Machine Learning Prediction of Parameters of Early Warning Scores in General Wards

In the event of illness or injury, patients are medically evaluated and initially treated in acute medical outpatient clinics, emergency rooms and surgeries. If medically indicated, care and treatment can also be provided in hospital. Depending on the severity of the illness and the main medical problem, this care is provided on hospital wards, which are primarily looked after by specific specialist disciplines and assigned to them in the form of clinical departments, for example.

As part of the inpatient stay, treatment and care is usually provided through ward rounds by the medical staff. However, ward rounds are spot checks of individual measured values at predefined times.

Qualified nursing staff carry out the agreed treatment plans and check the patient's general condition several times a day. In contrast to intensive medical monitoring, however, there is no continuous monitoring and therefore an aggravation of a patient's condition is not always immediately apparent. Furthermore, in addition to known complications of existing conditions, new or unexpected complications can also occur.

Although non-intensive care monitoring is based on discontinuous monitoring, incidents and complications can sometimes be life-threatening, especially if there is no immediate response to a deterioration in the patient's condition. Even if there are early warning systems such as scores, their ability to react is limited, partly due to the frequency with which they are collected.

In addition to patient-specific limitations of inpatient monitoring, such as patient cooperation in the sense of self-monitoring, medical limitations, such as the frequency of the survey, there are also economic limitations, such as the availability of staff who can be deployed for more frequent monitoring.

Although there are telemedical approaches to monitoring, setting these up is often limited both economically and by the additional training required, for example.

Even if threshold values are (or can be) defined for the measured data (vital signs, laboratory parameters, clinical impression and others), if these are exceeded or not reached, a consequence, e.g. a therapy step, can only be initiated retrospectively. In this situation, a pathophysiological change is already so far advanced that in many cases a compensation mechanism no longer functions adequately and turns into a decompensation situation. In this situation, the affected patients in a hospital ward are potentially in mortal danger.

One way of averting the dangers described above could be to use a reduced combination of monitoring methods compared to intensive care monitoring. At the same time, the use of artificial intelligence enables the automated evaluation of the collected data and can thus lead to the prediction of changes in parameters, which enables early alerting, i.e. before the occurrence of pathophysiological decompensation.

Study Overview

• Status: Active, not recruiting
• Conditions: Patient Safety
• Intervention / Treatment: Other: Parameters of Early Warning Scores

• Study Type: Observational
• Enrollment (Estimated): 3000

Contacts and Locations

Study Locations

• Austria
  • Upper Austria
    • Linz, Upper Austria, Austria, 4020
      • Johannes Kepler University, Kepler University Hospital

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

• Sampling Method: Non-Probability Sample

Study Population

Patients treated in general wards.

Description

Inclusion Criteria:

• Treated in general ward between 2024-10-01 and 2026-10-31 at the study center.

Exclusion Criteria:

• None.

Study Plan

How is the study designed?

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Area Under the Curve of the Receiver Operating Characteristic (AUC-ROC) for Prediction of Parameters of Early Warning Scores	Area Under the Curve of the Receiver Operating Characteristic (AUC-ROC) for Prediction of Parameters of Early Warning Scores	2024-10-01 to 2026-10-31
Area Under the Curve of the Precision-Recall Curve (AUC-PRC) for Prediction of Parameters of Early Warning Scores	Area Under the Curve of the Precision-Recall Curve (AUC-PRC) for Prediction of Parameters of Early Warning Scores	2024-10-01 to 2026-10-31
F-Beta Score with Beta = 1 (F1-Score) for Prediction of Parameters of Early Warning Scores	F-Beta Score with Beta = 1 (F1-Score) for Prediction of Parameters of Early Warning Scores	2024-10-01 to 2026-10-31
Confusion Matrix for Prediction of Parameters of Early Warning Scores	Confusion Matrix for Prediction of Parameters of Early Warning Scores	2024-10-01 to 2026-10-31

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
SHapley's Additive exPlanations (SHAP) Values for Prediction Models	SHapley's Additive exPlanations (SHAP) Values for Prediction Models	2024-10-01 to 2026-10-31
Prediction of Routine Laboratory Values	Routine laboratory values measurements collected in routine care; this comprises organ function measures such as liver function tests, and in turn more specifically aspartate-aminotransferase (ASAT), alanine transaminase (ALAT), cholinesterase (CHE). Predictions are made on their future values in the respective units they are measured in.	2024-10-01 to 2026-10-31
Prediction of Parameters Measured by Photophlethysmogram (PPG)	Future values of the photophlethysmogram (PPG) are predicted; this comprises heart rate, respiration rate, peripheral oxygen saturation Predictions are made on their future values in the respective units they are measured in.	2024-10-01 to 2026-10-31
Prediction of Medical Emergency Team or Emergency Critical Care Treatment	In case of deterioration, patients are examined by a medical emergency team (MET) at the ward they are admitted to, if necessary stabilized there or referred to a specialized emergency critical care (ECC) area for further examination and treatment. Both, MET and ECC activation are made on locally established protocols and defined threshold criteria such as the patient at risk-score. Predictions are made as categorical variable, either MET and/or ECC activation happen, or not (yes/no).	2024-10-01 to 2026-10-31
Prediction of Unplanned Intensive Care Unit (ICU) Admission	Intensive care unit (ICU) admission may happen in direct referral from a general ward, via MET referral or after ECC treatment. In either case, this variable will be predicted as a category, i. e. yes/no on ICU admission. An admission in an emergency due to deterioration is always considered unplanned.	2024-10-01 to 2026-10-31
Prediction of Electrocardiogram (ECG) Waveform	Future values of the electrocardiogram (ECG) waveform (millivolts at specific point in time) are predicted.	2024-10-01 to 2026-10-31

Collaborators and Investigators

• Sponsor: Kepler University Hospital
• Collaborators: RISC Software GmbH; innovethic eU; FiveSquare GmbH
• Investigators: Principal Investigator: Thomas Tschoellitsch, MD, Johannes Kepler University

Study record dates

Study Major Dates

• Study Start (Actual): August 15, 2025
• Primary Completion (Estimated): October 31, 2026
• Study Completion (Estimated): October 31, 2026

Study Registration Dates

• First Submitted: August 14, 2024
• First Submitted That Met QC Criteria: August 26, 2024
• First Posted (Actual): August 28, 2024

Study Record Updates

• Last Update Posted (Estimated): September 3, 2025
• Last Update Submitted That Met QC Criteria: August 26, 2025
• Last Verified: August 1, 2025

More Information

Terms related to this study

• Other Study ID Numbers: AIM-PEW-WAR

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No