Main project

Early identification of ventilator-associated pneumonia

Machine learning on data recorded by mechanical ventilators

• Healthcare

Data recorded by the ventilator without human intervention, together with clinical information on the patient, feed a machine learning model for the identification of ventilator-associated pneumonia.

Challenge

Ventilator-associated pneumonia (VAP) is one of the most frequent hospital-acquired infections in the intensive care unit. It increases morbidity, mortality, the duration of mechanical ventilation and the length of hospital stay. Current diagnostic criteria are neither sensitive nor specific enough for an early diagnosis, with a risk of both antibiotic overuse and delayed treatment.

Approach

The project differs from previous approaches in three respects:

• data are extracted directly from the mechanical ventilator during respiratory support, using a dedicated memory device;
• the models are designed for the time series of ventilator measurements;
• the reference diagnosis is established by experienced intensive care physicians with access to the complete patient information, including imaging and microbiology results.

Impact

An earlier and more accurate detection of VAP could allow timely and targeted treatment, better patient outcomes and significant savings for the healthcare system.

Publications

• Towards the identification of ventilator-associated pneumonia by ventilator-acquired data using machine learning: a prospective cohort study

  S. Mitrović, E. Bulleri, L. Ruinelli, A. Colombo, D. Malpetti, S. Toniolo, M. Scutari, G.A. Iotti, P. Merlani, L. Azzimonti, D. Novotni, A. Pagnamenta

  Manuscript, 2026 Under review