Data Science & Healthcare: Why Health Plans Need to Do Better with Data

Healthcare technology is the ultimate driver for lowering costs and improving the immediacy of clinical and non-clinical actions, and it's the payer's opportunity to effectively utilize health IT solutions and realize these technologies’ full potential.

One such area is data science – a term that saturates the market, yet many payers today still struggle to translate from a concept into an actionable tool. With the appropriate use of data science, health plans can make real-time decisions that directly impact a person’s long-term care and costs. But like many things in healthcare, technology is a tool and not a silver bullet.

To meet with success, payers must first develop robust data models and then put them into practice to positively impact member health daily, weekly, and monthly. To achieve a 360-degree view of their members, it’s non-negotiable that health plans must integrate data science directly into a unified platform. This gives them valuable sources of data that can be readily identified and leveraged to improve not just data models, but health outcomes.

The variety of the data and the speed at which one can access the insights from that data make up the critical elements of data science. Armed with this prescriptive insight, payers and providers can prioritize the next steps in care planning with a long-term view in mind.

So why is data science crucial to health plans, and where do they get their start?

“Payers are the ones that have this longitudinal view of a member’s health,” says Vital Data Technology President and CEO Matt D’Ambrosia. “Having that all-encompassing, 360-degree view of the member’s health means that they can build out models that enable understanding of what’s going to happen with particular members in the coming months.”

Key to establishing a holistic view of members, payers must eliminate a multitude of silos within their organizations, both in terms of data and personnel. According to D’Ambrosia, that inward-facing, isolated view is detrimental to the mission of putting data science and predictive models into action to positively impact a broader set of members and their health outcomes. What’s more, these models are not one and done; they must be continually run to generate further insight that ends up in clinical workflows.

“Building a predictive model that’s just for physicians means that we’re missing out on others who touch a member — social workers, case managers, disease managers, and members themselves, as well,” D’Ambrosia explains. “Organizations building these data science models need to enable each of these stakeholders to effect change based on the predictive model. It is key that all these individuals are on that same page in a way that makes sense to them.”

Understanding the complete picture of the member means understanding all the data - beyond social determinants of health and conventional data.

“There is so much more data readily available that’s not being leveraged,” D’Ambrosia maintains. “Intelligence is coming in from various data systems, and payers must find ways to marry all these data elements together. Engineering the features in a unique way that provides more data parameters to run algorithms against it will lead to an improved model.”

For instance, operational data can provide significant value. Information from health risk assessments and care management solutions can further bolster data generated by predictive models. The ability to bring disparate data together in real-time or near real-time provides a complete picture of a member available at the point of care for clinicians and other service providers to make the most well-informed decisions at the right time.

To see how the marriage of data science and predictive analytics leads to tangible results, one need only look at the work of D’Ambrosia and his team on preterm birth and the ability to identify members at risk of experiencing preterm labor at 20 weeks of gestation. In the group, more than ten percent had costs associated with preterm birth exceeding $500,000. Seventeen percent of the target population were at risk for preterm births, and nearly 60 percent lack care management. Earlier intervention by care managers reduced the risk of complications and overall costs.

“Originally, we were working on this preterm birth model and were predicting it at 28 weeks,” D’Ambrosia explains. “As we reviewed our work, we began to wonder how we could identify these members sooner. By continually revising and working on our models, we arrived at a 20-week model. It shows that if we can recognize them sooner, we can start to implement actions that will most likely have a positive impact in avoiding that preterm birth. That eight-week shift proved to have remarkable value.”

And preterm birth is just one out of the dozens of models needed to predict and affect change accurately. On top of that, D’Ambrosia’s team analyzes the impact of actionable intervention around disease progression that includes but is not limited to diabetes, chronic kidney disease, renal disease, and other long-term illnesses. They also build specific behavioral health models around substance abuse disorders, depression, and others to help understand their impact on clinical claims cost. And model outputs can be stacked with other model outputs – for example, a model on member engagement – so payers can identify valuable cross-sections of the population segments. These models are an essential service that all health plans would need to get a larger picture of their expenses and their members’ health.

Data science and predictive analytics are iterative processes requiring continued scrutiny and refinement. Health plans need a vendor that will build data science models and, more importantly, execute those models to uncover prescriptive insights on targeted populations.

Talk with one of our experts to learn more.