Background: Delirium prevention by multicomponent interventions is proven, with strong evidence confirming its effectiveness in both medical and surgical populations. Identifying high risk patients is very important for resource allocation and utilization. Risk factors for delirium are broadly classified into predisposing and precipitating factors. Combinations of these risk factors precipitate delirium and hence a prediction model is needed to risk stratify. As prediction models developed across the world have varied results, no one delirium prediction tool exists in widespread clinical use. Our goal is to systematically review delirium prediction tools developed in the last decade, identifying strengths and limitations, as we work towards one prediction tool that can be applied across care settings.
Methods: An experienced medical librarian developed and ran searches in Ovid-Medline, Embase, Psychinfo and Cochrane databases with date ranges from January 1, 2010 to September 1, 2018. Search terms included delirium, prediction model, risk model or score, hospital, inpatient and ICU. One investigator screened the shortlisted abstracts. Studies focusing only on risk factors or other prior testing tools, but did not involve the development of a risk prediction tool were excluded.
Results: 4465 abstracts populated the initial search and were further shortlisted to 570 abstracts that were manually reviewed, of which 82 were selected for full article review. A total of 14 different prediction tools were identified for further detailed review (Table 1).
Conclusions: Existing delirium prediction tools are derived mainly from small sample studies. A multitude of risk factors and combinations of these risk factors result in delirium; hence large population based studies are required. No prediction tool developed in the last 10 years has been externally validated in the US yet or is applicable across different clinical settings. A robust, but simple to use, clinical tool is required.