Main objective of the COMPANION project: Modeling the consequences of HOBSI using AI-supported causal models, simulation of target studies and data from real care practice.

The project focuses on the development of predictive models that are tailored to the German healthcare system. The aim is to improve treatment strategies and strengthen the system’s resilience to nosocomial infections and multidrug-resistant pathogens (MDROS).

1. data-based forecasting: KISS surveillance data is used to create time series forecasts to predict future trends in nosocomial infections and MDROS. This enables hospitals to identify infection peaks at an early stage, manage resource allocation (e.g. intensive care beds, staff) and prepare for potential outbreaks. The models take into account spatio-temporal and seasonal patterns, such as the increase in respiratory infections in the winter months, and enable proactive hospital management.

2. evaluation of therapeutic interventions: Target trial emulation simulates randomized trials using observational data to estimate robust causal relationships of therapeutic interventions. These models enable healthcare providers to predict infection trends and optimize resources, thereby strengthening resilience to nosocomial infections.

3. mathematical modeling of hypothetical intervention scenarios: Mathematical models will be developed to simulate the hypothetical impact of different therapeutic interventions on nosocomial bloodstream infections (HOBSI) and multidrug-resistant pathogens (MDRS). This approach is based on estimated parameters from previous subprojects and is used for decision making by assessing the cost-effectiveness of therapeutic interventions, such as new antibiotics and decolonization therapies.

Topics:

• Population and spatiotemporal modeling of KISS monitoring data
• Mathematical modeling of hypothetical intervention scenarios
• Development of causal strategies for machine learning
• Application of causal modeling in proof-of-concept cohort studies
• Simulation of target studies with routinely collected NUM-DIC data