- Poster presentation
- Open Access
PREDICT, Prediction of Delirium in ICU Patients: development and validation of a prediction model
© BioMed Central Ltd. 2010
- Published: 1 March 2010
- Metabolic Acidosis
- Multivariate Logistic Regression Analysis
- Prediction Rule
- Emergency Admission
Delirium is a serious and frequent disorder in the ICU, but a prediction model for critically ill patients is lacking.
We performed two prospective cohort studies in a 33-bed ICU for adult patients with use of the validated CAM-ICU, performed three times daily by well-trained ICU nurses. Twenty-three identified delirium risk factors were collected within 24 hours after admission and patients were followed during their ICU stay until death, discharge or development of delirium. Variables with P > 0.1 in univariate logistic regression, or with a prevalence rate <10% were excluded. Multivariate logistic regression analysis was followed by bootstrapping. Reduced coefficients were used on the second cohort for external validation. Also, nurses and physicians were asked to predict whether the patient would develop delirium during the ICU stay.
In the first study (February 2008 to February 2009) 2,116 consecutive patients and in the second study (May to September 2009) 748 patients were screened, of which 503 and 199 patients, respectively, were excluded for valid reasons. Ninety per cent of the total number of assessments that should have been performed was actually obtained and the Cohen's kappa for inter-rater reliability was 0.90. Out of 1,613 patients, 411 developed delirium and in the second study 171 out of 549 patients. Eight risk factors were excluded because of a prevalence rate <10% and one because of P > 0.1. After multivariate logistic regression analysis, 10 risk factors were included in the prediction model (age, APACHE II score, coma, diagnose group, increased urea, infection, metabolic acidosis, morphine use, sedation use, emergency admission) of which the area under the receiver-operation characteristic curve (AUC) was 0.87 (95% CI 0.85 to 0.89) and after bootstrapping 0.86. The cut-off point with the optimal predicted probability lies at 0.196 points. At this cut-off the sensitivity of the prediction rule is 80% and the specificity is 79%. Validation of PREDICT on the second cohort resulted in an AUC of 0.89 (95% CI 0.86 to 0.92). The AUC of the prediction of nurses and physicians was significantly lower, both 0.59 (95% CI 0.49 to 0.70).
We have developed and validated a delirium prediction model (PREDICT) with a high AUC that can be used in all adult ICU patients within 24 hours after ICU admission to predict the chance of delirium during their ICU stay and to guide preventive interventions.