Q&A: Chief health information officer talks value of emerging role

'The paradigm of patient-requested care is over. We need to know what they need, and it’s not hard to figure it out if you have data.'
By Skip Snow
06:17 AM
Share
Businesspeople

Terri Steinberg, MD, is the chief health information officer and vice president for population health informatics at Christiana Care Health System. She was the CMIO of the organization until this year but explains that in her new role she can address the needs of those not in the facility's beds or clinics significantly better.

[See also: Emerging role: How a CHIO harnesses data for population health]

Her department is expanding. Steinberg sees medicine's future as one in which data, care coordination teams, and evidence replace the notion of the heroic doctor.

[See also: What exactly is 'population health,' anyway?]

Below she speaks frankly about the value of care coordination and simple regression algorithms, and says she's a bit skeptical of the place that advanced analytics will play in community medicine in the near future.

Steinberg credits Christiana Care's ability to execute on value-based contracts more to its excellent care coordination than to her department's breakthrough ability to understand the true complexity of the data. Here's what else she had to say during a recent conversation with Healthcare IT News.

Analytics empowers care delivery organizations to proactively manage a population's health risk

Christiana Care focuses on those people who are between venues of care, so they are not in an office, they are not in a bed, or they have so many providers that nobody is really taking responsibility. The goal is to ensure everyone has evidence-based screening examinations. And if you are healthy, you maintain your health in a variety of ways. One of those is exercise.

A chief health information officer has a new way of looking at technology systems. It is not an inpatient system. It is not an ambulatory system. It is a risk management system driven by analytics in a way that is new to organizations. This has been a gigantic learning curve for my organization: the need to focus, not just on our patients, but also on individuals who may be somebody else's patients but for whom we are at financial risk.

Consumerism is the future of healthcare.

The paradigm of patient- or person-requested care is over. I think that what we have to assume is that people are not going to step up and ask for what they need in the new world of health. We need to know what they need, and it's not hard to figure it out if you have data that crosses the boundaries of healthcare entities.

Consumers today are just not as willing, especially young consumers, to make an appointment, wait a couple of weeks, and come in to see a doctor, wait for an hour in the waiting room. That has all changed. We find them using our smart technology systems. We may tell them: Enroll in this exercise program and we want to see your heart rates directly from your machine. Or teach them: Here is how you request a telemedicine visit.

As industry matures, Christiana Care considers replacing homegrown solutions.

Cerner is expanding its role and capability in this field. Cerner has a Hadoop data structure that we are trying to figure out how we use. Currently, our master person identifier is in our homegrown system. We are thinking of migrating that to the Cerner platform. Initially, we started with all of these systems off to one side; now we are trying to align our strategy so that over the next few years we can come closer to a unified and integrated technology structure.

Predictive analytics is hard.

It turns out the more diseases you manage and the more populations you manage, the more complex it gets. It is not just magic. You can't just sort of say, "Genie in a bottle, tell me what the risk is." We have 40 different disease states, a number of different insurance plans. What we discovered is we need to identify risk items. Then we need to look at each of our clinical programs.

You look at what happened in real life, and is the machine model that you picked actually doing what you want it to do? Let's say it is a pregnancy model; you are looking to avoid preterm. You find out the women who actually did have preterm labor. Did your risk model work? It's not so easy to generalize this to real life. I consider that an unproven experiment at this point.

Interventions are the key to population management.

You don't wait until the patient initiates contact; you reach out and you are proactive with an individual.

We manage comorbidities by excellent care coordination. It is the care coordinators who are able to react to the triggers and alerts that our predictive analytic system delivers. The key is to make sure that the people who need things are brought to the attention of people who can provide things.

We are really moving away from doctors being the air traffic controllers, to multidisciplinary care coordination being the air traffic control. The patient being moved to whatever location, venue, or service is appropriate at that point in time with the goal of optimization, of function, of cost of value, of quality.

The initial spearhead of the program was picked opportunistically and blossomed quickly.

We started with ischemic heart patients, because we wrote a grant proposal for that pathology. Our Center for Medicare and Medicaid Innovation (CMMI) grant was based on a data and integration data architecture. I imagined how this would work. Then I went shopping for a clinical use case to demonstrate this data integration architecture. I picked ischemic heart disease because a well-known researcher was interested in doing it.

What should be an institution's first step?

If I had to choose, it would be to implement really excellent systems to support care coordination. The ability to sift through complicated data and find the one piece that is going to change the world is less important. We are prioritizing it lower at Christiana care than implementing excellent systems that run routine and not fancy algorithms. As to calculating risk scores, I will continue to vet them, but I consider this an unproven experiment at this point.