Abstract Background Amiodarone treatment requires repeated laboratory evaluations of thyroid and liver function due to risk of side effects during long-term administration. Robotic process automation (RPA) represents the use of software robots to automate repetitive and routine tasks typically performed by humans. Purpose We aimed to develop an RPA using a diagnostic classification algorithm and clinical support system for follow-up of amiodarone treatment. Methods Based on international guidelines and clinical specialist advisory we constructed an RPA which was studied in a test environment in parallel with routine amiodarone treatment. Expert advice was integrated through a series of collaborative workshops, ensuring alignment of the algorithm with current best practices in thyroid and liver disease management. The RPA provided recommendations on the time interval to next laboratory testing and suggestions for patient management to the physician, who was serving as a human-in-the-loop responsible for clinical decisions. Results After iterative improvements the RPA was evaluated against physician orders (n=390 paired orders) without technical failures. The recommended time interval to next follow-up was (mean ± SD) 4.5 ± 2.4 vs. 3.1 ± 1.4 months for RPA vs. physician (p<0.0001). While normal laboratory findings resulted in recommended follow-up in 6 months by the RPA (72.2%) this was only ordered in 9.9% by the physician (Figure). Physicians’ orders clustered between 3-4 months (58.5%). All patients diagnosed with new side effects (n=12) were detected and communicated correctly by the RPA, whereas only 8 were correctly detected and responded to by the physician (Table). Conclusions An automated process constitutes a reliable alternative to the current fully manual management of amiodarone follow-up. It may reduce manual labour, frequency of laboratory testing and detect side effects with increased precision, thereby reducing costs and enhancing patient value. While the RPA shows promise, further research is needed to explore its adaptability to other treatment scenarios.
Read full abstract