Background: Hospital delirium is a common diagnosis associated with adverse patient outcomes including prolonged length of stay, higher healthcare costs, and increased morbidity and mortality. Prevention, early recognition, and timely intervention are critical to combatting the effects of delirium. Existing tools screen for the presence of delirium but are limited as they require manual completion and are reactive rather than proactive in nature. Low-touch predictive tools identifying patients at elevated risk of developing delirium are lacking in the literature and could help target proactive prevention interventions. The purpose of this study was to implement and validate an automatically calculated electronic health record (EHR)-based delirium risk prediction tool for medical inpatients by pulling data directly from routinely collected information without requiring additional inputs from front-line staff.
Methods: We implemented a simple delirium risk identification tool with elements extracted from the EHR which was previously validated among emergency department patients. The tool contained four variables, each associated with a point value, which combine to create a delirium risk score: age ≥75, recent use of psychotropic medications, diagnosis of dementia, and hearing impairment. We then performed a retrospective chart review to assess the association between the risk score and the development of delirium. For each audit, demographic information and the four risk score elements were extracted, along with additional variables such as sex, race, intensive care unit admission, and length of stay. To ensure capture of all cases, delirium was identified using two approaches: positive Confusion Assessment Method, and review of documentation for key terminology, including “delirium,” “encephalopathy,” and/or “altered mental status.”
Results: Among 112 consecutive patients aged ≥65 hospitalized on a general hospitalist service, delirium was diagnosed in 31%. Each element of the risk score was associated with higher odds of delirium, and for every 1-point increase in the risk score, odds of a diagnosis of delirium increased by 28%. A previously validated cutoff score of 3 yielded a sensitivity of 68.6% and specificity of 54.5%. Using this cutoff, the odds ratio for development of delirium was 2.62 (95% CI 1.13–6.08, p = .020).
Conclusions: A simple predictive score based solely on automated variables extracted from the EHR could support targeted proactive delirium prevention interventions. This method is an improvement compared to previously documented reactive scoring systems and could serve as the foundation for larger scale delirium prevention initiatives. Further refinement of the risk score and evaluation of its clinical utility in predicting delirium severity and informing preventative strategies is warranted.