Patients discharged from intensive care units (ICUs) to Normal Care Units (NCUs) are at high risk to develop complications, but face low, infrequent monitoring of vital parameters (VPs) and lower staff ratios. Accordingly, complications remain unobserved at their onset, and are more developed and more difficult to treat once diagnosed. For example, post-pancreatic surgery patients are at significant risk of developing anastomotic leakage, commonly occurring after ICU discharge, leading to ICU readmission, higher mortality, and increased hospital costs. Early detection of such complications is therefore critical.
Continuous monitoring of VPs can enhance the early detection of patient deterioration, thereby reducing intervention delays and the risk of failure to rescue. Existing systems pose challenges, however: (1) Alarm fatigue among medical staff develops as collected VP data is merely compared to top-down set thresholds, with alarms frequently triggered without the need for intervention; (2) staff workload and logistic complexity increase because one device monitors only one to three VPs, requiring multiple devices per patient. In addition, in the literature, prediction of adverse events has largely relied on retrospective data, with a gap for prospective studies using high-frequency VP data to warn physicians of specific complication development.
Our objective is to develop a software for early warning of post-ICU complication onset to support clinical decision making. This software is to use high-frequency VP data collected in NCUs using Rheo’s vital+ wearable, which measures all five VPs simultaneously and therefore is convenient for patients, nurses, and clinicians. Using these data, we will develop baseline machine learning models predicting complications (e.g., employing XGBoost, Random Forest). We will then test our baseline models prospectively with continuous model updates and validation. Ultimately, we will implement the best-performing model into software for clinical use, solving issues such as how to balance computational effort and continuous prediction updates, or how to visualize prediction model output best for nurses and physicians.
The OptiMaL project, funded by the HSG Health Forward program, prepares the above-described software development project with three preparatory studies: (1) clinical validation of the vital+ device, (2) a scoping review of existing wearables, their applications and their effectiveness, and (3) a pilot study to demonstrate the feasibility of an early warning system with a predictive machine learning model.
Dr. Justus Vogel, Andrés Gallardo, Enqi Fu
HOCH - Kantonsspital St. Gallen: Prof. Dr. med. Miodrag Filipovic, Prof. Dr. med. Urs Pietsch, Dr. Franz Boudriot, Milena Traber, Mathilda Osterwalder
Rheo: Philipp Vetter
HSG Health Forward Program
12 months (01.03.2026 – 28.02.2027)