With 720,000 cases annually in the U.S., and a staggering mortality rate between 25 – 50%, sepsis isn’t just life-threatening; it doubles as one of the country’s most expensive inpatient conditions, consuming more than USD 27 billion annually. What’s worse is that COVID-19, like other infections, can lead to sepsis – threatening to tip overwhelmed ICUs over the edge.

Working with IBM’s Data Science and AI Elite team, organizations such as Geisinger Health System have made tremendous leaps forward using inpatient clinical data to build models to predict – and prevent – sepsis mortality. Identifying which sepsis patients are at greatest risk can help providers prioritize care – and stave off risky, costly inpatient admissions.

With the increasing urgency facing today’s healthcare institutions, there’s more ground to cover. At Pittsburgh-based Highmark Health, the second-largest integrated healthcare delivery network in the country, a team of data scientists and researchers realized they could build a model from a source of patient information that might prove even more effective in time-critical cases: insurance claims data.

Unify the tools, processes and talent required for enterprise data and AI.

Objective: Develop a model and easily integrate and deploy insights into the existing clinical services application to identify high risk patients for sepsis and COVID-19 based on claims data.

It was an unexplored area for model building – a first-of-a-kind pursuit which, as promising as it sounded, would require Highmark to predict acute events months in advance using claims data from millions of members across multiple siloed data sources.

Brittany Bogle, IBM Senior Data Scientist and healthcare lead had significant expertise in similar data science engagements with other U.S. healthcare providers, so she knew the Highmark scenario well. But this time around the team had a new, integrated platform at their disposal that could handle Highmark’s complex and varied data sets – and even better – unite data scientists, architects and engineers who were collaborating on this first-of-a-kind project.

That new platform was IBM Cloud Pak® for Data with components for data modernization, DataOps and AI lifecycle automation including:

• IBM Data Virtualization
• IBM Watson Knowledge Catalog
• IBM Watson Studio, an enabler of ModelOps
• Explainable AI and Model Monitoring in Cloud Pak for Data

In a six-week proof of concept, Curren Katz, Highmark’s Director of Data Science R&D, teamed with Bogle and IBM to build a model, then score and identify patients likely to develop sepsis. The goal was to work within a three-month window for ingesting the claims data, giving clinical management teams time to develop action plans for intervention and hopefully, keep patients at highest risk out of the hospital.

“When we were building this, other people in the company heard about it and were talking about stories of people they knew and friends and relatives – so we really thought we had hit on a very important topic,” said Katz.

While Katz and her team were no strangers to building models, getting to the deployment stage caused some angst among Highmark’s most senior data scientists. Previously the organization’s architecture made the work cumbersome and clunky – stretching out for months, even up to a year. But with the new platform taking care of the heavy lifting, the IBM team turned over a deployed model in only a few short days. “The (IBM) data science elite team wanted to show me that this was possible and that I could tell our stakeholders across the company that we were going to have this model ready to deploy and ready to go into the clinical systems,” said Katz. “We wanted our care managers, nurses, and doctors to be able to access the findings and incorporate that into their work and reach out to patients. I think it was within a couple of days that IBM came back with a deployed model and I was kind of shocked.”

Katz and Bogle agreed that the early skepticism about tackling some of the biggest problems in healthcare quickly dissolved as the new platform enabled swift model deployment. The newly launched platform gives Katz the power to scoop up new research findings and contributors as COVID-19 evolves, changes and unveils new data.

“And that’s what this felt like: A platform where we can draw on all of the expertise in our company and build solutions that get ahead of problems, that give us insights into the future that we can act on,” said Katz. “That’s how we’re going to free people to be their best. And I think that’s where healthcare overall is really going forward: Keeping people healthy and being a partner in doing that.”

Advantages:

• Eliminates data silos.
• Provides trusted data source and reduces data preparation by cataloguing all the attributes in one place
• Integrates insights into the application workflow.
• Enables monitoring of insights for bias, trust and transparency.
• Reduced Highmark’s AI development and deployment lifecycle from 12 months to six weeks.

Unify the tools, processes and talent required for enterprise data and AI.