Machine Learning, News-2, Sepsis
Conditions
Keywords
Sepsis, Clinical Decision Support System, Machine Learning, NEWS2, Early Warning, lead time
Brief summary
This prospective observational study aims to objectively measure the lead-time (the time from the first KDS alert to sepsis diagnosis) of the NEWS2-based clinical decision support system (KDS) and compare its early warning performance with a machine learning model trained on 2000 patients and externally validated. The study seeks to answer the following main questions: How early does the NEWS2-based KDS provide an alert before sepsis diagnosis? Does a machine learning model, developed using logistic regression and externally validated in a prospective cohort, offer superior specificity and comparable sensitivity to KDS? Participants who are already receiving routine clinical care at Kocaeli City Hospital will have their vital signs and laboratory data monitored as part of standard practice. NEWS2 scores will be calculated automatically and the time of the first alert (T0) will be recorded. Sepsis diagnosis will be confirmed by an increase in SOFA score ≥ 2 (T1), evaluated by two independent and blinded physicians. Lead-time will be calculated as the difference between T1 (hours×60) and T0 (minutes). The machine learning model will be tested prospectively on this cohort, and its performance will be compared with KDS using sensitivity, specificity, F1 score, ROC-AUC, and accuracy.
Detailed description
Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection. Early recognition and treatment are critical for improving outcomes. The National Early Warning Score 2 (NEWS2) is widely used as an early warning system, but its lead-time (the time from alert to diagnosis) has not been objectively measured in prospective studies. This study aims to fill this gap by prospectively evaluating the lead-time of NEWS2-based KDS and comparing its performance with a machine learning model. The machine learning model was developed using 2000 patients from the PhysioNet Sepsis Prediction Challenge 2019 database and externally validated on a prospective cohort of 100 patients from Kocaeli City Hospital. The study will provide evidence on the comparative utility of traditional warning systems and machine learning approaches for early sepsis detection.
Interventions
Observational Study - No Intervention
Sponsors
Kocaeli Derince Education and Research Hospital
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Patients aged 18 years and older * Admitted to Kocaeli City Hospital Anesthesiology and Reanimation Clinic * Suspected infection at the time of hospital admission * Complete vital signs recorded * Informed consent obtained from the patient or legal representative
Exclusion criteria
* Patients under 18 years of age * Pregnant patients * Patients with chronic kidney disease requiring dialysis * Patients with a history of organ transplantation * Patients with incomplete data
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Lead-Time Calculated from NEWS2 KDS Alert and SOFA Score Change | From hospital admission to sepsis diagnosis, death, or discharge, whichever occurs first, assessed up to 14 days | Lead-time is defined as the time from the first KDS alert (NEWS2 ≥ 5) to the diagnosis of sepsis, confirmed by an increase in SOFA score ≥ 2 points. It is calculated using the formula: Lead-Time (minutes) = \[T1 (hours × 60)\] - T0 (minutes), where T0 is the time of the first NEWS2 measurement and T1 is the time of first SOFA increase ≥ 2. |
Secondary
| Measure | Time frame |
|---|---|
| Predictive Performance of Machine Learning Model vs KDS | Within 14 days of hospital admission |
Countries
Turkey (Türkiye)
Outcome results
None listed