A few years ago, when most of the Affordable Care Act’s major provisions came into play, Humana started setting concrete goals to help its members get healthier.
At the time, Humana’s leadership was starting to wonder: What if the healthcare industry dared focus on the health of the patient instead of the volume of care? The federal government was already introducing reimbursement models that incentivized payers and providers to actually keep people out of hospital beds in the first place. Humana saw opportunity paired with purpose. Its new journey began with a simple question: What does it mean to be more healthy?
The answer to that question was far less simple—and buried in data. Slawek Kierner, Humana’s Senior Vice President of Enterprise Data and Analytics, has been working to find it.
Turning raw material into predictive health insights
Over more than half a century of operation, Humana had slowly aggregated an enormous sum of customer data. From blood test results to patient surveys (even public Census data), the organization had the kind of information that could help forward-thinking data scientists quite literally change the face of healthcare. But all that data was scattered, disparate—far too disjointed to be usable for pattern identification or inference. Its true utility was predicated on centralization. In other words, Humana had all the raw material it could hope for. It just needed a hub powerful and secure enough to harness it.
So that’s exactly what Kierner and his team built. The data platform of Humana’s future was born.
With hundreds of thousands of data points now centralized, Humana’s data science teams—supplemented by powerful AI and machine learning tools—could now efficiently assess any number of data segments to tease out critical patterns and inferences that can help predict, for example, whether someone is at risk for developing a preventable medical condition. For the first time, the organization could actually see what makes people more healthy—and help make that a reality.
A key part of that initiative was building out a sophisticated machine learning layer on top of the data platform that enabled a wide variety of use cases. That layer allows Humana data scientists to assess, adapt, and document machine learning models built and deployed by their colleagues, rapidly accelerating innovation timelines. It also helped shape a solution called Control Tower—with the assistance of Boston Consulting Group (BCG), Humana built Control Tower to consolidate key information, insights, learnings, and recommendations from across the full scope of the organization’s clinical operations. Humana now uses Control Tower in the same way that, say, NASA relies on its Mission Control Center: as a centralized home for critical information that allows different specialists and leaders to immediately assess the health and efficiency of their respective operating fields and, frequently, make improvements.
But Humana’s new underlying data platform has an extensive list of use cases beyond Control Tower. Its data can be especially helpful in uniquely challenging circumstances. During the pandemic, for instance, this data helped Humana figure out which of its members needed transportation to a vaccine provider.
“We can predict, for every member, for every single day, do they have a food insecurity situation, do they have a transportation issue? Do they have a security challenge? Do they have financial strain?” Kierner says. “We use this to actively manage their care.”
From reactive to proactive care
Kierner believes such an approach parallels broader societal patterns that few would notice in their everyday lives. For instance, he chooses the anecdote of being stuck in an elevator. It’s exceedingly rare—and that’s because many elevators now send data to their operators in order to predict when maintenance is needed well before something goes wrong. “The same can be done in healthcare,” Kierner says. It’s swapping reactive care for proactive care.
Think about this for, say, pharmaceuticals. Using a bevy of key indicators from its centralized data pool, Humana’s machine learning models can predict which patients are unlikely to fill their prescription then prompt care specialists to make a phone call or send a text that gently reminds them to do so. Even the most effective outreach methodology can be predicted by such models. In other cases, perhaps the patient’s primary care physician is notified. In a reactive care model, interventions were difficult, if not impossible, to get ahead of. After all, patients are far more likely to end up in expensive emergency rooms when ailments like hypertension aren’t being properly treated. The key is getting ahead of it.
It sounds utopian—and in many ways could be. But a critical pitfall for machine learning use cases across all industries is the concept of equality: eradicating human biases from the model development process. Data might not lie, but machine learning algorithms are only as unbiased as the humans that train them. If a data scientist unintentionally teaches an algorithm to be biased, it will be—and the results can be disastrous. In a proactive and predictive care model, unintended bias could render entire communities invisible. As in all things AI, ethical and inclusive development is the difference between success and failure.
Humana had the foresight to plan for this, establishing a team focused on governance—ensuring that the appropriate treatment and range of options are offered to all of its customers, regardless of race, age, or gender. The organization has been so committed to transparency that it has even aggregated its AI principles and published them for public consumption. Still, the use of AI is uniquely challenging in healthcare; in the finance industry, for example, dedicated teams must constantly monitor against age or gender discrimination. But in healthcare, age, race, and gender can all be critical factors in prevention or treatment plans. Women of a certain age should receive routine breast cancer screenings. Men need their prostates checked. The list goes on, but prioritizing health equity through machine learning systems—a key Humana objective—while still incorporating critical and sensitive data is a delicate balance to strike.
“Age might increase your likelihood of having certain diseases,” Kierner adds. “We need to have that information in our model to be accurate. So you cannot just discard (it). But we can have proper testing, and we have the right team that looks at it. We have invested in technology, process, and people to ensure that whatever our algorithms recommend to do is always fair and ethical.”
In the long-term, Humana sees its use of data, artificial intelligence, and machine learning as a way to manage the health of its ever-growing enrolled community. For too long, payers have taken a hands-off approach, content doing business as usual. Humana went head-first into a new standard of care, one that recognizes that transparency and strategy can help people get the care they need.
And in the process, it’s helping the world understand what it truly means to be more healthy.
This story was produced by WIRED Brand Lab for Boston Consulting Group.
