The humans are still in charge, but the impact of AI and machine learning in healthcare is expanding. It’s being used in a variety of ways from the bench to the bedside — and even in the home. Here’s a look at a few of the uses.
Predictive analytics may be the most pervasive current use of machine learning and AI. While it’s still in its earliest stages, it enables doctors and clinicians to focus on providing better service and patient care by allowing them to intervene before a patient gets sick — or becomes sicker. It’s not just for physical health, however.
Beacon Health Options, a behavioral health management provider. Its clinicians use machine-learning tools to extract actionable insights from structured and unstructured patient data. “Our goal is to move from being a reactive model that solely looks at what has happened historically to being a much more predictive, proactive and targeted service provider,” Dr. Emma Stanton, associate chief medical officer, told Health IT Analytics.
Thanks to electronic health record (EHR) systems, healthcare organizations have a wealth of data. However, managing all this data — and extracting actionable insights — remains a challenge for all the players — payors, clinicians, provider organizations — and administrators.
But of course, raw data is just the starting point. There’s the challenge of providing context. TechEmergence illustrated this with an example of using machine learning to determine treatment for people with first- or second-degree burns: Based on the data, the machine may predict that second-degree burn victims will need only as much recovery time as those with first-degree burns. The problem: The machine doesn’t know that second-degree burn patients receive faster and more intensive care. Because it had no context, it assumed second- degree burns heal at the same rate as first-degree ones.
So once you have the data, the context and the machine-learning tools, you are good to go, right? Not really.
The point of machine learning in healthcare is, of course, to have actionable insights. That means you need to know what to do with the insights you glean. Lillian Dittrick, Fellow of the Society of Actuaries, offered this warning in Health IT Analytics: “Organizations will often develop great analytics solutions, but they have trouble communicating that information in an effective manner.”
Now let’s get down to specific examples of machine learning for healthcare applications
Tackling Diabetes and Heart Disease
Researchers from Boston University received a $900,000 grant from the National Science Foundation to develop and pilot a system to identify patients at risk of heart disease or diabetes. They will use “novel mathematical methods and the requisite algorithms” to predict the likelihood of hospitalization, readmission, worsening of the disease, etc. IT will use both EHR and real-time health data (from wearable, implantable and home-based diagnostic devices).
Clinicians use the predictions to provide personalized interventions. These could include increased monitoring, a change in the medication regime, updated care plans and an array of other options.
“Personalized, predictive healthcare is the future of medicine,” said Dr. Rebecca Mishuris, assistant professor of medicine at BU.“Our research is geared toward providing clinicians with powerful, interpretable data to achieve that goal. Physicians are drowning in data and administrative processes. Our research approach will help physicians manage the deluge of clinical and patient data to make decisions in a more systematic fashion.”
Making the ICU Smarter
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is taking machine learning to the ICU. Researchers described “ICU Intervene” in a paper presented at the 2017 Machine Learning for Healthcare Conference in Boston.
“The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment,” says Ph.D. candidate Harini Suresh, lead author on the paper. “The goal is to leverage data from medical records to improve health care and predict actionable interventions.”
ICU Intervene uses ICU data (everything from vital signs to patient demographics) to determine what kinds of treatments are needed for different symptoms. In short, the system uses deep learning to make real-time predictions. It also explains the reasoning behind those decisions. This allows the clinicians to understand — and if needed, overrule — machine-made decisions.
It “trained” using data from the MIT-developed MIMIC database, which includes de-identified data from roughly 40,000 critical care patients, according to MIT.
“Deep neural-network-based predictive models in medicine are often criticized for their black-box nature,” says Nigam Shah, an associate professor of medicine at Stanford University who was not involved in the paper. “However, these authors predict the start and end of medical interventions with high accuracy, and are able to demonstrate interpretability for the predictions they make.”
Intelligent Robotic Caregivers
The population is aging and we’re living longer; that’s going to put a tremendous strain on the healthcare system. And this “silver tsunami” is bringing with it an increase in diseases such as Alzheimer’s.
Enter the robot.
You may already be familiar with the use of robot companions and caregivers in Japan; the market is still small globally. Many of the robots are more companion than a caregiver, such as Hasbro’s robotic Joy for All Companion Pets. These pets will soon have a brain, thanks to a partnership with Brown University.
ARIES (Affordable Robotic Intelligence for Elderly Support) will add artificial intelligence capabilities to the pets. In addition to being companions, they could help older adults with simple tasks — for example, medication reminders or finding lost objects. This will be especially helpful to those with memory loss and/or mild dementia, according to Brown.
Moving further away from that uncanny valley is Denver University’s Ryan, an AI companion robot being developed by Associate Professor Mohammad Mahoor and students at the Ritchie School of Engineering and Computer Science. Ryan doesn’t really look completely human, but it’s getting there.
Ryan was created to help the elderly, especially those with dementia, but Mahoor notes that children with autism and those with mental disabilities may also benefit. Ryan can recognize who it interacts with and carry on conversations.
“Ryan can read people’s emotions through their facial expressions and then mimic it back,” Mahoor explains. “Ryan is an empathic robot, meaning that empathy is part of her character to support people socially and emotionally.”
Researchers plan to take blood samples of individual users to determine if the robot affects biomarkers of Alzheimer’s disease, HealthTech Magazine reports.
And a few more to wrap up
- Medication adherence: AiCure uses a tablet or smartphone’s camera to observe the patient taking medications. It’s not just a video for a human to review: An AI system analyzes the data from the camera to make sure the patient took the appropriate medication and took it correctly A small peer-reviewed study indicates it led to a 50 percent improvement in adherence to oral anticoagulants (blood thinners), MobiHealthNews reports.
- Drug discovery: AI and machine learning in healthcare can streamline and accelerate the drug-discovery process. Machine-learning algorithms can analyze large datasets from clinical trials, EHRs, genetic profiles, and other sources and then identify patterns and trends and develop hypotheses much more quickly than human researchers, according to a report from Deloitte. “For example, using the vast array of information on drug compounds and test results that exist within biopharma, AI has the potential to assess this information rapidly, in conjunction with researchers, to help facilitate drug discovery and develop new insights faster.”
- Earlier detection of cancer: Deep-learning medical diagnosis may lead to earlier detection of cancer, including pancreatic cancer, which has only a 7 percent five-year survival rate. Johns Hopkins is working on the latter. It’s training deep-learning algorithms to spot changes to tissue of the pancreas and nearby organs on a CT scan, reports Healthcare IT News.
- Machine learning in life insurance: Machine learning and AI can use data mined from various sources to see a fuller picture of their customers — without human bias. But on the other hand, based on this description from The Innovation Enterprise, it may be a little Big Brotherish. “Data from wearable devices such as Fitbit can track a customer’s activity, while analysis of their social media may give a more accurate idea of somebody’s lifestyle choices than they are willing to share. This will likely punish those who are unhealthier than they say, but it will also reward those who live healthier lifestyles.”
We’ve just skimmed the surface; machine learning in healthcare is poised to transform medicine. If you’re interested in learning how machine learning and AI can transform your business, we can help.