How human listening and machine learning are helping patients and doctors achieve better outcomes
August 16, 2019
Dr. Alan Glaseroff believes that patients who feel hopeless can improve their situations dramatically if empowered to manage their own conditions.
A Type 1 diabetic himself since 1983, his medical training had taught him several decades ago that, at best, he could expect to die young, with plenty of suffering along the way. To prove them wrong, he learned to manage the condition for himself, and soon his patients began learning alongside him.
Dr. Glaseroff built self-management programs around diet, physical activity and counselling for depression roughly eight years before the establishment accepted the value of “tight control” to improve outcomes.
Co-director of Stanford Coordinated Care (SCC) at Stanford University School of Medicine, Dr. Glaseroff focuses on patients burdened by multiple chronic diseases, so-called high-needs, high-costs patients. These patients may be just 5 percent of a normal population, but they can cost up to 50 percent of the population’s health care budget because of all the services they need.
His message is one of hope. “It’s not where you start, but where you end up that matters,” he says.
The doctor tells the story of a middle-aged patient, Theresa, to illustrate the idea that, although many patients first arrive at his clinic with barely a “shred of hope,” they almost always leave significantly improved. Theresa’s case, according to Dr. Glaseroff, was one of the most amazing turnarounds in his 40-year medical career.
Chain-smoking, full of self-loathing, vastly over-weight, and with out-of-control diabetes, Theresa told Dr. Glaseroff that the only things keeping her from killing herself were her two daughters and her religious faith.
Building patient trust and relationships by listening
Depression and unresolved trauma are common among patients with chronic conditions. Dr. Glaseroff took time to really listen to Theresa, as he does with all patients. For her, as for many others, this was the first time she had felt truly listened to by a health-care professional.
He remembers her as a tough case, frequently turning up late for visits, telling him what she thought he wanted to hear rather than the truth – that she hadn’t given up smoking as promised, for instance.
Therapy with one of the center’s specialists was key to her recovery. At just seven-years old, she had been severely shocked by the sudden death of her father, and had lived with her mother, who was herself traumatized by her husband’s death. As a result of Theresa’s traumatic childhood, it took longer than usual for Dr. Glaseroff and team to create the trust on which new healthy behaviors could be founded. When a breakthrough finally happened, it was dramatic.
“One day, she shot out of the therapy room, shouting, ‘I didn’t kill my father!’” says Dr. Glaseroff. From that day on, everything started to get better.
Theresa no longer smokes and has her diabetes and weight well under control. She describes herself as happier and healthier than ever before. Her children are studying successfully in college, and her own life is back on track, with friends and plenty of activities.
To Dr. Glaseroff’s delight, her goals go beyond being the best mother to her daughters. She’s also excited at the prospect of being ready for a new relationship after a decade of feeling unlovable.
“Focusing on patient goals, building trust by listening, not telling people what to do, but asking them what they think they should do, is what matters. And it turns out that what really drives successful work with high-cost, high-need patients is self-management,” says Dr. Glaseroff
How sophisticated health systems are using data
Dr. Gordon Moore, an old friend of Dr. Glaseroff, is a passionate believer in what he calls a “whole-person vs purely medical approach.”
Dr. Moore is also driven to help people who seem to carry the heaviest burden, and he has cared for many during his early career as a family physician. He too saw the greatest success with a focus on active listening to better understand patients and their lives.
In his role as a senior medical director at 3M Health Information Systems, Dr. Moore and his team work with health care delivery networks that are trying to improve the health of much bigger populations. Rather than listening and learning through one-on-one consultations, he helps health systems “listen at scale” – spotting patterns in the data at hand. Powerful algorithms filter “oceans of data” to uncover patient groups who are struggling – consuming significant resources but not seeing improvements in their health.
Why health systems need to understand how patients really live
“Health systems have to start seeing their patients as more than a diagnosis code,” says Dr. Moore, vehemently. In his experience, the more sophisticated the health system, the more insights they have about their patients and their real lives outside the facility.
Factors known as “social determinants of health” are vital information for health systems to gather, he explains. These include information such as people’s family support, the health and safety of their neighborhood, their cultural background, their socio-economic status and any history of depression.
Armed with a true picture of people’s vulnerabilities, health care teams can focus on programs likely to have the biggest impact on health outcomes in a community.
“Health systems have to start seeing their patients as more than a diagnosis code.”Dr. Gordon Moore, senior medical director, 3M Health Information Systems
Keeping track of discharged patients
Sending people back into the world after hospitalization is a clear danger point for patients and health systems alike. It’s important for hospital staff to have all the facts at their fingertips.
“Leaving the hospital to go back home can be a very risky, scary time for many patients. They’re often in pain, scared and feeble. If they don’t have the right support at home, there’s a high probability that their ability to manage will be overwhelmed,” explains Dr. Moore.
That’s illustrated in a study from health insurer Blue Cross® and Blue Shield® of Louisiana and The Baton Rouge Clinic (a medical corporation). They were alerted through data insights to a need to improve care for patients discharged from hospitals.
The problem was that they didn’t know enough about patients’ home lives to be able to pinpoint those who would need the most help. To fix that, they began systematically collecting more detailed information from patients when they arrived in the hospital, so they could quickly identify and categorize those most at risk.
“It was a simple closing of the loop,” Dr. Moore explains. “Discharged patients identified as at-risk were now much less likely to disappear into a void.” Transition-of-care nurses were assigned to check up on them and their status kept updated in the system.
Making difficult decisions easier with machine learning
Surgery is a field where knowing crucial facts up front can save lives. For example, if a surgical team were alerted in real time that the patient in front of them had a difficult-to-spot condition that makes them unusually vulnerable to surgical site infections, the surgeon could take extra precautions, like using very selective wound closure processes that are unnecessary for normal patients.
Dr. John Cromwell led a team that reduced post-operative surgical site infections in colorectal surgery patients by 74 percent. Insights from machine learning (computer systems that automatically ‘learn’ with experience) played a key part in this success.
Clinical Professor of Surgery at the University of Iowa, he describes his work as being on the front line – at the intersection of machine learning, natural language processing and clinical care. He sees great opportunity for machine learning as a tool to help guide very specific clinical decisions that can have huge impacts on outcomes.
He asks, for instance, “Do we really need so many blood transfusions?” He suggests instead that a more systematic approach to unmask anemia early, combining machine learning to highlight risk factors and non-invasive screening for anemia, might reduce risk and produce better outcomes than transfusions.
For now, he sees much of the “huge promise” of identifying conditions up front as being just around the corner – as we wait for hospitals to begin gathering data more consistently, in formats that allow quick and easy consolidation before we can “plug in” machine-learning tools.
Dr. Gordon Moore agrees, but sees a spark of hope in an innovative Natural Language Processing (NLP) research program at Massachusetts General Hospital.
“When someone comes into a hospital in a bad condition, there’s rarely enough time to find out much about them beyond what’s in their standard medical records,” he explains.
With the help of NLP technology, researchers showed that data from standard medical record fields can be quickly analyzed, together with unstructured, free-text notes about potentially critical psycho-social risk factors. Forewarned, medical teams could prepare more effectively, and struggling patients might discover for the first time that they are eligible for enrollment in a care coordination program that could really help.
Advances in both behavioral and computer sciences are significantly improving our ability to know more about people and their lives outside the health care system. This is already helping to inform where and how to provide more effective health services to help people like Theresa who are struggling most.
“We’re seeing big steps toward seeing patients as more than a diagnosis code,” says Dr. Moore.