GLM's in Data Science as a Tool in the Prediction of Delirium

Delirium is an acute neuropsychiatric dysfunction, prevalent in patients admitted to hospitals for inpatient and intensive care, being clinically characterized by attention deficit and clouding of the state of consciousness associated with cognitive disorders. Since this is a multifactorial manifestation that develops over a short period of time, it is usually underdiagnosed and neglected. Consequently, this disorder appears associated with high rates of mortality and morbidity, leading to a longer period of hospitalization. Additionally, delirium can be categorized, according to the motor activity profile, into two subtypes: hypo and hyperactive. Currently, there are studies and assessment instruments for the study and prediction of the disease at an earlier stage, based on a set of risk factors. However, in the context of a hospital emergency, time is essential, as well as a correct and early diagnosis to intervene as quickly as possible. In this context, the goal of this paper arises, which aims to implement a diversity of techniques to preprocess the data to perform the multinomial logistic regression. The ultimate goal is to identify the most effective data balancing technique for accurate prediction of delirium occurrence and its subtypes, based on the methodology of generalized linear models (GLMs), specifically multinomial logistic regression (MLR).