Analysis of Delirium in Intensive Care Patients using Bayesian Networks

Abstract Delirium is a multifactorial and complex syndrome commonly observed in intensive care unit (ICU) patients, significantly affecting outcomes, mortality, and healthcare costs. Despite numerous potential risk factors, its exact pathophysiology remains unclear. This study explores the use of Additive Bayesian Networks (ABNs) to analyse ICU patient data and identify delirium risk factors. Three cohorts with delirium-related symptoms were examined: general delirium, delirium tremens (DT), and hepatic encephalopathy (HE). Delirium was measured by the Intensive Care Delirium Screening Checklist (ICDSC) whereas the other cohort depend on the coded diagnosis. The analysis shows important connections, such as the two-way link between antipsychotic medication and delirium and the connections of ventilation and age to delirium. Some of the results are in line with research like the link between high levels of ammonia and HE and the strong link between infection, painkillers, and pain that remains the same across all three groups. However, novel insights, such as the reduced likelihood of delirium (ICDSC ≥4) with age and ventilation, challenge conventional understanding. This study shows how ABNs can be used to find complicated dependencies in the ICU. Future work should focus on addressing potential overfitting or unmodeled interactions in the dataset and exploring specific delirium subtypes, such as hypoactive and hyperactive delirium.