Machine learning improves risk prediction and prevention for pressure ulcers in the intensive care unit

Photo credit: The best photo for everyone

Below is a summary of “Explainable Artificial Intelligence for Early Prediction of Pressure Ulcer Risk,” published in the September 2024 issue of Intensive care by Alderden et al.

Individuals with hospital-acquired pressure injuries (HAPIs) impact outcomes in intensive care units (ICUs). Traditional risk assessment tools have limitations, while artificial intelligence models offer greater accuracy but lack transparency.

Researchers conducted a retrospective study to develop an artificial intelligence-based HAPI risk assessment model with an explainable artificial intelligence dashboard to improve patient interpretability both globally and independently.

They used an explainable artificial intelligence approach to examine intensive care patient data from the Medical Information Mart for Intensive Care. Predictor variables were limited to the first 48 hours after ICU admission. Multiple machine learning algorithms were tested, resulting in an ensemble “super-learner” model whose performance was calculated using the area under the receiver operating characteristic curve through five-fold cross-validation. An explainer dashboard was developed (using synthetic data to maintain patient privacy) with interactive visualizations for detailed model interpretation at global and local scales.

The results showed that the final sample included 28,395 patients with a HAPI incidence of 4.9%. The ensemble super learner model performed well (area under the curve = 0.80). The explainer dashboard provided interactive visualizations of the model predictions at global and patient levels and showed the impact of each variable on the risk assessment outcome.

The researchers concluded that the model and its dashboard provide clinicians with a transparent, interpretable, artificial intelligence-based risk assessment system for HAPIs that can enable more effective and timely preventative interventions.

Source: aacnjournals.org/ajcconline/article/33/5/373/32528/Explainable-Artificial-Intelligence-for-Early