67PREDICTORS OF DELIRIUM IN THE COMMUNITY: A CASE CONTROL ANALYSIS OF PRIMARY CARE RECORDS

Abstract

Introduction: Morbidity and mortality associated with delirium is well recognised. Delirium is a well-studied condition in hospitalised patients, where many risk factors for delirium have been identified. There is limited information and research available on risk factors and prediction of delirium in community-dwelling older adults. Delirium is frequently underdiagnosed. The aim of this study was to develop a risk stratification algorithm using primary care data to identify individuals who are at risk for an episode of delirium living in the community. Methods: We used the Clinical Practice Research Database, electronic primary care records from England. Predictors of delirium were identified in a case-control analysis of individuals aged ≥60 years, matched on age, gender and study entry year (1 January 2001 and November 2014). Conditional logistic regression with backwards deletion was used to identify risk factors from a list of 110 variables. The predictive model was built on a cohort of individuals aged ≥60 years on 1 January 2015, followed for up to 2 years. Logistic regression models estimated the probability of delirium within 1 and 2 years. Receiver operating characteristics were used to test model accuracy for the outcomes of delirium within 1 and 2 years, mortality within 1 and 2 years, and hospitalisations within 1 year. Results: The case control analysis included 17,286 cases and 68,321 controls and identified 55 risk factors of delirium. A cohort of 343,548 individuals (1,920 episodes of delirium in 1 year and 4,047 episodes of delirium in 2 years) was used to develop the model. An independent cohort of 85,887 individuals (451 episodes of delirium in 1 year and 996 episodes of delirium in 2 years) was used for validation. The model is a good predictor for 1 year delirium (validation dataset AUC 0.87 95% CI 0.85 to 0.88), 2-year delirium (validation dataset AUC 0.85 95% CI 0.85 to 0.86), and 1 year mortality (validation dataset AUC 0.85 95% CI 0.84 to 0.85) using the 1-year delirium prediction score. Conclusions: This is the first predictive model of community acquired delirium using primary care records. Clinically, this algorithm has important implications for early delirium diagnosis in the community. This may aid interventions, treatment and care pathways for elderly patients with delirium in the future.