The Key to Clinical and Financial Success? Atrius Health President Gives Out His Recipe

Atrius Health, the Newton, Mass.-based health system that has long been a pioneer in the value-based care movement, recently reported a strong financial performance for 2017, as it finished the year with a $24.4 million operating surplus—representing a turnaround of approximately $56 million from 2016.

Officials noted that a critical success factor for the health system, inclusive of more than 30 clinical locations in the state, was innovating in an array of different ways, including: increasing the use of predictive analytics to support population health management, moving care to lower acuity sites (ambulatory, home, and virtual), and working to improve access to care for high-risk behavioral health patients.

Looking at the revenue success Atrius has been able to achieve recently, what are some of the primary factors you would give credit to?

We have managed our expenses pretty aggressively, and there are three [sub]-areas within that: in the first category we have managed our operational costs aggressively; in the second bucket, we have managed our medical expense trends aggressively; and then targeted growth is the third bucket. And the underpinning of managing our expenses has been in the analytics space.

For example, we historically might review every one of our patients [via] a case manager. We [now] can develop analytical algorithms that would mainly identify our sickest patients, so rather than have to review everyone, we could review the top 1 percent, or 5 percent, of our sickest patients. And then we have interventions and plans, which helps us with medical expense management. We could then re-engineer the case management department, because we wouldn’t need as many nurse case managers since we would be doing less work. So the management of the medical expenses and our operating costs have to a large extent been built on the back of using analytics. 

The transition to value-based care has been difficult for all health systems, but Atrius has certainly been at the forefront of some key initiatives, such as your ACO (accountable care organization) models. What have been the core experiences and takeaways for you, in this shift to value?

We have a little bit of a different history [than most], based on our DNA. We started as a staff model HMO, so we have been doing population health before it was described as population health. We have organized all of our care and systems around individual patients, and then populations of patients, meaning we have developed disease registries, and identified those people who sit in the high-risk categories—either the top 1 percent, 5 percent, or the rising risk categories. Then have planned tailored interventions, including trying to manage and prevent [disease] for the general population. We have built most of this work around our Epic EHR [electronic health record], so there are a series of tools—be it order sets or reminders—that complements the analytics. And we have built the analytics into the EHR so it’s now analytics that are much more actionable.

One example is something we have built for an adult population, a model called CRISPI, our clinical risk predictor, which is based on about 60 variables [to predict a patient’s risk of hospitalization]. If you are a CRISPI patient and you call in, there is a purple flag across the top of the record, and anyone who sees the purple flag is told to [tell the patient to come right in], or if you are over 65-years-old, we send someone to your home if you feel that you can’t come in.

That action has helped us reduce our ED utilization to the lowest in the state. We have built a comparable CRISPI model for pediatrics, called “CRISPI Junior,” and that has allowed us to deploy a pediatric care facilitator to help us manage down our costs and improve patient outcomes.

The lessons learned are that you need: an EHR; a data repository that you can use to do analytics on; and the algorithms to layer in that will then feather back into the workflow, so at the point of care you can manage patients, and/or you can triage the work to case managers, population health managers, and healthcare facilitators.

We also have a large visiting nurse association, called VNA Care, and we coordinate with them. We register our sickest patients, our CRISPI patients, in something called the Care in Place program—a nursing outreach program, so in the event someone gets sick, we deploy a nurse or nurse practitioner to his or her home. That has helped tremendously in reducing ED visits. So we identify those CRISPI patients, get them registered in these programs, and then try to manage down total medical expenses by changing the way we provide care and also the location of care.

We have set up a hospitalization-at-home program as well, called Medically Home, in which we have essentially set up a hospital by virtue of putting a patch on a patient capable of doing the biometrics, and then [installing] an iPad and a phone with direct access to mission control. The idea here is to basically manage patients in their homes, with our VNA Care [providers] caring for these patients.

From an innovation standpoint, what other health IT products or services have been most important and beneficial for your organization?

Working with Johns Hopkins, we have identified indicators of frailty from free-text clinical notes. So this is for people at risk for falls, isolation, and things of that sort. We have also been using NLP [natural language processing] to find and then close care gaps, so if you have an underlying clinical condition such as heart failure, it is best practice to do a certain number of things and be on certain medications. We have used NLP to identify patients who are not getting those medications or treatments, and then we can intervene.

We have also been working with IBM Watson to develop a new product, which allows for reduced foraging in the medical record. It works by compiling a lot of clinical content for a given condition into a single screen, and then it allows you to compare to a similar cohort of patients, like your patient, to then enable you to pick the best treatment option—principally, which medications work best and the like.

NLP is exceedingly important and will be going forward. We have been using NLP to try to reduce unnecessary messages in the inbox, as we are looking to reduce physician burnout. A lot of burnout is due to piling on work through the EHR, so we have focused on the inbox in particular to reduce those unnecessary messages.

Another thing in pilot mode is that we are working with healthfinch to do automated medication refills. To a large extent, that will use a bot inside the EHR to do the things that would otherwise be done manually—checking for labs, for drug interactions, for the occurrence of your last visits, and making sure your dosages are correct, for example—and then making it possible in one review to click “refill” once the patient gives you the pharmacy to send it to. And that will reduce inbox messages by about 15 percent when we are at scale.

Reducing physician burnout is a major theme in healthcare these days. How can addressing this issue help change the organizational culture?

Reducing burnout, or what we call returning meaning and joy to the practice of medicine, is one of our top strategic priorities. And it’s a top priority because unless you have a happy and activated workforce, your patient care suffers. There is a reasonable amount of literature that suggests if your doctors are burnt out, the quality of care suffers and the risk of an adverse event goes up. There is a cost to burnout, in early retirements and turnover, which is quite significant.

We have created a whole department to focus on this; we have a three-part plan, following the framework of the Stanford University professional fulfillment model, which addresses the practice’s efficiency, the organization’s commitment to wellness, and then personal resilience. We have a body of work going on in all those buckets, with the biggest emphasis on improving practice efficiency.

There has been a re-training of our staff to have them work at the top of their license so the physician doesn’t have to do everything. The EHR has changed the workflow, so everything goes directly to the doctor now, as opposed to it going to many team members, which it did historically. We are trying to get back to team-based care and have the record triage messages so that only the work that needs to get to the physician gets there.  

It’s a complicated manner and it has to do with staff training, technology, and workflow. Getting to “joy perfection” is a long road. We have been on this path for three years, and I think we have done more than most, but we have not solved the problem. We have probably reduced 60 million clicks, but it’s still not enough. There is too much of an administrative need currently, and we are trying to work with regulators to reduce that demand, but it’s a heavy lift.

What are a few core pieces of advice you can offer CIOs, CMIOs and other clinical informatics leaders as they continue to forge ahead into this new world?

You need to have a great data repository, and you may need to assess the skillset of the people working in the organization. Data scientists become important and helpful. You need to be able to create disease registries and identify people, so you can plan interventions to improve outcomes and lower costs. It cannot be a “hint-and-hope” strategy; it has to be a structured approach.

A lot of this is built on an important data foundation, so you need to have the infrastructure that is the data repository, the tools to extract the data, and the mechanism to push it out to a group of people who can take action on it.