Mátyás Bukva, a research fellow at the Institute of Biochemistry of the HUN-REN Biological Research Centre, Szeged, was also honored at the event in the category of “Most Innovative Work in the Field of Healthcare Developments.” The title of the submitted project was: 'DoseLearn – Clinical protocol and ML-based predictive model for the personalized and objective planning of long-term sedato-analgesia and weaning.'

 

In paediatric intensive care, prolonged sedation and analgesia – primarily with opioids and benzodiazepines – are essential to enable mechanical ventilation and invasive procedures, but they carry substantial risks. Long-term exposure to these drugs can lead to tolerance, iatrogenic withdrawal syndrome, delirium, and a protracted tapering process. Prolonged benzodiazepine use in particular is strongly associated with withdrawal symptoms, delirium and longer intensive care stays, which makes reducing their use a key therapeutic goal. At present, the pace of dose reduction is largely based on clinical experience rather than standardized, data-driven recommendations, resulting in considerable variability in practice.

Clinically, this means that overly rapid tapering may cause instability and withdrawal symptoms, while overly cautious reduction can lead to unnecessarily long drug exposure, more side effects and extended stays in paediatric intensive care, and the expected duration of tapering remains difficult to predict for both families and clinicians.

The DoseLearn project addresses this problem in two steps. First, we implemented a protocolized, clonidine-based sedation and tapering guideline built on goal-directed sedation strategies, validated scoring systems and standardized monitoring. This led to a significant reduction in benzodiazepine use, which is particularly important because it is expected to lower the risk of withdrawal syndrome, delirium and prolonged intensive care stays.

In parallel, we established a structured, high-resolution clinical database and used these data to develop a machine learning model based on partial least squares (PLS) regression. This model allows us to estimate much more accurately how long the tapering process is likely to take and how quickly, in what step sizes, individual drug doses can be safely reduced; compared with the initial protocol-based estimates, the prediction error for tapering duration was reduced to roughly half. We are currently collaborating with the Department of Software Development at the University of Szeged to develop a clinically robust decision-support application that integrates this model into everyday bedside practice in paediatric intensive care.

 

Congratulations!