This observational retrospective cohort study with propensity score matching was conducted at Ramathibodi Hospital, a tertiary care and university hospital in Bangkok, Thailand. The study protocol was approved by the Ethics Committee of Ramathibodi Hospital, Mahidol University (COA MURA2020/505). The ethics committee waived the need for informed consent of each patient.
Study setting and population
The study enrolled patients 18 years and older who visited the Ramathibodi Hospital ED with suspected sepsis and were screened using the Ramathibodi sepsis protocol from August 1, 2019, to December 31, 2019. The exclusion criteria were as follows: (i) patient and/or family submission of a do-not-attempt resuscitation order, (ii) transfer to another hospital, (iii) treatment at another hospital before arrival at our ED, and (iv) missing vital signs, time of antibiotic administration, time of ICU admission, or mortality.
In Ramathibodi Hospital, during the pre-period (August 1, 2019, to October 29, 2019), patients with suspected sepsis were managed according to the standard of care defined by the sepsis-3 criteria. At the triage area, the nurse triggered a sepsis alert system in patients with an initial qSOFA score ≥ 2 or suspected sepsis and tracked patients to physicians for judgment following sepsis protocols. The REWs was provided in the ED on October 30, 2019, due to the sepsis committee of Ramathibodi Hospital. During the post-period (October 30, 2019, to December 31, 2019), patients were tracked following sepsis protocols, including REWs screening. If the REWs ≥ 4 points, the triage nurse alerted sepsis protocol, and patients were managed following the same sepsis protocols as those used during the pre-period. In addition, during the post-period, REWs were implemented as part of the continuity of care process, including monitoring and notifying physicians to evaluate patients.
The emergency physicians and nursing teams, the number of medical personnel per shift in the ED, and the process for administration of antibiotics were the same during the pre- and post-periods. All ED physicians and residents were trained in the use of sepsis protocols and in management of patients with sepsis.
Suspected sepsis was defined by qSOFA score ≥ 2 as a sepsis-3 criteria or physician judgment suspected infection in pre-period and REWs ≥ 4 in post-period that consists of lactate measurement, taking hemoculture and antibiotic administration. The REWs score was defined based on clinical parameters including systolic blood pressure, heart rate, respiratory rate, pulse oximetric saturation, body temperature, and mental status. Each parameter was scored as shown in Supplementary Online Table 1.
Patients included in this study were evaluated by emergency room staff. We recorded each patient’s demographic information (age, sex, and comorbidities), vital signs (systolic blood pressure, respiratory rate, body temperature, and pulse oxygen saturation), qSOFA score, and REWs in the triage area. Blood lactate levels were assessed upon admission. Data collected from hospital database software were used for this study. Additionally, the medical record was audited by an internal auditor in Ramathibodi Hospital, which uses medical record audit guideline 2021 by the Healthcare Accreditation Institute of Thailand (public organization).
The primary outcome of this study was the impact of the REWs on the administration of antibiotics within 1 h of presentation. The secondary outcomes were time from presentation to ICU admission and 28-day mortality rate among patients with sepsis treated according to standard of care (sepsis-3) and after addition of the REWs trigger tools to sepsis protocols.
The sample size required for this study was calculated based on a pilot study in the in-patient department. This study found that the proportion of patients with suspected sepsis who received antibiotics within 1 h as part of standard of care (sepsis-3) was 38%, and that following implementation of the REWs, the mean early warning score was 0.57. Therefore, this study required a sample size of 150 patients in each group assuming an alpha error of 0.05 (two-sided) and a power of 0.9.
Categorical variables were expressed as counts and percentages, and continuous variables were expressed as medians and interquartile ranges (IQRs). A propensity score (PS) generated from a multivariate logistic regression model was calculated for trigger tools. A total of 21 variables were used to generate the PS. The PS from the regression model (with nearest neighbor 1:1 matching without replacement) was derived with a caliper of 0.05.
For analysis of the pre-and post-propensity score matching groups, categorical variables were compared using Pearson’s chi-square test, and continuous variables were compared using Mann-Whitney U-tests. Multivariate Cox regression was used to assess associations between time to antibiotic administration and 28-day mortality.
Statistical analyses were performed using IBM SPSS Statistics for Windows, version 22.0 (Armonk, NY, USA), and STATA version 16.1 (StataCorp, College Station, TX, USA).