AI models help care managers prioritize outreach to highest risk patients.
Data Science

Best practices: Using AI in population health

June 23, 2020
AI models help care managers prioritize outreach to highest risk patients.
Shelley Riser

As population health practitioners know all too well, chronic disease is the leading cause of death and disability and the leader driver of the country’s $3.5 trillion in annual healthcare costs. Sixty percent of Americans have one chronic illness such as diabetes, asthma and heart disease and COPD, and a staggering 40 percent have two or more.

Close collaboration between clinicians and data scientists

I lead a clinical team of more than 130 nurses, engagement specialists, social workers and population health consultants. Together, we provide clinical care coordination to an eligible population of 750,000 patients, meaning my team is actively practicing population health. Many of the team are focused on interventions to engage chronic disease patients to prevent clinical deterioration and unnecessary complications and work toward the best patient outcomes and quality.

We work closely with Geneia’s Data Intelligence Lab (GDI Lab) as the data science practitioners create, test and hone analytic models designed to better manage chronic illnesses. In short, the GDI Lab prioritizes analytic models that:

  • Predict the onset of chronic disease for those who don’t have it yet
  • Predict complications of chronic disease for those who have a diagnosis

Take hypertension, for example

Hypertension affects one in three American adults, and the prevalence increases with age:

  • 20-44 years: 26 percent
  • 45-64 years: 59 percent
  • 65+ years: 78 percent

The annual medical costs associated with hypertension are approximately $131 billion. It is estimated that eliminating hypertension could reduce cardiovascular mortality by 30-38 percent. Poorly managed, the disease progresses to devastating complications such as stroke, end-stage renal disease and heart attack.

Hypertension model

Putting the GDI Lab hypertension model in practice

As Jasmine McCammon, Ph.D., principal data scientist, wrote in a blog published earlier this month, the GDI Lab created a model to predict hypertension complications.

Hypertension Complications Model
For those with hypertension this model predicts who is likely to experience a complication from three complication stages in the next 12 months.

The model predicts various stages of hypertension complication severity, including earlier signs and symptoms such as proteinuria and retinopathy, yielding binary classification with probability and risk buckets for each of the three complication stages.

Given the prevalence of hypertension, a health plan with 750,000 members can expect that approximately 225,000 are hypertensive. At the risk of restating the obvious, there are simply too many members with hypertension for care management to engage all of them.

Using the GDI Lab model helps my care management team identify where to start. In short, we use the GDI Lab’s hypertension complications model to align our interventional strategy with the risk level.

  • For the low-risk members, we use online tools and educational resources.
  • For medium risk, we use automated outbound calling (AOC) or interactive voice response (IVR).
  • We prioritize outreach to the highest-predicted-risk members. A population health engagement specialist will outreach to a patient and a nurse care manager completes an in-depth detailed assessment: family history, medical treatments, comorbidities and more.

Health plan with 750,000 members

  • 1 in 3 adult members with hypertension, increasing with age ~ 225,000
  • Hypertension model output – low, medium and high-risk buckets
  • Align interventional strategy with risk level
    • Low: Online tools and resources
    • Medium: AOC or IVR
    • High: Population health engagement specialist and/or nurse care manager outreach

Hypertension model output

Similarly, we use the heart failure complication model to identify members to target for enrollment in a remote patient monitoring program. ICYMI, Geneia’s 12-month scientific study of remote patient monitoring demonstrated impressive results:

  • Net reduction in hospital admissions: 76 percent for the pilot group versus 31 percent for the control group
  • Slowed disease progression: 58 percent (control) versus 29 percent (pilot)
  • Medication adherence: 18 percent higher pharmacy costs for control group
  • Improved patient experience: study participants report 96 percent overall satisfaction
  • Reduced per member per month medical spend by 50 percent

Best practices: Integrating analytics into population health and care management

The work by my clinical team to effectively integrate the GDI Lab analytics models into population health and care management has given rise to some best practices:

  • Involve clinicians as subject matter experts in model creation.
  • Integrate data and insights so everything is in the care management platform, ideally prioritized by patients most at risk.
  • Be sure care managers can see and create cohorts as well as drill down to patient-level data and record.
  • Align interventional strategy to risk level.
  • For those team members who speak directly to patients, it’s important to have scripting that is consistent and patient-friendly
  • Use models to determine gaps and/or improvements needed in care management offerings.

To learn more about the Geneia Data Intelligence Lab, click here.