By Allison Proffitt 

June 1, 2017 | Medical Informatics World kicked off last week in Boston with presentations on population health, telemedicine, and interoperability.

The United States gets a failing grade in population health, argued Harry Saag, medical director of the Greater New York City Practice Transformation Network with NYU Langone Medical Center, during a session on improving population health through patient stratification. The current standard of care for chronic illness management is to regularly check in every three months. “But what about the other 89 days of the year?” Saag asked.

As the payment models in the US healthcare system change, and financial outcomes are more tied to patient outcomes, no longer will one be able to be just a good clinician or a good data scientist to succeed, Saag said. Population health models depend on combining the two.

It's not an easy task. Simon Jones, a mathematician now with NYU Langone as well, tackled some of these tasks in the UK with the National Health System. Jones modeled population health and needed interventions in Birmingham, UK, and presented the findings to health officials. After show the officials slide after slide of graphs and charts, they just looked blank, Jones laughed.

Instead, the data needed to be reframed into “mood boards” visually summarizing the types of health issues common in different areas and the needed interventions. Unsurprisingly, population health needs and interventions were not aligned geographically. But by bringing both physicians and data scientists together, Jones’s team was able to bring the right education and solutions to the right groups of patients.

Jones reported even further disconnect between physicians, not just between physicians and data scientists. When developing a population health model, Jones and his colleagues found that it was very difficult to develop a model that both hospital physicians and family practitioners liked. They thought of their patients and their needs in different ways. Jones said that two groups of doctors representing both types of practice have been tasked with building their own population health model and then comparing them. But the results aren't in yet.

Emory’s No Show Solution

Haley Bolton, manager of business operations and patient access at Emory Healthcare had a different problem. Emory Healthcare serves a large population in Atlanta and the large system was struggling with missed patient appointments. First, the system instituted a $25 penalty for no-shows, but because of their existing need, didn’t apply to financial penalty to Medicaid patients.

There were still missed appointments, so the system began encouraging some overbooking. It didn't go well, though Bolton said Emory handled it far better than United Airlines. "Police didn't drag anyone out of their appointments," she said.

Bolton and her team were tasked with building a predictive data model to inform care planning and coordination.

The model considered patient age, distance from home to clinic, gender, history of missed appointments, lag time between when an appointment was scheduled and the appointment date, medical specialty, primary insurance, and the time of year.

It was complex. Initially they planned to have the patient call center handle booking based on the predictive model, but needed to turn the task over to nurses for more expertise. Doctors wanted patients’ co-morbidities to be considered as well, so that sicker patients would be prioritized. In the dashboard, doctors requested that the data be kept simple: which patients were on the schedule, how likely were they to miss an appointment, how sick were they.

The model is working. When deployed in a clinic with a high Medicaid population, the no-show rate fell by nearly 40%.

Bolton said she continues to refine the system. The team is working to add weather data to the model—“Atlanta shuts down in snow or ice,” she said—and they are working with the CDC to flesh out the co-morbidity data. Bolton is now expanding the model to predict best cadence to care, she said, rather than relying on standard three or six-month follow ups. The team is also planning to expand the model to better fill last-minute cancelations.

Alexa, Read My Medical File

Experimentation really does drive some of the best advances in healthcare, and John Halamka is in a constant state of experimentation. We fail a lot, the CIO at Beth Israel Deaconess Medical Center (BIDMC) told the audience during his keynote address, but we fail early.

Halamka and his team at BIDMC have been experimenting with moving several consumer technologies into healthcare. They've prototyped an Alexa-based patient bedside assistant, Halamka said, even adding their own code word to “wake up” the device: Ask BIDMC. From their hospital bed, patients will be able to say: “Ask BIDMC who is my doctor today?” And “Ask BIDMC what’s for lunch?”

Amazon doesn't currently sign business associate agreements (BAAs) for the Alexa platform, but they soon will, Halamka predicts. Could such a system then be used in the home? “Ask BIDMC Schedule an appointment.”

But Alexa is far from the only new product Halamka is experimenting with. He wonders if Amazon Dash-like buttons could be used to order services in the hospital. Wifi beacons could help patients find parking when they arrive at the hospital. Mobile apps can be developed for scheduling, physician questions, even to guide a family through the end-of-life process.

Halamka, who was one of the first participants in the Personal Genome Project, has no qualms with data stored in the public domain. Beth Israel hosts 7PB of patient-identified data in an Amazon data lake so far, Halamka said. He advocates for a “tamper-proof ledger” of all of your medical interactions. It wouldn’t be a true health information exchange, but at least a treating physician could look up your history and get a broad overview of which doctors in which specialties you have seen.

Halamka knows that won’t appeal to everyone; he rattled off a string of situations that would raise red flags. But he insists that machine learning has a necessary home in healthcare. Even with an electronic health record system internally (and a bespoke “care-management medical record” on top of the EHR to facilitate care throughout the system), Halamka says offices receive 18 inches of paper each day from outside the medical center.

They’ve trained Amazon machine learning to read those forms, giving the system 500 random patient forms to process. The system was more than 95% accurate, Halamka said, and it flagged any form that was troublesome to be processed by hand.

Patient Portal

At a panel on community health information exchanges, no one predicted scheduling appointments with Alexa, but the panel did lament the ways technology stands in the way of patient experience.

Patients don’t understand why healthcare is so clunky and hard to use, said Michelle Schneier of Iatric Systems. Privacy is important, but patients don’t see that; they see a horrible user experience.

Having a well-designed patient portal is my soap box, said Angie Bass, CEO of Missouri’s health information exchange, Missouri Health Connection (MHC). Patients must have access to their data! Bass recounts a story of a physician who literally duct-taped patient medical records to his patients’ chests when they left his office in an ambulance, so worried was he that their history might be missed. “That breaks my heart!” Bass said.