1. Study design
We conducted a retrospective cohort study by analyzing data from a prospective septic shock registry at an urban tertiary emergency department with annual admissions of more than 120,000 patients from 1 January 2014 to 31 December 2018. This registry was designed to enhance internal quality improvement of septic shock patients visiting emergency department and manage multicenter-cohort data of Korean Shock Society for future analysis [8]. The Institutional Review Board of the study facility approved the study (no. 2016-0548) and waived the requirement for informed consent because of its retrospective characteristics.
2. Data collection and definition of variables
This registry included all adult septic shock patients (≥ 18 years) consecutively diagnosed in the emergency department [9]. Infection was defined clinically by the emergency physicians on duty. In brief, they evaluated the systemic inflammatory response syndrome and quick sequential organ failure assessment (SOFA) criteria for every patient with suspected or confirmed infection [9]. In addition, we used a definition of septic shock as refractory hypotension (mean arterial pressure ≤ 65 mmHg) requiring vasopressors despite adequate fluid infusion or a blood lactate level of at least 4 mmol/L, based on a previous definition [10]. This registry excluded those who were transferred from other hospitals after proper resuscitation, transferred to another hospital because of no room for admission, had a “do-not-resuscitate” order, or refused to accept treatment because of non-medical issues, such as health care cost, fear of invasive procedures, or just not want to intensive care unit care (ICU) without definite reasons. We also excluded patients in this secondary analysis with microbiologically proven viral, fungal, and parasitic infections.
All enrolled patients were treated with protocol-driven resuscitation following the Surviving Sepsis Campaign guidelines [3]. In brief, aggressive fluids infusion and vasopressors were applied with blood pressure monitoring. Lactate levels were checked via venous or arterial blood gas analysis, and a central venous catheter was placed routinely for administering high dose vasopressors and repetitive sampling [11,12]. Blood cultures were obtained within 3 hours of recognition from samples at two or more different anatomical sites according to the local practice [13]. When indwelling catheters were present, one blood sample was obtained through the catheter, and the remainder were taken from different peripheral venous sites. Site-specific cultures, including urine, sputum, pleural, ascites, stool, and pus, were performed following the physicians’ decisions. Broad-spectrum empirical antibiotics were infused as soon as possible, and percutaneous or endoscopic drainages were aggressively conducted after image work-ups. The sites of infection were grouped as lower respiratory tract, urinary tract, gastrointestinal, and hepatobiliary. Minor sites, such as isolated blood stream infection, skin and soft tissue, and central nervous system were categorized others. The infectious sites were determined and recorded in the registry by the primary physicians on duty through the patients’ histories, physical examinations, and results of laboratory and imaging. We deemed fever without a definite focus as unknown.
Data regarding demographics, underlying diseases, initial vital signs, sites of infection, and clinical outcomes, such as intensive care unit (ICU) admission, requirements for a mechanical ventilator or renal replacement therapy, and duration of ICU stay and mechanical ventilation were obtained from the registry. The date of the patient’s death was brought out from the National Health Insurance Service in South Korea, and in-hospital and 90-day mortality were extracted. Subgroup analysis of in-hospital mortality according to site of infection was also investigated to confirm the clinical differences based on infection sites. Laboratory examinations, including white blood cell counts, hemoglobin, prothrombin time (international normalization ratio), lactate, and C-reactive protein, were also extracted. Source control, including all physical procedures to eliminate sources of pathogen, to draining abscess, to debride infected soft tissues, and to extract contaminated devices or foreign bodies, were recorded [14]. Escalation of antibiotics occurred whenever the initial antibiotics were changed to cover more extensive pathogens because of refractory shock or the identification of antibiotic-resistance pathogens. SOFA scores were calculated from the initial clinical and laboratory data on admission.
Moreover, microbiological culture results and the detection time of enrolled patients were extracted from electronic medical records and reviewed by three investigators (J.S.K., Y.J.K., and W.Y.K.). An experienced infection specialist made a final decision when there was disagreement for discriminating exact pathogen to contamination. Detection time was defined as the time to positive detection of the pathogens and calculated by subtracting the time of receipt in the laboratory from the time required to detect a positive culture. When growth was detected in both aerobic and anaerobic bottles, each time was recorded [see Additional file 1].
The primary outcome was 90-day mortality, and the secondary outcomes were ICU admission, ICU length of stay, mechanical ventilator or renal replacement therapy requirement, mechanical ventilation duration, and in-hospital mortality.
3. Statistical analysis
Descriptive statistics were stratified by culture results (i.e., culture-positive and negative). Baseline demographics, clinical characteristics, and outcomes were presented as frequency and percentage for categorical and median with interquartile range (IQR) for continuous variables. The Kolmogorov-Smirnov test was used to check the normality of the distribution. Categorical variables were analyzed using chi-squared or Fisher’s exact tests. Univariate logistic regression test was performed with potential risk factors which showed differences between two groups. Multivariate logistic regression model was conducted for the variables that had significance in univariate logistic regression analysis. Multicollinearity was confirmed by using regression analysis with calculating variance inflation factor values. Kaplan-Meier survival curves with the log-rank test were stratified by the culture results. The correlations between the severity of the septic shock and culture detection time were assessed by the Spearman rank correlation coefficient. We considered P-values less than 0.05 as statistically significant. Analyses were performed using SPSS Statistics for Windows, version 26 (IBM Corp., Armonk, NY, USA).