When every hour counts: Early sepsis detection possible with a small blood count - thanks to new AI-based methods

Launched in 2018 as an analysis and reporting system to improve patient safety through real-time integration of laboratory findings (AMPEL), AMPEL has since evolved and is now considered a digital infrastructure that enables clinical AI applications in routine care.

Since its launch over five years ago, the award-winning project at Leipzig University Hospital has been supporting nursing and medical staff in patient care by recognising critical situations in real time and thus significantly increasing patient safety. Automated alerts improve the availability and weighting of medical information. The AMPEL utilises simple calculations through to complex AI models and monitors all the necessary data live.

If septicaemia (blood poisoning) is suspected, the situation can become critical and every hour counts. The survival of this often fatal disease depends to a large extent on the earliest possible administration of antibiotics. The UKL's AMPEL project has now reached a milestone in the early detection of sepsis: using new machine learning methods, the team was able to develop an AI model and scientifically validate it at two other locations, in Germany and the USA.

"Our study on the prediction of sepsis on the basis of the complete blood count was accepted by the world's leading journal for laboratory medicine 'Clinical Chemistry'. We are demonstrating the potential of using AI methods and very few laboratory parameters that have already been collected in routine care to implement continuous screening for patients with the onset of sepsis," explains Dr Daniel Steinbach, physician and research associate at the Institute of Laboratory Medicine, Clinical Chemistry and Molecular Diagnostics and the Data Integration Centre at the UKL and member of the AMPEL core team. "And the fact that our AI model significantly outperforms the prediction of the established marker procalcitonin (PCT) at no additional cost should arouse great interest."

Laboratory values of the complete blood count always available, but rarely used

The fact that the data from a CBC could help to recognise sepsis at an early stage initially came as something of a surprise, even to the interdisciplinary AMPEL team. "The initial results showed that the AI models developed often rely on the data from the complete blood count. These are laboratory values that are always there, but are rarely taken into account," says Dr Steinbach. According to the UKL expert, these laboratory findings hardly play a role in the clinical care of sepsis detection, although they are determined in every hospital during almost every laboratory test. Specific laboratory parameters such as procalcitonin are more likely to be used.

Procalcitonin is a contradiction in terms, says Maria Schmidt, who is also part of the AMPEL core team: "Although it is used almost everywhere, studies regularly come to the conclusion that its predictive power is too low. What remains is a recommendation for guiding the right antibiotic therapy during the course of the disease." In their sepsis study, they have now been able to show that the predictive power of procalcitonin can be significantly improved using machine learning methods and in combination with the complete blood count, explains biometrician Schmidt.

The theory is now to be followed by practical application: "In laboratory medicine, I know of no AI model that has been tested as thoroughly and extensively as the sepsis model we have published. In the end, it's still just theory and only practical application will show whether and how much support it really provides. Fortunately, this is precisely one of AMPEL's core competencies," says Dr Steinbach.

Further development as an open source project

Martin Federbusch, a specialist in laboratory medicine, heads the AMPEL project and regrets that the UKL is still the only hospital in Germany with the kind of digital infrastructure that AMPEL has created. "So one important goal remains for us: the transfer to other locations," emphasises Federbusch. "Thanks to the broad support at our site, we are in a position to further develop the AMPEL AI platform independently."

Since this year, Prof Toralf Kirsten and his Department of Medical Data Science have been developing AMPEL as an open source project under the leadership of Leipzig University Hospital. The focus is on creating a non-profit AI infrastructure for patient care that fulfils the highest standards of adaptability, interoperability and transparency.

Link to the article in "Clinical Chemistry": https://doi.org/10.1093/clinchem/hvae001

Press release of the "University Hospital Leipzig" from 05 March 2024