Many Medicaid programs across the nation are seeking to foster cross-agency alignment to improve outcomes for Medicaid beneficiaries. In particular, integrated data sources from multiple agencies can be critical for informing policy and programmatic decisions, in addition to identifying disparities across populations. The Washington State Department of Social and Health Services is nationally recognized for its analytic capacity and successes in integrating data from several state agencies. David Mancuso, PhD, director of Washington State’s Research and Data Analysis Division (RDA), led the development of RDA’s Integrated Client Databases, a robust data repository that links Medicaid medical, behavioral health, and long-term care data with Medicare, social service, criminal justice, housing, child welfare, employment, and vital statistics data. This analytic resource has been used in Washington State to inform decisions about health and human service program operations and budget priorities. CHCS recently spoke with Dr. Mancuso about the division’s innovative approach to leveraging multiple data sources to improve outcomes and health equity in Washington.

Q: Many Medicaid programs are focused on partnering with other health and human services agencies to foster cross-sector alignment aimed at improving outcomes for beneficiaries. What do you see as the greatest opportunities and challenges in this area as it relates to data?

A: Opportunities for alignment exist where there are significant overlaps between enrollment in Medicaid and participation in other public programs, and where information from one delivery system could help identify risk factors, service needs, and intervention strategies that could improve outcomes for shared clients. Data linkage is never something we do for its own sake. Every linkage activity we perform is intended to further our ability to support the operation of health and human service delivery systems.

The opportunities are vast. Fifteen years ago this capability allowed us to get ahead of the curve in understanding the impact of the opioid epidemic, and helped make the case to significantly increase funding for substance use disorder treatment. A more recent example linked administrative data to measure adverse childhood experiences (ACEs) experienced by children enrolled in Medicaid, to show how ACEs, especially experiences related to child abuse and neglect, profoundly impact use of medical and behavioral health services in adolescence.

Washington State’s Health Home program has been another high impact area. The program provides intensive care management services to high-risk Medicaid beneficiaries. Integrated data were used to develop predictive modeling algorithms to define program eligibility; evaluate the impact of services on health outcomes and costs; and meet CMS requirements for quality reporting. Washington State’s program is the only one of its type in the country to earn shared Medicare savings from CMS.

There are several challenges that are intrinsic to doing this work, like building trust among data owners, gaining and maintaining support from agency leadership, and maintaining an analytic data infrastructure in a constantly evolving policy, program, and information technology system environment. From those challenges, our primary constraints are the availability of skilled staff, maintaining a portfolio of funded work that supports both data integration and analytics, and working with data owners to support data sharing allowed by law. By comparison, technology is a less significant constraint. Expecting investments in technology to be the solution to building organizational analytic capacity — in the absence of investments in highly skilled analytic staff — is a fundamental mistake.

Q: How do you build trust with internal and external partners to facilitate data-sharing progress?

A: Earning and maintaining trust depends first on maintaining rigorous quality standards, and second, on working closely with the partners who contribute to our integrated data environment.  If our partners and key external audiences — the Governor’s Office, the Legislature, health plans, providers, advocacy groups, community organizations, our clients, and the wider public — did not trust the quality of our analytic products, then we would be less useful to our state agency partners. Delivering high-quality analysis requires working closely with agency partners because they have the complementary program, policy, fiscal, IT system, and clinical subject matter expertise that is essential to inform our work. The importance of building effective relationships with agency subject matter experts cannot be overstated.

Q: Why is strong program leadership critical to the success of cross-agency collaboration?

A: Our division operates at its current scale because generations of leadership across multiple state agencies and program areas have recognized the analytic value that can be derived from integrated cross-system data and have been willing to fund our work on a per-project basis. Without ongoing leadership support, this capacity could wither and would be challenging to re-establish. Analysts have an obligation to produce high-quality work at a competitive cost and to maintain the trust of agency leadership and external stakeholders. If we do this, I am optimistic future generations of leadership will continue the scale of investments they are currently making.

Q: How is the role of data and data analysis changing in the overall operations of Medicaid?

A: We struggle with the use of the terms “data” and “analysis” as synonyms, which masks the critical role of the analyst in taking raw data and extracting knowledge from it. Ninety-five percent of our work is comprised of analytic activities or data management directly in support of analytics. We are rarely in the role of just providing “data” in a form where significant value is not being added to raw data in the products we create.

Our analytic activities have increased substantially over the past decade. We are doing predictive modeling to support clinical decision-making at a scale I would not have anticipated a decade ago. We are supporting value-based purchasing through an ever-increasing platform of health-related quality and performance metrics, including measures of social outcomes, such as employment, housing stability, and criminal justice.

There are also areas where we have seen less development than folks might expect, such as use of business intelligence tools (BI) or cloud technologies. Our data are too massively dimensional for the visualizations available through BI tools to be useful for our data scientists as they develop predictive models or simulate clinical trials in a program evaluation context. That said, we are significantly scaling up our investment in BI tools to communicate information to non-technical audiences. I am also sure we will have a more significant cloud presence within the next few years.

Q: How do you use data to understand disparities in health outcomes? What opportunities are there for data to inform policies that aim to build health equity?

A: Linked administrative data can be a powerful tool for understanding disparities in access to services, quality of care, utilization, and health outcomes. The most noteworthy example of work in our division that helped mitigate health disparities linked administrative data from birth and death certificates to identify racial disparities in infant mortality and sudden infant death syndrome (SIDS). Using additional linked information from the Pregnancy Risk Assessment Monitoring System, the team also looked at differences in sleep position by race/ethnicity. They found that infants whose mothers were African American experienced SIDS at far higher rates, and were less likely to be put to sleep on their backs, a known, modifiable risk factor. This finding led to outreach and partnering efforts with community leaders and a significant reduction in infant deaths.

More generally, using administrative data allows us to conduct health care quality measurement at a Medicaid population scale. This makes it possible to understand the experiences of communities that can be challenging to represent in data collected using survey methods. We can use linked administrative data to measure disparities and differences across many dimensions, including communities defined by race, ethnicity, language, or geography. And by linking to data from other delivery systems, we can identify profound disparities across communities in areas like homelessness, employment outcomes, and criminal justice involvement.

Q: What advice would you give to a state that is in the early stages of developing its capacity?

A: Consider starting with a modestly sized core team of skilled analysts and supporting data management staff who can prove to your authorizing environment that they can deliver value. My bias is toward staff with training in quantitative social sciences. Staff with this background are more likely to bring intellectual curiosity about your business, analytic programming skills, an understanding of the opportunities and challenges in drawing causal inferences from administrative data, and the ability to learn actuarial frameworks for measuring risk, utilization, and cost. I would also start with a very specific and manageable project that requires the integration of two or three data sources and expand from there. An expansive data integration project targeting too many data sources may quickly become a technology and data sharing project, without a primary focus on building analytic expertise to address specific information needs.

I would encourage investing in the quality and management of your Medicaid Management Information System and other key internal data sources. Your core business depends on these data systems, and they should always be your most important data sources. Lastly, be patient. Have reasonable expectations about the timeline for building capacity and the volume of work product that your team can create. That said, a skilled team of five people can develop the capacity to generate great value in relation to the cost of the investment.