Trial Review

Registering a new trial?

To achieve prospective registration, we recommend submitting your trial for registration at the same time as ethics submission.

The safety and scientific validity of this study is the responsibility of the study sponsor and investigators. Listing a study does not mean it has been endorsed by the ANZCTR. Before participating in a study, talk to your health care provider and refer to this information for consumers
Trial registered on ANZCTR


Registration number
ACTRN12622000197730
Ethics application status
Approved
Date submitted
10/12/2021
Date registered
4/02/2022
Date last updated
4/02/2022
Date data sharing statement initially provided
4/02/2022
Type of registration
Prospectively registered

Titles & IDs
Public title
Detecting serious infections early in the Emergency Department using data analytics
Scientific title
Early identification of sepsis in the Emergency Department using data analytics
Secondary ID [1] 306016 0
Nil known
Universal Trial Number (UTN)
Trial acronym
Linked study record

Health condition
Health condition(s) or problem(s) studied:
Sepsis 324651 0
Condition category
Condition code
Emergency medicine 322102 322102 0 0
Other emergency care
Infection 322103 322103 0 0
Studies of infection and infectious agents

Intervention/exposure
Study type
Observational
Patient registry
False
Target follow-up duration
Target follow-up type
Description of intervention(s) / exposure
Patients with presenting to Emergency Department from 1 July 2016 - 30 June 2021.
Information will be obtained from administrative datasets : Emergency Department Information System (EDIS) dataset and the Hospital Admission and Morbidity Datasets will be both used.
The triage free text information will be searched for key words to identify likelihood of admission to hospital with sepsis.
Variables being observed in patients are : description of presenting complaint, Emergency Department diagnosis, hospital admission diagnosis.
Intervention code [1] 322419 0
Early Detection / Screening
Comparator / control treatment
Patients without sepsis presenting to Emergency Department during the time period 1 July 2016 - 30 June 2021. Data source will be the Emergency Department Information System (EDIS)
Control group
Active

Outcomes
Primary outcome [1] 329861 0
The proportion of patient with an admission diagnosis of sepsis (Hospital admission diagnosis coding at time of hospital discharge), compared to emergency department presentation diagnosis.
Timepoint [1] 329861 0
At time of hospital discharge, an admission diagnosis of sepsis
Secondary outcome [1] 404093 0
Admission to intensive care using hospital admission dataset.
Timepoint [1] 404093 0
During index hospital admission
Secondary outcome [2] 404094 0
Diagnosis of sepsis at time of emergency department presentation using Emergency Department Information System.
Timepoint [2] 404094 0
At emergency department presentation
Secondary outcome [3] 404095 0
Rapid response team activation during hospital admission, using linkage to medical recrods.
Timepoint [3] 404095 0
During hospital admission

Eligibility
Key inclusion criteria
All patients who present to Sir Charles Gairdner Hospital Emergency Department from July 2016- June 2021 will be included for review
Minimum age
16 Years
Maximum age
No limit
Sex
Both males and females
Can healthy volunteers participate?
No
Key exclusion criteria
None

Study design
Purpose
Screening
Duration
Cross-sectional
Selection
Convenience sample
Timing
Retrospective
Statistical methods / analysis
The Emergency Department Information System (EDIS) dataset and the Hospital Admission and Morbidity Datasets will be both used.
Patients presenting to the Emergency Department from July 2016-2021 will be included.

The free text field from the EDIS dataset will be interrogated to determine if there are word/terminology that increases the likelihood of the subsequent hospital admission diagnosis being related to sepsis.
Sepsis definitions will be coded according to ICD-10AM definitions from Australian Commission on Safety and Quality in Health Care definitions of sepsis and infection (https://www.safetyandquality.gov.au/sites/default/files/2020-05/epidemiology_of_sepsis_-_february_2020_002.pdf)

Patients will be randomly allocated to fixed train (60%) , validate (20%) and test (20%) data sets. Building will be based on the paper by Horng et al (DOI: 10.1371/journal.pone.0174708).

Primary models will be constructed using machine learning and a linear support vector machine (SVM), to optimize the area under the ROC curve. The open source SVMperf software package will allow us to use a learning algorithm automatically controls for class imbalance by directly optimizing a lower bound on the AUC, rather than focusing on classification accuracy. For comparison purposes, we will create models using L2-regularized logistic regression, naïve Bayes, and random forests, using the open-source Scikit-Learn software. For all learning algorithms, model derivation will be first performed on the train data set. The validate data set will be used to optimize over model parameters. The test data set, a holdout sample, will be then used to test the internal generalizability of the model with the highest AUC on the validate data set. When we report train and validate results, we will also report them for the model with the highest AUC on the validate data set.

Data analysis:
Means with 95% confidence intervals will be reported for age. For a subset of patient admitted to ICU from Emergency Department, we will take the ANZICS physiological data: , temperature, heart rate, systolic blood pressure, and diastolic blood pressure, APACHE, lactate. . Medians with interquartile ranges will be reported for hospital and ICU admission days, and ICU days. Significance testing will performed using T-tests for parametric data, Wilcoxon rank sum for non-parametric data, and Fisher’s Exact test for proportions.
The area under the ROC curve (AUC) will be calculated for each of the four models to measure discriminatory power. We will report positive predictive value (PPV), sensitivity, and specificity at the optimal cutoff point that balances the tradeoff between sensitivity and specificity. This optimal cutoff point is defined as the threshold which maximizes Youden’s J statistic (Sensitivity + Specificity—1).

Recruitment
Recruitment status
Not yet recruiting
Date of first participant enrolment
Anticipated
7/02/2022
Actual
Date of last participant enrolment
Anticipated
Actual
Date of last data collection
Anticipated
Actual
Sample size
Target
5000
Accrual to date
Final
Recruitment in Australia
Recruitment state(s)
WA
Recruitment hospital [1] 21426 0
Sir Charles Gairdner Hospital - Nedlands
Recruitment postcode(s) [1] 36327 0
6009 - Nedlands

Funding & Sponsors
Funding source category [1] 310358 0
Hospital
Name [1] 310358 0
Sir Charles Gairdner Hospital
Country [1] 310358 0
Australia
Primary sponsor type
Hospital
Name
Sir Charles Gairdner Hospital
Address
Hospital Avenue
Nedlands, WA, 6009
Country
Australia
Secondary sponsor category [1] 311495 0
None
Name [1] 311495 0
Address [1] 311495 0
Country [1] 311495 0

Ethics approval
Ethics application status
Approved
Ethics committee name [1] 310012 0
Sir Charles Gairdner and Osborne Park Hospital Ethics Committee
Ethics committee address [1] 310012 0
2nd Floor
A Block
Hospital Avenue, Nedlands
WA 6009
Ethics committee country [1] 310012 0
Australia
Date submitted for ethics approval [1] 310012 0
Approval date [1] 310012 0
29/10/2021
Ethics approval number [1] 310012 0
RGS0000005033

Summary
Brief summary
Triage in the Emergency Department (ED) is an opportunity for a time critical point of identification of evolving severe sepsis. Current identification rates at triage are reported in the realm of 50-60% in the literature. Tromp et al undertook an education programme to improve identification of sepsis at triage, and their detection rates were 65%.

Techniques to identify sepsis earlier may reduce the time to administration of antibiotics, source control and other resuscitative measures to improve patient outcomes. However, often the triage nurse is working under time pressures, and has limited information available to them to assist with their decision making.

Hypothesis: Data analytic techniques may reveal early prompts to identify and place patients on sepsis pathways. Combining the demographic data and triage free text information inputted by the triage nurse could create prompts for the triage nurse to consider “is it sepsis?" earlier.

AIMS: To create a predictive likelihood of sepsis from key words in inserted text. In future, develop as decision aid within EDIS (Emergency Department Information System) or separate triage tool.

Methods: We will combine two large datasets - an Emergency Department dataset as well as the hospital admission dataset to create a way of determining whether information at the point of Emergency Department triage, may help predict the likelihood of subsequent diagnosis of a serious infection in a patient.

"Big data" analytics tools will be used, and the datasets will be split into train, validate and tes components.
Trial website
Trial related presentations / publications
Public notes

Contacts
Principal investigator
Name 116182 0
A/Prof Matthew Anstey
Address 116182 0
Intensive Care Department
Sir Charles Gairdner Hospital
Hospital Avenue, Nedlands 6009, Western Australia
Country 116182 0
Australia
Phone 116182 0
+61 864571010
Fax 116182 0
Email 116182 0
[email protected]
Contact person for public queries
Name 116183 0
Matthew Anstey
Address 116183 0
Intensive Care Department
Sir Charles Gairdner Hospital
Hospital Avenue, Nedlands 6009, Western Australia
Country 116183 0
Australia
Phone 116183 0
+61 864571010
Fax 116183 0
Email 116183 0
[email protected]
Contact person for scientific queries
Name 116184 0
Matthew Anstey
Address 116184 0
Intensive Care Department
Sir Charles Gairdner Hospital
Hospital Avenue, Nedlands 6009, Western Australia
Country 116184 0
Australia
Phone 116184 0
+61 864571010
Fax 116184 0
Email 116184 0
[email protected]

Data sharing statement
Will individual participant data (IPD) for this trial be available (including data dictionaries)?
No
No/undecided IPD sharing reason/comment
Data availability will be at the discretion of the approving HREC and the coordinating PI, and would need to be provided in a de-identified manner. Current approvals would not allow for data sharing.


What supporting documents are/will be available?

No Supporting Document Provided



Results publications and other study-related documents

Documents added manually
No documents have been uploaded by study researchers.

Documents added automatically
No additional documents have been identified.