“I think the goal and fear was, how is this bot going to affect our patients? Are our patients going to notice a difference and ask, “What is going on?” After all the testing and double-checking that CTG did, it was seamless.”
- Becky Kilen-Carter, Manager, Privacy and ROI, Gundersen Health System
Challenge
Gundersen Health System – part of Bellin and Gundersen Health System, which is becoming Emplify Health – supports communities in the Midwest through patient-centric care, spanning 7 hospitals and 65 clinics. Previously, Gundersen staff manually verified patient identity in their electronic health record (EHR) system when patients sent messages or requested medical records through Epic’s MyChart. Gundersen recognized the need to save time, improve cost savings, and test new technology to increase efficiencies.
Solution
Gundersen, powered by CTG, opted to pilot the high-volume task of patient verification for automation. To achieve this, CTG created a HIPAA-compliant bot that cross-referenced patient messages and recorded requests with existing patient data in the EHR. Once a patient was verified using demographic identifiers, access permissions to MyChart were granted via automatic email notifications containing verification codes.
Given the many anomalous situations that can arise during the patient verification process, the bot was trained to either handle or flag the following, among other scenarios:
- "Break the Glass” situations that necessitated record or care delivery before full verification
- Access to information for patients such as infants, the elderly, and those with disabilities
- Outdated records
- Patients with multiple identities
Building Trust with "Human in the Loop"
During the first two weeks of the bot’s launch, it was instrumental for staff to monitor and approve its activities and to verify accuracy. This was achieved through a Human-in-the-Loop (HITL) approach that combines supervised and active learning, where people play a crucial role in both training and testing the algorithm. To address this, CTG incorporated bot-generated reports displaying which messages had been addressed and which had been flagged. This process was paramount to establishing trust and a framework for addressing any niche issues prior to the move to full automation.
Outcome
After a two-week long HITL period, the bot was successfully transitioned to full automation. Since launch, patients have not flagged any treatment discrepancies when requesting access to MyChart records. As a result, more than 150 requests have been processed per week, freeing up hundreds of hours a month for staff to focus on higher-level tasks.