Success Story
Detecting Sepsis Early
National Data Stream
Data from 90,000 patients could help improve infection related outcomes in Swiss ICUs
The National Data Stream “Personalized, data-driven prediction and assessment of infection related outcomes in Swiss ICUs”, a follow-up to the Personalized Swiss Sepsis Study (PSSS) co-funded by PHRT and SPHN, aims to unravel the complexities of infections in intensive care units across Switzerland. “Our primary focus is on understanding infection outcomes in the ICU, with a particular emphasis on sepsis. We are working to detect sepsis early by identifying distinct patient phenotypes and gaining insights into mortality and long-term disability”, says Catherine Jutzeler of ETH Zurich, who co-leads the project with Adrian Egli of the University Hospital Zurich. Data is being collected from all five Swiss university hospitals— Basel, Zurich, Bern, Geneva, and Lausanne. However, obtaining ethics approval has been a complex process, as the multi-center project must meet the requirements of three different Ethics Committees across Switzerland, each with its own set of considerations.
Harmonizing terminology
While research is still ongoing, a major milestone has already been reached. “One of our biggest achievements was standardizing the documentation and tracking of antibiotic administration across institutions,” she says. What might seem straightforward is actually quite challenging, as each hospital has its own practices, terminology, and protocols regarding sepsis and infectious diseases. “Our machine learning models didn’t generalize well across different settings, meaning that we couldn’t apply a model trained on data from one institution to another,” Jutzeler explains. By establishing a unified approach to tracking antibiotic administration and creating an interoperability framework with SPHN, the project is now better positioned for more robust analyses and cross-institutional comparisons.
Harmonizing the data and refining the machine learning models remain key objectives. “We’re currently using data from the Personalized Swiss Sepsis Study,” she explains. “Once we receive the complete ICU data, we’ll apply—and, if necessary, adapt—the models we’ve developed.” Given the variability in treatment protocols and data collection methods across hospitals, one approach will be to deploy a tailored model for each facility. “While a universal model may not be feasible at this stage due to differences in the data, this approach allows us to better address the unique characteristics of each institution,” Jutzeler notes.
Continuing collaboration
Data from around 90,000 patients is set to be collected by the end of 2025, emphasizing the ambitious scope of the project. While data collection will continue beyond the conclusion of this phase, the project’s costs, including data management and infrastructure, will need to be supported by other sources as PHRT funding comes to an end.
The involvement of ETH in medical research is crucial. “The clinical expertise and domain knowledge the clinical partners bring to the table ensure that our analyses go beyond just number crunching and are truly meaningful for patient care,” says Catherine Jutzeler. The collaboration between ETH’s data scientists and the clinical hospitals forms the backbone of the project’s success. Establishing this network has been a remarkable achievement and will undoubtedly lead to further collaborations in the future. “I think PHRT did a fantastic job in making this possible,” Jutzeler adds. “It is essential that this initiative continues, building on the valuable insights that have already been gained.”
Prof. Dr. Catherine Jutzeler
