AI in the Exam Room: Combatting Physician Burnout and Improving Clinical Care

Microsoft is entering the exam room to help reduce physician burnout and improve patient outcomes through voice recognition and machine learning. Can it work?

Physician burnout is estimated to affect more than 50% of physicians, with clerical tasks, time pressures, and inefficient work environments as three of the largest contributing factors [1]. Microsoft, as part of its Healthcare NExT Initiative, is partnering with the University of Pittsburgh Medical Center (UPMC) to combat physician burnout in a new initiative entitled Project EmpowerMD. The first project of this collaboration is an Intelligent Scribe, a computerized audio processing program that not only transcribes patient-physician interactions, but also provides the physician with decision support tools and suggestions derived from machine learning on proper diagnosis and clinical management—a virtual AI consultant.

How does it work?

The Intelligent Scribe relies on voice detection of the patient-physician interaction, extracting key words and phrases from both the patient and the physician into the physician’s note and analyzing them using a variety of Microsoft’s cloud computing services, including Custom Speech Services and Language Understanding Intelligent Services [2]. It uses these words and phrases to make suggestions to the physician regarding further diagnostic questioning and even clinical next steps. Physicians are able to interact with the technology by accepting or rejecting suggestions, and the machine learning algorithm incorporates these actions into the algorithm to “learn” from them.

See below for a video released by Microsoft describing the technology:

https://www.youtube.com/watch?time_continue=127&v=c6exHAzNwy4

Given the rapidly expanding body of medical knowledge and the various forms that this information takes (e.g. clinical trials, reviews, public databases, clinical guidelines, etc.), physicians will require these AI decision support tools more and more to be able to deliver the most up-to-date, evidenced-based care possible while still maintaining meaningful relationships with patients. Furthermore, the Intelligent Scribe platform will allow the integration of actual behaviors of numerous physicians during real patient interactions, creating a dynamic program of best practices in the field.

Not only could this program help reduce administrative documentation burden for physicians and improve the patient-physician interaction, but it also has the potential to improve patient outcomes by helping physicians synthesize enormous amounts of clinical and scientific data in its decision support capabilities. With potential healthcare savings from artificial intelligence initiatives estimated well into the billions of dollars annually [3], there is clearly an economic argument for continued investment in this area as well. Indeed, investors seem to agree with Microsoft’s expansion of these artificial intelligence and cloud computing initiatives [4].

Looking ahead

Artificial intelligence in medicine is here to stay. In an effort to move broadly and swiftly in this space, Microsoft, as part of the Healthcare NExT Initiative, is employing artificial intelligence and machine learning in a whole host of other healthcare contexts, including cancer genomics, health information security and compliance, global health, and medical diagnostics, trying to establish itself as a trustworthy leader in healthcare artificial intelligence and cloud computing, according to Peter Lee, the Corporate Vice President of Microsoft Healthcare [5]. The company’s long-term vision for the initiative is “a learning system that incorporates data from longitudinal medical records, medical devices, genomics, population health, research papers, and more” [2]. In order to keep up with fast-approaching rivals in this space, such as Amazon, Google, and Apple, in addition to the numerous healthcare AI startups popping up, Microsoft will need to continue establishing and fostering relationships with academic leaders like UPMC while ramping up its own internal research and development in this area. Furthermore, in addition to the provider side, Microsoft should consider partnerships with health insurers, who are currently sitting on vast quantities of claims data from which artificial intelligence and machine learning could derive key health insights.

Questions still remain about the efficiency and accuracy of these and other digital scribe initiatives. What will the medicolegal implications of these emerging technologies be, and how will Microsoft and others handle backlash associated with the unavoidable medical errors resulting from this technology? How will Microsoft unlock the potential of this technology while ensuring security and confidentiality in compliance with the Health Insurance Portability and Accountability Act (HIPPA)? The company is keenly aware of this and other data security issues and appears to be addressing them [6]. And finally, what other healthcare domains should Microsoft explore with this technology?

Patients and physicians are eagerly awaiting answers to these questions, hoping that the responses will provide value in the form of more personalized clinical encounters, reduced physician burnout, and improved patient outcomes.

(715 words)

Photo from Steelcase. https://www.steelcase.com/research/articles/topics/healthcare/exam-rooms-that-empower-people/, accessed November 2018.

References

[1] Rothenberger, D. “Physician Burnout and Well-Being.” Diseases of the Colon & Rectum, 60(6) (2017), pp.567-576.

[2] Microsoft. “Project EmpowerMD: Medical conversations to medical intelligence.” https://www.microsoft.com/en-us/research/project/empowermd/, accessed November 2018.

[3] Kalis, B., Collier, M., and Fu, R. “10 Promising AI Applications in Health Care.” Harvard Business Review, May 10, 2018. [https://hbr.org/2018/05/10-promising-ai-applications-in-health-care], accessed November 2018.

[4] Ranks, T. “3 Fascinating Companies Using AI to Forge New Advances in Healthcare.” TheStreet, May 18, 2018. [https://www.thestreet.com/investing/three-intriguing-companies-using-artificial-intelligence-14591053], accessed November 2018.

[5] Lee, P. “Microsoft’s focus on transforming healthcare: Intelligent health through AI and the cloud.” Official Microsoft Blog, February 28, 2018. [https://blogs.microsoft.com/blog/2018/02/28/microsofts-focus-transforming-healthcare-intelligent-health-ai-cloud/], accessed November 2018.

[6] Pohar, D. “Four paths to voice innovation through healthcare provider and technology partnerships.” Microsoft Research, February 28, 2018. [https://www.microsoft.com/en-us/research/project/empowermd/articles/four-paths-voice-innovation-healthcare-provider-technology-partnerships/], accessed November 2018.

Previous:

The Benefits of Additive Manufacturing at BIA

Next:

Pinterest Needs to Improve Revenue Per User Ahead of IPO

Student comments on AI in the Exam Room: Combatting Physician Burnout and Improving Clinical Care

  1. First of all, this is great to see. Doctors have so little time with patients these days, and its not good for anyone. For the medical error issues, my question would simply be, are the errors statistically better or worse when the system is used? How is the overall benefit to the student population vs. an increase in errors if this is the case?

  2. Very interesting read. I am a strong believer in applying machine learning more extensively in health care to both improve the working conditions of physicians and health care professionals, but also support them in decision making for even better health outcomes. According to a report released by McKinsey, AI clinical support reduces the number of wrong diagnoses. The reality is, that AI can spot patterns based on patient medical data, enabling doctors to make their decisions based on a more comprehensive analysis. Another interesting application is that of personalized clinical pathways. By using algorithms and pattern recognition on data of previous patients as well as your individual patient data, AI can enable personalized clinical programs for patients to improve health outcomes. I.e. based on your symptoms and your demographics, treatment B is the preferential treatment based on previous patients as it offers the highest expected success rate. [1]

    However, for application of AI to be efficient you need access to all the patient medical data. In Sweden a main issue is that due to different IT systems of different health care providers there is no one place with the full data set. This severely limits the ability of AI.

    As you mention there are questions related to efficiency and accuracy of the methods. I believe it should be used as a supporting tool for doctors to ensure that it is safe and accurate, as well as ensuring patients feel comfortable with it. Moreover, AI must be incorporated in the curriculum of medical schools around the world to build awareness and understanding of the technology. Lastly, I believe it is crucial for Microsoft as well as hospitals incorporating the technology to work closely with and collaborate with regulators and medical associations to develop standards for the use of AI in clinical applications.

    [1] “Digitizing healthcare in Sweden”, McKinsey & Company, June, 2016, https://www.mckinsey.com/industries/healthcare-systems-and-services/our-insights/digitizing-healthcare-in-sweden, accessed November 2018.

  3. Awesome essay! It’s reassuring to see that our technology and healthcare corporate communities are working together on one of the more challenging issues that we face as a society – fixing/improving our healthcare system. This morning I found out that an old colleague of mine opted to have a knee replacement surgery done by a robot vs. a traditional knee surgeon, which shocked me as I wasn’t aware of how far along machine learning was among the different verticals in healthcare. I’m curious if Intelligent Scribe played a role in my old colleagues surgery… I would be curious how Intelligent Scribe works / is integrated with the electronic health record systems (EHR) within hospitals. I know these software platforms (epic, all scripts, etc.) are complex and are often hard integrate with. It seems as though Intelligent Scribe would benefit an incredible amount from integrating with these systems. In any regard, great work; I really enjoyed reading this.

  4. Great read. I see the potential benefit of this technology — helping to offload some of the onerous charting requirements physicians face in combination with providing support in clinical decision making. I have two concerns, one of which will probably be ameliorated over time. First, I imagine there will be a fair number of errors during early adoption. If Microsoft rolls this out too soon and the number of errors are significant (be it in quantity or, certainly, in quality), I think they risk losing physician buy-in very quickly. Nothing is more frustrating than using a scribe service only to find out that you have to go back at the end of the shift/day and correct multiple documentation errors made by the scribe. In these cases, I know from personal experience that many physicians will elect not to have a scribe at all and remain skeptical.
    Second, I believe this technology cannot fall into the same trap as current EHRs. Many current electronic systems are created to meet insurance/government metrics, and, I would argue, all else is secondary (including measuring meaningful clinical metrics, streamlining physician workflow, and improving the patient experience). As a result, data entry has become a large part of physicians’ workflow, and the information collected has not led to the potent clinical insights that were hoped for or promised.

  5. Good piece. Agree with the relevance of physician burnout (unfortunately compounded by physician shortage). You briefly mention the ramification of medical errors on Microsoft but I think another point here is once health insurers get involved (and they will at any opportunity to improve/better standardize treatment) will they mandate physicians to act on the software’s recommendations? If so, then at what point does loss of decision making autonomy only serve to worsen the same burnout that this software was trying to solve in the first place?

Leave a comment