Telemedicine and connected health have a reputation for being hard, but the 2020 pandemic has suddenly taken away all other options. In New York City, the hardest hit area so far, a hospital chain in just a single week put together a creative approach to monitoring patients at home.
The pandemic crisis struck NYC Hospitals + Health, the largest U.S. public health system, in the middle of a different telehealth project they were implementing with Lumeon, a patient communications and care coordination automation platform. I have written before about one of their collaborations, and Colin Hung has written about the COVID-19 response in another Healthcare IT Today article. Hung did not explain how an accurate model was developed, though, and I hope to fill in that gap in this article.
Predicting COVID-19 Risk: Both Easy and Hard
With the intensification of the pandemic, the hospital chain and Lumeon realized that they needed to pivot fast and come up with a way to reduce admissions of people whose cases of COVID-19 could be managed at home.
We don’t know much yet about COVID-19, but we know that the majority of victims, even if they feel absolutely awful, can recuperate on their own. A hospital would offer the less ill patients little more than hand-holding, plus monitoring them for signs that they’re getting worse. One sole symptom determines whether victims need hospitalization: shortness of breath. Wouldn’t it be great if the monitoring could be done at home?
But other characteristics of the COVID-19 scourge make home monitoring tricky. People who seem to doig OK can go downhill in just a few days to a point where they’re in serious danger. Furthermore, people with a selection of various health problems are much more in danger than people who lacked those conditions. Respiratory problems are obvious markers of risk, as are advanced age and obesity, but some other conditions add risk too.
What could Lumeon do to create a home monitoring solution? The company offers a collection of technology capabilities and design services circling around workflow. Basically, they help health care institutions rationalize their processes, ensure that they’re followed faithfully, and automate them where possible. This adds more dimensions than are offered by purely analytical models, such as one recently announced from an Israeli HMO.
Making the Model
Robbie Hughes, CEO of Lumeon, and Dr. Gajan Srikanthan, Lumeon’s Director of Clinical Pathways, recently took half an hour out of their busy schedules to brief me on the solution. Srikanthan highlighted the difficulty of remote monitoring: because systems in the past tended to look only at symptoms and vital signs, and had little knowledge of the patients’ comorbidities or clinical context. The results was a one-size-fits-all. In the current pandemic environment, with hospital staff already nearly at the breaking point, such a simplistic approach could potentially result in overhospitalization, detracting from the critical care other patients need.
Yet the solution that Lumeon and NYC Hospitals + Health worked out is currently monitoring about 500 to 600 patients, mostly people who came into the emergency room. Very few have been readmitted. The system works so well that New York City recommends it to people who call its 311 hotline to report symptoms, and soon it will be adopted by other hospitals.
Impressed as I was by this success, I found it even more amazing that the developers pulled it off with so little data to go on. We just don’t have, at this early stage, clinical models that help predict the outcome of COVID-19 infections among people with comorbidities even though Epic customers are working on one as my colleague Anne Zieger reported. All that the developers could use as input was the experiences reported by hospital staff.
They started with the Charlson Comorbidity Index, a validated and widely used model for stratifying comorbidity. The clinicians modifyied its scoring criteria in light of known risk factors in the specific context of COVID-19. The resulting model was used to calculate a customer’s baseline risk. This, plus the different combination of responses a patient could give about their breathing, enabled the system to calculate a daily risk status, which was further validated and calibrated by the hospital clinicians.
What’s happening here? How can clinicians develop an accurate model based on local experience? Whatever happened to “big data”?
This is not the first time I’ve found that health care institutions don’t need big data to do useful analytics. In fact, researchers have found that data from one hospital or neighborhood may produce incorrect results when applied to another. Because environments differ so much from place to place, as well as the way data is used, an organization sometimes does best sticking to the data it collected. Of course, when large, diverse data sets are available, useful insights can also be derived from analyzing them.
Putting into Practice
Deployment of the NYC Hospitals + Health model used pretty simple technology; the sophistication is mostly in the processes used.
Baseline risk, as described earlier, is calculated from the patient’s electronic record together with a form that the patient fills out to list their comorbidities. The patient is then prompted by SMS messages at regular intervals to report how they’re feeling (see Figure 1). The interactions can currently take place in English or Spanish, and other languages are being added. Thus, while baseline risk is calculated once, the overall risk of a poor outcome is determined day to day.
The system applies the model to decide whether a clinician should be brought in to discuss the patient’s condition with them. If someone fails to respond to a prompt, a clinician is also quickly notified and can take further steps to contact the patient.
There you have it: a robust telemedicine system developed in one week (albeit as an outgrowth of another project). Hughes and Srikanthan spent a little while with me tossing around ideas about how health care systems could develop innovative telehealth programs even in the absence of a crisis. Basically, they said, “Where there’s a will, there’s a way.” The COVID-19 crisis spurred everyone at the hospital to overcome old habits and to drop their resistance to new solutions.
So there may be several things we can learn from this case study:
- Surprisingly useful analytical models can be based on one institution’s data.
- Once you understand what data you’re looking for, you may be able to collect it using simple, low-cost technologies.
- Once everyone throughout an organization is on board, deeply grasping their need and the benefits of the solution, the organization can make tremendous progress.
Telehealth is in its infancy, but it may turn out easier to develop than most of us thought.
Note: Lumeon is a sponsor of our Pay-HIT-Forward program.