The Risk for ICU Delirium (RIDE) Models: Cross-Validated LASSO Logistic Regression Based on Admission Acuity
CCCF ePoster library. Cherak S. 11/13/19; 283374; EP116
Ms. Stephana Cherak
Ms. Stephana Cherak
Login now to access Regular content available to all registered users.

You may also access this content "anytime, anywhere" with the Free MULTILEARNING App for iOS and Android
Abstract
Rate & Comment (0)
ePoster
Topic: Retrospective or Prospective Cohort Study or Case Series

Cherak, Stephana J.1,2,3,4; Brown, Kyla N.1,2,3; Soos, Andrea2,5, Stelfox, Henry T.1,2,5; Fiest, Kirsten M.1,2,3,4
 
1Department of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada
2Department of Critical Care Medicine, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada
3O'Brien Institute for Public Health, University of Calgary, Calgary, AB, Canada
4Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada
5Alberta Health Services, Calgary, AB, Canada
 


Introduction: Patients in the intensive care unit (ICU) are at increased risk of developing delirium1. ICU patients with delirium are more likely to die or have long-term problems2. Over half ICU patients will develop delirium during ICU stay; incidence depends on admission acuity3. Delirium prediction models allow clinicians to forecast individuals at higher risk for delirium to facilitate early implementation of prevention measures4,5. There are several available delirium prediction models but with limitations in study design, application and reporting of statistical methods, overlapping data capture, and non-systematic assessment of delirium outcome6. These limitations compromise clinical utility.

Objective: Develop efficacious and feasible risk prediction models for incident delirium based on ICU admission acuity.

Methods: Three administrative datasets from inpatient health services visits from 14 medical-surgical ICUs across Alberta Canada were merged to capture ICU cohort and diagnostic information 5 years prior to ICU stay. Only variables readily available in clinical practice and feasible to administer were explored as predictors. Only data available prior to delirium onset were used. Delirium was identified by structured, twice daily assessments of the Intensive Care Unit Delirium Screening Checklist (ICDSC). Model development followed rigorous methods to counter low sample size and overly optimistic model performance using least absolute shrinkage and selection operator (LASSO) logistic regression7,8. Investigation of model performance adhered to TRIPOD Statement for statistical performance reporting9,10. Model performance was investigated by cross-validation, cross-validation-ROC curve analysis with bootstrap corrected 95% confidence intervals, and decision curve analysis for clinical utility.

Results: Total of 16,005 adult >=18 years patients admitted between Jan 1, 2014 and Jun 30, 2016 were screened, from which 8,878 patients (84.7%) were eligible (elective post-surgery, n=795, 9.0%; emergency post-surgery, n=1,892, 19.4%; no-surgery, n=6,359, 72.0%). Main reasons for exclusions were death in ICU (14.4%), <24 hrs ICU stay (12.9%), and sustained coma (9.1%). Mean ICDSC compliance was 72.8%. Three separate risk prediction models for delirium incidence by admission acuity were developed. Delirium incidence was as follows: elective post-surgery, 33.1% (95% CI 29.8-36.5%); emergency post-surgery, 46.2% (95% CI 43.8-48.6); and no surgery, 53.0% (95% CI 51.8-54.3). All models demonstrated excellent calibration. Although delirium incidence differed by admission acuity, cross-validated-ROC curves remained good: elective post-surgery, ROC 0.65 (95% CI 0.57-0.66); emergency post-surgery, ROC 0.70 (95% CI 0.67-0.72); no surgery, ROC 0.76 (95% CI 0.74-0.77). Clinical utility of all models is high—they are all both feasible and efficacious at predicting those at risk. Across range of reasonable threshold probabilities each model is superior and never worse than any alternative strategy.

Conclusion: A patient's risk for delirium during ICU stay can be predicted at admission based on admission acuity using the RIDE model. The developed models have ability to enhance clinical care. Predicting high risk individuals is important but, in these individuals, delirium may already be anticipated. The maximum value of these models will be obtained by aiding in prediction of moderate risk patients, where risk of delirium may be more ambiguous.
 


Image Image Image

References
 
1.  Arumugam S, El-Menyar A, Al-Hassani A, et al. Delirium in the Intensive Care Unit. J Emerg Trauma Shock. 2017;10(1):37-46.
2.  Cole MG, McCusker J, Bailey R, et al. Partial and no recovery from delirium after hospital discharge predict increased adverse events. Age Ageing. 2017;46(1):90-95.
3.  Kishi Y, Kato M, Okuyama T, et al. Delirium: patient characteristics that predict a missed diagnosis at psychiatric consultation. Gen Hosp Psychiatry. 2007;29(5):442-445.
4.  Adams ST, Leveson SH. Clinical prediction rules. BMJ.2012;344:d8312.
5.  Ryan DJ, O'Regan NA, Caoimh RO, et al. Delirium in an adult acute hospital population: predictors, prevalence and detection. BMJ Open. 2013;3(1).
6.  Lindroth H, Bratzke L, Purvis S, et al. Systematic review of prediction models for delirium in the older adult inpatient. BMJ Open. 2018;8(4):e019223.
7.  Tibshirani R. The lasso method for variable selection in the Cox model. Stat Med. 1997;16(4):385-395.
8.  Tibshirani R, Bien J, Friedman J, et al. Strong rules for discarding predictors in lasso-type problems. J R Stat Soc Series B Stat Methodol. 2012;74(2):245-266.
9.  Collins GS, Reitsma JB, Altman DG, Moons KGM, members of the Tg. Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD Statement. Eur Urol.2015;67(6):1142-1151.
10. Moons KG, Altman DG, Reitsma JB, Collins GS, Transparent Reporting of a Multivariate Prediction Model for Individual Prognosis or Development I. New Guideline for the Reporting of Studies Developing, Validating, or Updating a Multivariable Clinical Prediction Model: The TRIPOD Statement. Adv Anat Pathol. 2015;22(5):303-305.
 

    This eLearning portal is powered by:
    This eLearning portal is powered by MULTIEPORTAL
Anonymous User Privacy Preferences

Strictly Necessary Cookies (Always Active)

MULTILEARNING platforms and tools hereinafter referred as “MLG SOFTWARE” are provided to you as pure educational platforms/services requiring cookies to operate. In the case of the MLG SOFTWARE, cookies are essential for the Platform to function properly for the provision of education. If these cookies are disabled, a large subset of the functionality provided by the Platform will either be unavailable or cease to work as expected. The MLG SOFTWARE do not capture non-essential activities such as menu items and listings you click on or pages viewed.


Performance Cookies

Performance cookies are used to analyse how visitors use a website in order to provide a better user experience.


Save Settings