4CPS-225 Implementation of a computer algorithm in the electronic health record to identify inpatients with untreated atrial fibrillation to optimise stroke prevention

Background and importance Stroke prevention in atrial fibrillation (AF) is a major indication for oral anticoagulants. Appropriate use of anticoagulants reduces the relative risk of stroke by approximately 66%. Despite an abundance of evidence for this treatment, up to 40% of AF patients do not receive anticoagulant therapy. Decision support systems have shown promise in increasing guideline adherence to reduce therapeutic omissions. Aim and objectives We aimed to develop and validate a screening tool to identify untreated AF inpatients. Material and methods A computerised screening tool was developed integrating the following data from the patient’s electronic health record (EHR): demographic, laboratory and medication data, ECG reports and allocation to specific care programmes. A decision process was applied, which consisted of (1) determining whether AF was present, (2) calculating the CHA2DS2-VASc score and (3) determining whether anticoagulant treatment was present during hospitalisation and/or in the pre-admission therapy. Subsequently, based on these three steps, a priority score was assigned to the patient, from 0 (no risk) to 5 (highest level of risk). A validation study was done to assess the accuracy of this approach. Criterion and tool validity were ascertained by determining specificity and sensitivity, compared with a manual check of the EHR. Consistency regarding the priority score was determined by estimating Cohen’s kappa (κ). Results For the validation, 800 inpatients were included. The specificity and sensitivity of the tool for identification of patients with AF were 87.6% and 95.1%, respectively. Overall specificity and sensitivity for identification of AF patients with a CHA2DS2-VASc score ≥2 was 72.7% and 97.7%, respectively. Specificity and sensitivity to determine the presence of anticoagulants was 97% and 87%, respectively. There was good agreement between the priority score obtained by the pharmacist after EHR review and the one generated by the screening tool (κ unweighted 0.74; κ equal weighted 0.66). Conclusion and relevance This screening tool to identify untreated AF inpatients was found to be reliable and valid with a high sensitivity. To further improve specificity, future investigations might focus on better digital structuring of patient data. Our future goal is to implement the AF screening tool in clinical practice to improve the use of preventative therapy and reduce the significant burden of stroke. References and/or acknowledgements Conflict of interest No conflict of interest