Dual autoencoders modeling of electronic health records for adverse drug event preventability prediction

Elderly patients are at increased risk for Adverse Drug Events (ADEs). Proactively screening elderly people visiting the emergency department possibility of their hospital admission being drug-related helps to improve patient care as well prevent potential unnecessary medical costs. Existing routine ADE assessment heavily relies on a rule-based checking process. Recently, machine learning methods have been shown be effective in automating detection ADEs, however, most approaches used only either structured data or free texts feature engineering. How better exploit all available EHRs predictive modeling remains an important question. On other hand, automated reasoning preventability ADEs is still nascent line research. Clinical information 714 ED-visit with labels was provided ground truth by Jeroen Bosch Ziekenhuis hospital, Netherlands. Methods were developed address challenges applying engineering heterogeneous data. A Dual Autoencoders (2AE) model proposed solve problem imbalance embedded existing training Experimental results showed that 2AE can capture patterns minority class without incorporating extra process balancing. yields adequate performance and outperforms more mainstream approaches, resulting AUPRC score 0.481. We demonstrated how employed analyze both unstructured from electronic health records purpose preventable prediction. The algorithm effectively learn group phenotype imbalanced