Have you ever searched for something on the Internet only to see ads pop up for the same topic for days or weeks afterward? These are Google’s machine learning algorithms, hard at work memorizing your interests and optimizing the content they show you based on your online search activity. It might feel intrusive, but the mechanics behind these algorithms play an essential role in business analytics, marketing, and product development.
During the past decade or so, big data has become a big deal, and today, its implications are far reaching. In the corporate world, implementing big data and data analytics means that businesses of all sizes are able to tailor their marketing and products in ways that are most likely to drive consumer interest and sales. In the medical community, it means something else—the opportunity to shape the future of healthcare, improving therapies and outcomes as well as patient experiences.
Two leaders in healthcare data analytics, L. Nelson Sanchez-Pinto, MD, MBI, and Omar Badawi, PharmD, MPH, FCCM, will speak at the 49th Critical Care Congress on some of the future implications of big data and machine learning and what they potentially mean for healthcare research, development, and delivery. One of their key takeaways: Even though big data and machine learning may sound like brand-new concepts, their foundations should be familiar to most clinicians. At their roots, big data and data analytics are just new ways of gathering and analyzing data and then using those results to drive innovative treatments and therapies aimed at improving healthcare. As tools, the combination of big data and data analytics stands alongside clinical trials and quality improvement projects. And, as in the corporate world, in healthcare they have far-reaching and potentially transformative implications.
“Fundamentally, great advances in medicine are usually preceded by powerful observations,” Dr. Sanchez-Pinto notes. “Some of the most important advances come from observing how care is provided, learning when certain approaches work and when they don’t, and then trying to improve care based on those observations with a new approach.
“In medicine, both big data and data science help us make those powerful observations,” he says. “You can think of big data as a digital representation of healthcare, while data science is the powerful microscope used to study that data, gain insights, generate new knowledge, and make powerful observations. This is what we call a ‘learning healthcare system.’ Data science and clinical informatics tools will be a fundamental part of this learning healthcare system, especially in the ICU.”
These observations and the therapies that come from them promise improvements for almost every area of medicine. Both Drs. Sanchez-Pinto and Badawi agree that critical care stands to reap significant benefits, primarily because intensive care units (ICUs) typically gather a lot of data from the patients they serve.
“Critical care has a major advantage in data science work because ICUs are so rich in data,” Dr. Badawi notes. “ICU patients generally have high-resolution data available due to their intensive monitoring. In this population, the potential benefits are dramatic because critically ill patients are also the least stable, and they have the highest risk of mortality and complications.
“Having the capability to gather and analyze large datasets also enables clinicians to make observations on a much bigger scale, and it’s that capability that holds a lot of promise for advances in critical care medicine,” Dr. Sanchez-Pinto notes.
“We collect and digitize thousands and thousands of data points in our ICUs every day, and that data is ready to be studied,” he says. “We also usually deal with very different types of diseases and syndromes, and within those diseases and syndromes, we know that there are different subgroups or phenotypes that behave differently. If we want to study a particular phenotype of a particular disease or syndrome, we may have only seen a few hundred patients with it in our ICU over the course of several years. But if you take a big-data approach and collect and analyze data from many ICUs, you can start to generate datasets with thousands or hundreds of thousands of examples from which to learn.”
Although most researchers recognize that contributions from big data and data analytics will likely shape the way standard therapies are delivered in the future, today the use of big data in routine clinical care is still in its nascent stages. Still, expectations are optimistic, and both Drs. Sanchez-Pinto and Badawi agree that it will not be long before data-driven models find their way into regular clinical practice. Furthermore, it is very likely that these models will focus on more than standardizing therapies and the delivery of care.
“I believe data science is going to have a tremendous impact on simplifying clinician workflow and reducing documentation needs,” Dr. Badawi says. “Clinicians should be working at the top of their license as much as possible, and I anticipate the administrative burden currently facing clinicians will be dramatically reduced in the future.
“I also think many clinicians would be surprised to learn that, in recent years, an open, collaborative culture has sprung up in health data science, and it continues to grow,” he adds. “Crowdsourcing and sharing of code and data has become much more common. Publicly available ICU databases such as the MIMIC and eICU Collaborative Research Database managed by MIT’s Laboratory of Computational Physiology have literally been used by thousands of researchers across the globe. An entire ecosystem has been developed to share code in order to improve transparency and reproducibility and to reduce the burden on researchers so they don’t feel compelled to reinvent the wheel with each study.”
With all its potential for positive impact, there is one thing about data science that Dr. Sanchez-Pinto wants to make crystal clear: Artificial intelligence is not going to be taking over medical care any time soon.
“A lot of terms have been used to describe the data revolution in medicine, and all of these concepts tend to be grouped together under the umbrella term of artificial intelligence,” Dr. Sanchez-Pinto says. “But let’s be clear: The data-driven systems that are being developed are not intelligent. The intelligence is in the humans designing them and the humans interacting with them and making the actual care decisions. We are nowhere near the era when these systems will autonomously do any care. The future of medicine is about how we use data science and data-driven systems to enhance our decisions and our care; it is not about being substituted by machines.”
As for the future, Dr. Sanchez-Pinto says that, while the role of data science will continue to expand, its goal will remain squarely focused on providing more targeted care and more successful outcomes.
“Every day, we are performing natural experiments on patients. We base our decisions as much as we can on the best evidence available, but at the end of the day, when I start norepinephrine on a patient with septic shock, I’m still going to wonder if it will work right away or not. It’s a constant trial and error,” he says. “If it doesn’t work, that gives me more information and then I can try something different. This is happening all the time, millions of times per day in ICUs around the world.
“The amount of knowledge we are generating through those natural experiments is vast,” he notes. “If we don’t go back and learn from those natural experiments and make the powerful observations that someday will help us recognize that a given medication will work better on a specific type of disease or subgroup of patients, then we will be failing ourselves and those we care for. We owe it to them.”