As the Mount Sinai Health System, a hospital network in New York City, moves rapidly into at-risk arrangements with multiple payers, technology and practices to support risk adjustment have become increasingly important because it impacts the amount of any shared savings identified.
In a recent interview, Mike Berger, vice president of population health informatics and data science at Mount Sinai, talked about getting help to gather its accountable care organization’s unstructured data and identify risk adjustment opportunities.
Berger said the health system has about 300,000 lives at risk now at various levels, from shared savings to full risk to bundled payments — the full gamut of ways providers can start taking on risk. “Our need to connect to data both within and outside our system is becoming more critical,’ he said.
Risk adjustment has been particularly challenging, he said. In the past, it was done by manual chart pulls, and the payers all offered some help in different ways.
“As an ACO, you don’t just work with one payer. We have 10 payer relationships, and each one might have a different method,” Berger said. To help the ACO do risk adjustment, some would give Mount Sinai data they pulled from claims; some would offer chart abstraction manual resources, some of which worked and some of which didn’t. Others were actually putting documentation specialists in the larger practices to do chart pulls right there to look for opportunities to do education. “There was a mish-mash of ways payers were trying to do this, but it just didn’t scale for us,” he said. “We have so many payers, we needed to control our own destiny and partner with someone who could help us both with the technology and the resourcing component to help us with risk adjustment.”
Mount Sinai turned to a solution from Health Fidelity called HF360 Data Acquisition to help gather both structured and unstructured clinical data in an EHR-agnostic manner.
Part of the challenge is educating physicians that participating in value-based care means changing the way they document and putting more detail in structured parts of the EHR. For instance, previously diagnoses and diagnosis codes were for the physicians and not for the payer. But as we move toward population health, the diagnosis code, even more than the procedure code, drives the acuity of the population, Berger explained.
Many physicians document so they can open the chart back up to review. They know the patient has had an amputation. They don’t need to document that every year. “But the reality is that the amputation is an important part of the calculation of their panel,” he said. “If their panel looks healthier than it actually is, it will impact the performance of the ACO and affect the doctor’s own top line. Tools like Health Fidelity help us highlight where we are seeing those deficiencies and doing it at scale instead of us doing it one doctor and one chart at a time, which was effective but completely inefficient.”
The Health Fidelity tool has some natural language processing (NLP) tools to help extract unstructured data from physician notes. Another tool uses optical character recognition (OCR) in what Berger calls an extremely innovative solution to a thorny interoperability problem.
“At Mount Sinai, we are a big academic medical center, but like most of America, we have an Epic system that is not necessarily easy to get data out of en masse, he said. “If you need data out of Epic, you take a ticket and get in line.”
The OCR tool goes into Epic with a basic automation script and sends things to the virtual printer that uses OCR technology. “So using a basic data science skill set is allowing them to interface all the data we need out of Epic in something of a brute force manor, but it is really quite elegant and novel,” he said. “And it works. It saved us both the time and effort of trying to go to the Epic team to try to build an interface and get it in a priority queue. We have been able to bypass that and we are now scaling it up for other EHRs. One we are almost finished with is eClinicalworks.”
Berger noted that there are impressive tools on the market that build risk adjustment documentation tools into Epic. “But because as an ACO we are much bigger than just Epic, having a tool that just works in Epic doesn’t really work for us,” he said. “Health Fidelity allows us to be more EHR-agnostic.”
An ongoing challenge involves the different kinds of technology and levels of sophistication around technology that ACO participants have. “The dream of interoperability is not here. We are probably still in the nightmare of the lack of interoperability,” Berger said. “We have 1,000 voluntary physicians participating. They might be mom-and-pop shops where the nephew does IT, or they might be federally qualified health centers or multispecialty practices that have a real budget for technology. They have a wide range of maturity of being able to get to the data. We are trying to educate our physicians about the value of better documentation, the value of risk adjustment and how it impacts the success of an ACO. Regardless of the technology, we are still climbing the wall of the learning curve.”
And risk adjustment is just one arm of population health, Berger noted. In terms of quality reporting and teeing up best practice alerts so doctors are closing out care gap opportunities, Epic gives you the tools to do all that, he said. “We have implemented several alerts, but now we are trying to figure out how to align their best-practice alerts with our ACO payer contracts so that we have good alignment and they are firing off at the right time. If someone is a commercial payer member, we want that member’s care gap alerts to fire off, not the same one that the Medicare or Medicaid patient gets.”
