49th ESCP virtual symposium on clinical pharmacy 19.10.2021–21.10.2021 Clinical pharmacy, working collaboratively in mental health care

Background and Objective: After cardiovascular diseases and cancer, dementia is the third most frequent cause of death in Switzerland among patients aged 65 years and older.A recent cohort study conducted in a regional hospital in Switzerland showed that a high anticholinergic drug burden in older patients is significantly associated with increased in-hospital mortality.We aimed to develop and validate a risk prediction model for in-hospital mortality in older patients with dementia and consider their exposure to anticholinergic drugs as a predefined predictor.Design: Electronic health record data within the first 24h of admission was collected from 2015-2018.Included were inpatients C65 years old, hospitalised C48h, diagnosed with dementia according to ICD-10 codes.Outpatients and patients with stays [24h in the intensive care unit were excluded.The outcome was in-hospital mortality.The developed model was based on Naı ¨ve Bayes using data from 2015-2016 as training set, the years 2017-2018 served as validation set.The candidate variables were obtained with the Minimum Redundancy Maximum Relevance (MRMR) variable selection algo-rithm and backward selection.Synthetic Minority Oversampling Technique (SMOTE) was applied within the training-set.Performance measures of interest were AUC, sensitivity and specificity.Results: In total, 2,595 patients were included.The training set contained 1266 patients (mean age 83.2±6.54 years, 60% female) of which 63 (5.0%) died during hospitalisation.Variables selected for prediction were: self-care index, delirium observation screening score, sodium, CRP, creatinine, GFR, chronic heart failure, hemi-/paraplegia, COPD, cancer, diabetes and the anticholinergic burden scales Cancelli, ACL, ADS and SCDL.Further, blood pressure, polypharmacy, age, BMI, and temperature were included.A first validation for the final model using the training set showed an AUC of 0.872, a sensitivity of 90.44% and a specificity of 68.50%.Using the validation set, we obtained an AUC of 0.771, a sensitivity of 63.01% and a specificity of 73.65%.Conclusion: The developed prediction model for in-hospital mortality showed an acceptable performance.It might be a useful tool to identify patients with dementia with a high risk for in-hospital mortality in order to implement interventions and management strategies during hospitalisation to improve patient outcome, however, further validation is required.