Table 1.

Problems of integrating prognostic modelling into decision-making processes in the ICU

Problem	Solutions
Generalisability of predictive models	Add variables describing specific legal, cultural and professional norms as well as available resources, which influence care processes, to prediction models
Prediction of functional outcome and quality of life in very old patients	Suitable training samples with respect to chronic conditions (e.g. frailty, multimorbidity) and impairments (e.g., assisted living) typical for very old age
Interpretability of model predictions	Incorporate explainability metrics
Alignment of data-informed decisions with patient's preferences	Document patient’s goals and expectations, share decision-making
Uncertainty of outcome predictions for individual patients	Time-limited trials with sequentially updated predictions, shared decision-making integrating the burden of treatment