Preventing Delirium Takes a Village: Systematic Review and Meta-Analysis of Delirium Preventive Models of Care
BACKGROUND: Each hospital day of delirium incurs greater healthcare costs, higher levels of care, greater staff burden, and higher complication rates. Accordingly, administrators are incentivized to identify models of care that reduce delirium rates and associated costs.
PURPOSE: We present a systematic review and meta-analysis of delirium prevention models of care.
DATA SOURCES: Ovid MEDLINE, CINAHL, Cochrane Database of Systematic Reviews, EMBASE, and PsycINFO.
STUDY SELECTION: Eligible models of care were defined as provider-oriented interventions involving revision of professional roles, multidisciplinary teams, and service integration. Included studies implemented multidomain, multicomponent interventions, used a validated delirium instrument, and enrolled a control group to evaluate efficacy or effectiveness.
DATA EXTRACTION: We extracted data on study design, population, model of care, outcomes, and results.
DATA SYNTHESIS: A total of 15 studies were included. All but two studies reported reduction in delirium or its duration, and 11 studies reported statistically significant improvements. Using random effects models, the pooled odds ratios of delirium incidence were 0.56 (95% CI: 0.37-0.85) from three randomized controlled trials, 0.63 (95% CI 0.37-1.07) from four pre–post intervention studies, and 0.79 (95% CI: 0.46-1.37) from three additional nonrandomized studies.
CONCLUSIONS: Several models of care can prevent delirium. In general, higher quality studies were more likely to demonstrate statistical significance of an effect. The diverse models of care included here explored interventions adapted to specific care settings, especially by addressing setting-specific delirium risk factors. These care models illustrate a range of promising strategies that deserve growing recognition, refinement, and implementation.
© 2019 Society of Hospital Medicine
One researcher (AK) screened articles by title for relevance. Relevant articles were then divided among four authors (AK, MO, NF, and OB), and the abstracts were screened for eligibility. The authors reviewed the full texts of any potentially eligible studies. Each full text was assigned to two authors for full review. Discrepancies were adjudicated by conference among all authors. In addition, references within all full-text publications were scanned for potential additional articles.
The inclusion criteria for review of full-text articles required English-language description of a model of care with multiple interventions, delirium reported as an outcome, and presence of a comparator group.
“Model of care” was defined by the Cochrane Effective Practice and Organization of Care Review Group as follows: (1) revision of professional roles, including shifting of professional roles or expansion of roles to new tasks; (2) creation of clinical multidisciplinary teams or addition of new members to the team who collaborate inpatient care; (3) delivery of multiple interventions across multiple domains (ie, studies involving a single intervention such as physical therapy or targeting a single domain such as sleep were excluded); and (4) formal integration of services whereby teams work together in collaboration with existing services to enhance care.17 For this review, we required that studies include a comparator group so that effectiveness of the intervention could be assessed. Quality improvement studies that lacked a comparator group were excluded.
Delirium incidence was the primary outcome and was evaluated by meta-analysis. Heterogeneity was assessed using I2 and visual inspection of forest plots. I2 values of 25%, 50%, and 75% represent low, moderate, and high heterogeneity, respectively. The studies were pooled according to study type as follows: randomized controlled trials, pre–post design, and other nonrandomized prospective studies. Random effects models were used to calculate estimates using the Comprehensive Meta-Analysis software (Version 3, Biostat, Englewood, New Jersey), which also generated forest plots.
Risk of bias was assessed using criteria established by the Cochrane Collaborative Review Criteria, which lists six categories of potential bias: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective reporting.17 Each study was assessed by two authors (either MO and AK or MO-P. and OB) for bias and a numerical value was assigned to each of the six categories as follows: 1 = low risk, 2 = unknown/moderate risk, and 3 = high risk. Where scorers disagreed, all authors jointly conferred, and a consensus score was given. The values for each of these six categories were added to create a composite risk-of-bias score for each study, with 6 being the lowest possible score and 18 the highest. Overall risk was classified as follows: <9 = low risk, 9-12 = moderate risk, and >12 = high risk.
RESULTS
Study Selection Process
An initial literature search identified 352 articles. After reviewing the titles, 308 articles were excluded for irrelevance, and 44 abstracts were screened for eligibility. We excluded 27 articles upon abstract review, and the full texts of 17 were obtained for detailed review. In addition, we identified another 10 potentially eligible articles through review of references and obtained full texts of these as well. Of the 27 full-text articles reviewed, 15 were included in this systematic review, 10 of which were suitable for meta-analysis. The Figure shows the PRISMA flow chart.
