A retrospective cohort study was performed in the ED of Chiang Mai University Hospital, Thailand from November 2017 to December 2018. Chiang Mai University Hospital is a tertiary-care, non-profit, university hospital. The estimated number of patients receiving healthcare service is 70,000 patients per month. Of these, roughly 40 patients with sepsis receive care in the ED per month. Inclusion criteria were patients aged older than eighteen years old diagnosed with sepsis by the Third International Consensus definition for sepsis (Sepsis-3) and received at least one dose of the single antimicrobial agent in an ED. The exclusion criteria were patients referred from another hospital with prior intravenous antibiotic exposure, and patients admitted at another hospital where the treatment chart or electronic record database cannot be reviewed to assess treatment outcomes.
The patients were recruited using the ICD-10 referred for sepsis (include A41, R65, R57) from the hospital electronic medical records (EMR; provided by Digicard 2007® and SMI®). The chart review was performed by data collectors, consisting of health care providers and technicians. We re-assessed the SOFA scores of all patients from the pooled data, included baseline SOFA scores before the current visit and current SOFA scores at emergency room disposition. All intravenous antibiotics given to these patients were counted as empirical antibiotics. If no intravenous antibiotics were given in an ED, these patients were excluded from the study. The patients were identified as an immunocompromised host with at least one of the following conditions: (1) immunosuppression (defined as viral immunosuppression, neoplastic disease, immunosuppressive drugs including steroids, chemotherapy, or congenital immunosuppression), (2) active hematologic malignancy (i.e., still requiring treatment), and (3) active neoplasm (i.e., a neoplasm that has not been resected, still requires treatment, or with metastasis)(22, 23). Probe thermometers were used to evaluate body temperature; the probes were placed in the armpit as the most convenient site. The relationship between axillary and core temperature measurement was a moderate to high correlation (r = 0.53–0.90)(24−26). The temperature was recorded in Celsius degree units.
Patients were divided into two groups according to laboratory culture information: the culture-positive sepsis (CPS) group and the culture-negative sepsis (CNS) group. All patients were assessed for the appropriateness of empirical antibiotic use (EAU) by pre-defined criteria before the clinical outcomes were assessed. Antibiotic use in compliance with the hospital’s Clinical Practice Guidelines (CPG) was also assessed. Patients included in our study were almost treated in compliance with the hospital sepsis practice guideline, developed from the recent updates in Survival Sepsis Campaign – intravenous antibiotic was given within one hour, adequate intravenous fluids were given, and vasopressor was given as indicated. The hemocultures and cultures from source-specific sites were collected by registered nurses to ensure the adequacy of culture results.
Appropriate antibiotic use in the CPS group was judged based on the antibiotic sensitivity data on bacteria isolated from suspected sites of infection. The empirical antibiotic was considered to be appropriate if it covered all causative bacteria. For the culture-negative sepsis group (CNS group), the appropriateness was judged by response to antibiotic treatment, defined as highest body temperature less than 38 degrees Celsius or white blood count less than 10,000 cells per mm3 after treatment (27). The empirical antibiotic was considered to be appropriate if the patient became responsive to treatment at 48–72 hours. All appropriateness of empirical antibiotic use was assessed by the project leader. The overuse of empirical antibiotics was also assessed. The EAU defined as overuse if using the broad-spectrum antibiotic without indications of the hospital’s CPG.
The treatment outcomes included 30-day mortality, length of hospital stays, length of ICU stays, and total cost of hospital admission. These outcomes were collected using the data recorded in the EMR. The survival analysis was done in both the CPS and CNS group. Time origin was the day of each hospital admission. All-cause mortalities were counted as an event. The hospital discharge for end-of-life care was also counted as an event only if there was no hospital visit after that disposition. Censoring criteria include patient survival to hospital disposition, except for end-of-life care.
The analysis plan according to a previous study(18). The sample size calculation was assessed by in-hospital mortality of culture-positive sepsis patients with an appropriate antibiotic group (24.2%) compared to an inappropriate antibiotic group (38.8%), at a ratio of patient 0.37. To have power at 80% and α-level at 0.05, we enrolled 420 culture-positive patients. Among these, defined as 305 appropriate EAU and 115 as inappropriate EAU. The data was collected until culture-positive patients met with the calculated number. Because there was no data about the impact of appropriate empirical antibiotics in culture-negative sepsis patients, the sample size calculation was not done in this patient group.
Study data were collected and managed using REDCap electronic data capture tools hosted at the Faculty of Medicine, Chiang Mai University. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies(28, 29). The REDCap data was limited access for the related users and the user rights were determined as the data collector, outcome accessor, and project manager. The identifier data was removed before exporting from REDCap. Pooled project data in REDCap was deleted at three months after the final analysis. The data were managed and analyzed by Microsoft Excel® 16.29.1 for Mac (License no. 02954-038-529776) and Stata/MP® 16.0 for Mac (Serial no. 501609250008)
Statistical Analysis
Data analysis used descriptive statistics to measure central tendency and variability. We assessed normality by measuring skewness and kurtosis, histogram, and Kolmogorov-Smirnov tests. The normally distributed numerical variables were presented as mean ± SD, and other numerical variables as median (interquartile range). Comparison between the study groups was performed; Chi-square and Fisher’s exact test were used for categorical data. Unpaired t-test, ANOVA, Wilcoxon Rank Sum Test, and Kruskal-Wallis test were used for continuous data. The risk ratio was used to measure 30-day mortality between appropriate and inappropriate EAU group. The survival analysis was performed as a univariate analysis using the log-rank statistic, Kaplan-Meier estimate graph, and Cox analysis model.