How AI Supports Low-Risk Member Identification, Care Management

How AI fits into low-risk care management

For years, Collins and his team at BCBSRI have relied on a resource utilization bands system in order to stratify their member populations. The system that they used was considered one of the best of its kind.

However, Collins wanted more from his population health management tools.

“It may be the best one out there, but it is still a static system,” he explained.

The technology could stratify members based on the risk of hospitalization, for example, but it was difficult for the payer to use this system to assess other risk factors such as potential behavioral healthcare conditions.

“We've added a lot to it over time, but, understanding that we're trying to be excellent at care management, we needed to have the best identification system,” Collins shared.

This pursuit of a better identification strategy was what led BCBSRI to invest in AI solutions—more specifically, a machine learning system.

“One of the best things about machine learning is that it's an evolutionary thing; it's not just a static risk identification system,” said Collins.

Knowing when to expend resources in order to intervene in a low-risk scenario is a challenging task for health plans, but it is one that machine learning and AI systems can help solve, according to Collins.

“What AI has the opportunity to do is to start to find those needles in the haystack from those lower-risk people who are following patterns that the people with chronic diseases followed,”  Collins said. “And then maybe you can be more judicious in your application of care management resources.”

The machine learning solution that BCBSRI chose will be incorporated at touchpoints all along the member care journey, focusing on members whose health risks could increase.

The tool is easy to customize so that BCBSRI can observe specific member populations such as those that have heart conditions or struggle with weight gain.

Artificial intelligence technologies excel at identifying patterns of behaviors within a population, beyond human capabilities. That is why payers have implemented these tools so successfully to fight unpredictable fraud schemes.

Principles for using AI effectively in low-risk care management

AI and machine learning systems do not provide solutions, but rather a sense of what path payers should follow to improve care management and preventive care services for certain populations.

“It provides you direction, but you have to be ready to act on that information and adapt your programs to adjust,” Collins emphasized.

“One has to be able to adapt one’s program to follow the evidence of what the machine learning tells you that you should be doing in terms of who you should be identifying. You should be an expert in managing what you should actually do with them. Adopt AI when you have a problem that you want to solve and you're willing to really change your thinking about how to solve that problem and be informed by the process.”

The approach is similar to the way in which industry experts have enjoined payers to maintain a proper perspective on the role of AI and technology in integrating behavioral and physical healthcare.

“Technology needs to be a tool to accomplish the mission,” Michael Renzi, DO, president of healthcare delivery at Capital District Physicians' Health Plan (CDPHP), told HealthPayerIntelligence on the subject of behavioral-physical healthcare integration. “It's not the backbone for mission success. It's about people and process. And the people and process have to be enabled by technology.”

BCBSRI faced particular challenges in implementing these technologies as a smaller health plan.

In a state with less than 1,060,000 residents, the payer’s membership of over 400,000 members makes it the largest insurer in Rhode Island. But in an industry in which large companies boast double-digit millions in member populations, this payer would be classified as small outside of its own territory.

Smaller health plans with a more narrow member pool do not have access to as much data as larger plans.

“The output of machine learning is only as good as the information that goes into it and the more data, the more informed it is” Collins said. “And the more you can bring in other data sources—disparate sources perhaps—the better.”

Collins suggested that small health plans turn to vendors with larger pools of data at their disposal.

This approach was key for BCBSRI as the payer developed its behavioral healthcare intervention, HealthPath.

“They have the benefit of access to much bigger data sources, claims-based sources,” Collins explained, referring to BCBSRI’s vendor partner. “They can use much larger data sets that they have access to identify and to test the hypothesis that there is a better way to identify the serious mental illness population in a commercial claim set.”

Once payers have identified low-risk member populations, they can employ digital solutions for light touch interactions with members like virtual wellness programming. These solutions may be lower cost and, when they are vendor-facilitated, will not draw as heavily on payers’ resources.

For example, BCBSRI recognized that seniors who develop osteoarthritis often undergo joint replacement surgeries. Collins and his team knew that there were ways to mitigate the effects of osteoarthritis that did not require invasive, expensive surgery.

The payer collaborated with its machine learning vendor to identify members who might develop osteoarthritis in order to engage them in a digital physical therapy program for joint health.