Researchers developed a clinical decision support system that uses neural processes to forecast vital signs and determine when patients can switch from intravenous to oral antibiotics, according to peer-reviewed research published in Nature Communications.
Timely transitions from intravenous to oral antibiotics reduce catheter-related infections, shorten hospital stays, and decrease healthcare costs. However, one in five patients in England remains on intravenous therapy despite meeting switching criteria. Existing clinical decision tools frequently learn from historical medical choices, which can reproduce delays and inconsistencies seen in routine hospital care.
To bypass past clinical habits, the new model uses neural processes to predict vital sign trajectories probabilistically. The system identifies switch-readiness by checking physiological forecasts directly against clinical rules, then ranks patients to help staff prioritise medical reviews. This structure provides interpretable results, adjusts to updated medical guidelines without retraining, and preserves clinical judgement.
Validation tests covered two separate hospital records environments: 6,333 intensive care encounters in the United States and 10,584 encounters across a UK academic hospital group. Across these datasets, the forecasting system selected 2.2 to 3.2 times more relevant patients for review than random selection.
Magnus Ross, Ingemar J. Cox, and Vasileios Lampos from University College London led the study alongside co-authors from University College Hospital London and the University of Copenhagen. The research received grant funding from the EPSRC Digital Health Hub for Antimicrobial Resistance and the National Institute for Health and Care Research UCLH Biomedical Research Centre.
