Pressure injury surveillance in the intensive care unit: Development, validation, and clinical application of a...

Pressure injuries (PIs) are one of the most common hospital-acquired complications in patients admitted to an intensive care unit (ICU). Existing PI surveillance systems are known to have multiple limitations, including poor data quality. Previous studies investigating PI acquisition in critically ill populations have had relatively small sample sizes and have not considered when PIs occur during the ICU admission.

The objective of this study was to develop and validate an algorithm capable of performing PI surveillance in the adult ICU population and to use this algorithm to describe the number of patients with PIs identified in the cohort and clinical characteristics associated with PI development.

A multicentre, retrospective cohort study was conducted across five Australian ICUs from June 2017 to June 2023. Natural language processing techniques were used to develop the surveillance algorithm, which detected PIs documented in the progress notes. The surveillance algorithm was externally validated using a combination of clinical codes and manual chart review. Data from the algorithm were then linked to the Australian and New Zealand Intensive Care Society Adult Patient Database to obtain the clinical characteristics for the cohort.

Data from 40,033 patients were included in the study, including over 120 million free-text fields. The surveillance algorithm demonstrated satisfactory overall performance (F1 score: 0.743-0.749) on external validation. PIs were identified in 8.35% (n = 3344) of the cohort, with 63.19% (n = 2113) of these acquired during the ICU admission. PIs acquired in the ICU tended to occur early in the ICU admission, with 70.3% (n = 1486) identified within the first 5 days. Patients identified as having a PI had a higher severity of illness (median Acute Physiology and Chronic Health Evaluation III score: 67 vs 49) and were more likely to have received invasive ventilation (57.1% vs 32.5%).

Natural language processing can be used to perform PI surveillance in the ICU and facilitate the conduct of observational research. Future studies should consider integrating the finding that ICU-acquired PIs occur early in the ICU stay into quality improvement programs aimed at reducing PI incidence.