GlobalHealth Transforms Member Outcomes with Predictive Analytics

An HMO combines predictive and prescriptive analytics with human outreach to reduce emergency room encounters by 18%.
Data Analytics

In the age of Big Data, healthcare organizations face enormous opportunities-and challenges-to uncover data insights for improving patient and member outcomes. GlobalHealth, an Oklahoma-based health maintenance organization (HMO), is one such organization that embarked on a journey three years ago to apply predictive and prescriptive analytics to its data to better understand its members' needs.

The catalyst that sparked the initiative was a member who fell into a diabetic coma, explains David Thompson, chief operating officer of GlobalHealth. "We were discussing the case and we asked ourselves if there was something we could have done differently to prevent it," Thompson says.

"Although there's no way to know for certain, we realized that if we had all of this data in one central point, we likely could have made a difference in this member's outcome." This realization sparked the decision to invest in tools to identify at-risk members and reach out to them before their conditions escalate. GlobalHealth selected VitreosHealth, an analytics provider, to help it analyze its data sets and build algorithms with which it could segment members who may be at risk of health emergencies.

What was clear early on was "we didn't have all the data we wanted, but we didn't let that stop us from building on it [the data] and taking action," Thompson recalls. The company started with aggregated medical claim information and pharmacy data that VitreosHealth layered with socioeconomic and demographic information like education levels and household income. Additionally, the company increasingly added information from electronic health records (EHR) to its predictive model.

GlobalHealth's next step was to run numerous tests to assess the accuracy of its predictive model. The model considers predictive risks such as disease-specific risks, composite risk, and utilization risk combined with outcomes like hospitalizations and ER visits to understand a member's state of health. Members were categorized as either critical, high utilizers (of benefits), hidden risk, or healthy and unknown (e.g., new and young members with short medical histories). VitreosHealth ran the data through regression analysis tests and clinical team members vetted the diagnosis codes.

For example, it was discovered that a small percentage of members were being incorrectly identified as chronic diabetics. These members may have been subscribed a steroid and had blood tests taken shortly after, and were recorded as having high or low blood sugar levels which thereby triggered a diagnosis code for hypertension or prediabetes. "We adjusted the algorithm to eliminate false positives like that," Thompson explains. "It's not so much the data is wrong, but that we have to fine-tune our algorithms to interpret the data in a precise way."

GlobalHealth launched its proactive outreach program in January 2014. Since then, the company has seen an 18 percent reduction in emergency room encounters and can predict nearly 70 percent of its hospital admissions. The organization has nearly 50,000 members and has contacted approximately 7,000 members through its outreach program. About 20 percent of members need extensive care while the rest are relatively healthy, but "we won't stop at reaching out to 20 percent of the population," Thompson says. "Our goal is to capture all of the care gaps that exist in the population." Having human outreach managers contact members over the phone is also critical. Just sending an email or letter with the information would not be enough, Thompson adds. "It's a two-pronged approach," he says. "The data is critical to better prioritize outreach and focus on specific care gaps and complexities.

But the data alone isn't enough to tease out all the variables. It also takes an educated conversation to make an impact in helping people better utilize their healthcare." As an example, Thompson describes a nurse outreach manager who called a member who had been flagged with complex diabetic conditions and considered a critical case. When the care manager spoke with the member, it was clear the member understood how to control the disease but his blood sugar level was "still out of whack," Thompson says. The care manager took the initiative to send the member a new glucometer and it turned out the earlier glucometer had been malfunctioning and over administering insulin.

"The only way to solve the issue was with a conversation and deductively identifying the issue," he adds. For other organizations that want to build a similar outreach program, Thompson's advice is to "realize the data will never be perfect, but take action on what you have. And at the same time, "remember that it takes more than data to have an impact. It also takes high touch member services."