Background: Intensive care unit (ICU) delirium is one of the most common clinical syndromes that results in many adverse events that affect patients, families, and hospitals. To date, there has been no tool for effectively predicting the occurrence of delirium in emergency intensive care unit (EICU) patients.
Methods: We conducted a retrospective cohort study and constructed a prediction model for 319 patients in EICU, who met our inclusion criteria. We analyzed the relationship between patients’ clinical data within 24 hours of admission and delirium, applied univariate and multivariate logistic regression analyses to select the most relevant variables for construction of nomogram models, then applied bootstrapping for internal validation.
Results: A total of five variables, namely stomach and urinary tubes, as well as sedative, mechanical ventilation and APACHE-II scores, were selected for model construction. We generated a total of five sets of models (three sets of construction models and two sets of internal verification models), with similar predictive value. The optimal model was selected, and together with the 5 variables used to construct a nomogram. The AUC of the MFP model in all patients was 0.76 (0.70, 0.82), whereas that in non-elderly patients (< 60 years old) for the full model was 0.83 (0.74, 0.91). In elderly patients (≥ 60 years old), the AUC of the MFP model was 0.82 (0.73, 0.91).
Conclusion: Overall, the five-marker-based prognostic tool, established herein, can effectively predict the occurrence of delirium in EICU patients.
Keywords: area under curve, delirium, emergency intensive care unit, model, prediction