The Intensive Care Unit: Tomorrow and Beyond

The intensive care unit (ICU) is less than a century old. Since its inception in the 1950s, every decade has brought remarkable advancements in technology, and at the same time, increasing patient acuity and complexity. The coming decade is perhaps a unique moment in critical care with the explosion of clinical data and the advent of artificial intelligence (AI). If AI achieves its full potential, its implementation into clinical care could be as transformative to the ICU as the introduction of the ventilator—which brought with it great promise in addition to risks and new problems. Although we have considered implications of AI in previous editorials,^1,2  we return to the topic now to focus on how AI may influence the role of the bedside clinicians in the ICU. Artificial intelligence will not only have an impact on how critical care is delivered in the coming decade, but also on how the fundamental experience of delivering that care as a clinician may transform. Despite this transformation, the essential traits of the bedside critical care clinician will endure and only become more important. Technological aptitude, breadth and depth of clinical knowledge, and empathy for the sickest and most vulnerable patients are values that will be even more important—and potentially under threat—in the ICU of tomorrow as it embraces the full potential of AI to improve care.

Today, care traditionally provided at acute care facilities is already moving into the home, from urgent care³  to low-acuity home hospitalization.⁴  As a result, the acuity in the ICU is only expected to increase. Patients will not be admitted to the ICU for monitoring for the risk of decompensation—only for the need for advanced diagnostics and care not yet safe to deliver outside of the ICU. Technology will become less invasive and provide more autonomous “closed-loop” support. An example might be the development of pathogen-resistant invasive and noninvasive devices that provide continuous diagnostics such as moment-to-moment blood gas and chemistry measurements; these devices may even be integrated into the ICU bed itself.⁵  Other examples include ventilators and other life-support devices that make breath-to-breath or beat-to-beat adjustments autonomously to maximally personalize care to the individual patient, potentially reducing duration of critical illness and the need for sedation and analgesia.^6,7 

Robust descriptions exist of what such an ICU system could look like,^8,9  but what is often missing is how the role of clinicians would be different in this futuristic setting. To understand that, we must first consider how today’s critical care clinician already struggles to manage an overwhelming amount of clinical data. Today, much of the critical care clinician’s time—regardless of their role—is spent generating and synthesizing data. We collect and compile data from myriad sources; the electronic medical record (EMR) houses laboratory data, radiology images, notes from other clinicians, medication lists, and more. Bedside devices provide their own data, and if they contribute data to the EMR at all, they rarely do so instantaneously; the best way to know the patient’s ventilator settings is still to go and look at the ventilator. The same goes for intravenous medication and tube feeding pumps, dialysis machines, circulatory assistance devices, and certain external monitoring devices, among other technologies. Meanwhile, data on body fluid outputs—from urine to fluids from various drains—are generally measured manually. The patient and family provide important data through physical manifestations as well as communicating symptoms and concerns. If collected at all, these data are often recorded as unstructured text in clinician notes. Altogether, patients already generate considerably more than 1300 data points a day in the modern ICU¹⁰ ; that number will only increase with the deployment of more devices providing continuous monitoring. We do the best we can, collecting, reading, interpreting, and acting upon all the data—a task that feels Sisyphean even under the best of circumstances and is impossible under the time and staffing constraints posed by modern care. More data points pose an increasing challenge for clinicians, especially in an environment where one patient’s signal is another’s noise. How can we possibly manage and interpret all of this information on our own?

Artificial intelligence will likely do it better,^11–13  and although the technology may be unfamiliar to bedside clinicians, cautious optimism is better than blind fear.¹⁴  Simply put, AI can often do more than the human brain with a given set of data; when AI “reads” an electrocardiogram, it is using every millisecond of data, not simply algorithmically scanning for the specific patterns that we are trained to look for.¹⁵  Beyond collecting and synthesizing more information more quickly than any human, AI can emphasize signals buried in the noise that a human cannot detect, making it easier for clinicians to respond early and appropriately to subtle clinical changes or deteriorations. This concept is familiar to clinicians practicing in facilities using EHR-based early warning systems that scan vast amounts of patient data to identify harm as it is occurring.¹⁶  But the systems in place now are primitive compared to what could be achieved by collecting more data and using AI to its full potential.

Of course, AI tools are fraught with uncertainty, and there are many rightful concerns about reliability, patient safety, and equity.¹⁷  We understand that these tools have the potential to not only carry our own biases forward, but enhance them.¹⁸  Although it is quick and easy to get answers from a chatbot, building clinical protocols optimized to be able to handle a diverse set of patients and implementing them is much harder.^13,19  Moreover, in many cases, building these tools requires a “gold standard” to train AI models—but such gold standards are often lacking in critical care. As George Washington University intensivist Dr Gutierrez wrote, “The greatest challenge when creating a clinical machine learning model lies in identifying the gold standard to be used in the classification. A great deal of what we see and do in medicine is highly subjective, and unanimity of opinion is seldom found among intensivists.”²⁰  Without safeguards, it is easy to imagine a patient suffering from avoidable harm or even death due to these problems.

This brings us to the essential role of the ICU clinician in the next decade. The amount of data generated at the bedside will undoubtedly increase above its current level, which is already unmanageable by human clinicians alone. To achieve the benefits these data can provide, however, bedside clinicians will need to be heavily involved in the development of bold new AI tools. It is also essential that bedside clinicians be involved in the deployment and improvement processes, just as we needed to be present in the transition to EMRs. Bedside clinicians—especially ICU nurses—are the sentinels of patient safety; this role will only become more important as AI has an increasing presence in the unit. Today’s generation of ICU nurses will be of critical importance in taking ICUs across the bridge from today into the ICU of the future with AI. They will know best which new AI tools are working, which ones are not, which ones are problematic, and which ones should be developed to fill in gaps. The most valuable contribution of the bedside nurse is and will still be that nurse’s attention on the patient; managing an intravenous pump or continuous venovenous hemofiltration machine will become mundane, automated tasks that don’t require your level of skill, knowledge, and heart. Instead, you will watch over the device-patient interface, which continues to expand—and you will be responsible for recognizing when the supposedly autonomous systems fail. In commercial aviation, for example, the advent of autopilot features was problematic during design and implementation but ultimately succeeded, transforming the human pilot’s role without diminishing it in the least.^21,22 

The advent of AI offers many opportunities, but it will come with immense challenges that can be solved only by critical care clinicians, who must be part of the conception, design, implementation, and improvement of AI tools at every step. We have spent decades working with invasive, automated, life-sustaining technology and trusting it, from cardiac telemetry monitors to mechanical ventilators and ventricular assistance devices; the fundamental challenge that digital technology faces when it meets physical reality has not yet been completely met. Technology still often requires a human “system supervisor”²²  to see the big picture and keep the machine running properly. The nature of our jobs as clinicians will surely transform, but our roles will not. More than ever, the critical care clinicians of the future will need to be competent, knowledgeable, empathetic—we will need to be human—to best leverage and manage modern technologies while maintaining empathy. After all, there is more at risk than even the lives of our patients—our own humanity is also at stake.