Curtis Kennedy, MD, PHD, a practicing physician at Texas Children’s Hospital wants to ignite a conversation around the future of healthcare based on real world needs:

I am a physician that cares for children with severe, life-threatening conditions. I work in a Paediatric Intensive Care Unit (PICU), where my job is helping children survive when their bodies are being ravaged by disease, and new complications are routinely arising – from the diseases afflicting them and the therapies I employ to treat them.

It is critical for me to recognize when a patient’s condition is worsening, because early detection and appropriate intervention increases their chance of survival. It also reduces the magnitude of symptoms and shortens the time they must spend in the hospital.

In addition, I am expected to teach residents and fellows (residents subspecializing in Paediatric Intensive Care) how to interview parents, examine patients, think about possible problems, and develop a plan to diagnose and treat disease.

I must communicate with parents, so they understand what is wrong with their child, what we are doing to treat their diseases and seek their informed input as to which strategic approach and plan of care best fits their families’ values – while complying with legal and ethical standards.

I need to manage patient flow into and out of the PICU and prioritize the utilization of scarce resources – weighing the risks and benefits of each possibility.

These other jobs demand time and attention. They distract me from focusing on my patients, as each patient also distracts me from my other patients. A distraction at the wrong time could have disastrous implications for the fragile children I care for.

I want to paint a vision of what I believe are realistic, achievable goals for hospital monitoring technology to help physicians like me improve care in the future.

Monitoring, fundamentally, is the process of repeating observations on something of interest, so that its status is known, and changes can be detected. We monitor faces when we talk to people, and water while we are waiting for it to boil.

In the hospital, “monitoring” typically means the measurement of vital signs: heart rate, blood pressure, respiratory rate, temperature, and often, oxygen saturation of the blood.

But in reality, it is more than that because monitoring involves two parts: the interesting and the boring.

Texas Children's physician ignites talk on AI monitoring healthcare

Texas Children’s Hospital, where Curtis Kennedy practices.

Interesting things have our attention: we think about them; expect to see change; and can create stories about them. Boring things are ignored: we do not expect to see change, so rarely look for it. This behavior is human nature, and it is logical and efficient.

When boring things change, it is usually unexpected, and it often becomes extremely interesting. Unfortunately, in the hospital setting, this often means a risk of death (like a “code event”) in a patient that was “stable”.

Consider the case of the physical examination. Technically, when caregivers examine a patient, they are also monitoring them – but that action is less frequently interpreted in that context. Why? There are multiple reasons, but one that applies here is that a (complete) physical exam is comprehensive: the whole body is examined (checking in ears), even if the problem is focused (stepped on a nail).

While it is useful in narrowing possible diagnoses when there is uncertainty, and in screening for and detecting unrelated problems, repeated observations of “normal” are inefficient and uninformative.

In the PICU, however, problems are rarely focused: they are varied. Also, changes are frequent and progress rapidly, while new issues arise routinely. Currently, this is addressed in several ways:

More nursing resources are devoted to care for sicker patients. PICU nurses typically have 1-2 patients, whereas general inpatient nurses may have 4-8;

Vital signs are continuously measured and “right now” data is displayed on the bedside monitor; and multiple other measurements (labs, life-support device settings, etc.) are repeatedly taken to determine if patients are getting better or worse.

In essence, we currently employ a “more monitoring / more data” strategy to help us care for critically ill patients. However, it takes time and effort to not only document measurements, but also to find that documentation and interpret it in a context specific to the patient.

It is not uncommon for “important” details to be missed because they were being interpreted in a context of “boring” rather than “interesting.”


Routine monitoring technology has not fundamentally changed since its introduction 50 years ago: wires still connect electrodes on patients to machines that perform measurements.

In critically ill patients, however, these wires present a constant nuisance to patients and caregivers alike. They are a pain! Beyond mere nuisance, they restrict movement into and out of the bed – a fact that has a legitimate risk of leading to other complications, such as hospital acquired pneumonias and life-threatening blood clots. In young children, they also pose a strangulation risk.

With existing technology, the measurement device can be embedded within the electrode, eliminating the need for a wire connecting to a separate machine that performs the measurement. The electrodes also house the circuitry necessary to transmit the data wirelessly to the hospital’s information system.

This paradigm is likely to be the foundation for the next wave of medical monitoring devices. Two effects of this advancement are worth mentioning.

“In the not-too-distant future, we will be able to monitor a hundred more parameters, a hundred more times per day.”

Firstly, Continuous monitoring will no longer be constrained to intensive care environments. Patients will be able to be monitored per existing intensive care unit standards from regular hospital beds (or even from home).

And secondly, Parameters will not be limited to vital signs. Accelerometers will likely be embedded within the electrode, providing objective, quantitative motion data that could measure activity levels to determine disease status / response to therapy.

Even the signals themselves will be analyzed to find hidden patterns. Humans can measure a pulse rate fairly easily, but accurately quantifying beat-to-beat variability in each heart rate is something that can only be done by a machine.

Monitoring in the future is likely to routinely track such features. Furthermore, biosensors placed on the skin will continuously measure chemistries that are currently measured intermittently from blood samples, sent to a lab, and processed by expensive machines. This technology also currently exists, although its introduction into clinical use will likely lag existing monitoring technology.

Finally, monitoring patients may not require any “touch” at all. Video amplification of color and motion can already detect pulse and respiratory rate, and it is possible to measure chemicals in the air we breathe out. So, future monitoring devices may be able to learn much about our health simply by “watching” and “smelling” us.


Technology will enable more precise measurements to be taken more frequently, on an ever-increasing number of parameters. It will also reduce the amount of effort required to document measurements.

The pace of this evolution is increasing exponentially, and in the not-too-distant future, we will be able to monitor a hundred more parameters, a hundred more times per day, for the same or less cost and effort than is currently required.

Texas Children's physician ignites talk on AI monitoring healthcare

What is not keeping pace, however, is our ability to make meaningful use of all of this data. Important findings are routinely overlooked because they are buried in a sea of “normal” noise, i.e. data that are not changing and are uninformative.

We simply cannot manually process all the data our patients generate, even though we are expected to. This is one area where decision support must evolve – helping us see what needs to be seen.

In the future, monitoring will continue to produce more and more data while we humans will have reached our ceiling in being able to decipher and interpret that data.

The data will be useless to us without the assistance of decision support tools that continuously screen through it and highlight features that we need to see to help us do our job in providing the best care possible for our patients.

If my physician perspective speaks to you, whether you agree or disagree, please let me know in the comments section below. I want to make this into an interactive conversation including everyone, because it will impact everyone.

Author bio:
Curt Kennedy is a pediatric intensivist at Texas Children’s Hospital in Houston, Texas. His primary interest is in prediction modeling using time series analyses that characterize deteriorations preceding many cardiac arrests in critically ill patients. His secondary interest is in automated screening for and detection of potentially actionable problems and the prevention of avoidable harm in critically ill patients. His aim is to equip bedside caregivers with decision support tools that bring them meaningful information across multiple channels, making it easy to use in all cases and hard to miss in cases where stakes are high. Curt personally codes all facets of the decision support platform, including: automated data extraction from the clinical interface of the Epic EMR (an admitted hack, but a current necessity), regular expression parsing to translate quasi-formatted text into relational data, SQL import/ export routines, and the creation of multiple channels of decision support output, including web, email, pager, and SMS text messaging.

Ignite the discussion!

Share your comments in the box below, or email [email protected], share this post on social media, and be sure to view Curt’s work alongside other thought leaders in the AIMed Magazine.