Prognostic Modelling of Mortality in Chronic Critical Illness After Traumatic Brain Injury

Background: Advances in intensive care have markedly improved survival from acute critical illness. Nevertheless, the subsequent trajectory of these patients is heterogeneous: while most recover and are eventually discharged, approximately 10% remain dependent on life-support systems, forming a distinct group classified as chronic critical illness (CCI). These patients experience prolonged ICU stays, high mortality, and poor long-term outcomes. Prognostication in CCI remains challenging, as traditional severity scores based on admission data seem to lose prognostic accuracy progressively over longer ICU stays. This is particularly relevant in traumatic brain injury (TBI), where patients constitute a significant proportion of the CCI population and require specialized prognostic approaches. Objective: To develop and validate prognostic models for in-hospital mortality in patients with TBI who progress to chronic critical illness, comparing the performance of a traditional admission-based (left-aligned) model with a novel dynamic (right-aligned) model utilizing data from the week preceding the outcome. Methods: A real-world data analysis was conducted using the Russian Intensive Care Dataset (RICD v2.0). The cohort included 430 ICU admissions of adult TBI patients with a stay of ≥7 days. Multivariable logistic regression was used to develop two nomograms: one using parameters from ICU admission and another using data from 7 days prior to discharge or death. Model performance was assessed via ROC analysis, sensitivity, specificity, and predictive values. Results: The left-aligned model, based on admission data (coronary artery disease, multiorgan failure, CRP), showed moderate discriminative capacity (AUROC 0.720). In contrast, the right-aligned model, incorporating dynamic parameters from the pre-outcome period (lymphocyte count, platelet count, urea, CRP), demonstrated excellent predictive performance (AUROC 0.889), with 90.0% sensitivity and 98.6% negative predictive value. A high score on the right-aligned nomogram was associated with a 19.7-fold increased risk of mortality within the subsequent week. Conclusions: For patients with CCI following TBI, a dynamic prognostic model based on data from the immediate pre-outcome period significantly outperforms traditional admission-based models. The high negative predictive value of the right-aligned model provides a reliable tool for identifying patients with a low short-term risk of mortality, supporting a paradigm shift towards dynamic risk stratification in chronic critically ill patients.