AI in Healthcare

Introduction

Artificial intelligence (AI), machine learning (ML), predictive analytics, and the Internet of Things (IoT) are the buzzwords promising to change the face of healthcare. Their value and benefits are undeniable — and they are key to realizing the vision of patient-centered care. However, before realizing the true benefits of these technologies, hospitals need to be careful to not put the information analysis cart before the data-bearing horse.

To build a successful patient-centered program, hospitals must get access to the data feeding the model and visualizations. This is especially important in critical care environments, where

more than five million people are admitted in the U.S. annually and over 500,000 die every year

often without warning. To help reduce these numbers and save more lives, critical care teams need as much data about a patient as they can get, as fast as possible, along with the relevant context to take action.

---

Creating the foundation for patient-centered care and AI

A typical patient entering acute care will often go from the ambulance, to the emergency department, to the operating room, to the ICU, to the step-down unit. During this journey, medical devices are continuously capturing mission-critical data. That volume of data increases dramatically once these patients enter ICU environments and other critical care areas such as cardiology, neurology, and neonatology.

An average patient could be connected to eight to 10 devices, generating

more than 800,000 data points every hour.

For many patients, the clinical decisions informed by this large volume of data could alter their outcomes.

The challenge is much of the information is

not saved or stored.

When seconds count, this is a real challenge for the doctors, nurses, and care teams that have to quickly assemble the pieces of a complex puzzle to make crucial, real-time decisions.

Quality in, quality out. Generating not just more data, but better access and better visualization of that data has to be a top priority for the system to work.

The most important step in the journey to better care for our most medically complex patients and the creation of AI in Healthcare is to provide the data care teams need, accurately — and as fast as possible for optimal results. After all, real-time care decisions, analytics, and data transformation are only as good as the information that informs their models.

Overcoming Device Deficiencies

In a perfect world, data would be harnessed, analyzed, and used to drive real-time care. Patients’ information would be examined to identify patterns that signal worrisome changes or life-threatening events. All data sources would be combined to add value to other patient data to build out a more precise, personalized picture of a patient’s health. The patients would essentially be the wellspring of data to expedite care and intervention and feed the models for patient-centered AI in Healthcare.

The problem? The biomedical devices monitoring patients who are clinging to life generate some of the largest quantities of data in healthcare today. Yet these devices largely operate in isolation, disconnected from one another and the hospital IT infrastructure. It is no exaggeration to say that biomedical device data is itself in need of critical care.

---

Over-Extending the Electronic Medical Record

Acute care teams are also frustrated by the fact that patient history and trends - as well as most machine learning- and AI-based predictive analytics and early warning scores - rely on the electronic medical record (EMR) as their primary data source. EMR data has achieved major breakthroughs in care and is a prerequisite for clinical summaries, data modeling, analytics, and ultimately, precision medicine.

The challenge is, EMRs were never intended to manage the continuous real-time, high-fidelity time series waveform data needed for complete trend analysis and real-time AI. Rather, they were designed to store limited data types as snapshots at discrete intervals with low sampling rates to support reimbursement and general trending.

---

When seconds matter, wasted time may literally mean

the difference between life and death for thousands of critically ill patients.

It is not an overstatement to say that life support is also needed for data access, especially in critical care.

Complementing the need for more data and better access to data is the need to deliver data with context, directly to caregivers within their clinical workflow. Real-time surveillance with robust patient history is a crucial mandate in critical care. This will ensure that healers can intervene quickly to reduce risk and save lives.

---

Without some way to unify individual patient data, from the EMR, fused with beat-by-beat waveforms from all connected devices, care teams will continue to spend more time as detectives trying to put together the pieces of a complex puzzle than caring for their patients.

The first step in the data journey is solving these challenges of harnessing already-available data — and making it available to care teams to take action. Assuming we have solved these issues, we then have the foundation needed to add AI to that data set to further reduce risk and save more lives.

---

The key to Healthcare in AI, however, is to have a complete system that enables its deployment back into the clinical environment. Data aggregation can be accomplished, but how that data is accessed is the bigger problem to be solved.

Anyone can do a single analytic. What we really need is a single architecture; a single time series data engine; a single platform that enables vendor-agnostic patient surveillance and the ability to develop and deploy individual hospital developed analytics, at scale.

Conclusion

To successfully implement AI in healthcare and at scale, hospitals must solve the data crisis through three crucial pillars: more data, better data; data delivery with context; and open data with tools for transformation. Technology can break down silos to capture the single data point, the waveform, the beat, from the devices connected to the patient — and do it in real time.

The information is available. Harnessing it is the challenge; leveraging it is imperative.