Clinical characteristics, organ failure, inflammatory markers and prediction of mortality in patients with community acquired bloodstream infection

This was a retrospective, cohort study of patients hospitalised at the National Hospital for Tropical Diseases (NHTD) (a tertiary referral infectious disease hospital) in northern Vietnam between January 2011 and December 2013. As a referral centre, this hospital often receives patients with specific infections (eg. central nervous system infections), complicated infections and those with severe infections who have failed on treatment elsewhere. Additionally, at the time of the study, the hospital had not establish a separate emergency department and intensive care unit (ICU), therefore we refer to the critical care unit (CCU) for the unit with both ventilated and unventilated beds, available haemodynamic support and renal replacement therapy. A convenience sampling method was used to select medical notes from the list of all hospitalised patients with positive bacterial blood cultures during the study period. The inclusion criteria were having a blood culture, taken within 48 h of hospitalisation (to any institution) for the current admission, positive for a recognised pathogen according to the US CDC’s National Healthcare Safety Network (NHSN) list [11]. Patients with infection with more than one bacterium were excluded, as were cases considered to be pseudobacteraemia [12].

Data was extracted from patients’ medical notes using a case-report form that captured patient demographics, reported history of prior medical illness, clinical manifestations, laboratory results, inflammatory markers within the first 24 h of admission to the study hospital and outcome at hospital discharge.

BSI with concurrent meningitis was defined in bacteremic patients who had cerebrospinal fluid examination within 24 h of blood drawn for microorganism isolation showing at least one of the following criteria: (1) turbid appearance; (2) leukocytosis (> 100 cells/mm3) or (3) leukocytosis from 10 to 100 cells/ mm3) and either an elevated protein (> 100 mg/dl) or decreased glucose (< 40 mg/dl) [13]. BSI with concurrent pneumonia was confirmed by radiology within 24 h of blood drawn for microorganism isolation. Gastrointestinal tract infection and urine tract infection were defined by the CDC/NHSN Surveillance Definitions for Specific Types of Infections [14]. Sequential [Sepsis-related] Organ Failure Assessment (SOFA) score, and quick SOFA (qSOFA) score were calculated using the worst parameters recorded within the first 24 h of admission to the study hospital and missing values were considered to be normal [6]. Organ dysfunction was defined by organ-specific SOFA scores ≥1. Failure of kidney function was further evaluated using the RIFLE criteria with RIFLE-F (Failure) defined as patients with a serum creatinine greater than three times the age adjusted upper limit of serum creatinine [15]. Neutropenia was defined as an absolute neutrophil count < 1500 cells/mm³, severe anemia as hemoglobin concentration was < 80 g/L and thrombocytopenia as a platelet count below 100 × 10³ cells/mm³.

The outcome at hospital discharge was defined as death for those who died in hospital or were palliatively discharged (discharged home for palliative care with the expectation of an early death, as per common practice in Vietnam) and ‘survived’ in all other cases.

Data was analysed using IBM SPSS Statistics for Windows (IBM Corp., Armonk, NY). Depending on the distribution, continuous data were presented as mean (95% confidence interval) or median (interquartile range) and categorical data as number (percentage). To evaluate the predictive value of SOFA, qSOFA score, white blood cell counts, C-Reactive Protein (CRP) and procalcitonin levels, a receiver operating characteristic (ROC) curve and the area under the curve (AUC) were calculated along with the sensitivity, specificity, positive and negative predictive values, positive and negative likelihood ratios associated with the cut-off value that gave the highest difference between sensitivity and (1-specificity) (Youden index). Since procalcitonin level was obtained by a semi-quantitative test that was only quantitatively measured for levels under 100 ng/mL, the result of “above 100 ng/mL” was considered as 100 ng/mL. The Mann Whitney U test and Kruskal Wallis test were used to analyze continuous variables and the Chi-square and Fisher’s exact were used for bivariate analyses as appropriated. Logistic regression models were used to calculate unadjusted odds ratios (ORs) and 95% confidence intervals (CIs) for associations between clinical, laboratory characteristics and case fatality rate. Comparison of AUC between different ROC curves was performed using a nonparametric approach [16]. Cox proportional hazards regression was used to identify variables that predicted clinical outcomes. Variables for inclusion were selected by review of the literature (age, HIV infection status, neutropenia, SOFA score and aetiology of CABSI). All tests were two-tailed and differences were considered statistically significant at p values ≤0.05.

Concurrent meningitis was confirmed in 18.3% (72/393) and pneumonia in 24.9% (98/393) of patients, with both conditions occurring in 2.8% (11/393) of patients with CABSI. Gram-positive organisms were isolated from 72.2% (52/72) of those with meningitis, Enterobacteriaceae in 18.1% (13/72) and non-Enterobacteriaceae Gram-negative bacteria in 9.7% (7/72), see Table 1 for details. Concurrent meningitis was found in 6.5% (13/200) patients with Enterobacteriaceae BSI, 9.0% (7/71) patients with non-Enterobacteriaceae Gram-negative bacteria BSI and 45.2% (52/115) patients with gram positive BSI. In the 98 cases of pneumonia the causative pathogens isolated from blood were Enterobacteriaceae (46.9%, 46/98), non-Enterobacteriaceae Gram-negative bacteria (25.5%, 25/98) and Gram-positive bacteria (27.6%, 27/98) (see Additional file 1: Table S1).

A further 8.4% (33/393) of patients presented with or developed at least one abscess, of which 2 patients (6.1%) had 2 abscess foci and 1 patient (3.0%) had 3 abscess foci in different locations. The locations were 16/37 (43.2%) liver, 8 (21%) skin, 4 (10.8%) muscle, 3 (8.1%) brain, 3 (8.1%) spleen, 2 (5.4%) lung, 1 (2.7%) eyelid. Endocarditis was confirmed by echocardiography in 4.1% (16/393) patients. These were due to Staphylococcus aureus (7/16, 43.8%), viridans group streptococci (4/16, 28.6%), Enterococcus species (3/16, 18.8%), Klebsiella pneumoniae (1/16, 6.25%) and Pseudomonas aeruginosa (1/16, 6.25%). The distribution of aetiology by the foci of infection is presented in Additional file 1: Table S1.

On the day of admission to the study hospital, BSI patients had a median SOFA score of 3 (IQR 1–7) and 78.1% (307/393) of patients had dysfunction of at least one organ. The SOFA score differed significantly between patients admitted direct to CCU (median of 7, IQR 4–12) and medical wards (median of 2, IQR 0–4) (p < 0.001). Enterobacteriaceae BSIs accounted for most cases with SOFA score above 12 (74% or 28/38, vs. 48% (172/355) in patients with SOFA score ≤ 12, p = 0.003) or more than 3 organ dysfunctions (60% or 47/78, vs. 49% or 153/315 in patients with ≤3 organ dysfunctions, p = 0.065). qSOFA score was ≥2 in 28.6% (71/248) of patients that were initially admitted to medical wards. The unadjusted associations between clinical factors and case fatality are presented in Table 1.

The proportions of patients with white blood cells count < 4 × 10⁹/l, 4–12 × 10⁹/l and > 12 × 10⁹/l were 11.8% (46/389), 47% (183/389) and 41.1% (160/389) respectively. Neutropenia, severe anaemia and thrombocytopenia was presented in 7.2% (28/388), 8% (31/389) and 43.4% (169/389) patients with BSI on admission to the study hospital, respectively. The proportion of patients with RIFLE-F, increased lactate, procalcitonin or CRP were not significantly different when classified by bacterial aetiology (Enterobacteriaceae, non-Enterobacteriaceae and Gram-positive). Lactate levels were only available for 73 patients and 52/73 (69.9%) patients had lactate level ≥ 2 mmol/L on admission. The median lactate level increased significantly with increasing SOFA score; from 1.23 mmol/L (IQR: 0.99–2.04 mmol/L) in those with a SOFA score < 6, to 7.89 mmol/L (IQR: 3.7–10.38) in those with a SOFA score > 12 (p < 0.001, Kruskal Wallis test). There was no significant difference between median lactate levels in those with qSOFA < 2 and qSOFA≥2 (p = 0.055, Mann-Whitney U test). Laboratory factors associated with case mortality was showed in Table 2. There was no unadjusted associated between procalcitonin and CRP levels with case fatality rates.

The overall case-fatality of CABSI was 28% (110/393), of which 71.8% (79/110) occurred within 7 days of admission to the study hospital. Case fatality rates in patients with CABSI due to Enterobacteriaceae, non-Enterobacteriaceae Gram-negative bacteria and Gram-positive bacteria were 33.5%, 25.6% and 20%, respectively. Among the most common isolates, the case fatality was 35.2% (31/88) in K. pneumoniae, 32.8% (21/64) in Escherichia coli, 9.3% (4/43) in Streptococcus suis, 15.4% (6/39) in Stenotrophomonas maltophilia and 32.4% (11/34) in S. aureus. The case-fatality in patients directly admitted to CCU was 46.9% (68/145).

Organ dysfunction was associated with higher risk of in-hospital mortality (33.2% patients with at least one organ dysfunction on admission to the study hospital vs 9.3% patients without, p < 0.001). Case fatality rate increased with increasing SOFA score (Fig. 1). The mortality in patients with 1, 2, 3 and more than 4 organs dysfunction was 17% (17/100), 19.5% (16/82), 31.9% (15/47) and 69.2% (54/78), respectively. The highest case fatality rates were observed in patients with cardiovascular, respiratory and central nervous system (CNS) dysfunction, 80% (52/65), 60.9% (53/87), 52.4% (55/105), respectively. qSOFA < 2 was associated with lower mortality compared with qSOFA ≥2, 18.8% (26/138) vs 40.6% (63/155), p < 0.001. Among patients admitted directly to CCU, SOFA performed well at predicting in-hospital mortality (AUC 0.858, 95%CI 0.793–0.922) while qSOFA was a poor predictor (AUC 0.692, 95%CI 0.605–0.780) in this population. However, outside of CCU, regardless of eventual CCU admission, both SOFA and qSOFA had poor predictive validity (AUC 0.667, 95%CI 0.577–0.758 and AUC 0.527, 95%CI 0.424–0.630, respectively).

Table 3 shows the prognostic validity of the Youden index of SOFA, qSOFA, WBC, CRP and PCT on admission to NHTD in all patients. The SOFA score was more accurate than qSOFA in predicting mortality (AUC = 0.795 vs 0.658, p < 0.001); PCT (AUC = 0.703), and WBC (AUC = 0.642) was more accurate than CRP (AUC = 0.561) in predicting mortality (P < 0.001 and P = 0.0498 respectively).

In unadjusted association analysis, higher mortality was associated with male sex, any previous hospitalisation, direct CCU admission, history of HIV infection, moderate severe liver diseases and. Analysis, male sex, previous hospitalisation prior to NHTD admission, any history of chronic disease, history of moderate or severe liver diseases, HIV infection, organ dysfunctions on admission, neutropenia, haemoglobin < 80 g/L, RIFLE failure (RIFLE-F), hypoalbuminemia (< 30 g/L), elevated liver enzymes (> 2 times the upper limit of normal), thrombocytopenia (< 100 × 10³/mm³) and were associated with increased in-hospital mortality (p values < 0.05). In Cox regression proportional hazards model, HIV infection (HR = 3.145, p = 0.001), neutropenia (HR = 2.442, p = 0.002), SOFA score 1-point increment (HR = 1.19, p < 0.001) and infection with Enterobacteriaceae (HR = 1.722, p = 0.037) were significant risk factors for in-hospital mortality (Table 4).