The prevention of delirium in ICU represents a major unmet need because of its high prevalence, under-diagnosis, and independent association with adverse outcomes including higher mortality and accelerated cognitive decline.

Moreover, the absence of diagnostic biological parameters and the fluctuating characteristics of delirium create a condition that lacks a diagnostic gold standard.

In this setting, AI, large language models, ML and NLP all offer a real and novel opportunity to develop new operational standards for delirium prediction and diagnosis and open the door to better patient selection for interventional trials.

In a cohort of 12,609 ICU patients, Young et al. used NLP to study 200,000 thousand progress notes and 69 million words.

The notes were searched for words indicating the presence of disturbed behaviour and possible delirium.

Read More