Background
Sepsis is a leading cause of morbidity and mortality among hospitalized patients and results in a significant healthcare burden [
1]. In the United States, severe sepsis and septic shock remain the dominant cause of death in intensive care units (ICUs), with approximately 900,000–3,000,000 cases yearly [
2]. Economically, sepsis was ranked in the top four most expensive conditions, costing an aggregate of $20,298,000 yearly in US hospitals [
3,
4]. Mortality from severe sepsis and septic shock remains unacceptably high, ranging between 20 and 40% depending on the severity of illness [
5]. Blood culture remains an essential test in the care of sepsis patients [
6]. The results of blood culture can be used to guide the adjustment of empiric antibiotics regimen [
7]. Blood culture positivity has been suggested as a surrogate marker of bacterial load, as evidenced by the worse prognosis in sepsis patients with a shorter time from blood culture incubation to growth detection [
8]. However, isolation of specific microorganisms by blood culture remains challenging [
9]. In approximately one third to two thirds of sepsis patients, no specific organism can be identified by culture [
10].
Culture-positive patients were thought to have a higher burden of infection and therefore worse outcomes [
11,
12]. However, contrary to the general belief, recent studies have shown that culture-negative and culture-positive sepsis patients are generally comparable in resource utilization and outcomes [
13,
14]. To date, only a few studies have addressed the epidemiology and outcomes of culture-negative sepsis [
2,
10]. Several questions remain to be answered. First, previous studies typically included patients with severe sepsis or septic shock rather than the whole spectrum of sepsis [
2,
14]. Second, previous studies only focused on in-hospital or 30-day mortality.
We therefore extended the analysis to include the whole spectrum of Systemic inflammatory response syndrome (SIRS)-defined sepsis patients and followed the patients until 90 days to see if blood culture positivity may affect the mid-term outcome in sepsis patients.
Methods
Patient population
The investigation was a retrospective multicenter cohort study. During a 24-month period from 1 January 2015 through 31 December 2016, the charts of all patients who were admitted to the emergency department, hospital floor, or ICU with a presumed systemic infection and who received blood cultures were reviewed. We included patients who fulfilled the criteria of systemic inflammatory syndrome and excluded patients with missing data or do-not-resuscitate (DNR) orders. We also excluded patients transferred from other hospitals or patients lost to follow-up.
Collection of clinical variables
To verify the presence of infection as a cause for admission, participating investigators from the three locations independently examined all the collected medical records. The clinical and laboratory data were collected at the time of admission. The definition of the variable collected has been described in our previous work. Briefly, we defined SIRS as two of the following: temperature > 38 °C or < 36 °C, heart rate > 90 beats/min, respiratory rate > 20 breaths/min or PaCO
2 < 32 mmHg (4.3 kPa), white cell count > 12,000 cells/μl or < 4000 cells/μl, and 10% immature/band forms. We defined severe sepsis as sepsis associated with organ dysfunction and septic shock as having a systolic blood pressure < 90 mmHg requiring vasopressor therapy. Organ dysfunctions assessed at admission were based on sequential organ failure assessment (SOFA) score with modifications for use outside an ICU setting, which are defined as follows: 1) altered consciousness, a Glasgow Coma Scale score of < 14 or a decrease in the score of at least 2 if a primary central nervous system injury was present, 2) acute respiratory failure, a pulse oxygen saturation < 90% at admission or requirement of mechanical ventilation, 3) acute kidney injury, a serum creatinine level > 2.0 mg/dL or, in the case of preexisting renal dysfunction, a doubling of previous serum creatinine values, 4) acute hepatic dysfunction, an increase in bilirubin concentration > 2 mg/dL and the occurrence of coagulation disorders with INR > 1.5, 5) acute hematologic dysfunction, an decrease in platelet count < 100 × 10
3/μL, and 6) acute cardiovascular dysfunction, the presence of hypotension requiring vasopressor. We also applied quick sequential organ failure assessment (quick SOFA) for patient severity classification [
15]. The criteria for quick SOFA include altered level of consciousness (Glasgow coma score < 14), hypotension (blood pressure < 100 mmHg), or tachypnea (> 22 breaths per min). In patients with dementia or delirium, altered mental status was used instead of GCS.
The conditions of the patients were being recorded for at least 90 days, which includes the duration of hospital stay and clinic appointments after discharge. For all eligible patients, the following data were collected using a standard case report form: demographic features, the presenting co-morbid medical states, initial vital signs, requirements for assisted ventilation or hemodialysis, diagnosis during admission and discharge, and the identity of blood-isolated microorganisms. In the final discharge diagnosis, the occurrence and origins of a focal infection was categorized as lower respiratory tract infection, hepatobiliary tract infection, urinary tract infection, soft tissue and skin infection, intra-abdominal infection, and other infections. We reviewed past medical records to collect the comorbidity information. Diabetes mellitus was defined by discharge diagnosis with use of hypoglycemic agents or insulin. Chronic pulmonary disease include chronic obstructive pulmonary disease (COPD), asthma, bronchiectasis, and chronic bronchitis. Hemiplegic stroke was defined as stroke with hemiplegia or being wheelchair bound. Previous myocardial infarction and congestive heart failure was defined by discharge diagnosis and use of related medications. Chronic liver disease includes chronic viral hepatitis or liver cirrhosis. Chronic kidney disease includes stage four or stage five diseases. Malignancies include both solid and hematologic cancer.
Blood culture and determination of bloodstream infection
During the study period, anaerobic and aerobic blood cultures were routinely used for patients with presumed systemic infection. Blood cultures were performed using the BacT/ALERT FA and BacT/ALERT (bioMérieux, Marcy-l’Etoile, France) bottles. Each bottle was incubated into the BacT/ALERT 3D instrument until signs of positive culture or a maximum of 5 days have passed. All positive bottles (including both aerobic and anaerobic BCs) were systematically plated on blood agar plate (BAP)/eosin methylene blue (EMB) agar, chocolate and anaerobic blood agar and incubated under anaerobic (10%CO2 + 5%H2 + 85%N2; Concept 400, Ruskinn Technology, Guiseley, UK) and aerobic conditions. All isolated were identified from the colonies using a standard algorithm using growth on differential agar and biochemical tests, followed by definitive identification with Phoenix 100 (Becton Dickinson, Franklin Lakes, New Jersey, USA). Clinically significant bloodstream infection (BSI) was defined as at least two sets of positive blood cultures from separate sites, one set positive for a Gram-negative bacterial pathogen or one set positive for a Gram-positive pathogen in a patient with an intravascular device and clinical compatibility. It was deemed as a true infection if the microorganisms isolated from the blood cultures included Staphylococcus aureus, Streptococcus pyogenes, Streptococcus agalactiae, Streptococcus pneumoniae, Escherichia coli and other members of the family Enterobacteriaceae, Pseudomonas aeruginosa, the Bacteroides fragilis group, and Candida species. Contamination was presumed if blood cultures were positive for Propionibacterium acnes, Corynebacterium spp., or Bacillus spp. In addition, when Coagulase-negative staphylococcus was isolated from only one of at least two sets of blood cultures, it was also regarded as contamination. Primary BSI was defined as a laboratory-confirmed BSI that was not secondary to an infection at another body site, whereas secondary BSI was designated as BSI attributable to a primary site of infection. This study was approved by the Research Committees and Institutional Review Boards for all institutions, and it met the criteria for exemption from informed consent.
Statistical analysis
Absolute counts with relative frequencies were reported for categorical variables, and the median with 25th and 75th percentiles was reported for continuous variables. Univariate comparison was performed with Chi-square tests for categorical variables and Mann-Whitney U tests for continuous variables. We evaluated the impact of blood culture positivity on the early and late mortality of sepsis. Given known differences in patient characteristics between blood culture-positive and blood culture-negative patients, inverse probability of treatment weighting (IPTW) was employed to control confounding caused by baseline general health disparities in the two groups of patients [
16‐
18].
The IPTW approach weights individuals based on their propensity score, with those patients with a high likelihood of positive culture weighted lower; its goal is to create weighted individual samples where the measured baseline covariates match between the two groups [
19,
20]. In this study, the PS was defined as the conditional probability of culture-positive compared to culture-negative results for the index episode of sepsis. By creating a logistic regression model with 19 covariates representing patient demographics and features that are proven to be related to blood culture positivity and the results, we estimated a PS for blood culture-positive sepsis. For the model of PS component variables and their respective weights, see Supplementary
Table 2. To visualize the balance of covariates after PS matching, we plotted the standardized mean differences of all covariates (Supplementary
Figure 2). We used the IPTW to evaluate the logistic regression model in the IPTW weighting study and obtained an estimation that is independent of the possible confounding of all component variables in the PS. The IPTW-weighting study used robust sandwich variance estimators to establish a reliable confidence interval. All tests were 2- tailed, and
p values < 0.05 were considered statistically significant. The 90-day survival curve was plotted, and the cumulative survival between culture-positive and culture-negative sepsis patients was compared with the log rank test. All analyses were conducted using SAS Version 9.4 (SAS Inc., Cary, NC, USA). A 2-sided
p-value < 0.05 was viewed as significant.
Discussions
In this multicenter cohort, we found that culture-positive sepsis patients have a higher comorbidity burden and higher clinical severity than culture-negative sepsis patients, but the 30-day mortality was comparable between the two groups. The culture-positive patients, however, had worse outcomes than the culture-negative patients between 31 and 90 days. The results were similar in patients with SIRS-defined sepsis, severe sepsis, or septic shock. Persistent staphylococcal bacteremia appeared to be a contributing factor to the increased risk of late mortality among patients with culture-positive sepsis.
Several studies have evaluated the impact of culture positivity on the outcome of sepsis patients without consistent events [
2,
10,
11,
13]. In a prospective observational cohort study including 1001 medical ICU patients with severe sepsis, Phua J. et al. showed significant differences between culture-negative and culture-positive sepsis, including fewer comorbidities, milder severity of illness, shorter hospitalizations, and lower mortality for culture-negative patients [
11]. However, the culture results status was not independently associated with mortality in multivariable analysis. Later, in a trinational retrospective cohort study (
n = 8670), Kethireddy S. et al. showed that culture-negative septic shock patients experienced similar ICU survival (58.3% vs. 59.5%;
p = 0.276) and overall hospital survival (47.3% vs. 47.1%;
p = 0.976) to culture-positive septic shock patients [
2]. Additionally, Sigakis M.J.G. et al. utilized 8 years of patient records in a large academic medical center comprising 9288 culture-negative patients (89%) and 1105 culture-positive patients (11%) to show that a positive culture was not independently associated with mortality in sepsis defined by SIRS (OR: 1.01, 95% CI, 0.81–1.26), sequential organ failure assessment (SOFA) score (OR: 1.13, 95% CI, 0.86–1.43), and quick SOFA score (OR: 1.05, 95% CI, 0.83–1.33), respectively [
13]. While most of these studies found no difference in survival between culture-positive and culture-negative patients, Gupta, S. et al. conducted a large cohort study using the Nationwide Inpatient Sample (NIS) comprising 6,843,279 admissions of patients with severe sepsis and showed that 47.1% of patients had culture-negative results [
10]. Those with culture-negative severe sepsis had more comorbidities and acute organ dysfunction (respiratory, cardiac, hepatic, and renal dysfunction), and culture-negative severe sepsis was independently associated with in-hospital mortality (OR, 1.75; 95% CI, 1.72–1.77). Our study corroborates with Phua J. et al’s finding that patients with culture-positive sepsis have higher severity and mortality than those with culture-negative sepsis; however, there was no difference in 30-day mortality between the two groups of patients after IPTW weighting [
11]. Between 31 to 90 days, culture-positive sepsis has a significantly higher mortality than culture-negative sepsis.
None of the previous studies reported the survival curves between two groups of patients, so we could not compare our late mortality difference results with those of previous studies. We suspect that persistent
Staphylococcus aureus bacteremia (SAB) is an important contributing factor to the increased late mortality of culture-positive sepsis patients as compared with culture-negative sepsis patients. Excluding patients with documented SAB attenuated the late mortality risk of culture-positive patients and narrowed the survival difference between culture-positive and culture-negative patients to a statistically nonsignificant status. Persistent SAB, described as SAB that persists for 3–7 days, is a well-documented problem [
22,
23]. Persistent bacteremia causes high rates of morbidity and attributable mortality [
24]. Reported independent risk factors for persistent bacteremia included community-onset bacteremia, bone and joint infection, central venous catheter-related infection, metastatic infection, suppurative thrombophlebitis, infective endocarditis, and methicillin resistance [
2,
23]. It has been shown that bacteremia may persist for ≥7 days even after removal of CVC, so early source control and higher vancomycin trough levels should be implemented [
25]. Duration of antibiotic treatment has been shown to be a key factor to the survival of patients with persistent SAB. Abbas M. et al. evaluated the effect of duration of therapy (DOT) on mortality and relapse in patients with SAB. In a retrospective, single-center cohort study including 225 patients with uncomplicated SAB and 305 patients with complicated SAB. After carefully accounting for confounding and immortal time bias, they concluded that in patients with complicated SAB, DOT > 14 days was associated with a higher survival rate than DOT < 14 days. Other causes can explain the comparable mortality of culture-negative sepsis patients with culture-positive sepsis patients despite the more severe clinical presentations.
In total, 18 patients were excluded from final analysis, of which 7 were excluded due to transferring to other hospitals during the follow-up period, 6 lost follow-up, and the remaining was excluded due to missing data. The outcome information of patients transferred to other hospitals or lost to follow-up was not available. However, we recorded the reasons for transferal. As the three participating hospitals are the largest medical centers locally, the reasons for transferring to other hospitals were the geographic proximity of caregivers rather than the deterioration of clinical conditions. Therefore, these patients lost to follow-up are highly unlikely to have the worst outcome to bias our analysis. On the contrary, including these patients lost to follow-up may even improve the overall outcome.
Strengths and limitations
The results of this study should be interpreted in light of its strengths and weaknesses. The multi-centered Asian patient cohort that included patients from intensive care units, emergency departments, and hospital wards increases the generalizability and clinical applicability of the study in Asian populations. The use of an IPTW approach instead of PS matching or traditional regression increases the statistical efficiency and allows the estimation of population average effect. The IPTW-weighted survival curve enables the survival differences of the two comparison groups to be visualized over time, allowing the time-dependent impact of culture positivity on different stages of sepsis to be identified. There are, of course, limitations in our study. First, because of the observational character of the study, there is always a possibility for unmeasured confounding factors. Second, information on prior exposure to antibiotics before hospital admission, an important determinant of culture positivity, was lacking. The outcome impact of culture positivity may also involve the possible increased use of prehospital antibiotics in the culture-negative patients. Third, unlike previous studies that included all types of microbiological culture results, we only focused on the initial blood culture results. Gupta, S. et al.’s study, which utilized an administrative database, was the only other study that showed increased mortality associated with culture-negative sepsis [
10]. In that study, culture positivity was defined as the presence of any ICD-9 code for all kinds of microbiology diagnoses. It is likely that the immortal time bias played a role in their exceptional findings, as patients with longer length of hospital stay tend to receive more microbiological tests and hence more microbiological diagnoses. The immortal time bias was not amenable to regression analysis. Most of the positive blood culture results (97.7%) of our study came from the initial blood cultures. Therefore, the risk of immortal time bias in this cohort is minimal.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.