Geneia clinicians help improve the usability of our predictive analytic models
Data Science

How Geneia data scientists collaborate with clinicians

March 10, 2021
Geneia clinicians help improve the usability of our analytic models.
Lead Principal Data Scientist

Health plans and their value-based care partners use our AI-driven data models, cloud-based technology, and analytic insights to identify and engage the right populations with the right care, improving cost, quality and health outcomes. Physicians, nurses, care managers and other clinicians are the most likely end-users of our predictive models.

That’s one of the most important reasons the data scientists at the Geneia Data Intelligence Lab (GDI Lab) regularly consult with Geneia clinicians as we create and perfect our models. It’s not enough to create highly accurate predictive models; they need to be usable by the people who manage populations of patients.

Heart Failure Complications Model

Take our heart failure complications model*+, for example.                                                    

Heart failure, defined as the inability of the heart to pump enough blood to meet the demands of the body, is a very significant chronic condition with increasing prevalence, a high burden on the healthcare system and poor prognoses. Two percent of adults and six to 10 percent of people age 65 years or older have heart failure. It’s a leading cause for hospitalization of seniors and one of the top 10 most expensive inpatient conditions. More than a third of people die in the first year after diagnosis.

GDI Lab clinical collaboration heart failure complications

Complications of heart failure are complex and multi-factorial.

There are many models out there that can predict hospitalizations or mortality due to heart failure. To us, particularly in the case of predicting mortality, these approaches offer more limited actionable insight in terms of preventive care. In consultation with Geneia’s clinicians, the areas we chose to target for prediction are not the severe complications with devastating outcomes, but rather the early signs and symptoms of an exacerbating condition, such as edema and shortness of breath. These complications can be swiftly addressed with straightforward treatment and management strategies. In other words, our analytic insights help health plans and providers identify and engage heart failure patients with remediable complications in the hope of preventing further exacerbation.

Hypertension Complications Model

We took a similar approach with the GDI Lab’s hypertension complications model*+.

As you may know, hypertension or high blood pressure affects one in three American adults. The prevalence has been rising across all ages, even children, and increases with age:

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

GDI Lab clinical collaboration hypertension statistics

As my colleague, Natalie Benner, RN, Geneia’s principal clinical transformation consultant, recently wrote, “Hypertension can cause a host of serious problems. Undetected and uncontrolled hypertension can lead to:

  • Heart attack
  • Stroke
  • Heart failure
  • Kidney disease or failure
  • Vision loss
  • Sexual dysfunction
  • Angina
  • Peripheral artery disease.”

We defined three stages for hypertensive patients. The complications steadily increase in severity. So for stage 3, they are very severe, e.g. end stage renal disease, heart failure, stroke and more. 

GDI Lab clinical collaboration hypertension complications

The GDI Lab’s hypertension complications model predicts, for those with hypertension, which patients are likely to experience a complication from each stage in the next 12 months. The model uses variables like age, total medical costs, outpatient visits and emergency department visits to yield binary classification with probability and risk buckets for each of the three complication stages.

Natalie helped the data scientists understand, “Knowing which members are at risk for Stage 1, 2 and 3 hypertension complications helps to prioritize patient activation efforts.” Health plans with limited care management resources can use this model to align their interventional strategy with a member’s risk level, targeting nurse-mediated care management to those most likely to experience a stage 2 or 3 complication in the coming year. While several models exist that predict hypertension complications, they are focused on more severe outcomes that we categorized as stage 2 or 3 complications. The ability to predict stage 1 complications is novel and affords an opportunity to intervene earlier, when actions are more likely to be effective.

Opioid Abuse and Overdose Model

For those who have received at least one opioid prescription, Geneia’s opioid model*+ predicts that person’s likelihood of an opioid abuse diagnosis or an overdose in the coming six months. There are a number of factors that make Geneia’s opioid model unique. Geneia’s model:

  1. Uses 22 variables whereas others use as many as 200 data points to achieve comparable predictive accuracy. Fewer variables make it easier to explain the model’s results to physicians and their patients.
  2. Combines the likelihood of an opioid abuse diagnosis and an overdose event. Many models identify one of those outcomes.
  3. Can be used with children and adolescents. Opioid models typically are created for people age 18 or older.

During the development process for this model, Geneia clinicians conveyed the importance of having a model with two outputs: an opioid abuse diagnosis or an overdose event. They also helped us understand the need for a model that can predict a child’s or adolescent’s likelihood of opioid misuse.

In short, Geneia clinicians collaborate with the data scientists throughout the model creation process, offering critical input on which variables to consider and the associated codes within claims, which prediction task or output(s) to prioritize to improve usability by clinicians, and much more.

*Certain data used in this study were supplied by International Business Machines Corporation. Any analysis, interpretation, or conclusion based on these data is solely that of the authors and not International Business Machines Corporation.

+Predictive models, by their very nature, contain certain assumptions. This is not an attempt to practice medicine or provide specific medical advice, and it should not be used to make a diagnosis or to replace or overrule a qualified healthcare provider's judgment.