‘It’s not easy to be a start-up in health care’: Family doctor’s AI tool aims to ease burnout

Physicians may heal, but they must also write … and write. Physicians are required to document each patient visit. If a physician sees 20 patients in a day, that physician is required to complete 20 notes detailing the concerns of the patient, physical exam findings, results of recent investigations and plans for further management. This administrative burden has been noted time and time again as a cause of burnout for family physicians.

However, artificial intelligence (AI)-based tools are starting to complete this documentation in clinics across the province.

Noah Crampton, a family physician-turned-entrepreneur, is developing artificial intelligence-based software to aid physicians in clinical note taking. The software, Autoscribe, generates a transcript of the physician-patient conversation and translates that to a corresponding note in the Electronic Medical Record (EMR). A human scribe edits the note in its final stage. While a private entity, this start-up represents one way in which innovation is being introduced into health care locally in Ontario. Within a publicly funded health-care system, the role of such private entities is not clearly delineated. While they represent avenues of innovation, physicians need to critically appraise these new technologies and start-ups seeking to usher them in.

Crampton says Autoscribe started as a research project that he and his team decided to commercialize after noting how accurately the AI could create clinical notes from voice transcripts. Although this presented an exciting advancement, innovating technology within health care often is met with skepticism from health-care providers and patients alike.

“We make these advanced sophisticated models in our universities but getting them into the market is a huge leap,” Crampton says. “Ultimately, there is tension because this is a market-based solution to a public health-care problem,” and Canadian physicians are not in the habit of looking to companies to solve their workflow issues.

The conservative nature of medicine is such that physicians are hesitant to adopt new technologies without a significant evidence-base around it. One government solution to the reduced uptake of innovative technology in health care is the development and funding of the Canhealth network. This government-funded network connects private companies to networks of hospitals and clinics to have their proposed technologies implemented.

“It’s not easy to be a start-up in health care,” Crampton says with a chuckle partway through our discussion. “Investors have asked us, what are you doing in Canada?” he pauses. We both seem to contemplate how the United States, alleged free-market utopia, would certainly be an environment more conducive to marketing such technology. “But I feel quite strongly about the Canadian health-care system.”

Artificial intelligence-based tools have been emerging in medicine across specialties, in areas of diagnostics and risk-stratification. Within primary care, AI-powered chatbots have been harnessed to answer patient questions regarding mental health. As AI-based technologies emerge, it is important that clinicians and other health-care providers actively involve themselves in their development.

“We are still at the early phase (of development) such that we can shape it,” says Andrew Pinto, an unpaid advisor to Crampton’s start-up who has collaborated on evaluating AutoScribe and is the AI lead for EXITE (EXploring Innovative TEchnologies in Family Medicine) at the University of Toronto, a pioneer in the adoption of AI in primary care. “I would encourage clinicians to not be passive recipients.”

“I would encourage clinicians to not be passive recipients (of new technologies).”

Lessons learned from the development of medical technologies like EMRs can help smooth the way for development of AI-based tools. EMRs, for example, have changed the practice of clinical medicine significantly, allowing for robust documentation of clinical interactions and investigations. However, physicians often were not involved in co-designing these platforms. As a result, EMRs often were found to be unfriendly for the end-users, the clinicians themselves.

“Looking back, I think we were very passive in receiving them. Companies said that they can come up with these systems and we just said OK,” Pinto notes. “Ultimately, EMRs have generated way more work for us.” Engaging physicians in co-designing new technologies is imperative in ensuring that they serve the needs of patients and clinicians rather than any private entity, Pinto says.

The benefits of having physicians involved are evident in Autoscribe. As a practicing family physician, Crampton is familiar with the reality of clinical medicine and nuances of medical culture.

When discussing the involvement of the human scribe, he says, “Some companies just have the physician edit the AI-generated notes, which lowers costs. But ultimately, if you have a busy clinic and you’re having to edit these notes yourself, how much value is really being added?” The difference between software giving a physician an edited note and one that must be combed through may seem small, but the labour of editing adds up throughout the day. For clinician developers, such considerations are key in designing software that is maximally useful to other physicians.

Any new technology in medicine, especially one that collects and manipulates patient data, must contend with concerns about patient privacy.

Embedded with EMRs, “(Autoscribe) is built to the same security standard as the EMR itself,” Crampton affirms. The preservation of patient-physician confidentiality remains a prominent consideration. Conversations between patients and their family physicians are not always purely medical. There are discussions of children and partners, updates about recent life events and more. These conversations are picked up by Autoscribe. As such, at the outset of each patient interaction, the physician is prompted to check off a box stating that the patient has consented to using the tool. When, how and how often consent is obtained for involving these tools requires careful consideration.

Aside from scaling Autoscribe up, Crampton says he is hoping to expand its functions. For example, after the note is complete, the software could help facilitate next steps in patient care, such as auto-preparing prescriptions and referrals. “We are currently solving an administrative problem. But this is the foundational piece to much more support for the workflow.”

Becoming comfortable in assessing the risks and benefits of such technologies and appropriately interrogating the companies pitching them will allow physicians to make the most informed decisions about integrating them into their care.

Says Pinto: “For-profit entities will operate in a way that meets their needs. It is important to ask, what are their motives? And what ultimately is going to help patient care? And where is the proof? Show me that your new fancy technology will make a difference.”