Background: Delirium is an acute disorder of attention and cognition that is common, serious, costly, under-recognized, and often fatal in the elderly. Approximately 80% of delirium cases are missed in the ED by clinical gestalt alone. We conducted an observational study from a hospital-wide database to externally validate three delirium prediction models: Delirium Risk Score, Risk Prediction Rule, and Susceptibility Score.
Methods: This was an observational cohort study to evaluate the diagnostic accuracy of three previously developed delirium prediction models compared to a gold standard for diagnosis of delirium. We included patients aged 65 years and older who presented to our ED from 2014 to 2017 and were hospitalized. All patients were evaluated by bedside nurses using the Delirium Observation Screening Scale (DOS) twice daily while hospitalized as part of standard institutional protocol. We extracted variables to examine the three prediction models within seven days of ED arrival. We defined a positive DOS as the gold standard and also examined ICD9/10 diagnoses to test its robustness. The predictive ability of the feature(s) in a model was summarized using the area under a receiver operating characteristic curve, or AUC.
Results: We identified 4,745 visits with a positive DOS and 1,316 patients with a diagnosis of delirium from ICD9/10 codes from a total of 14,526 encounters. The c-statistics of these prediction models ranged from 0.70 to 0.80 when compared to the DOS, and 0.64 to 0.69 when compared to the ICD9/10 diagnosis (Table). In our cohort, Delirium Risk Score predicted DOS positive delirium with a c-statistic of 0.8. The sensitivity, specificity, positive and negative predictive values were 96.5 (95% CI 95.8-97.1), 36.9 (95% CI 35.6-38.1), 48.7 (95% CI 47.5-49.9), and 94.3 (95% CI 93.4-95.3).
Conclusion: In this external validation study, the delirium risk score had the highest diagnostic accuracy, sensitivity and negative predictive value to predict delirium. The delirium risk score is a useful tool to detect and rule out delirium in the acute care setting.