Pressure injury prediction models for critically-ill patients should consider both the case-mix and local factors
Abstrakt
Pressure injuries in intensive care unit (ICU) patients are associated with unfavourable outcomes Models for predicting ICU-acquired pressure injury have been developed using machine learning and classic regression techniques, respectively resulting in areas under the receiver operating characteristic curve (AUC) of 0.79 and 0.89 , but also resulting in large differences among predictors identified. The latter may, at least partially, be associated with the fact that these models were not built for distinct subgroups within the highly-heterogeneous ICU population.
Alderden and colleagues (2018) used surgical and cardiovascular-surgical ICU patient data; Ladios-Martin et al. (2020) used data from a mixed ICU population. Such generalised approaches often originate from the researchers' lack of sufficient data for distinct ICU subpopulations.