Background
Within the last decade, the concept of healthcare-associated (HCA) infection has been introduced, and HCA infection has been described as an epidemiological category different from both community-acquired (CA) and nosocomial infection [
1,
2]. Most importantly, mortality in HCA infection seems to be generally higher than that in CA infection, and similar to that in nosocomial infection [
2‐
4]. However, there are conflicting results regarding whether HCA infection is an independent risk factor for mortality in bloodstream infection [
1,
5]. A few pathogens have been studied in terms of HCA infection with
S. aureus dominating the research, and these studies reported inconsistent data concerning the impact of HCA infection on mortality [
6‐
13]. For gram-negative bacteria, the data on the impact of HCA infection on mortality were conflicting, as well [
9‐
12].
Klebsiella pneumoniae is one of the most important gram-negative bacteria clinically, and
K. pneumoniae bloodstream infection (
KpBSI) has a mortality rate of about 20% [
11,
14‐
16]. Classically,
KpBSI was simply classified into CA and nosocomial infections depending on bacteremia onset time: within 48 hours and after 48 hours of admission, respectively, and the different characteristics of CA-
KpBSI versus nosocomial
KpBSI have been well evaluated [
11,
14,
15,
17,
18]. CA-
KpBSI is usually associated with liver abscesses in patients with diabetes in East Asian countries, such as Korea and Taiwan [
19‐
22]. On the other hand, nosocomial
KpBSI presents as primary bacteremia and/or pneumonia in patients with severe underlying diseases like malignancies. Thus, nosocomial infection has a higher mortality than CA infection [
11,
14,
15,
17,
18]. However, there have been few studies of HCA-
KpBSI [
11‐
13]. Therefore, we aimed to evaluate the impact of HCA infection on mortality and to compare the clinical characteristics of HCA and CA infection in patients with community-onset
KpBSI.
Discussion
In a previous study, we demonstrated that nosocomial
KpBSI was different from CA-
KpBSI [
18]. However, the healthcare system has changed dramatically and this simple dichotomy is no longer appropriate for
KpBSI in the current clinical setting. In this study we showed that HCA-
KpBSI accounted for over 50% of community-onset
KpBSI and HCA-
KpBSI had different clinical characteristics from CA-
KpBSI in terms of underlying disease, infection focus, antimicrobial susceptibility and treatment outcome. To our knowledge, this is the largest multicenter study comparing the clinical characteristics of HCA-
KpBSI and CA-
KpBSI [
11,
12].
Cancer was the most common associated condition in HCA-
KpBSI. In contrast, diabetes mellitus was the most common associated condition in CA-
KpBSI. The distribution of underlying disease in HCA-
KpBSI was similar to that in nosocomial
KpBSI, except for the frequency of chronic liver disease. While we found previously that the frequency of this disease did not differ between CA-
KpBSI and nosocomial
KpBSI [
18], it was more common in HCA-
KpBSI than in CA-
KpBSI (27% vs
. 16%;
p = 0.003) in the present study. The latter finding is similar to that of a study performed in Taiwan, although in the Taiwanese study the difference was not statistically significant (liver cirrhosis in HCA-
KpBSI [14.0%] vs. CA-
KpBSI [10.6%];
p = 0.339) [
12].
The primary site of infection was identified in 87% of community-onset
KpBSI. The most common source was the pancreatobiliary tract (33%), followed by liver abscess (20%). Liver abscess was more frequent in CA-
KpBSI than in HCA-
KpBSI. Compared to nosocomial
KpBSI, in which liver abscess was very rare (0% to 2%) [
11,
14,
18], HCA-
KpBSI was quite frequently associated with liver abscess (13%). Peritonitis was fairly frequent (>10%), more so in HCA-
KpBSI than in CA-
KpBSI in our analysis; in contrast Wu
et al. found only a few (<5%) of intra-abdominal infection and no difference in frequency between HCA-
KpBSI and CA-
KpBSI [
12]. Our higher frequency of peritonitis may be due to the prevalence of chronic liver disease caused by hepatitis B or C virus in Korea, which increases the occurrence of spontaneous bacterial peritonitis [
27,
28].
More of the HCA-
KpBSI isolates than of the CA-
KpBSI isolates were resistant to antimicrobial agents. Over 10% of the former were not susceptible to ciprofloxacin and 9% were not susceptible to one of the extended-spectrum cephalosporin. Although frequent antimicrobial resistance could affect the inadequacy of the initial choice of antimicrobial agent, there was no difference in rate of inappropriate empirical antimicrobial therapy between the HCA-
KpBSI and the CA-
KpBSI (6% vs. 8%;
p = 0.266). This result could have arisen because in cases of healthcare-associated infection clinicians may have taken into account frequencies of antimicrobial resistance when selecting the initial antibiotic. Actually, fewer HCA-
KpBSI than CA-
KpBSI (6% vs. 10%;
p = 0.088) were started on quinolones while more were started on piperacillin-tazobactam (Table
1).
Regarding empirical treatment, the proportion of patients treated inappropriately (7.2% of total patients) was much lower than was observed in other studies, which showed that over 20% of patients were treated inappropriately [
9,
29,
30]. This discrepancy might have been the result of differences in the definition of ‘appropriate empirical treatment’, because the definition we used was less strict than those in other studies [
31‐
33]. In addition, broad-spectrum antimicrobial agents, such as 3
rd generation cephalosporins or carbapenems, were frequently used empirically in our study (84.6% in CA infection, 74.7% in HCA infection). Considering that only 3.3% of organisms in CA infection and 9.3% of organisms in HCA infection were non-susceptible to extended-spectrum cephalosporins, the use of broad-spectrum antimicrobial agents also might have influenced the lower proportion of patients with treated inappropriately. However, other East Asian studies of
K. pneumoniae bacteremia also demonstrated a similar proportion of patients treated with inappropriate empirical therapy [
12,
14,
18].
There was a significant difference of 30-day mortality rate between HCA-
KpBSI and CA-
KpBSI in this study (22% vs. 11%;
p = 0.001). High Charlson’s weighted index of co-morbidity (≥3), high Pitt bacteremia score (≥4), neutropenia, polymicrobial infection and inappropriate empirical antimicrobial therapy were found to be independent risk factors for mortality. However, HCA infection itself was not a significant risk factor for 30-day mortality in multivariate analysis. This finding is consistent with the recent report from Taiwan and a bloodstream infection study dealing with gram-negative bacteria [
9,
12]. There are several explanations for this result. First, in our study, underlying disease and acute illness, which are classical risk factors for outcome of infectious disease, may have been so severe as to have attenuated the effect of HCA infection on mortality [
23,
24,
34]. Second, whether the infection focus was removable, and was or was not removed, may have affected mortality more than whether the infection was HCA or not [
35]. Liver abscess and pancreatobiliary infection can be classified as infections with removable foci, as opposed to pneumonia or primary bacteremia. In our study, percutaneous or internal drainage was performed in cases of liver abscess and obstructive pancreatobiliary infection, and these kinds of infection were found to be independent protective factors for mortality, as in the previous studies [
11,
14,
18]. Third, as indicated by Friedman
et al., the definition of HCA infection which we used in this study may have been excessively broad since the definition was based on the U.S. medical system [
1]. Unlike the U.S., South Korea has started a national health insurance system in 1977 and extended it nationwide in 1982. Consequently, there is a tendency for more people to access the medical system and be classified as HCA infection in Korea. Such national differences in healthcare systems could complicate the unambiguous identification of patients with HCA infections so as to be able to evaluate the actual effect of HCA infection on mortality. Therefore we need further studies using a more accurate and consistent definition of HCA infection that accords better with variations in clinical practice.
Our study had several limitations. First, there was a potential bias because it was performed retrospectively. Second, it was conducted in tertiary-care and university-affiliated hospitals and there was a large proportion of cancer patients in both the CA-
KpBSI and HCA-
KpBSI groups. Hence, we cannot extrapolate our result to community-based institutions. Third, we did not evaluate the ‘attributable’ mortality due to
KpBSI, so that some of the deaths in our study may not have been related to
KpBSI. However, efforts to designate outcomes as ‘attributable’ to infection are often subjective and inconsistent. We therefore employed an unambiguous definition, namely 30-day mortality rate, for evaluating treatment outcomes. Fourth, because we did not review the patients’ previous exposure to antimicrobial agents, we could not determine the influence of that factor on the acquiring resistant organisms and treatment outcome. Additionally, we did not collect data on the variation of antimicrobial therapy, such as duration or dosage; therefore, we could not take into account these issues, which could affect the analysis of risk factors for mortality. However, upon examining the mortality rate in other studies, our data are comparable; therefore, the regimen and duration of therapy we used were also likely to be similar to others [
12,
14].
What clinicians actually want to know is that HCA infection needs a specific work-up process or treatment. Accordingly, our study can provide useful information. First, we could see the difference of infection focus according to the epidemiological category more clearly by separating HCA infection and CA infection in comparison to our previous report [
18]. Peritonitis is more commonly associated with HCA infection than with CA infection in the present study, while the frequency of peritonitis in CA infection was relatively high (20.4%) and was not different from that observed in nosocomial infection in the previous study [
18]. However, when we classified more precisely we could see that peritonitis occurred in much lower frequency in true CA infection. In addition, we could identify the infection focus of all but 8.8% of the patients with true CA infection (in previously defined CA infection, 25.7% patients were unidentified for infection focus) [
18]. Second, comparing the resistance rate to antimicrobial agents in HCA infection with CA infection can help clinicians choose an initial antimicrobial agent in treating patients of each subset, which means not only should we consider broad-spectrum antimicrobial agents in HCA infection but also that we may not need to start broad-spectrum antimicrobial agents, such as extended-spectrum cephalosporins or carbapenems, in CA infection. Based on our data, quinolone can be a drug of choice in treating true CA-
KpBSI. Finally, even though we showed many differences between HCA infection and CA infection, we did not find HCA infection to be an independent risk factor for mortality in
KpBSI, which confirmed that the already known risk factors for mortality (severity of underlying disease, inadequate empirical therapy and severity of acute illness) are more important predictors of mortality in
KpBSI [
12,
14].
Competing interests
There are no potential conflicts of interest for any authors.
Authors’ contribution
All authors conceived of the study. YHJ, MJL, HYS collected the data. KHS and YHJ carried out data analysis and interpretation. NHK, JHH, JYP, PGC, WBP, ESK, SWP, KUP, HBK, NJK, ECK and MDO carried out data interpretation. YHJ and KHS drafted the manuscript. All authors have read and approved the final manuscript.