Key Points
Question
Does electronic sepsis screening based on quick Sequential Organ Failure Assessment score (qSOFA), compared with no screening, reduce the mortality of patients admitted to hospital wards?
Findings
In a stepped-wedge, cluster randomized trial at 45 wards (clusters) in 5 hospitals in Saudi Arabia, 29 442 patients were in the screening group and 30 613 in the no screening group. Accounting for period, clustering within wards, hospitals, and COVID-19 status, electronic sepsis screening significantly reduced in-hospital mortality within 90 days compared with no screening (adjusted relative risk, 0.85).
Meaning
In hospitalized ward patients, electronic sepsis screening compared with no screening resulted in reduced 90-day in-hospital mortality.
Abstract
Importance
Sepsis screening is recommended among hospitalized patients but is supported by limited evidence of effectiveness.
Objective
To evaluate the effect of electronic sepsis screening, compared with no screening, on mortality among hospitalized ward patients.
Design, Setting, and Participants
In a stepped-wedge, cluster randomized trial at 5 hospitals in Saudi Arabia, 45 wards (clusters) were randomized into 9 sequences, 5 wards each, to have sepsis screening implemented at 2-month periods. The study was conducted between October 1, 2019, and July 31, 2021, with follow-up through October 29, 2021.
Intervention
An electronic alert, based on the quick Sequential Organ Failure Assessment score, was implemented in the electronic medical record in a silent mode that was activated to a revealed mode for sepsis screening.
Main Outcomes and Measures
The primary outcome was 90-day in-hospital mortality. There were 11 secondary outcomes, including code blue activation, vasopressor therapy, incident kidney replacement therapy, multidrug-resistant organisms, and Clostridioides difficile.
Results
Among 60 055 patients, 29 442 were in the screening group and 30 613 in the no screening group. They had a median age of 59 years (IQR, 39-68), and 30 596 were male (51.0%). Alerts occurred in 4299 of 29 442 patients (14.6%) in the screening group and 5394 of 30 613 (17.6%) in the no screening group. Within 12 hours of the alert, patients in the screening group were more likely to have serum lactate tested (adjusted relative risk [aRR], 1.30; 95% CI, 1.16-1.45) and intravenous fluid ordered (aRR, 2.17; 95% CI, 1.92-2.46) compared with those in the no screening group. In the primary outcome analysis, electronic screening resulted in lower 90-day in-hospital mortality (aRR, 0.85; 95% CI, 0.77-0.93; P < .001). Screening reduced vasopressor therapy and multidrug-resistant organisms but increased code blue activation, incident kidney replacement therapy, and C difficile.
Conclusions and Relevance
Among hospitalized ward patients, electronic sepsis screening compared with no screening resulted in significantly lower in-hospital 90-day mortality.
Trial Registration
ClinicalTrials.gov Identifier: NCT04078594
This trial compares the effect of electronic sepsis screening vs no screening on mortality among hospitalized patients.
Introduction
The early recognition and treatment of sepsis is essential to improving patient outcomes.1,2,3 As such, clinical practice guidelines recommend sepsis screening programs for hospitalized patients.4 But supporting evidence is scarce. Observational studies have suggested that sepsis screening is associated with improved processes of care such as serum lactate measurement and the administration of antibiotics, intravenous fluids, and vasopressors and may be associated with reduced mortality.5,6,7,8,9 Small randomized clinical trials, however, have shown inconsistent results.10,11,12,13 Sepsis screening may be associated with increased antibiotic use and greater occurrence of multidrug-resistant organisms and Clostridioides difficile.14 Recently, sepsis screening was identified as a research priority that should be addressed using a cluster randomized trial.15
The Stepped-wedge Cluster Randomized Trial of Electronic Early Notification of Sepsis in Hospitalized Ward Patients (SCREEN) evaluated whether electronic sepsis screening using the quick Sequential Organ Failure Assessment (qSOFA) score, compared with no screening, among patients admitted to hospital wards would reduce 90-day in-hospital mortality.16
Methods
Settings and Oversight
The study was conducted in the 5 Ministry of National Guard–Health Affairs hospitals in Saudi Arabia. These general hospitals serve the National Guard members, their families, and the community and use the same electronic medical record (EMR) system (BESTCare).17 The study protocol and statistical analysis plan are in Supplement 1 and were previously published.16,18 The hospital characteristics are summarized in eTable 1 in Supplement 2.19 This study was approved by the Ministry of National Guard–Health Affairs Institutional Review Board. Informed consent was waived because of the system-based nature of the study.
The Electronic Alert
In the study wards, blood pressure and respiratory rate values were entered into the EMR every 4 hours and the Glasgow Coma Scale score every 12 hours (eTable 2 in Supplement 2). The EMR generated an alert on each vital sign input if 2 or more of the qSOFA components (systolic blood pressure ≤100 mm Hg, respiratory rate ≥22 breaths/min, or Glasgow Coma Scale score <15) were met within a 12-hour window based on the recent values. Although calculated and recorded in all patients, the alert could be in silent mode or revealed to the treating team depending on group and period assignment.20
Study Design
This study was designed and reported as per the guidelines outlined in the CONSORT extension for stepped-wedge, cluster randomized trials.21 We included adult medical, surgical, and oncology wards. Other ward-level and patient-level eligibility criteria are found in eTable 3 in Supplement 2.16 Before the trial, 45 eligible wards (clusters) were randomized into 9 sequences, of 5 wards each, using a computer-generated nonstratified concealed list to transition from the no screening group to the screening group.
In the screening group, the intervention included (1) activating the electronic alert system to revealed mode, (2) training of physicians and nurses, and (3) feedback to ward medical and nursing leads about each ward performance in acknowledging the alerts.
After a baseline period of 2 months in which the alert was in the silent mode in all wards, the alert was activated to revealed mode in a new sequence every 2 months until the alert became active in all sequences (Figure 1). The ward allocation remained concealed and was disclosed 1 month before the implementation of screening in a given sequence to allow training. The alerts appeared in the EMR as pop-up messages for the nurse and physician, accompanied by a visual and audible alarm on a handheld device carried by the ward charge nurse (eTable 2 in Supplement 2).16 The alert prompted the nurse to acknowledge the alert, assess the patient, and communicate with the covering physician. It prompted the physician to acknowledge the alert by documenting an assessment of whether the patient had sepsis or not. The acknowledgments were time-stamped.16 The nursing and medical staff training focused on timely assessment, vital sign documentation, and early sepsis management. Dashboards and monthly reports were used for feedback and displayed the number of alerts and the percentage and timing of acknowledgment by nurses and physicians for feedback.16 The hospital administration and quality management department were engaged in the training and feedback process.
Figure 1. Ward and Participant Flow in the SCREEN Trial.
Numbers in boxes for each period and sequence represent the total number of patients and mean (SD) per ward. SCREEN indicates Stepped-wedge Cluster Randomized Trial of Electronic Early Notification of Sepsis in Hospitalized Ward Patients.
aTwo wards, 1 in each of sequences 3 and 9, were excluded after randomization, because they were backup nonoperational wards at the beginning of the study then were converted to intensive care units (ICUs) with the COVID-19 pandemic.
bData were excluded from 1 ward in sequence 5 in periods 5 to 10 and from 1 ward in sequence 7 in periods 6 to 10 because they were converted to ICUs. In addition, 1 ward in sequence 3 was converted to an ICU between June 1, 2020, and August 31, 2020; therefore, it contributed data partially in periods 4 and 5.
cDuring the first wave of COVID-19 cases, total admissions to most wards declined substantially; therefore, 2 consecutive periods (starting June 2020) were extended from 2 to 3 months each to account for the decline in cluster size.
The alert activation was ward-based. On transitioning from the no screening to screening ward, the alert was activated for all upcoming patients, but it also included existing patients who were carried over from the no screening period. For patients who were transferred from the screening to no screening ward or vice versa, the alert was silent or revealed according to the ward assignment.
The primary analysis population included all eligible patients admitted to eligible wards. Patients were assigned to the first admission ward, even if they were subsequently transferred to another ward. Patients were enrolled between October 1, 2019, and July 31, 2021; the last follow-up date was October 29, 2021. When the COVID-19 pandemic began, some wards were designated as COVID-19 wards, and others were converted to intensive care units (ICUs).22 Data from wards were excluded for the duration of use as intensive care. During the first wave of COVID-19, total ward admissions declined substantially; therefore, 2 consecutive periods (starting June 2020) were extended from 2 to 3 months each.
Descriptive Data and Outcomes
Baseline characteristics, intervention data, and other features were extracted from the EMR16,18 (eMethods and eTables 4-7 in Supplement 2). Process measures, such as serum lactate testing, intravenous fluid ordered, and blood, respiratory, and urine culture collected, were described in the 12 hours after the alert in the screening and no screening groups. To assess the cumulative effect of repeated alerts, these measures were also evaluated during the entire ward stay.
The primary outcome was 90-day in-hospital mortality. Secondary outcomes included ICU admission, rapid response team activation, code blue activation, vasopressor therapy, mechanical ventilation, incident kidney replacement therapy, new multidrug-resistant organisms, and new C difficile infection up to 90 days in the hospital. Hospital length of stay was censored at day 90, and ICU-free and antibiotic-free days were calculated in the first 90 days (eTable 8 in Supplement 2).
Statistical Analysis
A reduction of in-hospital mortality in the primary analysis population from a baseline of 3.13% to 2.46% required a sample size of 65 250 patients (mean of 145 patients per ward per period) with 80% power, α of .05, and an intracluster correlation of 0.22.18
The effect of screening on the primary outcome was analyzed using a generalized linear mixed model, and the jackknife method was used to estimate standard errors to account for grouping within wards. We used binary distribution with a log-link function to estimate the adjusted relative risk (aRR) with 95% confidence intervals (CIs).23,24 We accounted for periods and nested clustering within wards as random effects and hospitals and COVID-19 status as fixed effects. If the model failed to converge, we reported the adjusted odds ratio (aOR) from the mixed-effect logistic regression (eMethods and eTable 9 in Supplement 2).
The effect of screening on process measures and categorical secondary outcomes was tested using a similar approach to the primary outcome. The effect of screening on continuous secondary outcomes was tested using a mixed-effect Poisson or negative binomial model with similar adjustments to the primary outcome analysis.
To address between-cluster contamination, we conducted a sensitivity analysis excluding patients in the no screening group admitted within 90 days before transitioning to the screening group. We also conducted a sensitivity analysis restricting the duration of observation to 14 days, thus evaluating 14-day in-hospital mortality. Other sensitivity analyses are outlined in the statistical analysis plan in Supplement 1 and the eMethods in Supplement 2. We evaluated for effect modification of the primary outcome in prespecified subgroups, including patients aged 65 and younger and older than 65 years; patients with and without documented infection source; patients admitted to medical, surgical, oncology, and mixed wards; and patients with and without COVID-19 (eTable 10 in Supplement 2). In a secondary analysis, the population was restricted to those alerted patients, and the primary and secondary outcomes were assessed with a similar approach.
Analyses of process measures, secondary outcomes, and subgroups were adjusted for multiplicity by reporting the false discovery rate (FDR).25 Statistical tests were 2-sided, with a statistical significance level set at 5%. All analyses were performed using SAS version 9.4 (SAS Institute).
Results
Participants
Of 45 wards randomized, 2 wards were excluded after randomization, and the study was implemented in 43 wards (Figure 1). Of 62 378 patients admitted to the study wards, 60 055 were eligible. Patients in the screening group (n = 29 442) and no screening group (n = 30 613) were similar in age, sex distribution, comorbidities, and baseline physiologic data (Table 1; eTable 11 in Supplement 2). However, patients in the screening group were more likely to be admitted to medical or mixed wards and to have a diagnosis of COVID-19 (11.8% vs 5.2%).
Table 1. Baseline Data of Patients in the Screening and No Screening Groups in the Primary Analysis Populationa.
| Screening (n = 29 442) | No screening (n = 30 613) | |
|---|---|---|
| Age, median (IQR), y | 59 (39-67) | 59 (40-68) |
| No. | 29 090 | 30 137 |
| Sex, No./total (%) | ||
| Female | 14 534/29 427 (49.4) | 14 893/30 596 (48.7) |
| Male | 14 899/29 427 (50.6) | 15 697/30 596 (51.3) |
| Admitting ward, No. (%) | ||
| Medical | 11 682 (39.7) | 10 422 (34.0) |
| Surgical | 8294 (28.2) | 12 150 (39.7) |
| Mixed (any combination) | 6670 (22.7) | 4635 (15.1) |
| Oncology | 2796 (9.5) | 3406 (11.1) |
| Hospital admission source, No. (%) | ||
| Emergency department | 20 443 (69.4) | 20 372 (66.6) |
| Otherb | 8999 (30.6) | 10 241 (33.4) |
| Transferred to the ward from ICU, No. (%) | 1552 (5.3) | 1526 (5.0) |
| Comorbid conditions, No. (%)c | ||
| Any comorbidity | 11 327 (38.5) | 11 521 (37.6) |
| Kidney | 9856 (33.5) | 9784 (32.0) |
| Gastrointestinal | 9566 (32.5) | 10 102 (33.0) |
| Immunocompromised state | 6688 (22.7) | 4826 (15.8) |
| Cardiovascular | 1996 (6.8) | 2228 (7.3) |
| Diabetes | 1946 (6.6) | 2284 (7.5) |
| Neurological | 1974 (6.7) | 3024 (9.9) |
| Respiratory | 1246 (4.2) | 1338 (4.4) |
| Charlson Comorbidity Index score, median (IQR)d | 2 (2-4) | 2 (2-4) |
| Source of infection on admission, No. (%)e | ||
| No documented infection source | 22 018 (74.8) | 23 717 (77.5) |
| Documented infection source | 7424 (25.2) | 6896 (22.5) |
| COVID-19f | 3462 (11.8) | 1589 (5.2) |
| Respiratory infection | 2790 (9.5) | 2198 (7.2) |
| Other infections | 1984 (6.7) | 2374 (7.8) |
| Urinary tract infection | 1425 (4.8) | 1408 (4.6) |
| Intra-abdominal infection | 1059 (3.6) | 1170 (3.8) |
| Skin and soft tissue and cardiovascular infection | 404 (1.4) | 489 (1.6) |
| Vital signs on admission, median (IQR)g | ||
| Lowest systolic blood pressure, mm Hg | 110 (101-122) [n = 29 275] | 110 (101-123) [n = 30 431] |
| Lowest diastolic blood pressure, mm Hg | 57 (51-64) [n = 29 253] | 57 (51-64) [n = 30 419] |
| Highest heart rate, beats/min | 93 (84-103) [n = 29 315] | 92 (83-102) [n = 30 467] |
| Highest respiratory rate, breaths/min | 20 (20-22) [n = 27 553] | 20 (20-22) [n = 30 461] |
| Highest temperature, °C | 37.0 (36.9-37.2) [n = 29 313] | 37.0 (36.9-37.2) [n = 30 469] |
Abbreviation: ICU, Intensive care unit.
Refer to eTable 11 in Supplement 2 for more information on patients’ baseline characteristics.
Nonemergency department admission source included outpatient clinics, operating rooms, and transfers from other hospitals.
Data on comorbid conditions were extracted based on documented International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) codes, as defined in eTable 4 in Supplement 2.
For the Charlson Comorbidity Index, age and each comorbidity category are weighted from 1 to 6, and the sum of the weights produces the score. A score of 0 indicates an absence of known coexisting conditions, and higher scores indicate a greater disease burden.
No documented infection source was defined as no documented ICD-10-AM code for respiratory, urinary tract, skin, soft tissue, cardiovascular, intra-abdominal, or other infections. Other infection sources included central nervous system infections, connective tissue infections, joint infections, meningococcal disease, and osteomyelitis (eTable 5 in Supplement 2). Some patients had more than 1 source of infection.
COVID-19 was diagnosed based on reverse transcription polymerase chain reaction testing anytime during the index hospital admission or the preceding emergency department visit.
Lowest values for systolic and diastolic blood pressure values and highest values for temperature, heart rate, and respiratory rate were documented within the 12 hours before or after admission to the ward.
Alert Information, Subsequent Procedures, and Treatments
In the screening group, 9447 alerts occurred in 4299 of 29 442 patients (14.6%), with a median of 3 alerts (IQR, 1-6) per patient. Of these, 1954 patients (45.5%) had a documented infection source and 3494 patients (81.3%) had received antibiotics prior to the alert. In the no screening group, 8418 alerts occurred in 5394 of 30 613 patients (17.6%) with a median of 2 alerts (IQR, 1-2) per patient. Of these, 1915 patients (35.5%) had a documented infection source and 3923 patients (72.7%) had received antibiotics prior to the alert. Other demographic, physiologic, and laboratory features of patients with alerts are provided in Table 2 and eTables 12-14 in Supplement 2.
Table 2. Prealert, Alert, and Postalert Data in the Screening and No Screening Groups in the Alert-Only Population.
| Screening (n = 4299) | No screening (n = 5394) | Difference (95% CI) | Unadjusted RR (95% CI) | Adjusted RR/OR(95% CI)a | P value | FDRb | |
|---|---|---|---|---|---|---|---|
| Source of infection on admission, No. (%)c | |||||||
| No documented infection source | 2345 (54.6) | 3479 (64.5) | |||||
| Any infection | 1954 (45.5) | 1915 (35.5) | |||||
| COVID-19d | 931 (21.7) | 380 (7.0) | |||||
| Respiratory infection | 955 (22.2) | 768 (14.2) | |||||
| Other infections | 576 (13.4) | 636 (11.8) | |||||
| Urinary tract infection | 389 (9.1) | 548 (10.2) | |||||
| Intra-abdominal infection | 96 (2.2) | 165 (3.1) | |||||
| Skin and soft tissue and cardiovascular infection | 63 (1.5) | 104 (1.9) | |||||
| Antibiotics ordered before the alert, No. (%) | 3494 (81.3) | 3923 (72.7) | |||||
| Total No. of alerts | 9447 | 8418e | |||||
| Alert count per patient, median (IQR) | 3 (1 to 6) | 2 (1 to 2) | |||||
| Time from ward arrival to first alert, median (IQR), h | 23.1 (4.4 to 86.0) | 19.4 (3.5 to 69.4) | |||||
| Alert acknowledgment by a nurse, No. (%)f,g | |||||||
| Not acknowledged | 1144 (26.6) | 4736 (87.8) | |||||
| Acknowledged | 3155 (73.4) | 658 (12.2) | |||||
| Time to acknowledgment, min | |||||||
| 0-15 | 1710 (39.8) | 43 (0.8) | |||||
| 16-60 | 618 (14.4) | 21 (0.4) | |||||
| >60 | 827 (19.2) | 594 (11.0) | |||||
| Alert acknowledgment by a physician, No. (%)f,g | |||||||
| Not acknowledged | 1126 (26.2) | 4707 (87.3) | |||||
| Acknowledged | 3173 (73.8) | 687 (12.7) | |||||
| Time to acknowledgment, min | |||||||
| 0-30 | 526 (12.2) | 13 (0.2) | |||||
| 31-120 | 554 (12.9) | 13 (0.2) | |||||
| >120 | 2093 (48.7) | 661 (12.3) | |||||
| Physician assessment as sepsis, No./total (%) | |||||||
| Yes | 1105/3173 (34.8) | 169/687 (24.6) | |||||
| No | 2068/3173 (65.2) | 518/687 (75.4) | |||||
| Procedures and treatments following the alert, No. (%) | |||||||
| Lactate tested | |||||||
| Within 12 h following the first alerth | 612 (14.2) | 556 (10.3) | 3.9 (2.6 to 5.3) | 1.38 (1.24 to 1.54) | aRR: 1.30 (1.16 to 1.45) | <.001 | <.001 |
| From first alert and throughout stay on the wardi,j | 1551 (36.1) | 1540 (28.6) | 7.5 (5.7 to 9.4) | 1.26 (1.19 to 1.34) | aRR: 1.21 (1.14 to 1.29) | <.001 | <.001 |
| Intravenous fluid ordered | |||||||
| Within 12 h following the first alerth | 617 (14.4) | 364 (6.8) | 7.6 (6.4 to 8.9) | 2.13 (1.88 to 2.41) | aRR: 2.17 (1.92 to 2.46) | <.001 | <.001 |
| From first alert and throughout stay on the wardi,j | 2744 (63.8) | 2113 (39.2) | 24.7 (22.7 to 26.6) | 1.63 (1.57 to 1.70) | aRR: 1.61 (1.54 to 1.67) | <.001 | <.001 |
| Blood culture collected | |||||||
| Within 12 h following the first alerth | 597 (13.9) | 667 (12.4) | 1.5 (0.2 to 2.9) | 1.12 (1.01 to 1.24) | aRR: 1.04 (0.94 to 1.16) | .42 | .51 |
| From first alert and throughout stay on the wardi,j | 2592 (60.3) | 2691 (49.9) | 10.4 (8.4 to 12.4) | 1.21 (1.17 to 1.25) | aOR: 1.40 (1.27 to 1.55) | <.001 | <.001 |
| Respiratory culture collected | |||||||
| Within 12 h following the first alerth | 459 (10.7) | 470 (8.7) | 2.0 (0.8 to 3.2) | 1.23 (1.08 to 1.38) | aRR: 1.10 (0.97 to 1.25) | .14 | .18 |
| From first alert and throughout stay on the wardi,j | 1957 (45.5) | 1829 (33.9) | 11.6 (9.7 to 13.6) | 1.34 (1.28 to 1.41) | aOR: 1.39 (1.28 to 1.52) | <.001 | <.001 |
| Urine culture collected | |||||||
| Within 12 h following the first alerth | 468 (10.9) | 528 (9.8) | 1.1 (−0.1 to 2.3) | 1.11 (0.99 to 1.25) | aRR: 1.03 (0.92 to 1.17) | .53 | .53 |
| From first alert and throughout stay on the wardi,j | 2200 (51.2) | 2340 (43.4) | 7.8 (5.8 to 9.8) | 1.18 (1.13 to 1.23) | aRR: 1.14 (1.09 to 1.19) | <.001 | <.001 |
| New antibiotic orderedk | |||||||
| Within 12 h following the first alerth | 773 (18.0) | 979 (18.2) | −0.2 (−1.7 to 1.4) | 0.99 (0.91 to 1.08) | aRR: 0.97 (0.89 to 1.06) | .50 | .53 |
| From first alert and throughout stay on the wardi,j | 3473 (80.8) | 3892 (72.2) | 8.6 (7.0 to 10.3) | 1.12 (1.10 to 1.14) | aOR: 1.45 (1.32 to 1.60) | <.001 | <.001 |
Abbreviations: FDR, false discovery rate; OR, odds ratio; RR, relative risk.
The effect of the screening compared with no screening on these process measures was tested using a generalized linear mixed model accounting for periods and nested clustering within wards as random effects and hospitals and COVID-19 status as fixed effects, and the results were reported as aRRs with 95% CIs. If the model failed to converge, aORs from the mixed-effect logistic regression were reported.
The FDR accounts for multiplicity by calculating the expected proportion of tests with false-positives at a specified rank of a set of tests.
No documented infection source was defined as no documented International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification code for respiratory, urinary tract, skin, soft tissue, cardiovascular, intra-abdominal, or other infections. Other infection sources included central nervous system infections, connective tissue infections, joint infections, meningococcal disease, and osteomyelitis (eTable 5 in Supplement 2). Some patients had more than 1 source of infection.
COVID-19 was diagnosed based on reverse transcription polymerase chain reaction testing anytime during the index hospital admission or the preceding emergency department visit.
The alerts in the no screening group were mostly silent alerts, but also included some revealed alerts. There are 2 reasons for this. Upon transitioning from no screening to a screening ward, the activation of the alert was applied to all upcoming patients, but also included existing patients at the time, who are carried over from the no screening period. In addition, for patients who were transferred from one ward to another, the alert followed the geographic location.
These data are based on first alert.
The acknowledgments of the nurse and physician were time-stamped; acknowledgment of physician included documenting an assessment whether the patient had sepsis or not. Acknowledgments in the no screening group were related to the alert activation being ward-based. On transitioning from no screening to a screening ward, the activation of the alert was applied to all upcoming patients, but also included existing patients at the time, who are carried over from the no screening period. In addition, for patients who were transferred from one ward to another, the alert followed the geographic location.
Prespecified process measure analysis.
Post hoc process measure analysis.
Censored at day 90.
A list of antibiotics is provided in eTable 6 in Supplement 2.
In the screening group, alerts in 3155 patients (73.4%) were acknowledged by nurses and in 3173 patients (73.8%) by physicians, of which the physicians considered 1105 of 3173 patients (34.8%) to have sepsis. In the no screening group, alerts in 658 patients (12.2%) were acknowledged by nurses and in 687 patients (12.7%) by physicians (Table 2).
In the 12 hours following the first alert, patients in the screening group were more likely to have serum lactate tested (aRR, 1.30; 95% CI, 1.16-1.45; FDR < 0.001) and intravenous fluid ordered (aRR, 2.17; 95% CI, 1.92-2.46; FDR < 0.001; Table 2). Patients with an alert compared with patients with no alert had higher risk of 90-day in-hospital mortality (1150 of 9693 patients [11.9%] compared with 748 of 50 362 patients [1.5%]; OR, 8.93; 95% CI, 8.12-9.82).
Primary Outcome
The crude 90-day in-hospital mortality was not different in the screening group (937 of 29 442 patients [3.2%]) compared with the no screening group (961 of 30 613 patients [3.1%]) (difference, 0.0%; 95% CI, −0.2% to 0.3%). When adjusting for period, clustering within wards, hospitals, and COVID-19 status in the primary analysis, the screening group had lower 90-day in-hospital mortality compared with the no screening group (aRR, 0.85; 95% CI, 0.77-0.93; P < .001, Table 3; eTables 15 and 16 in Supplement 2). The adjusted treatment effect corresponded to a number needed to screen of 206 patients (95% CI, 140-425).
Table 3. Primary and Secondary Outcomes in the Screening and No Screening Groups in the Primary Analysis Populationa.
| Screening (n = 29 442) | No screening (n = 30 613) | Difference (95% CI) | Unadjusted RR (95% CI) | Adjusted RR/OR (95% CI) | P value | FDRb | |
|---|---|---|---|---|---|---|---|
| Primary outcome: 90-d in-hospital mortality, No./total (%)c | 937/29 442 (3.2) | 961/30 613 (3.1) | 0.0 (−0.2 to 0.3)d | 1.01 (0.93 to 1.11) | aRR: 0.85 (0.77 to 0.93)e | <.001 | |
| Secondary outcomes, No. (%) | |||||||
| ICU admission | 1753 (6.0) | 1550 (5.1) | 0.9 (0.5 to 1.3) | 1.18 (1.10 to 1.26) | aOR: 1.03 (0.94 to 1.13) | .50 | .50 |
| Rapid response team activation | 1402 (4.8) | 1330 (4.3) | 0.4 (0.1 to 0.8) | 1.10 (1.02 to 1.18) | aOR: 0.96 (0.89 to 1.04) | .27 | .30 |
| Code blue | 296 (1.0) | 188 (0.6) | 0.4 (0.3 to 0.5) | 1.64 (1.36 to 1.96) | aOR: 1.24 (1.02 to 1.50) | .03 | .04 |
| Vasopressor therapyf | 862 (2.9) | 844 (2.8) | 0.2 (−0.1 to 0.4) | 1.06 (0.97 to 1.17) | aRR: 0.86 (0.78 to 0.94) | .002 | .004 |
| Mechanical ventilation | 814 (2.8) | 700 (2.3) | 0.5 (0.2 to 0.7) | 1.10 (1.05 to 1.15) | aOR: 0.93 (0.84 to 1.03) | .18 | .22 |
| Incident kidney replacement therapy, No./No. (%)g | 1294/28842 (4.5) | 1088/30168 (3.6) | 0.9 (0.6 to 1.2) | 1.24 (1.15 to 1.35) | aOR: 1.20 (1.11 to 1.31) | <.001 | <.001 |
| New multidrug-resistant organismsh | 526 (1.8) | 598 (2.0) | −0.2 (−0.4 to 0.1) | 0.91 (0.81 to 1.03) | aOR: 0.88 (0.78 to 0.99) | .03 | .04 |
| New Clostridioides difficileh | 186 (0.6) | 147 (0.5) | 0.2 (0.0 to 0.3) | 1.32 (1.06 to 1.63) | aOR: 1.30 (1.03 to 1.65) | .03 | .04 |
| Beta estimate (95% CI) | |||||||
| Hospital length of stay, di | |||||||
| Median (IQR) | 3.7 (1.8 to 7.8) | 3.9 (2.0 to 8.1) | |||||
| Mean (SD) | 7.7 (12.4) | 7.8 (12.3) | 0.1 (−1 to 0.3) | NA | −0.10 (−0.12 to −0.07) | <.001 | <.001 |
| ICU-free days up to day 90j | |||||||
| Median (IQR) | 90.0 (90.0 to 90.0) | 90.0 (90.0 to 90.0) | |||||
| Mean (SD) | 86.9 (15.7) | 87.0 (15.6) | 0.1 (−0.2 to 0.3) | NA | 0.005 (0.002 to 0.008) | <.001 | .004 |
| Antibiotic free-days up to 90 dk | |||||||
| Median (IQR) | 88.0 (85.0 to 90.0) | 88.0 (85.0 to 90.0) | |||||
| Mean (SD) | 83.8 (16.1) | 83.9 (16.0) | 0.1 (−0.4 to 0.2) | NA | 0.004 (0.001 to 0.006) | .001 | .004 |
Abbreviations: FDR, false discovery rate; ICU, intensive care unit; NA, not applicable; OR, odds ratio; RR, relative risk.
The effect of the screening compared with no screening on categorical outcomes was tested using a generalized linear mixed model accounting for periods and nested clustering within wards as random effects and hospitals and COVID-19 status as fixed effects, and the results were reported as adjusted RRs with 95% confidence intervals (CIs). If the model failed to converge, adjusted ORs from the mixed-effect logistic regression were reported. The effect of the screening compared with no screening on continuous outcomes was tested using a mixed-effect Poisson or negative binomial models with adjustments that were similar to what was performed in categorical outcomes analyses. More details about definitions of outcomes are provided in eTable 8 in Supplement 2 and additional details on the analysis of the primary outcome are provided in eTables 15, 16, and 17 in Supplement 2.
The FDR accounts for multiplicity by calculating the expected proportion of tests with false-positives at a specified rank of a set of tests.
For the analysis of the primary outcome, the model converged using the variance components structure and the corresponding intracluster correlation was 0.23.
The mean difference is 0.04% and was rounded to 1 decimal point.
The adjusted treatment effect corresponds to a number needed to screen of 206 patients (95% CI, 140-425).
Vasopressors included norepinephrine, epinephrine, and phenylephrine.
Incident kidney replacement therapy was calculated on all patients except those receiving kidney replacement therapy on admission to the ward, defined as having received intermittent dialysis or continuous kidney replacement therapy before or within 48 hours of arrival to the ward.
Patients were identified as having new multidrug-resistant organisms or Clostridioides difficile based if flagged as such by infection control practitioners for isolation precaution purposes after 48 hours of arrival to the ward.
Hospital length of stay was censored at day 90.
ICU-free days calculation was calculated within the 90 days of check-in to the ward. For multiple admissions to the ICU, we used the first ICU admission date and the last ICU discharge date. For patients who died within 90 days, ICU-free days were considered as 0. Data on ICU-free days were missing for 56 patients in the screening group and 45 in the no screening group.
Antibiotic-free days calculation was within the 90 days starting from check-in to the ward, counting days with no administration of intravenous antibiotics. For patients who died within 90 days, antibiotic-free days were considered as 0.
Secondary Outcomes
There was a reduction in vasopressor therapy (aRR, 0.86; 95% CI, 0.78-0.94; FDR = 0.004) and an increase in code blue activation (aOR, 1.24; 95% CI, 1.02-1.50; FDR = 0.04) and incident kidney replacement therapy (aOR, 1.20; 95% CI, 1.11-1.31; FDR < 0.001) comparing the screening vs the no screening group. There was no significant change in ICU admission, mechanical ventilation, or rapid response team activation. There was a reduction in the rates of new multidrug-resistant organisms and an increase in new C difficile (Table 3). There was a significant but not clinically meaningful increase in antibiotic-free days, reduction in hospital length of stay, and increase in ICU-free days (Table 3).
Sensitivity Analyses and Subgroup Analyses
Sensitivity analyses were consistent with the primary analysis (eTable 17 in Supplement 2). There was no statistically significant heterogeneity of treatment effect between the study groups across the prespecified subgroups by age, documented infection source, and COVID-19 status (Figure 2). The treatment effect was greater in oncology wards than in medical, surgical, and mixed wards. In post hoc analyses, there was no heterogeneity in the treatment effect by period (eTable 16 in Supplement 2). In the alert-only population, there was no association between screening compared with no screening and in-hospital 90-day mortality (aRR, 1.04; 95% CI, 0.93-1.16). Screening was associated with greater ICU admissions, rapid response team activation, code blue, vasopressor therapy, mechanical ventilation, incident kidney replacement therapy, and hospital length of stay. There was no statistically significant association with multidrug-resistant organisms, C difficile infection, or ICU-free or antibiotic-free days (eTable 18 in Supplement 2).
Figure 2. Results of Prespecified Subgroup Analyses of the Primary Outcome (90-Day In-Hospital Mortality).
The effect of screening compared with no screening was tested using a generalized linear mixed model accounting for periods and nested clustering within wards as random effects and hospitals and COVID-19 status as fixed effects, and the results were reported as relative risk with 95% confidence intervals. The size of each square is proportional to the subgroup sample size. Two-sided P values for interaction are reported.
aThe false discovery rate accounts for multiplicity by calculating the expected proportion of tests with false-positives at a specified rank of a set of tests.
bPatients with documented infection source included those with International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification codes for respiratory infection, urinary tract infection, skin, soft tissue infection or cardiovascular infection, intra-abdominal infection, or other infections.
cFor the subgroup based on COVID-19 status, the same model was used with the exception of not including COVID-19 status as a fixed effect.
Discussion
In this stepped-wedge, cluster randomized trial, electronic sepsis screening among hospitalized ward patients, compared with no screening, reduced 90-day in-hospital mortality.
This trial used a stepped-wedge design, one commonly used to evaluate service innovations delivered at the cluster level to avoid pitfalls observed with pre-post studies.21,26 By incorporating the research question within clinical practice, this design facilitated implementation at 43 wards at 5 hospitals. One inherent complexity of this design is confounding with time.27 As a result, the overall treatment effect could only be assessed by adjusted rather than crude analysis. As the COVID-19 pandemic started after trial launch, the baseline mortality increased in later periods corresponding with COVID-19 waves, resulting in 2-fold more COVID-19 cases and higher-risk patients in the screening group compared with the no screening group (eFigure in Supplement 2). This confounding may bias the results toward the null, underestimating the treatment effect. This confounding was accounted for by adjusting for periods and COVID-19 status. Interestingly, subgroup analysis demonstrated that screening was similarly effective in patients without or with COVID-19 and across study periods.
The current study highlights specific aspects of electronic screening with qSOFA among patients admitted to hospital wards. First, the alert identified patients who were at high risk of death. Second, among patients with alerts, physicians made an assessment of sepsis in one-third of cases, reflecting the nonspecific nature of qSOFA. Third, the reduction in mortality was observed consistently among patients with or without documented infection, suggesting that the screening effect was not limited to patients with sepsis. Fourth, the alert prompted more frequent measurements of lactate and orders of fluids in the first 12 hours. There was no significant increase in the collection of body fluid cultures and orders of antibiotics in the first 12 hours after the alert. This is likely due to antibiotic treatment before the alert and rate of sepsis in the alert population. Fifth, the study protocol mandated communication between nurses and the medical team after an alert, perhaps improving care coordination and early recognition.
Some patients randomized in the no screening group underwent screening. This could be due to a ward transitioning to the intervention or if the patient was transferred between wards. Such between-cluster contamination, which may bias the results to the null, would underestimate the treatment effect. These factors, among others,28,29 may also have contributed to less-than-complete acknowledgment of alerts by nurses and physicians in the screening group, and to the presence of some alert acknowledgments in the no screening group. To address contamination, sensitivity analyses were performed after excluding patients admitted to the no screening wards within the 90 days before the transition to an active ward, and to evaluate 14-day in-hospital mortality as an alternate outcome.
A potential unintended consequence of sepsis screening is antibiotic overuse.14 Although antibiotics were ordered more frequently in the screening group in this study, the antibiotic-free days increased, albeit minimally, and rates of new multidrug-resistant organisms were lower. The observed increased incident kidney replacement therapy is hypothesis generating, and when combined with increased C difficile rates, raises a question as to whether it might have been related to changes in the antibiotic spectrum used, given the reported association of certain antibiotics with C difficile and acute kidney injury.30 It is unclear whether the increase in code blue, which was not associated with an increase in ICU admissions or the need for mechanical ventilation, could represent an increased escalation of care or a true increase in cardiac arrest.
There was a lower overall rate of alerts in the screening group compared with the no screening group. And yet, alerted patients in the screening group had more alerts per patient, higher rates of infection documented, and antibiotic treatment administered. These differences could have many explanations. First, the documentation of vital signs to populate qSOFA may have changed during the study. If the training, audit, and feedback of the intervention led to greater precision of vital sign documentation, this could affect the behavior of the alert. Because the alert itself is a postintervention variable, analyses on alert-only patients should be considered exploratory. It remains unclear whether in this alert-only population, the absence of a treatment effect from screening is related to the case mix of patients, subsequent process of care and organ support, or even possible harm.
The study had many strengths. It was a multicenter, large, cluster randomized trial testing an electronic sepsis screening tool with communication loop. The intervention is continuous, low cost, reliable, reproducible, unbiased, and sustainable.31 Several approaches enhanced implementation, including training, dashboards, and feedback. EMR data collection was near complete. The primary analysis was conducted on the as-randomized population on which the screening system was implemented, and was consistent in many sensitivity analyses and subgroups.
Limitations
The study had several limitations. First, there was no consensus tool for sepsis screening, and many tools, including qSOFA, may be viewed as nonspecific, early warning scores. In a meta-analysis of 26 studies (62 338 patients), qSOFA had the highest prognostic accuracy for predicting mortality compared with Systemic Inflammatory Response Syndrome criteria or the National Early Warning Score.32 Second, the COVID-19 pandemic had a complex, time-varying impact on the event rate in the trial. COVID-19 status and period were adjusted for, but residual confounding could not be excluded. Third, some wards were converted to ICUs and were excluded from the analysis, reducing the sample size. Fourth, the antibiotic choices were not evaluated for appropriateness and whether they could explain the increased incident kidney replacement therapy and C difficile. Fifth, multidrug-resistant organisms and C difficile were evaluated based on infection control data, which did not differentiate between colonization and infection. Sixth, because COVID-19 was defined based on reverse transcription polymerase chain reaction testing during the hospital admission or the preceding emergency department visit, a small group of patients who were classified as non–COVID-19 might have tested positive in other health care settings.
Conclusions
Sepsis screening using an electronic alert system reduced 90-day mortality among patients admitted to hospital wards.
Section Editor: Christopher Seymour, MD, Associate Editor, JAMA (christopher.seymour@jamanetwork.org).
Trial Protocol and Statistical Analysis Plan
eMethods
eTable 1. Participating Hospitals Characteristics
eTable 2. Sepsis Electronic Alert System Workflow
eTable 3. Eligibility Criteria
eTable 4. The ICD-10-AM Codes for Comorbid Conditions
eTable 5. The ICD-10-AM Codes Used for Source of Infection
eTable 6. List of Antibiotics Based on the Hospital Formulary
eTable 7. Keywords for Multi-drug Resistant Organisms and Clostridium difficile
eTable 8. Definitions of Outcomes
eTable 9. Variations from SAP
eTable 10. Hypotheses of Subgroup Analyses
eTable 11. Additional Baseline Data of Patients in the Screening and No-Screening Groups in the Primary Analysis Population
eTable 12. Baseline Data of Patients With Alerts in the Screening and No-Screening Groups
eTable 13. Additional Data on Alert Information and Pre-Alert and Post-Alert Clinical Data, Laboratory Data and Procedures and Treatments of Patients With Alerts in the Screening and No- Screening Groups
eTable 14. Post Hoc Analysis for the Effect of Screening on Having an Alert
eTable 15. Primary Outcome Analysis Model
eTable 16. Primary Outcome, 90-Day In-Hospital Mortality, in the Screening and No-Screening Groups in the Primary Analysis Population by Period
eTable 17. Sensitivity Analyses
eTable 18. The Association of Screening With Primary and Secondary Outcomes Among Patients With Alerts in the Screening and No-Screening Groups
eFigure. Trends in the Number of Coronavirus Disease 19 (COVID-19) Cases and Mortality over the Study Periods
eReferences
Nonauthor Collaborators. SCREEN Trial Group and the Saudi Critical Care Trials Group
Data Sharing Statement
References
- 1.Dellinger RP, Levy MM, Rhodes A, et al. ; Surviving Sepsis Campaign Guidelines Committee including The Pediatric Subgroup . Surviving Sepsis Campaign: international guidelines for management of severe sepsis and septic shock, 2012. Intensive Care Med. 2013;39(2):165-228. doi: 10.1007/s00134-012-2769-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rivers E, Nguyen B, Havstad S, et al. ; Early Goal-Directed Therapy Collaborative Group . Early goal-directed therapy in the treatment of severe sepsis and septic shock. N Engl J Med. 2001;345(19):1368-1377. doi: 10.1056/NEJMoa010307 [DOI] [PubMed] [Google Scholar]
- 3.Zhang D, Micek ST, Kollef MH. Time to appropriate antibiotic therapy is an independent determinant of postinfection ICU and hospital lengths of stay in patients with sepsis. Crit Care Med. 2015;43(10):2133-2140. doi: 10.1097/CCM.0000000000001140 [DOI] [PubMed] [Google Scholar]
- 4.Evans L, Rhodes A, Alhazzani W, et al. Surviving Sepsis Campaign: international guidelines for management of sepsis and septic shock 2021. Intensive Care Med. 2021;47(11):1181-1247. doi: 10.1007/s00134-021-06506-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Arabi YM, Al-Dorzi HM, Alamry A, et al. The impact of a multifaceted intervention including sepsis electronic alert system and sepsis response team on the outcomes of patients with sepsis and septic shock. Ann Intensive Care. 2017;7(1):57. doi: 10.1186/s13613-017-0280-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Manaktala S, Claypool SR. Evaluating the impact of a computerized surveillance algorithm and decision support system on sepsis mortality. J Am Med Inform Assoc. 2017;24(1):88-95. doi: 10.1093/jamia/ocw056 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Sawyer AM, Deal EN, Labelle AJ, et al. Implementation of a real-time computerized sepsis alert in nonintensive care unit patients. Crit Care Med. 2011;39(3):469-473. doi: 10.1097/CCM.0b013e318205df85 [DOI] [PubMed] [Google Scholar]
- 8.Raines K, Sevilla Berrios RA, Guttendorf J. Sepsis education initiative targeting qSOFA screening for non-ICU patients to improve sepsis recognition and time to treatment. J Nurs Care Qual. 2019;34(4):318-324. doi: 10.1097/NCQ.0000000000000379 [DOI] [PubMed] [Google Scholar]
- 9.Joshi M, Ashrafian H, Arora S, Khan S, Cooke G, Darzi A. Digital alerting and outcomes in patients with sepsis: systematic review and meta-analysis. J Med Internet Res. 2019;21(12):e15166. doi: 10.2196/15166 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Shimabukuro DW, Barton CW, Feldman MD, Mataraso SJ, Das R. Effect of a machine learning-based severe sepsis prediction algorithm on patient survival and hospital length of stay: a randomised clinical trial. BMJ Open Respir Res. 2017;4(1):e000234. doi: 10.1136/bmjresp-2017-000234 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Tarabichi Y, Cheng A, Bar-Shain D, et al. Improving timeliness of antibiotic administration using a provider and pharmacist facing sepsis early warning system in the emergency department setting: a randomized controlled quality improvement initiative. Crit Care Med. 2022;50(3):418-427. doi: 10.1097/CCM.0000000000005267 [DOI] [PubMed] [Google Scholar]
- 12.Hooper MH, Weavind L, Wheeler AP, et al. Randomized trial of automated, electronic monitoring to facilitate early detection of sepsis in the intensive care unit. Crit Care Med. 2012;40(7):2096-2101. doi: 10.1097/CCM.0b013e318250a887 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Semler MW, Weavind L, Hooper MH, et al. An electronic tool for the evaluation and treatment of sepsis in the ICU: a randomized controlled trial. Crit Care Med. 2015;43(8):1595-1602. doi: 10.1097/CCM.0000000000001020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Hiensch R, Poeran J, Saunders-Hao P, et al. Impact of an electronic sepsis initiative on antibiotic use and health care facility-onset Clostridium difficile infection rates. Am J Infect Control. 2017;45(10):1091-1100. doi: 10.1016/j.ajic.2017.04.005 [DOI] [PubMed] [Google Scholar]
- 15.De Backer D, Deutschman CS, Hellman J, et al. ; Surviving Sepsis Campaign Research Committee . Surviving Sepsis Campaign research priorities 2023. Crit Care Med. 2024;52(2):268-296. doi: 10.1097/CCM.0000000000006135 [DOI] [PubMed] [Google Scholar]
- 16.Arabi YM, Alsaawi A, Al Zahrani M, et al. ; SCREEN Trial Group . Electronic early notification of sepsis in hospitalized ward patients: a study protocol for a stepped-wedge cluster randomized controlled trial. Trials. 2021;22(1):695. doi: 10.1186/s13063-021-05562-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Arabi YM, Al Ghamdi AA, Al-Moamary M, Al Mutrafy A, AlHazme RH, Al Knawy BA. Electronic medical record implementation in a large healthcare system from a leadership perspective. BMC Med Inform Decis Mak. 2022;22(1):66. doi: 10.1186/s12911-022-01801-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Arabi YM, Vishwakarma RK, Al-Dorzi HM, et al. ; SCREEN Trial Group . Statistical analysis plan for the Steppedwedge Cluster Randomized trial of Electronic Early Notification of sepsis in hospitalized ward patients (SCREEN). Trials. 2021;22(1):828. doi: 10.1186/s13063-021-05788-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Al-Qahtani S, Al-Dorzi HM, Tamim HM, et al. Impact of an intensivist-led multidisciplinary extended rapid response team on hospital-wide cardiopulmonary arrests and mortality. Crit Care Med. 2013;41(2):506-517. doi: 10.1097/CCM.0b013e318271440b [DOI] [PubMed] [Google Scholar]
- 20.Healthcare Information and Management Systems Society (HIMMS) . Leveraging predictive analytics and EHR data for prediction and early detection of sepsis. Accessed January 5, 2024. https://www.himss.org/sites/hde/files/media/file/2022/12/15/mngha-kamc-r-success-story-sepsis-e-alerts-final-121322-1.pdf
- 21.Hemming K, Taljaard M, McKenzie JE, et al. Reporting of stepped wedge cluster randomised trials: extension of the CONSORT 2010 statement with explanation and elaboration. BMJ. 2018;363:k1614. doi: 10.1136/bmj.k1614 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Al-Dorzi HM, Aldawood AS, Almatrood A, et al. Managing critical care during COVID-19 pandemic: the experience of an ICU of a tertiary care hospital. J Infect Public Health. 2021;14(11):1635-1641. doi: 10.1016/j.jiph.2021.09.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hussey MA, Hughes JP. Design and analysis of stepped wedge cluster randomized trials. Contemp Clin Trials. 2007;28(2):182-191. doi: 10.1016/j.cct.2006.05.007 [DOI] [PubMed] [Google Scholar]
- 24.Shakeshaft A, Doran C, Petrie D, et al. The effectiveness of community action in reducing risky alcohol consumption and harm: a cluster randomised controlled trial. PLoS Med. 2014;11(3):e1001617. doi: 10.1371/journal.pmed.1001617 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B. 1995;57(1):289-300. doi: 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
- 26.Hemming K, Haines TP, Chilton PJ, Girling AJ, Lilford RJ. The stepped wedge cluster randomised trial: rationale, design, analysis, and reporting. BMJ. 2015;350:h391. doi: 10.1136/bmj.h391 [DOI] [PubMed] [Google Scholar]
- 27.Hemming K, Taljaard M, Forbes A. Analysis of cluster randomised stepped wedge trials with repeated cross-sectional samples. Trials. 2017;18(1):101. doi: 10.1186/s13063-017-1833-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Schorr C, Odden A, Evans L, et al. Implementation of a multicenter performance improvement program for early detection and treatment of severe sepsis in general medical-surgical wards. J Hosp Med. 2016;11(suppl 1):S32-S39. doi: 10.1002/jhm.2656 [DOI] [PubMed] [Google Scholar]
- 29.Harrison AM, Gajic O, Pickering BW, Herasevich V. Development and implementation of sepsis alert systems. Clin Chest Med. 2016;37(2):219-229. doi: 10.1016/j.ccm.2016.01.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Luther MK, Timbrook TT, Caffrey AR, Dosa D, Lodise TP, LaPlante KL. Vancomycin plus piperacillin-tazobactam and acute kidney injury in adults: a systematic review and meta-analysis. Crit Care Med. 2018;46(1):12-20. doi: 10.1097/CCM.0000000000002769 [DOI] [PubMed] [Google Scholar]
- 31.Thompson DS, Oberteuffer R, Dorman T. Sepsis alert and diagnostic system: integrating clinical systems to enhance study coordinator efficiency. Comput Inform Nurs. 2003;21(1):22-26. doi: 10.1097/00024665-200301000-00009 [DOI] [PubMed] [Google Scholar]
- 32.Wang C, Xu R, Zeng Y, Zhao Y, Hu X. A comparison of qSOFA, SIRS and NEWS in predicting the accuracy of mortality in patients with suspected sepsis: a meta-analysis. PLoS One. 2022;17(4):e0266755. doi: 10.1371/journal.pone.0266755 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Trial Protocol and Statistical Analysis Plan
eMethods
eTable 1. Participating Hospitals Characteristics
eTable 2. Sepsis Electronic Alert System Workflow
eTable 3. Eligibility Criteria
eTable 4. The ICD-10-AM Codes for Comorbid Conditions
eTable 5. The ICD-10-AM Codes Used for Source of Infection
eTable 6. List of Antibiotics Based on the Hospital Formulary
eTable 7. Keywords for Multi-drug Resistant Organisms and Clostridium difficile
eTable 8. Definitions of Outcomes
eTable 9. Variations from SAP
eTable 10. Hypotheses of Subgroup Analyses
eTable 11. Additional Baseline Data of Patients in the Screening and No-Screening Groups in the Primary Analysis Population
eTable 12. Baseline Data of Patients With Alerts in the Screening and No-Screening Groups
eTable 13. Additional Data on Alert Information and Pre-Alert and Post-Alert Clinical Data, Laboratory Data and Procedures and Treatments of Patients With Alerts in the Screening and No- Screening Groups
eTable 14. Post Hoc Analysis for the Effect of Screening on Having an Alert
eTable 15. Primary Outcome Analysis Model
eTable 16. Primary Outcome, 90-Day In-Hospital Mortality, in the Screening and No-Screening Groups in the Primary Analysis Population by Period
eTable 17. Sensitivity Analyses
eTable 18. The Association of Screening With Primary and Secondary Outcomes Among Patients With Alerts in the Screening and No-Screening Groups
eFigure. Trends in the Number of Coronavirus Disease 19 (COVID-19) Cases and Mortality over the Study Periods
eReferences
Nonauthor Collaborators. SCREEN Trial Group and the Saudi Critical Care Trials Group
Data Sharing Statement
