Moving SDOH Programs Beyond Referrals to Measurable Outcomes

Can you please share your name, title, and a brief overview of your professional history?

Rakshith Yashvanth: I'm Rakshith Yashvanth, senior product manager at ZeOmega, where I work on artificial intelligence (AI) and social care capabilities within care management, utilization management, and population health platforms used by health plans and Medicaid managed care organizations.

My work in social care has spanned both sides of the problem: identifying who has a social need and connecting that individual with services. On the identification side, that has meant combining assessment, claims, and public data to develop a view of social risk at both the community and individual levels. On the delivery side, it has involved closed-loop referral management and the tools health plans use to manage community-based organization networks, including routing referrals, matching members with organizations that have capacity, tracking whether services were delivered, and meeting requirements such as CalAIM's closed-loop referral mandate.

Why do many social determinants of health (SDOH) programs lose momentum after health plans identify members' social needs and issue referrals?

Yashvanth: Largely because the referral moves outside the system that created it. Within a health plan, a member with an identified need sits within a workflow that has a clear owner and record. Once the referral reaches a community-based organization, however, it typically arrives at a partner with no contractual relationship with the plan and no funded time to document what happened next.

That is not a failure of will. It is simply the arrangement most health plans and community organizations have with one another. Plans contract with clinical providers to document care and pay them for doing so, but the equivalent structure rarely exists on the social care side. As a result, referral volume rises because sending a referral costs almost nothing, whereas confirmed resolution lags because reporting back requires staff time that no one has budgeted.

North Carolina's NCCARE360 network offers particularly clear evidence of this. Duke researchers compared 2 identical 5-month periods, 1 year apart, at a large Durham County health system using the same closed-loop platform and the same community organizations. During the first period, a temporary program reimbursed those organizations for the food assistance and case management they provided. Of 3220 referrals, 88% were confirmed as resolved. After that funding ended, referral volume fell to 860, and the resolution rate decreased to 30%. The technology was identical during both periods. What changed was whether the work was being paid for.

There is also a diagnostic problem underlying this issue. A referral is marked as resolved only when the receiving organization reports it, so a 30% resolution rate could mean members were not served or that they were served by organizations without the staff time to document it. Those are opposite problems requiring opposite responses, and the data collected by most programs cannot distinguish between them.

That ambiguity is often what ends the program. Returns on social care investments accrue over 2 to 3 years, whereas budgets renew annually, and Medicaid churn means a member may no longer be enrolled when savings emerge. A program that cannot demonstrate what it delivered has no case to make during the next budget cycle. It rarely loses funding because it definitively failed; it loses funding because it cannot demonstrate that it worked.

How can Medicaid plans confirm that members receive the services to which they were referred while avoiding additional administrative burden for community-based organizations?

Yashvanth: It helps to be precise about what creates the burden. It is not simply the number of fields on a form. An organization asked to provide even a single data point that it is not funded to produce, through a portal it must learn and maintain, still experiences that requirement as a cost it cannot absorb. Reporting becomes sustainable when it is funded and occurs through a channel the organization already uses.

The first design choice is to ask only for information that answers the question. Confirmation requires only a handful of coded fields: referral status, date, closure reason, and little more. California's closed-loop referral requirement is instructive. A plan can close a referral as "services received" once the provider confirms delivery through a single coded element selected from a defined list, with free-text documentation required only for exceptions. No assessment scores, outcome measures, or narrative are required.

Whether the service changed an outcome is a separate question, and one the plan can answer using claims and eligibility data it already holds. When plans incorporate that question into the confirmation transaction, the burden increases and the smallest organizations may drop out.

The second design choice is to reduce the variability in reporting formats that a plan contributes. In markets with several Medicaid plans, a community organization's burden scales with the number of plans it serves, not necessarily with the number of fields each plan requests. Five plans asking 3 questions through 5 separate portals creates substantially more burden for an understaffed organization than 5 plans asking 8 questions through a single shared channel.

California addressed this by standardizing the reporting file at the state level and then limiting what plans could require: plans cannot require tracking data through any channel other than that file. When closure information is missing, the plan must first review its own claims and encounter data before sending the organization a follow-up request.

That last practice does not require a mandate. For services that are billed, a plan already knows the service occurred when it pays the claim. Any plan can reconcile that information internally rather than asking the community organization to report it again.

The third design choice is to meet community organizations where they are technologically. A larger organization with its own client tracking system can transmit data directly. Organizations already using a shared referral platform such as Findhelp or Unite Us can update referral status there. A small organization with no software needs a reporting option that does not require new technology, such as a short web form or a telephone call that someone at the health plan records.

Each of these approaches requires setup, and that cost is real whether it is borne by the health plan, a state capacity-building program, or a philanthropic grant. What matters is that the cost is budgeted at the outset. When it is not, the network ultimately retains only organizations capable of absorbing those costs themselves.

What contracting and accountability standards can plans use to measure community organizations' performance without limiting participation by smaller, resource-constrained organizations?

Yashvanth: The first consideration is that audit, accounting, and compliance requirements carry fixed costs, and those costs do not change with the size of the contract. A requirement that a large organization absorbs as part of routine operations can make a small contract not worth accepting.

As a result, the sophistication of what a health plan requires can shape which organizations participate in its network, whether or not that was the intent. Scaling the complexity of requirements in proportion to contract value is one of the most useful standards a plan can adopt. It works best as a deliberate policy rather than as an exception that smaller organizations must request individually. Many small organizations do not have staff whose role includes negotiating with health plans.

For performance measurement, a standard that translates well across organization sizes is service fidelity: Was the service delivered? Was it delivered to the correct member? Was it delivered within the agreed time frame?

A 10-person organization can meet that standard, and a plan can verify it using data it already receives. It is also a fair standard because it holds organizations accountable for what they can directly control.

Whether a member's circumstances improve during the following year depends on many factors beyond any single service. A performance standard based solely on those downstream outcomes would hold the organization accountable for factors it may not be able to influence.

What data infrastructure, governance, and outcome measures are needed to demonstrate that social care interventions improve health outcomes and reduce avoidable health care utilization or costs?

Yashvanth: It is important to distinguish between 2 questions that are often treated as one. Confirming that a member received a service is a narrow question that can be answered with a status field. Demonstrating that the service changed an outcome is an entirely different question and requires longitudinal data, a comparison group, and a realistic time horizon. Plans frequently invest in the first and assume it will produce the second.

From an infrastructure perspective, the key requirement is linkability. Confirmation data stored in a referral platform but disconnected from claims and eligibility data at the member level cannot support an analysis of health care utilization. Resolving member identity across those data sources may not be the most visible part of the work, but it is a prerequisite for everything that follows.

Linkage, however, depends on permission, and that is where governance becomes important. Most community organizations are not entities covered by the Health Insurance Portability and Accountability Act (HIPAA), so the default assumption in many organizations is that service-level data cannot flow back to the health plan. Programs can therefore stall during legal review before anything is built.

California addressed this issue directly by clarifying that health plans, as covered entities, may share information needed to close the referral loop with noncovered entities for treatment and care coordination purposes without separate member authorization. No change in law was necessary. The issue was how existing rules were being interpreted, and that experience provides a useful reference point for other states working through the same question.

For outcome measurement, the central principle is to select measures that are closely connected to what the service is intended to accomplish. Consider a transportation benefit. It may ultimately reduce avoidable health care utilization, but only through a long causal chain: rides lead to attended appointments, attended appointments may improve disease control, and improved disease control may reduce exacerbations.

Not every member who receives transportation attends the appointment, and not every member who attends an appointment experiences improved health. A strong effect at the beginning of that chain therefore becomes more difficult to detect at the end. Outcomes also emerge on different timelines. Appointment attendance may change within weeks, whereas emergency department utilization patterns may take a year or longer to change.

By the time a plan examines outcomes at the far end of that chain, numerous other factors may have affected the member, making attribution difficult. If measured solely by emergency department visits at 12 months, a program that worked may appear ineffective. If measured by missed appointments, the effect may be larger, appear sooner, and be easier to attribute.

Downstream health care utilization is a reasonable outcome to examine over several years. It is a poor basis for judging a program after only 12 months.

There is also one constraint that no individual health plan can address on its own. Medicaid churn means members frequently disenroll before the outcome measurement period is complete. Members who remain enrolled throughout the full measurement period tend to be more stable, so analyses limited to those members describe a narrower population than the program originally served.

It is important to acknowledge that limitation. Attribution across health plans and over multiple years is a design challenge at the state or federal level, and it remains unresolved.

© 2026 HMP Global. All Rights Reserved.

Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of First Report Managed Care or HMP Global, their employees, and affiliates.