Validity of a screening method for delirium risk in older patients admitted to a general hospital in the Netherlands

Abstract

Objective: Delirium is an impactful, frequently occurring complication in older hospital patients. Consequently, risk stratification of delirium was included in a set of mandatory safety measures in general hospitals in the Netherlands. This risk stratification contains three consensus-based questions that have not been validated. Therefore, we evaluated their predictive performance and examined whether other routinely collected patient data can improve the prediction of delirium. Method: Using data from a continuous data registry from a general hospital, the prediction of the three questions was compared with the occurrence of delirium in 3786 older patients. Regression models were fitted that included other patient-related delirium risk factors. The performance was expressed by discrimination and calibration. Results: Delirium occurrence was 16.8%. The three questions, a regression model with the three questions, a full model and a reduced model – including the three questions, age, use of glasses, number of medications and Katz-ADL – showed sensitivities of 0.88, 0.88, 0.92 and 0.91 and specificities of 0.52, 0.52, 0.53 and 0.54, when treated as dichotomous models respectively. The three risk models had C-statistics of 0.81, 0.86 and 0.86, with excellent p-values of the U-statistics. Conclusion: The three risk-stratification questions show promising results but substantial overprediction (49% predicting positive). Further validation should be done outside the Netherlands, given the potential bias as a result of clinical activities following the risk stratification. The reduced model shows excellent calibration performance, indicating good prediction in each individual patient. In clinical practice, this advantage adds to clinical reasoning.