In Oklahoma, a Regional HMO Deploys Predictive Analytics to Drive Better Health Outcomes

Oct. 4, 2016
GlobalHealth, an Oklahoma City-based HMO, implemented a predictive analytics platform combined with human outreach to improve its care management program, resulting in an 18 percent reduction in ER encounters and a 22 percent reduction in readmissions.

Two years ago, GlobalHealth, an Oklahoma City-based health maintenance organization (HMO), began a proactive outreach program utilizing predictive and prescriptive analytics as part of its care management efforts. The goal was to identify plan members whose health was most likely to change for the worse in the next 12 months in order to intervene to reduce medical costs and improve health and wellness for plan members.

GlobalHealth is a regional HMO that covers more than 45,000 individuals in all 77 Oklahoma counties and its membership includes state and education employees, federal employees, municipal employees, Medicare Advantage members and private employers.

According to a case study released by GlobalHealth, the organization has seen notable results since implementing a predictive analytics platform from VitreosHealth as an Insights-as-a-Service (IaaS) delivery model for population risk models for predictive and prescriptive health data insights. GlobalHealth essentially combines the data insights with human outreach to better understand its members’ needs.

Since the initiative began in early 2014, GlobalHealth has experienced an 18 percent reduction in emergency room encounters and emergent hospital admissions among the target population as well as a 22 percent reduction in readmissions. GlobalHealth also has seen per-member, per-month (PMPM) medical costs for that target population reduced by 16 percent, and, more generally, spread across all members, there has been a 6 to 8 percent reduction in PMPM medical costs. In 2015, the organization realized $10 million annual savings, and GlobalHealth can now predict nearly 70 percent of its hospital admissions, according to David Thompson, GlobalHealth chief operating officer.

Back in 2013, GlobalHealth executive leaders recognized the need to utilize predictive analytics technology and prescriptive health insights to identify at-risk members and strengthen its care management program in order to provide better health outcomes for its members.

Thompson specifically recalls a meeting three years ago with staff members and executives discussing case management rounds that proved to be the impetus for the analytics initiative. During the meeting, the discussion turned to a plan member who suffered a recent diabetic coma. “The member was a relatively healthy, young person who did have diabetes and other conditions. So we asked ourselves if there was anything we could have done to prevent this. The directors of case management said yes, we could have prevented it, if we had done this or that at this point in time, so we went through the case in reverse chronology to piece together the patterns of that case and we realized it would be helpful to have all this data in one central, manageable place. And that was the lightbulb moment when we decided we needed an analytics engine that could help predict these types of events. In the past few years we have been working on building predictive modeling into our care management techniques.”

Thompson continues, “Because we are a small plan, we had a lack of available, combined sophisticated data sets and we were challenged with identifying members for these negative health events, or what we considered to be avoidable, manageable events, such as emergent hospitalizations or readmissions. We just didn’t have great data to in order to intervene to prevent that. We’d always been good at the point of service and managing care when a member is hospitalized and transitioning them back home and identifying the community resources around them. But we were not good at predicting the diabetic comas or the members who had co-morbid, complex conditions which often manifests into a future unhealthy event.”

At the same time, Oklahoma also is a state with significant health concerns as it is consistently ranked near the bottom of all 50 states based on the healthiness of its population. Oklahoma’s adult obesity rate is 33 percent, which is the sixth highest rate in the nation, according to data from Trust for America’s Health and the Robert Wood Johnson Foundation. Nearly 22 percent of adult Oklahomans reported having a mental health issue and close to 10 percent experienced a substance abuse issue, according to Mental Health America. And, among Oklahoma’s adult population, 24 percent are smokers, compared to the national average of 19 percent, according to 2015 data from the Centers for Disease Control and Prevention.

Thompson says for organizational leaders the first step on this journey was to examine options for building an analytics program. The organization’s internal information technology team believed it could develop the necessary data infrastructure, however, it was estimated it would take two years to build. So, GlobalHealth executive leaders reviewed the options offered by external vendors.

One of GlobalHealth’s priorities was finding an analytics partner with a solution that could seamlessly integrate and evolve with the organization’s existing IT structure.

“We needed a partner who could do several things. One, have a strong data architecture and legitimate interfacing capabilities,” Thompson says. “A lot of companies say they are doing predictive modeling, but they’re not doing anything that couldn’t be done in house, as it’s still reactive and closing care gaps, but it doesn’t impact the full spectrum. From a vendor standpoint, we wanted a partner who was willing to customize and be flexible, and was on the bleeding edge.”

According to Scott Vaughn, president and CEO of GlobalHealth, while most health plans look at the members who have generated the most costs historically and focus on those members, GlobalHealth wanted a predictive analytics solution that would zero in on those members, as well as identify members who haven’t generated high costs in the last 12 months but are very likely to have serious health complications in the near future.

Once the analytics partner, VitreosHealth, was selected, GlobalHealth set about the work of analyzing its data sets and building algorithms to segment members who may be at risk of health emergencies. 

Using the predictive analytics solution provided by VitreosHealth, GlobalHealth has been able to reduce patient populations’ health complications by anticipating their actions. Thompson says GlobalHealth uses a novel approach to risk stratification to identify opportunities. The analytics platform leverages data sources including electronic health records, claims, health risk assessments, socioeconomic and wellbeing data for predictive risk and prescriptive care management.

“We’re able to look at specific characteristics of plan members, such as where they live, whether they are compliant with their referrals and compliant with medications, state-wide education levels, behavioral and mental health diagnosis, or characteristics that will likely result in a worse health state if we don’t intervene in order to help manage those plan members from a clinical standpoint,” Thompson says.

Traditionally, organizations use historical utilization to understand population risk and cost differences with respect to variations in care, non-compliance to evidence-based care guidelines and opportunities for care management, according to the GlobalHealth case study. In this model, the population is typically stratified as critical, high utilizers (of benefits), moderate risk and healthy.

GlobalHealth’s model uses a multi-dimensional risk stratification approach using predictive risks, such as disease-specific risk, composite risk and utilization risk, combined with outcomes like hospitalizations and ER visits to understand a member’s state of health at any point in time. In this model, members are stratified as either critical, high utilizer, hidden risk (high risk relative to care being received) and healthy and unknown.

According to GlobalHealth, those classified as critical account for 50 percent of the total population spend and have high-cost intervention in progress. High utilizers includes those with non-clinical risk, such as socioeconomic and accessibility, and non-compliant members with avoidable ER visits. Hidden risk members have diseases that are not well-managed with high-cost intervention on the horizon and are headed toward the “critical” stratification if unmanaged. And, for members in the healthy and unknown classification, it is assumed there is some hidden unknown such as new and young members with short medical histories.

Using this approach, GlobalHealth reviews the changes in the state of health risk from year to year to identify drivers of changes in costs and risks, or, essentially, to identify “mover populations” or members moving from the “hidden risk” status to “critical” or from “healthy and unknown” to “high utilizers.” By identifying these “mover populations,” GlobalHealth identified the cohorts of population to target for care management programs, and designed the right care management programs supported with the optimal resources, Thompson says.

GlobalHealth, working with VitreosHealth, ran the data through regression analysis tests and clinical teams vetted the diagnosis codes. Those tests enabled GlobalHealth to fine-tune the algorithms to eliminate any false positives and better interpret the data.

The health plan launched its proactive outreach program in early 2014 as a pilot project. In 2014, the outreach program targeted and engaged about 3,000 plan members, and that number has risen to about 7,000 plan members today, according to Thompson. Within the next six months, the organization plans to increase this number to 10,000 members, which will represent approximately 20 percent of GlobalHealth's total patient population.

While the data analytics technology is one piece of the program, Thompson says another critical piece is proactive outreach by nurse and clinician outreach managers who contact plan members over the phone to address any care gaps or to connect members with the community resources they may need. “We help the member navigate the healthcare system,” Thompson says.

Initially, GlobalHealth used its existing case management staff for the program but wasn’t getting the kind of traction it needed, so the organization created a separate unit dedicated to member outreach. GlobalHealth’s vendor partner created monthly dashboards for each care management program to better understand the performance drivers and also created an app that prioritizes the plan members to call and incorporates the members’ detailed clinical and socioeconomic information, Thompson says.

From a technology standpoint, one key to the data analysis and outreach program’s success has been ongoing performance tracking using the analytics dashboards which enables fine-tuning of the care management plans, Thompson says.

“Some key lessons that we’ve learned through this process of implementing predictive analytics is that your data is never going to be perfect. If you wait for it to be perfect, you’re going to miss out on an opportunity to intervene and make a difference,” Thompson says. “Many times we are identifying members at risk for a health event, using this model, a sophisticated regression analysis, and it looks at types of members, their zip code, demographic and diagnostic categories, and based on that, we know they have a specific progression through time. If you intervene sooner rather than later, it’s going to have a bigger impact.”

Thompson continues, “Another big lesson we learned is that if you don’t dedicate resources, you will only get so much value out of it. During the pilot years, we were splitting up resource. For the nurses, half their workload was managing referrals and the other half of their workload was doing proactive calls, and then we quickly realized that to create efficient and effective programs, you have to dedicate the resources to it.”

For organizations wanting to build similar care management programs that utilize predictive analytics, Thompson says organizations need to have the agility to evolve, he says. As an example, in the course of this work, GlobalHealth executive leaders combined departments to work together instead of in silos and reorganized the corporate structure by hiring and creating new positions.

Moving forward, the organization’s executive leaders plans to continue to evolve the data analysis and outreach program by bringing in more data sources, including clinic, laboratory and imaging center medical records, to further refine the analysis, Thompson says. This year, the organization has a plan in place to reduce healthcare costs by 10 percent, representing $25 million in savings.