PB1134 - A Swiss National Initiative to develop and validate a multi-modal, multi-source, multi-centric data-driven approach for antithrombotic adverse drug events detection in older inpatients

Background: Hemorrhages associated with antithrombotics are a major concern, yet underreported in older inpatients. To enhance the monitoring and risk management of these high-risk medications, there is a need to develop automated detection systems that utilize electronic medical record (EMR) data to identify related adverse drug events more effectively.

Aims: To develop an automated detection system for severe (SH) and clinically significant (CSH) hemorrhages associated with antithrombotic prescription in inpatients > 65 years, using various sources of structured and unstructured information from the EMRs of three University hospitals in Switzerland.

Methods: We analyzed EMR data from older inpatients admitted between January 1, 2015, and December 31, 2016. We developed rule-based algorithms to detect SH and CSH, as defined by adapted ISTH criteria. These algorithms utilized structured data, included diagnostics and procedures coded with the International Classification of Diseases (ICD-10-GM) and the Swiss Classification of Surgical procedures; laboratory data; and medications prescribed. We applied natural language processing (NLP) to 7513 discharge letters. Our key outcomes were the cumulative incidence of SH and CSH, and the relative contribution of each source to these outcomes.

Results: Analysis of data from 36,039 patients revealed a cumulative incidence of 8.3% for SH, with 1.0% resulting in in-hospital deaths, and a 18.2% cumulative incidence of CSH. ICD-10 GM codes detected 28.3% and 35.3% of SH and CSH, respectively. The relative contribution of the presence of an antihemorrhagic agent were 1.3% for SH, a drop or a low hemoglobin value and a transfusion added 66.8% and 62.1%, 6.4% and 2.6% to SH and CSH detection, respectively. The combination of the rule-based and NLP detection algorithms enabled the detection of 12.2% SH and 31.3% CSH.

Conclusion(s): Automatic detection of hemorrhages combining multi-sources is an innovative approach with the potential to enhance patient safety and quality of care by facilitating identification of adverse events.