How OurHealth pinpointed $150 million in ER savings by using Indiana Medicaid data

The organization won a HIMSS Indiana Chapter Hackathon, finding where ER visits could have been treated in the primary care setting and offering guidance for all healthcare organizations.
By Bill Siwicki
11:56 AM
OurHealth uses Medicaid data

Indiana partnerships present data and insights to work on better health outcomes in the state at the MidWest Fall Tech Conference. Credit: Twitter

OurHealth, a provider of employer-based clinic solutions including onsite and near-site primary care clinics, identified more than $150 million worth of savings from non-emergent and potentially primary care treatable visits in a study of 25 aggregate, de-identified Indiana Medicaid data sets going back several years.

In doing so, it won a 2017 hackathon sponsored by the Indiana Chapter of HIMSS. And it showed how careful use of analytics can look to the past to point the way forward in improving care and trimming costs. OurHealth also provided a place to start to deep dive for millions of dollars in potential savings and, more important, work toward a better patient experience for the Medicaid population, Norris added.

[Also: Clinical analytics market set for big growth worldwide]

“The goal of the event was to engage the data science and healthcare community to get value out of this data release immediately,” said Brian Norris, vice president of data and analytics at OurHealth. 

To that end, OurHealth looked at the broad set of Indiana data released and, given that emergency department utilization is high in many populations, the team set out to understand if there was any potential opportunity within the Medicaid data set.

“If you think of the typical emergency department experience, a person can spend hours going through the process, which leads to a ton of lost productivity and a much higher cost of care,” Norris said. “We started with a question, ‘How much of the emergency department spend might be avoidable? Then leveraged a number of data sets and external probabilities from a New York University algorithm to answer this question.”

The millions of dollars that were spent on abdominal pain are one example. What if those patients had someone they could call and get an opinion prior to going to the emergency department? Would they have gone to the hospital? Likely not, in most cases, Norris contended.

In the patient populations served by OurHealth – those self-insured by their employer – avoiding the emergency department creates huge savings. In these populations, a trip to the emergency department could cost the patient five to 10 times more than what a visit to a primary care provider would cost. And had they visited an OurHealth onsite or near-site clinic, that would likely have been free or cost very little.

Regarding the algorithm, in 2000, New York University developed an Emergency Department Profiling Algorithm that sought to answer the following questions among New York’s Medicaid patients: “What proportion of emergency department cases could be treated in a primary care setting?” and “How much emergent emergency department use is preventable or avoidable with timely and effective primary care?”

This algorithm classifies emergency department visits into four groups: non-emergent; emergent/primary care treatable; emergent, emergency department care needed, preventable/avoidable; and emergent, emergency department care needed, not preventable/avoidable.

“We sought to understand how these groups may be prevalent within the released emergency department data sets,” Norris explained. “We mapped the diagnosis codes across these data sets, with some data clean up, to the NYU probabilities that the diagnosis fell within each of the four groups. Once mapped, we further mapped the diagnosis codes to the Agency for Healthcare Research and Quality Clinical Classification System Codes for both ICD-9 and ICD-10, allowing us to group the diagnosis codes into clinically relevant groups in our data visualization.”

Once the OurHealth team had the mapping completed, it began its data exploration. First, it sought to answer the question, “How much spend and claims volume fall into each of these categories?” To do this, the team assigned the spend and claims volume for a diagnosis to the category which had greater than 50 percent probability that it was associated with one of the four specified groups above that group. The team identified more than $150 million worth of non-emergent and potentially primary care treatable visits.

“If even half of this was correct, based on a deeper analysis of claims, it posed a significant savings opportunity,” Norris said.

Norris offers advice to healthcare provider organizations seeking such savings and considering working deeper with the data they have.

“My advice is to start with, ‘What’s the question?,’” he said. “Many times, initiatives start with the endgame in mind and many healthcare organizations start with that in mind. Because of this, they spend millions of dollars more than they need to and much more time answering simple questions that could have a huge impact on healthcare cost reduction, or quality improvement, or both.”

It’s very important to start with the right question and work one’s way from there. As a former ICU nurse, Norris said he did this in practice intuitively.

“I believe this can be brought to the approach in analytics within an organization,” he said. This includes a blend of clinical and claims data at fine levels with a high degree of interoperability between the systems needed to deliver a high-quality patient experience. Health IT plays a significant role in this; however, technology should help answer questions and be part of the overall solution.

“Many times, healthcare organizations get in a cycle of buying everything on the shelf and end up with solutions looking for problems,” he said. “If organizations start with the question and work their way to the solution, these organizations will find success far more often than the former situation.”

Twitter: @SiwickiHealthIT
Email the writer: