Duke University Health Systems is composed of a world-renowned hospital and healthcare network. Duke University Hospital, one of their three hospitals in North Carolina, is nationally ranked in 14 adult specialties.
Technology continues to transform healthcare. From reducing administrative burden to improving patient care to optimizing workflow to facilitate better medical decisions, AI and machine learning are leading to improvements felt by clinicians and the patients with whom they interact.
Eagerly adopting the latest and greatest AI solutions might be a viable strategy for eCommerce or consumer-driven industries, but it is not optimal for the healthcare industry. Duke University Health System has a unique approach to its use of artificial intelligence, prioritizing in-person consultations with doctors over prematurely defaulting to digital interactions.
Additionally, they take a measured approach to implementing machine learning models into their workflows. Notably, though, this doesn’t represent an unwillingness to innovate but is representative of how healthcare leaders need to approach change management: embrace changes only if they lead to cost savings or improve quality and safety as verified in data.
Emerj Senior Editor Matthew DeMello recently sat down with Dan Buckland, Medical Director of Duke University Health System, on the ‘AI in Business’ podcast to talk about his organization’s distinctive, skeptic-friendly approach to AI adoption. The following article examines three key takeaways from their conversation:
- Applying mature artificial intelligence solutions: Minimize negative impact on patient care by only using artificial intelligence tools or algorithms that have been proven.
- Determining the appropriate place for virtual care in healthcare: Using telemedicine visits when in-person experiences are holding back patients.
- Designing interoperable healthcare solutions to reduce friction: Reducing patient view of friction created by implementing new technology into healthcare workflows without sacrificing patient agency and awareness.
Listen to the full episode below:
Guest: Dan Buckland, Medical Director of Duke University Health System
Expertise: acute care diagnostic and therapeutic medical technology development, data science, human-robot interaction
Brief Recognition: Dan Buckland is an Assistant Professor at Duke University and also Deputy Human System Risk Manager at NASA. Buckland is also Director of the Duke Acute Care Technology Lab (DACTL). Previously, Dan worked as a physician for a clinical practice group affiliated with Duke Health.
Applying Mature Artificial Intelligence Solutions
Buckland begins his appearance on the podcast by explaining how Duke Healthcare takes a very measured approach to implementing new technologies into its healthcare workflows.
He talks about how Duke professionals are looking at technologies that already have an established knowledge base, including information about how the technology has failed at other institutions with their specific use cases. Buckland further mentions how this allows them to use lessons learned from different industries, including related healthcare industries.
“We find the actual artificial intelligence, the technology portion of applying things into a healthcare system, is only 10 to 15% of solving the problem itself,” says Buckland. He explains the end user of the technology in the healthcare system is 80 to 85% of the actual implementation of that problem.
Buckland prefers a machine learning model or technology to be solid and proven. In this way, if an implementation does work, it is apparent that it is the result of his implementation. He explains this makes his troubleshooting problem-solving more straightforward.
Determining the Appropriate Place for Virtual Care in Healthcare
Buckland doesn’t necessarily treat telemedicine as a last resort, although it was evident his approach to it is influenced by his experience as a practicing emergency physician. He acknowledges the aspects of in-person visits between a doctor and patient that are lost in digital interactions, including being able to smell the patient or lay hands on them during an examination.
Buckland also explains that there is a spectrum of care in healthcare, from phone calls to in-person visits. He highlights that phone calls, which have existed for decades in healthcare, are still a potent telemedicine modality.
He points out that adding a video feed sometimes introduces an extra bit of complexity that isn’t always necessary. In some cases, it can cause unnecessary friction. He describes a scenario where a patient or provider is having trouble logging on to a virtual visit. As a result, what could have been a brief phone call turned into an IT session to ensure both the patient and provider could connect to the video meeting.
“What we find through the pandemic and through all the recent technologies that have made it easier to deploy telemedicine is that the use cases for telemedicine, as seen as video interaction between a provider and a patient, is not as wide as we had hoped.” Emails, phone calls, and asynchronous messaging between patients and doctors can help fill in the gaps created in the absence of direct interactions.
According to one study, telemedicine visits likely provide comparable clinical outcomes to in-person visits. However, this is limited to certain medical conditions. The quality of healthcare is not as clear for conditions related to infection management, including conditions like urinary tract infections and sinusitis, which aligns with Buckland’s insight on what type of data a doctor can capture from direct interaction with patients.
“Having that backup option of telemedicine for when it is needed to prevent an in-person visit is valuable,” says Buckland. “It’s just not as valuable as we had hoped it was in 2020 the way we thought it was going to be.”
Designing Interoperable Healthcare Systems to Reduce Friction
When asked about how technology creates friction in the patient experience, Buckland provides essential insight. He explains how the friction in the healthcare industry falls roughly into two categories:
- Friction in the healthcare experience that the patient is directly aware of and
- Friction that the patient isn’t aware of the results when a new technology is introduced into the healthcare workflow.
Buckland tells Emerj that when staff need to be trained on new technologies, processes can be further complicated by high employee turnover. Hospitals also need to re-establish the efficiency of the workflow after they switch to a new technology.
Buckland continues, explaining how they try to conceal resource allocation decisions that happen on the backend from patients. He describes a scenario where a patient is informed what time their surgery is.
That remains unchanged, but the process behind the scenes might have changed due to the implementation of new systems. This process can be further complicated not only by the required learning curve the new system requires but also by high staff turnover resulting in the onboarding of new employees. He goes on to quantify the issue introducing new technology presents, “The amount of human factors in healthcare, both in safety and just implementation, is 80% of all of these technology problems.”
For many industries, fax machines are an afterthought. Electronic data interchange has replaced fax machines in many industries, including manufacturing, eCommerce, and logistics, but that hasn’t been the case in healthcare.
Buckland explains to the podcast audience why and how fax machines are a mainstay in healthcare and the primary modality at the center of technology built for the healthcare industry.
He succinctly summarizes the extent this legacy component has on the development of new technology as “every new technology has to be backward compatible to a fax machine.”
According to a 2017 study from the National Library of Medicine, approximately 63% of doctors have been using the fax machine as a primary communication method as recently as 2012.
He also mentions that there is a more significant effort for IT systems to talk to one another but that it’s not as advanced as they would have expected by this point. When patients have to navigate two separate healthcare systems, if they have different EMR systems, the commonality between them will be fax, explains Buckland.
The main hurdle to achieving interoperability is due to the fact that medical record software can’t communicate with software in other hospitals because of its proprietary nature.
Interoperability in healthcare is difficult to achieve. The capability for integration is necessary to make sure that healthcare providers have a complete clinical picture of patients as they move between hospital systems.
