Early deaths associated with community-acquired and healthcare-associated bloodstream infections: a population-based study, Finland, 2004 to 2018

Background Bloodstream infections (BSI) cause substantial morbidity and mortality. Aim We explored the role of causative pathogens and patient characteristics on the outcome of community-acquired (CA) and healthcare-associated (HA) BSI, with particular interest in early death. Methods We used national register data to identify all BSI in Finland during 2004–18. We determined the origin of BSI, patients´ underlying comorbidities and deaths within 2 or 30 days from specimen collection. A time-dependent Cox model was applied to evaluate the impact of patient characteristics and causative pathogens on the hazard for death at different time points. Results A total of 173,715 BSI were identified; 22,474 (12.9%) were fatal within 30 days and, of these, 6,392 (28.4%) occurred within 2 days (7.9 deaths/100,000 population). The 2-day case fatality rate of HA-BSI was higher than that of CA-BSI (5.4% vs 3.0%). Patients who died within 2 days were older than those alive on day 3 (76 vs 70 years) and had more severe comorbidities. Compared with other BSI, infections leading to death within 2 days were more often polymicrobial (11.8% vs 6.3%) and caused by Pseudomonas aeruginosa (6.2% vs 2.0%), fungi (2.9% vs 1.4%) and multidrug-resistant (MDR) pathogens (2.2% vs 1.8%), which were also predictors of death within 2 days in the model. Conclusions Overrepresentation of polymicrobial, fungal, P. aeruginosa and MDR aetiology among BSI leading to early death is challenging concerning the initial antimicrobial treatment. Our findings highlight the need for active prevention and prompt recognition of BSI and appropriate antimicrobial treatment.


Introduction
Bloodstream infections (BSI) are a major health concern. According to a systematic review published in 2013, nearly 2 million BSI episodes and 250,000 deaths from BSI were estimated to occur annually in North America and Europe combined [1]. Despite advances in antimicrobial therapy and intensive care, mortality of BSI remains high, with 1-month case fatality rates of ca 10-19% for community-acquired BSI (CA-BSI) and even higher rates, 17-28%, for healthcare-associated BSI (HA-BSI) [2][3][4][5][6].
Previous studies have shown that a marked proportion, approximately one third of BSI-associated deaths occur early, within 2 days after the positive blood culture specimen [7][8][9]. To our knowledge, only one previous population-based study has reported 2-day case fatality rates of BSI, yet this study presents data from 1992-97 and consists of solely CA-BSI [7]. Possible predisposing factors for early mortality have not been studied; however, delayed diagnosis, inappropriate antimicrobial treatment, and the patient's overall condition may contribute to an early fatal outcome of BSI. In general, the initial antimicrobial treatment of BSI is challenging since the confirmation of the causative pathogen and its susceptibility are typically not known until the second day after collecting blood cultures. Thus, some BSI-associated deaths occur before the definitive results from the blood cultures are known.
Our recent population-based study from Finland showed a twofold increase in the incidence and mortality of BSI during 2004-18, which is presumably related to ageing of the population and the rising burden of underlying medical conditions [10]. However, the 30-day case fatality rate remained steady over time (12.6-13.0%). In the present study, we used the same national laboratory-based surveillance data linked to other national registers to analyse all BSI in Finland during 2004-18 with the aim of evaluating the role of different factors, such as comorbidity and causative pathogens, on the outcome of BSI of both community and healthcare origin, with focus on early death (≤ 2 days).

Study setting and population
The healthcare system in Finland (population: 5.2 million in 2004 and 5.5 million in 2018 [11]) is organised into 20 healthcare districts, which include a total of five tertiary care hospitals, 15 secondary care hospitals and numerous primary care hospitals. All clinical microbiology laboratories notify bacterial and fungal isolates from blood samples, i.e. BSI, to the National Infectious Disease Register (NIDR). The notifications are reported electronically and include the following information: specimen date, pathogen, patient date of birth, sex, place of residence, and national identity code. Multiple notifications of the same pathogen containing the same identity code, i.e. referring to the same person, are merged into a case if they occur within 3 months of each other.
In this retrospective cohort study, we used NIDR data to identify all BSI in Finland during 2004-18. Only BSI with valid identity codes were included in the study (n = 587 excluded) (flow chart of the data in [10], Figure 1). Duplicate notifications (same specimen date, pathogen, and identity code) were excluded (n = 155). Information on the hospitalisation of the patient, including origin of the infection, i.e. CA-BSI vs HA-BSI, and current and prior (within 1 year) diagnosis codes were obtained by linkage to the National Hospital Discharge Register (HILMO).

Definitions
BSI was defined as the occurrence of viable bacteria or fungi in the blood evidenced by positive blood cultures. Polymicrobial BSI was defined as isolation of more than one bacterial or fungal species in blood cultures within 2 days.
BSI was classified as healthcare-associated (HA) if the first blood culture was obtained more than 2 days after admission to the hospital or within 2 days of discharge [12]. Patients with a BSI who were transferred from other healthcare facilities, including nursing homes, were also classified as having a HA-BSI. Patients with CA-BSI had not previously been hospitalised and their blood culture specimen was obtained within 2 days of hospital admission.
Comorbid illness was defined by using a validated algorithm for the Charlson comorbidity index (CCI) based on the International Classification of Diseases, 10th revision [13,14]. Three levels of comorbidity were defined by the CCI scores: low (score 0) for patients with no reported underlying diseases included in the CCI, medium (score 1-2) and high (score > 2) [15,16].

Multidrug-resistant pathogens
The interpretation of susceptibility data of causative pathogens was performed in the clinical microbiology laboratories by using the CLSI standard for samples collected before year 2011 and, afterwards, according to the EUCAST clinical breakpoints [17].

Outcome
The case fatality rate at 2 or 30 days (1 month) after withdrawal of the first blood specimen with a positive culture of a particular patient was determined from the Population Information System by linkage with the identity code. All deaths occurring within 2 days after collection of the specimen yielding the first positive blood culture (day 0) were referred to as early deaths.

Analyses and statistics
The average annual 2-day and 30-day mortality rates (early and overall deaths/100,000 population, respectively) were calculated according to the total number of deaths and population during 2004-18; population data was retrieved from Statistics Finland [11]. Univariate analysis of categorical variables was done with the chi-squared test, using Yates's correction or Fisher's exact test, as appropriate. The differences in distributions between continuous variables were tested by the Kruskal-Wallis test. A time-dependent Cox model was applied to evaluate the effects of patient characteristics and causative pathogens on the hazard for death within different time points (within 2 or 3-30 days after the positive blood culture). Data was analysed using SPSS Statistics version 25 (IBM) and Stata 16 (StataCorp).

Predictors of early death
Patients who died early, within 2 days, were older than those who were alive at day 3 (median age: 76 vs 70 years) (

Risk factors for death of community-acquired and healthcare-associated bloodstream infections within 2 or 30 days
To evaluate possible risk factors for BSI fatality, we conducted a time-dependent Cox model demonstrating the effect of patient characteristics and causative pathogens on the hazard for death in two outcome groups, death within 2 or 3-30 days; CA-BSI and HA-BSI were analysed separately. Sex and CCI, together with the most common causative pathogens of all BSI (E. coli and S. aureus) and those associated with the highest 2-day case fatalities (P. aeruginosa, fungi and polymicrobial BSI), were included in the model. A clustering effect of healthcare districts was taken into account by robust standard errors, and calendar year was included as a continuous variable. The effect of age on risk of death was presented by a 10-year increase in age. Increasing age and CCI were risk factors for death in both outcome groups and among both CA-BSI and HA-BSI ( Figure); the highest hazard ratio (HR: 3.19) was noted among patients with CA-BSI leading to death within 3-30 days with high CCI (Figure,

Discussion
Our population-based study of over 170,000 BSI in Finland during 2004-18 offers a comprehensive overview on the outcome, particularly early death, of both CA-BSI and HA-BSI. The 30-day case-fatality was 12.9%, and nearly one third of these deaths occurred early, within 2 days. We noted higher 2-day and 30-day case fatalities for HA-BSI compared with CA-BSI. Older age and a greater burden of comorbidities were associated with early BSI mortality. BSIs with early fatal outcome were more often polymicrobial and caused by P.

Figure
Hazard ratios for death of healthcare-associated bloodstream infections (n = 50,483) and community-acquired bloodstream infections (n = 123,232) according to patient characteristics and causative pathogens in different outcome groups, Finland, aeruginosa, fungi and MDR pathogen compared with other BSI.
We conducted the time-dependent Cox model separately for HA-BSI and CA-BSI based on the knowledge that hospitalised patients are older and have more severe underlying conditions than patients who have acquired a BSI in the community. Also, the treatment guidelines for empiric antimicrobial therapy differ between CA-BSI and HA-BSI based on the differences in causative pathogens. Our model allowed direct comparison of hazards for death concerning causative pathogens of BSI in two outcome groups, as the effect of increasing age and burden of comorbidity was similar between the groups. Although E. coli and S. aureus were the most common causative pathogens of all BSI and of those leading to death within 2 days, they were not predictors of early death. However, P. aeruginosa aetiology was distinctly associated with increased risk for early death, whereas polymicrobial and fungal BSI were associated with a fatal outcome both within 2 days and 3-30 days. Polymicrobial BSI accounted for over 10% of all BSI leading to early death, and P. aeruginosa and fungal BSI together for over 15% of HA-BSI leading to early death. The empiric antimicrobial therapy of these BSI is a challenge for clinicians, as the lack of antifungal or broad-spectrum coverage may contribute to early fatality. In fact, a previous study showed that inappropriate initial treatment of P. aeruginosa BSI was associated with increased hospital mortality [18]. In our former population-based case series of BSI leading to early death in Southern Finland in 2007, empiric antimicrobial therapy was inappropriate in nearly 30% of the BSI; in 12% of CA-BSI and in 50% of HA-BSI [19]. These inappropriately treated BSI leading to early death were mainly caused by intrinsically resistant Gram-negative bacteria (most commonly P. aeruginosa). Consistently, previous reports demonstrate that healthcare-associated status in general is a predictor of ineffective empiric antimicrobial treatment of BSI [20] and of 30-day BSI mortality [5,21].
In our study, the 2-day case fatalities for CA-BSI and HA-BSI were 3.0% and 5.4%, respectively; 30.4% of the deaths among CA-BSI and 26.2% among HA-BSI occurred early. In a survey of community-onset BSI from Calgary, Canada during 2003-07, 38% of the deaths occurred by day 2, equalling a 2-day case fatality rate of 4.7% [8], whereas a slightly higher 2-day case fatality rate (7.2%) was presented in an older population-based cohort study of CA-BSI from North Jutland, Denmark, during 1992-97 [7]. The 30-day case fatality rate observed in the present study (12.9%) is comparable to rates in recent population-based reports (12.8-16.9%) [5,22,23]. Our 2-day and 30-day case fatalities were higher for HA-BSI than for CA-BSI. This is in line with findings from a Swedish study spanning from 2000-13 that demonstrated a 30-day case fatality rate of 17.2% for hospital-acquired BSI and 10.6% for community-onset BSI [5]. In our study, the patients who died early had more underlying diseases and were older compared with other BSI patients, which was also noted in our previous case series [19]. In a Danish survey, only 2% of the BSI patients who died early had no predisposing underlying condition, and these patients were older than survivors [24]. Similarly, former studies have shown that rising age and comorbidity are associated with 30-day BSI mortality [5,7,21,[25][26][27].
Overall, the 2-day case fatality rate of BSI in our study decreased slightly during 2004-18, from 4.1% to 3.3%, possibly indicating advancements in recognition and accuracy of empiric antimicrobial treatment of BSI. In a Danish survey of CA-BSI during 1992-97, the 2-day case fatalities for given pathogens were considerably higher compared with those of CA-BSI in our study; for S. pneumoniae 9.3% vs 3.6%, respectively, for S. aureus 9.0% vs 3.1%, for E. coli 6.3% vs 2.0%, and for polymicrobial BSI 10.7% vs 5.9% [7]. The descending overall 2-day case fatality rate noted in the present study might reflect changes in causative pathogens of BSI leading to early death over time, such as the increase in the proportion of E. coli BSI but also the decline in S. pneumoniae BSI. The observed reduction in S. pneumoniae BSI is probably associated with the introduction of the pneumococcal vaccine to the childhood immunisation schedule [28,29]. We observed a distinct ascending trend in the percentage of ESBL-E. There are certain limitations in our study. Firstly, we did not have information on possible delays in recognition of the infection and commencement of the treatment, nor data on whether the antimicrobial therapy was appropriate. Delayed and ineffective initial treatment are associated with increased BSI mortality, as demonstrated in previous reports [26,[32][33][34]. In our previous study of BSI leading to early death, the time from symptom onset to administration of antimicrobial therapy was longer in CA-BSI compared with HA-BSI referring to probable delays in seeking medical care [19]. Secondly, we lacked data on detailed clinical features, such as severity of infection (e.g. the respiratory tract as a focus of infection), and information on the role of BSI in the chain of morbid events and on the main cause of death. However, among patients who died early, BSI may have been at least a contributing factor. We did not have data on patients' underlying medical conditions other than the ones included in the CCI, nor information on a possible do-not-resuscitate (DNR) order, which likely have influenced the outcome. In fact, nearly one third of the BSI patients who died early in our previous case series had either a rapidly fatal underlying condition or a prior DNR order indicating a poor overall condition, and it is probable that most of these deaths were inevitable [19]. Furthermore, we lacked information on behavioural predisposing factors, such as overweight and smoking behaviour, which may affect the outcome [35]. Thirdly, we had limited information on antimicrobial resistance of the causative pathogens, for example, no susceptibility data was available concerning P. aeruginosa and Acinetobacter sp. Lastly, it is possible that some HA-BSI were inaccurately interpreted as CA-BSI since the hospital discharge register contains data on day surgery only (not all outpatient invasive procedures) and on direct transfers between healthcare facilities. Moreover, the timeframe, blood cultures obtained within 2 days of hospital discharge, for the definition of HA-BSI in the study was quite strict possibly leading to underestimation of HA-BSI.

Conclusion
In view of ageing of population in Finland, as in most other industrialised countries, BSI will constitute a major health burden in the future with a risk of fatal outcome, especially among vulnerable patients, elderly people and those with severe comorbidity. In our study, a notable proportion of BSI patients died early, and probably at least some of these deaths were inevitable. However, one fifth of those who died early had no recorded underlying medical condition, which emphasises the importance of rapid recognition of BSI and prompt initiation of adequate antimicrobial treatment according to the origin of the infection. The 2-day case fatality rate of BSI might potentially be used as an indicator of effectiveness of the healthcare system and the treatment chain. The 2-day and 30-day case fatalities were higher for HA-BSI in the present study underlining the need for considerable efforts in prevention of BSI in healthcare facilities. Active surveillance of causative pathogens and their resistance trends is beneficial when composing local guidelines for empiric antimicrobial therapy of BSI. Although the proportion of BSI caused by MDR pathogens was low in our study, the growing problem of antimicrobial resistance causes concern worldwide. Further studies are needed to evaluate the impact of the COVID-19 pandemic on the incidence, outcome, and causative pathogens of BSI, particularly on the occurrence of resistant pathogens.