The Administration for Community Living, the organization under the Department of Health and Human Services that supports communities of older adults as well as those with disabilities, is using artificial intelligence and machine learning to identify risks of adult mistreatment and predict maltreatment.
Doing so involves collaborating across groups and leveraging multiple data sources and data sets, including information that already existed, to develop algorithms and dashboards that ultimately drive decision-making.
“When you're dealing with technology issues that broadly address issues of data and tools like AI, having a working group mitigates against the risks that are arising, regarding the impact of bias on machine learning,” said Scott Cory, CIO at the organization, during an ACT-IAC community of interest meeting earlier this month.
In conducting the research, Cory noted that the agency focused on identifying data sets that specifically looked at what information was available around abuse, neglect and exploitation, as well as potential risk factors. The National Adult Maltreatment Reporting System enabled ACL to know where cases of abuse, neglect and maltreatment have been reported, which provided a foundational set of data.
Cory’s team had to take a closer look at the data before it influenced the development of ACL’s algorithm.
“We were really concerned about the sources of data. We took additional steps to ensure the data was clean to allow homogeneity,” Cory said.
The group chose machine-learning algorithms to parse the data for potential risks. The agency also had subject matter experts review the results and make adjustments to provide confident outcomes. Following the review and results, the team leveraged data visualization tools to create understandable models that drive decision-making.
“We anticipate that this work may create algorithms that may have greater use at the state and local level than it would at ACL. The outcomes of this specific project were different than we expected and different areas where the benefits accrue,” Cory said. “We will continue to look at county-level data based to identify value within the county as a geographical unit of identifying potential risk factors.”
Following the initial results, Cory added that ACL will leverage data from the Centers for Medicare and Medicaid Services tied to adult protective service information.
“This data would allow us to ascertain whether or not there are patterns in payment and engagement that can be related to adult protective services,” he noted.
ACL will partner with CMS and the Centers for Disease Control and Prevention to understand adult abuse, neglect and maltreatment. It will be the first time the agency uses CMS data to do so, Cory said.
“This is work the ACL cannot do alone. The nature of our relationships with our grantees doesn’t provide us with the information we need to effectively do this. It requires ongoing and consistent efforts with our partners, CMS and CDC,” Cory said. “We’re always looking for opportunities to partner with organizations that are willing to work with us.”