Communication and Algorithmic Decision Making in a Virtual Healthcare Context
Extended Abstract

Like in many healthcare workplaces, the VICU staff face frequent changes in the virtual platforms they are expected to master. During the course of a single staff meeting, the staff were trained to use a new virtual scheduling system to sign up for shifts; reminded to prompt their patients to complete a new educational module program targeted to high-risk patients; and congratulated for their rapid progress ordering Covid-19 tests for patients on a new e-platform. A large change for the VICU staff occurred this past winter, when administrators decided not to renew their contract with the patient warning system, the Rothman Index.

The Rothman Index [1, 2] is a proprietary closed source algorithm that uses vital statistics, lab results, and other health related information to predict health deterioration. In the VICU, nurses would receive alerts from the Rothman Index and make decisions about whether to send those alerts to providers on the hospital floor. Administrators are considering a transition to Epic's early warning system – the Epic Deterioration Index – that is already available and free through the hospital's Epic platform. Administrators cite the limited efficacy of the Rothman in improving patient health outcomes as a key reason for the transition.

As the VICU team transitions away from the Rothman Index, practitioner communication and accountability are also undergoing changes. The director of the VICU has noted that without the Rothman alerts, VICU staff do not have as regular opportunities for outreach to floor personnel, reducing both practitioner accountability and opportunities for patient advocacy. The director also noted that relationships with practitioners on the floor are “always a little dicey,” echoing prior research that has found that workers tend to be resistant to the experience of being monitored by tele-ICU staff [12, 13]. The team has been working strategically to develop a consultative and mentoring relationship with floor nurses, however, especially new hires. Positioning themselves as a resource for new nurses can facilitate a more collaborative relationship while also helping VICU staff to provide support and accountability for those who are most in need. 

VICU nurses also reported that while they often receive requests to monitor a particular patient from floor nurses, they then enter their database to find that the patient's vitals were entered many hours ago. They noted that vitals are taken and recorded on the physician's schedule, typically every 8 hours, which does not reflect the needs of the VICU team. Thus, they can do little to support a nurse that has requested monitoring but has not input vital information recently. These findings echo recent research on early warning systems that has found one of the largest detriments to success was delayed information input by nurses [14] and demonstrate how tele-ICU's can disrupt communication workflows [15].

In the aftermath of the Covid-19 pandemic, practitioners in the VICU have found their work has shifted substantially towards providing direct outreach to ambulatory patients, especially Covid-19 patients who are recovering at home. Primary duties include information intake and entry into digital tools and regular monitoring of patients through daily protocols and checklists to determine need for hospital readmission. Nurses report that their one-on-one patient contact has increased significantly as a result.

This increased patient contact has also provided opportunities for patient advocacy. One VICU nurse frequently directed floor nurses to reduce ambient noise from telemetry alerts when it was clearly agitating a mental health patient. During a staff meeting, another VICU nurse argued for revised protocols for patients at risk of self-harm, noting that following many of these patients into the bathroom seemed unnecessary. The team discussed building more flexibility into protocols to empower floor nurses to make contextual decisions about a patient's needs. While neither of these examples emerge from algorithmic decision making per se, they do demonstrate how virtual monitoring, alongside algorithmic monitoring, can provide more contextual patient information that enables situated patient advocacy and care and counteracts algorithmic biases [9, 10].