Previous Submission

Capacity Command Center: Machine Learning & Hospital Management at Johns Hopkins

Next Submission

A top US hospital is taking substantial steps in the use of machine learning to transform healthcare delivery

Machine learning will become an indispensable tool for clinicians seeking to truly understand their patients. As patients’ conditions and medical technologies become more complex, its role will continue to grow, and clinical medicine will be challenged to grow with it. As in other industries, this challenge will create winners and losers in medicine. But we are optimistic that patients, who generously — if unknowingly — donate the data underlying algorithms, will ultimately emerge as the biggest winners as machine learning transforms clinical medicine. (Obermeyer 2016, NEJM)

A Broken Winning Streak but a Promising Future

In 2012, a 21-year winning streak was broken when Johns Hopkins Hospital was displaced as the #1 ranked hospital in the United States [1]. In the past few years, Hopkins has embraced new technology and processes that may help them reclaim their #1 spot and, more importantly, lead the way in the transformation of their field.

Healthcare offers some of the most impactful opportunities to apply machine learning. Artificial intelligence has the potential to improve patient outcomes, provide a better patient experience, and dramatically increase hospital efficiency and cost savings [2]. A report by Accenture estimates that AI applications can potentially create $150 billion in annual savings for the United States healthcare economy by 2026 and that the AI health market is expected to reach $6.6 billion by 2021—representing a compound annual growth rate of 40 percent [3].

The Capacity Command Center

In 2016, Johns Hopkins Hospital teamed up with GE Healthcare Partners to launch a Capacity Command Center- a dedicated space, staff, and set of tools aimed at transforming how the hospital delivered care [4]. The Command Center leverages prescriptive and predictive analytics, machine learning, natural language processing, and computer vision to convey key information to decision makers in real-time [5]. The system translates about 500 messages per minute from 14 different IT systems to provide these decision makers with actionable information, empowering them to coordinate services and to reduce bottlenecks, patient wait times, and risks [6].  The introduction of the Command Center was as much about people and process as it was about technology. Jeff Terry, CEO of GE Healthcare Partners, describes it as “an excuse to get decision makers in the same room and an impetus for a hospital to transform its processes [4].”  The new process fosters a completely new approach to collaboration by collocating 24 staff who would previously have been distributed [7].

Initial results are promising. The hospital reports gains in several key domains [6]:

  • Patient transfers from other hospitals: 60 percent improvement in ability to accept complex patients from other hospitals
  • Ambulance pickup: critical care team dispatched 63 minutes sooner to pick up patients from outside hospitals
  • Emergency Department: Emergency room patients admitted to the hospital are assigned a bed 30 percent faster
  • Operating room: Transfer delays from the operating room after a procedure reduced by 70 percent

In the longer term, this adoption of technology and transformation of process represents an important first step and a foundation for future expansion of AI applications into initiatives like telemedicine and population health initiatives, which are already being tested in similar command centers [5]. Hopkins’ choice of entry point (compared to more complex and controversial forms of machine learning decision support) and emphasis on process and people as well as technology position them well for these future developments. The team utilized a highly collaborative approach  in the design of the Command Center, and been very deliberate to focus on enabling and enhancing but never questioning the front-line providers. This approach been critical in getting buy-in and cooperation from some of the skeptical clinician leaders [5]. It is also a critical line of defense against some of the potential unintended consequences presented by the application of machine learning in the medical context [2].

Key Considerations

While technologies like those used by the Capacity Control Center at Johns Hopkins present undeniably exciting economic and health opportunities, there are also some important risks and considerations for managers overseeing their adoption and expansion.

Data in medicine are fundamentally different from those in other fields, so it is critical that managers pay specific attention to what differentiates medical data and ensure that technology be adapted to suit this unique context. Medical data are subjective, based on provider opinions and patient descriptions, selective, driven by patient’s decisions of what to pursue, and event-based, recorded around clinical visits and hospitalizations, which are also subject to patient behavior [8]. It is particularly critical during this adoption phase that managers and institutions hold these tools to the highest standard of evaluation until they prove superior clinical outcomes.

Finally, as machine learning plays an increasingly prominent role in medicine, it will substantially impact the role of the providers. It is critical that organizations plan and invest in ensuring that their care providers’s skills keep pace with this new technology, so they are equipped to leverage its opportunities while avoid its risks [9].


Machine learning promises improved health outcomes as well as financial gain. Does Hopkins have a responsibility to share what they learn with other hospitals since it might improve patient outcomes, even if it this comes at the expense of competitive advantage?

What will the role of the physician (and other providers) become in the presence of really robust machine learning decision support?


(799 words)


  1. U.S. News & World Report. (2018). Best Hospitals – National Rankings. [online] Available at: [Accessed 10 Nov. 2018].
  2. Obermeyer, Z., & Emanuel, E. J. (2016). Predicting the Future – Big Data, Machine Learning, and Clinical Medicine. The New England journal of medicine, 375(13), 1216-9.
  3. Accenture (2017). Artificial Intelligence: Healthcare’s New Nervous System. Insight Driven Health. [online] Available at: [Accessed 10 Nov. 2018].
  4. Rubenfire, A. (2016) ‘Command centers help manage flow’, Modern Healthcare, 46(48), p. 0028. Available at: (Accessed: 10 November 2018)
  5. Frost & Sullivan (2018). Global; Visionary Innovation Leadership Award – Hospital Command Centers. Industry Research Analysis: Healthcare.
  6. Analytics Magazine. (2018). Johns Hopkins Hospital opens capacity command center – Analytics Magazine. [online] Available at: [Accessed 10 Nov. 2018].
  7. Analytics Magazine. (2018). Executive Edge: Command center analytics revolutionize healthcare – Analytics Magazine. [online] Available at: [Accessed 10 Nov. 2018].
  8. Mullainathan, S. and Obermeyer, Z. (2017). Does Machine Learning Automate Moral Hazard and Error?. American Economic Review, 107(5), pp.476-480.
  9. Cabitza, F., Rasoini, R. and Gensini, G. (2017). Unintended Consequences of Machine Learning in Medicine. JAMA, 318(6), p.517.



6 thoughts on “Capacity Command Center: Machine Learning & Hospital Management at Johns Hopkins

  1. Great article! The early metrics from the Command Center program seem incredibly promising, particularly when it comes to the quality and speed of healthcare delivery.

    As you raised in the article, AI and machine learning will undoubtedly change medicine from the provider perspective. Additionally, in order for machine learning to truly be integrated into front-line healthcare delivery, the nature of relationship between the provider and the patient will also need to change. It seems to me that the patient perception will be critical here, as patients have power (particularly in the US healthcare system) to dictate how they receive treatment. The real test of increased technology in healthcare will come when misdiagnosis or critical failure arrives at the hands of AI or machine learning processes. At that point, patients could demand more care from providers, and less machine involvement, necessitating a reversal of the amount of technology involved in delivery.

  2. Very interesting read. It would be interesting to see how much of the benefit that John Hopkins saw is a result of getting the 24 folks in the same room and how much is attributed to advanced algorithms that enhance doctor synthesis. What would also be interesting to consider is how does the algorithm adapt with time. You had mentioned in your post that one of the risks is that medicine is very subjective. I also think it is always evolving. Machine learning algorithms favor getting to a steady state and enhancing the algorithms confidence levels over absorbing new and upcoming medical research. I wonder how the industry plans to rectify the two.

  3. Alexa, it is very interesting to see machine learning implemented in a hospital setting, particularly at such a renowned institution as Hopkins. I do believe that institutions like Hopkins should share their learnings with other institutions to encourage better healthcare management and operations, particularly because it can help with the systemic issue of rising healthcare costs. An extension of this question is whether or not GE Healthcare Partners should offer some of these services and technologies at a discounted rate or at cost for hospitals that are struggling to stay afloat. A different question from a leadership and management perspective is how a “command center” should be launched in institutions — should implementation be slow and steady with one department tested at a time? Or should the system be applied broadly and all data fed in at once? How should hospitals take the recommendations from the command center? And what quality initiatives can be combined with the learnings from these tools that can improve service, reduce costs or errors, and ultimately hopefully improve patient outcomes?

  4. Alexa, this is a good topic. Health care costs in the U.S are extremely high. In leveraging machine learning Johns Hopkins was able to become more efficient. My hope is that as more machine learning and artificial intelligence is embraced by health care facilities, efficiency will continue to increase and health care will cheaper for patients.

  5. Very interesting! I love this model. To answer your question, I don’t think that physicians will ever be fully displaced. They (perhaps assisted by AI diagnostic tools) will be the frontlines of patient care because it seems that patients want it that way. But when it comes to hospital stays, I think this is a genius idea. There is no need for doctors to go on rounds (using valuable time) or for equal nursing resources to be devoted to all kinds of patients. If patients can be monitored remotely in one center, the highest risk patients will receive the proportionately correct amount of care relative to others (and low-risk patients less) as the hospital is effectively enjoying economies of scale in the actual practice of monitoring. Now that we have remote sensors and high-speed connections, having a centralized control room seems like it will lead to great efficiencies.

  6. Great article! I’m curious the how AI and telematics can be used to increase the ability for more patients to be released from hospitals and monitored closely at lower-cost facilities or even home, depending on the severity of the issue.

    If you could ever get past the liability issues that plague health care, this would have the potential to lower cost and even potentially improve results. Hospitals are cess pools…

Leave a comment