (n = 1080), then we have a 30-day in-hospital case fatality risk of 8.5%, which is considerably lower. In-hospital mortality has been shown to be a biased measure of mortality in population-based studies [Reference Laupland27] and our results provide further support of this bias. In-hospital mortality does not accurately reflect mortality in population-based studies because any deaths in patients that were never hospitalised or in those that died after discharge but before 30 days are not captured. Bacterium-attributable mortality where deaths are confirmed to be a result of the BSI is a less frequently used definition of mortality, and is often not practical or even possible to confirm. Patients with BSIs often have comorbidities and commonly this information is determined from medical record review, which makes the accurate attribution of cause of death to the BSI difficult. Therefore, all-cause mortality without restriction to in-hospital is the best definition for mortality in population-based studies and this is often done with a 30-day post-infection timeframe [Reference Laupland27].
The factors with significantly higher odds of 30-day mortality in E. coli BSIs in our study included cases ⩾65-years-old, healthcare-associated or nosocomial BSIs, infection foci other than urogenital, and CCI ⩾ 3. We used a generalised estimating equation for the logistic regression models to account for the lack of independence due to 59 patients having more than one E. coli BSI during the study and therefore, correctly estimated coefficients and standard errors. Managing the lack of independence is important, because E. coli BSIs from the same patient will be more similar to each other than those from other patients. Our findings are consistent with previous studies, including those that identified associations between increasing age and higher case fatality risks [Reference Laupland7, Reference Abernethy12, Reference Blandy15, Reference Melzer and Petersen28]. Also similar to our findings, studies established that nosocomial E. coli BSIs are associated with increased mortality [Reference Laupland7, Reference Williamson8, Reference Yoon29], and specifically, the odds of mortality are higher with nosocomial than healthcare-associated when compared to community-acquired [Reference Laupland7, Reference Williamson8]. Our study is also consistent with others in identifying that a BSI focus other than urogenital is associated with increased case fatality risk [Reference Laupland7, Reference Abernethy12, Reference Blandy15, Reference Melzer and Petersen28, Reference Kang30–Reference Ortega32]. Some previous studies found an association between higher CCI or Elixhauser comorbidity score and increased mortality [Reference Blandy15, Reference Ha31]. We used CCI in the multivariable logistic regression model and found that a higher CCI was associated with higher odds of 30-day mortality. We went further and evaluated common comorbid conditions with univariable logistic regression models to gain insight into comorbidities that might be contributing to the CCI mortality association. Escherichia coli BSI cases with diabetes mellitus, cancer, chronic pulmonary disease, congestive heart failure, myocardial infarction, dementia, mild liver disease, and cerebrovascular disease all had significantly higher odds of mortality, which is important from a clinical perspective. Other variables, including sex, ESBL, ciprofloxacin resistance, MDR, and season are not consistently significantly associated with mortality in the literature; some previous studies found that males, ESBL production, ciprofloxacin resistance, MDR, and winter were significantly associated with higher mortality, whereas others, including our study, did not [Reference Laupland7, Reference Sawatwong11, Reference Abernethy12, Reference Blandy15, Reference Melzer and Petersen28–Reference Ha31, Reference Denis33, Reference Camins34]. The inconsistent associations for sex, AMR, and season between studies are likely due to differences in healthcare practices, population, and geography. Studies also found significant associations between inappropriate initial antimicrobial therapy and higher mortality [Reference Ortega32, Reference Denis33] and this could be explored in future western interior studies.
Longer LOS is associated with increased hospital costs and therefore is an important measure of the burden of disease [Reference Naylor13]. When exploring burden of disease, post-infection LOS is the appropriate LOS measure since it only covers LOS during the BSI. Blandy et al. estimated factors associated with long post-infection LOS in E. coli BSIs, but used hospital-based data [Reference Blandy15]. They included E. coli BSIs from three teaching hospitals in West London, United Kingdom from January 2011 to June 2015 [Reference Blandy15]. In our study and theirs, having an ESBL-producing or third-generation cephalosporin-resistant E. coli BSI significantly increased the odds of having a long post-infection LOS [Reference Blandy15]. This is an important finding because third-generation cephalosporin resistance in E. coli has been increasing globally [35]. Third-generation cephalosporin-resistant E. coli is considered a critical priority pathogen for research and development of new antibiotics by the World Health Organization and a serious threat to human health by the Centers for Disease Control and Prevention [35, 36]. Blandy et al. found that patients with nosocomial E. coli BSIs had significantly higher odds of long post-infection LOS compared to those with community-onset [Reference Blandy15]. Similarly, we found that cases with nosocomial E. coli BSIs had significantly higher odds of long post-infection LOS compared to those with community-acquired and healthcare-associated. There were two significant associations identified by Blandy et al. that were not significant in our analysis, patients being 60-years-old or older, and ciprofloxacin resistance, which both had greater odds of having a long post-infection LOS [Reference Blandy15]. In both studies, sex and urinary focus of BSIs were not associated with long post-infection LOS [Reference Blandy15]. It is interesting that a urogenital focus of infection was protective for mortality but did not impact the odds of having a long post-infection LOS. We found that having a CCI ⩾ 3 increased the odds of having a long post-infection LOS; comparatively, Blandy et al. did not find a significant association between Elixhauser comorbidity score and long post-infection LOS [Reference Blandy15]. Differences in the comorbidity indices used by the two studies and the number of categories used for the categorical comorbidity index variable likely contributed to the contradictory results. By exploring individual comorbidities with univariable analyses, we found that E. coli BSI cases with diabetes mellitus, chronic pulmonary disease, congestive heart failure, renal disease, dementia, cerebrovascular disease, or diabetes mellitus with end organ damage had significantly higher odds of having a long post-infection LOS. In future studies, we could explore an alternative approach to modelling the post-infection LOS data using a Cox proportional hazard model with death as a competing risk for being discharged alive from hospital.
Our study had some limitations that should be noted. We did not have access to data regarding patient history (e.g. or hospital costs). The case data were collected retrospectively through medical record review and therefore, are limited to details contained within the medical records, however, it was performed by an experienced infectious disease specialist (KBL). We used the current CLSI clinical breakpoints at the time of AST to interpret the results as susceptible, intermediate, or resistant, and therefore it is possible that some of the breakpoints changed slightly during the study period. Although our study is population-based, it is only from one geographical area in a high-income country with publicly funded healthcare and this should be taken into consideration when attempting to generalise our results to a different population [14, 37].
In conclusion, our population-based study presented detailed analyses of E. coli BSIs in the western interior area of BC and explored several different perspectives of burden of disease in E. coli BSIs at the population-level. Our study demonstrated the substantial burden of E. coli BSIs with crude incidence and 30-day mortality rates of 59.1 E. coli BSIs and 6.8 deaths/100 000 person-years, respectively. We identified having a nosocomial E. coli BSIs and a CCI ⩾ 3 as risk factors for both 30-day mortality, and long post-infection LOS in patients that survived to discharge. Whereas, older cases, a non-urogenital BSI focus and healthcare-associated BSIs had significantly higher odds of 30-day mortality, and ESBL-producing E. coli BSIs were a risk factor for long post-infection LOS in patients that survived to discharge. Several individual comorbidities, including diabetes mellitus, chronic pulmonary disease, congestive heart failure, cerebrovascular disease, and dementia, were risk factors for both 30-day mortality, and long post-infection LOS in patients that survived to discharge. Future research should explore which of these risk factors could be effectively targeted with interventions to reduce mortality in E. coli BSIs, and post-infection LOS in patients that survived to discharge.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0950268821001874
Financial support
No external funding was received as direct support for this project. MCM's PhD programme was supported by a Banting and Charles Best Canadian Graduate Scholarship Doctoral Award from the Canadian Institutes of Health Research, a Brock Doctoral Scholarship from the University of Guelph, an OVC Scholarship from the Ontario Veterinary College at the University of Guelph and funding from the Genome Research Development Initiative through the Government of Canada. The funding organisations were not involved in any stage of the project.
Conflict of interest
MCM reports a scholarship from the Canadian Institutes of Health Research and funding for a Federal Student Work Experience Programme placement from the Government of Canada – Genome Research Development Initiative, both related to her PhD programme, but not directly related to the submitted work.
Ethical standards
The Interior Health and University of Guelph Research Ethics Boards approved this study and granted waivers of individual informed consent (2013-14-052-I and 2018-10-050, respectively). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Data availability statement
The dataset analysed during the current study is available from the corresponding author on reasonable request.
Open access
