Impact of Severity of Illness on Health State Transitions During Intensive Care Unit Admission: Application of Markov Multi-State Transition Modeling

Abstract

There is insufficient understanding of the relationship between the severity of physiologic illness and the transitions in cognitive and functional health states of patients during the course of an intensive care unit (ICU) stay. To this end, the fundamental aim of this project was centered on developing a predictive model for state transitions during ICU admission in order to determine the relevance of indices of illness severity and other potentially modifiable risk factors that may inform clinical decision-making. The motivating data for this work was derived from the BRAIN-ICU (Bringing to Light the Risk Factors and Incidence of Neuropsychological Dysfunction in ICU Survivors) study. Five health transition states (normal, delirium, coma, death, and discharge) were considered in the parent study, as were clinical indices of illness severity such as the Sequential Organ Failure Assessment (SOFA) and the Acute Physiology And Chronic Health Evaluation (APACHE) scores. The transition states constitute multiple end points laden with scientific information that can be elucidated by sophisticated modeling approaches now afforded by the advent of advanced statistical computing. Since the current state of a patient may be related to his/her previous states, the BRAIN-ICU data was analyzed with accommodation for multiple outcome categories (the transition states) by relating state-transition probabilities to patient covariates and past states via a polytomous regression with Markov structure. This analysis strategy addressed competing risk explicitly by assessing the effect of previous states while evaluating the motivating question of the impact of severity of illness.