25.30 — Medicaid in Detail

A woman works two part-time jobs, has two children, and is covered by Medicaid. Her hours increase for three months over the summer. Her income for those months crosses a threshold. A renewal notice goes to the address she moved away from in March. She does not respond, because she never saw it. Her coverage ends — not because she became ineligible, but because a form was not returned.

When the pandemic-era rule requiring states to keep people enrolled ended in 2023, and states worked through the backlog of renewals, a large share of the people who lost coverage lost it for exactly that reason: procedural, not substantive.

That single failure mode explains why Medicaid eligibility and renewal systems are one of the largest and most consequential areas of government health IT work in the United States, and why the technical details in this chapter matter to real people.

What Medicaid is, structurally

Medicaid is a joint federal and state programme covering low-income people. The federal government sets minimum rules and pays a share of the cost; each state runs its own programme within those rules.

The federal share varies by state, calculated so that poorer states receive a higher percentage, and it is substantially higher for the expansion population created by the Affordable Care Act. That funding structure is the reason states behave the way they do: every dollar a state spends draws down federal money, and every eligibility or benefit choice has a state budget consequence.

A state's programme is defined by its state plan, plus waivers. Section 1115 demonstration waivers let a state try something outside the normal rules if it serves the programme's objectives; other waiver authorities allow mandatory managed care enrolment or home and community-based services for people who would otherwise need institutional care. Waivers are why two states' Medicaid programmes can look genuinely different, and why a vendor's product must be configurable rather than fixed.

Enrolment across Medicaid and the closely related Children's Health Insurance Program has run in the tens of millions — around 80 million people at its recent peak, falling after the post-pandemic renewals worked through. Treat any specific figure as a moving number and check the current published tracker before quoting it to a client.

Who is eligible

Eligibility is organised into categories, and the category determines both the rules applied and the systems that handle the case.

Children, at higher income limits than adults, with the Children's Health Insurance Program covering those just above the Medicaid threshold.

Pregnant women, also at higher limits, with coverage continuing for a period after birth — many states have extended this to a full twelve months postpartum.

Parents and caretaker relatives, usually at low income limits.

Adults under the expansion, in states that adopted it, up to 138 percent of the federal poverty level.

Aged, blind and disabled individuals, which is a different world: these determinations involve assets as well as income, disability determination, and often long-term care needs.

And dual eligibles — people covered by both Medicare and Medicaid, typically older or disabled people with low incomes. They are a small share of enrolment and a large share of spending, because they are the sickest population in the system, and coordinating their two sets of benefits is a permanent policy and engineering problem.

Two methodologies run in parallel and this distinction shapes every eligibility system. Most non-disabled applicants are assessed using a standardised income definition based on tax rules, with no asset test. Aged and disabled applicants are assessed under older rules involving income and asset limits, and often a spend-down mechanism where medical expenses reduce countable income.

Some states also cover the medically needy: people whose income is too high but whose medical costs are so large that they qualify once those costs are deducted.

How an application is actually decided

The steps below are what an eligibility system does, and each one is a place where people fall out of coverage.

Application through a state portal, the federal marketplace, by paper, by phone or in person. The single-streamlined-application requirement means one application should be able to route a person to Medicaid, CHIP or marketplace subsidies as appropriate.

Verification, and this is the heart of the system. States must first try to verify income, identity, citizenship or immigration status, and residency using electronic data sources — federal income and wage data, social security records, state wage databases — before asking the applicant for documents. The design principle is deliberate: every document a person has to produce is a chance to lose them.

Where electronic data conflicts with what was stated, or is unavailable, the applicant gets a reasonable opportunity period to provide documentation, with coverage generally provided in the meantime where the rules allow.

Then determination, notice, and enrolment, including selecting or being assigned a managed care plan in most states.

Two provisions exist to prevent gaps and both are worth knowing. Retroactive coverage can cover medical bills incurred in a period before the application, subject to state rules. And presumptive eligibility lets qualified entities — hospitals, clinics — temporarily enrol someone who appears eligible so that care can start immediately while the full determination proceeds.

Renewals and recertification: the part that decides everything

Coverage is not permanent. Eligibility is redetermined, normally once every twelve months for most groups.

The federally required approach is to renew automatically wherever possible. The state must first attempt to redetermine eligibility using data it already has or can obtain — an ex parte renewal — and only if that fails may it send the enrollee a pre-populated renewal form to complete. The enrollee then has at least thirty days to respond, and if coverage is terminated for failure to respond, there is a reconsideration period — typically ninety days — during which they can submit the information and have coverage restored without a new application.

Understanding why this design exists makes you far more useful on these projects. Every step that requires a human to receive mail, understand it, gather documents and return them by a deadline loses people who are still eligible. The measure of a good renewal system is therefore the ex parte rate: the proportion of renewals completed automatically with no action required from the person. States vary enormously on that number, and improving it is the single highest-impact change available in this whole domain.

The practical engineering problems behind it are entirely familiar. Data matching against imperfect sources with names, dates of birth and addresses that do not agree. Address quality, since people move often and mail is the primary channel. Household composition, which changes and is hard to verify. Income that is variable, seasonal or from gig work, where a monthly snapshot misrepresents annual eligibility. And notice content that has to be legally sufficient and actually comprehensible.

The post-pandemic unwinding demonstrated all of this at national scale. Continuous enrolment protections ended in 2023, states worked through an accumulated backlog of renewals, and millions lost coverage — a substantial proportion for procedural reasons rather than because they had become ineligible. It is the clearest available evidence that administrative design is a determinant of health coverage.

How care is delivered and paid for

Two arrangements exist and most states use both.

Fee-for-service, where the state pays providers directly for each service.

And managed care, where the state pays a health plan a fixed amount per member per month and the plan is responsible for arranging care. The large majority of Medicaid enrollees are in some form of managed care.

The plan types have names you will meet in data models. Comprehensive risk-based plans cover the full benefit. Limited-benefit plans cover one area such as behavioural health, dental or transport. Primary care case management pays a small fee for care coordination while keeping services on fee-for-service. And managed long-term services and supports cover nursing home and home-based care for people who need daily help.

Special plans exist for dual eligibles, designed to coordinate the Medicare and Medicaid benefits that would otherwise be administered separately.

Managed care carries its own federal rulebook, and the requirements it imposes are the source of most of the recurring work. Rates must be certified as actuarially sound. Networks must meet adequacy standards for access and travel time. Plans must meet a medical loss ratio standard, meaning a defined minimum share of the payment must go to care rather than administration. There is a defined grievance and appeal process ending in a state fair hearing. An external quality review organisation independently assesses plan performance every year. And plans must submit encounter data — a record of every service delivered, in claim format, even though the plan was paid by capitation rather than per service.

That last requirement deserves emphasis because it is a perennial data quality problem. Encounter data is how the state knows what care was actually delivered for its money, how rates are set for future years, and how quality is measured. Plans have historically submitted it late, incompletely and inconsistently, and improving encounter data completeness is one of the most valuable and least glamorous projects in Medicaid analytics.

Drugs in Medicaid

Medicaid gets statutory rebates from manufacturers as a condition of having their drugs covered, calculated from formulas based on the manufacturer's average price and its lowest price to other purchasers, with additional rebates when a drug's price rises faster than inflation.

States run preferred drug lists and may require prior authorisation for non-preferred products, and many negotiate supplemental rebates on top of the federal ones.

For a pharmaceutical client, the consequence is that Medicaid pricing interacts with every other price they set, because the rebate formula references their other prices. This is one of the reasons drug pricing decisions are made by specialised teams with their own analytics and are treated as extremely sensitive.

The systems, and what makes them distinctive

A state Medicaid programme runs on an eligibility and enrolment system; a claims and encounter processing system, historically called the Medicaid management information system; provider enrolment and screening; care management; reporting; and integration with plans, providers and federal partners.

Three features distinguish this work from commercial healthcare IT.

Federal funding comes with conditions. States receive enhanced federal funding for system development on condition that systems meet federal standards, are built in modules rather than as monoliths, and are certified against defined criteria. This is why Medicaid modernisation programmes are structured as a series of module procurements rather than one large replacement, and it is a genuine opportunity for smaller vendors.

Everything is reported upward. States submit detailed eligibility, claims and encounter data to the federal government in a defined national data set, and its quality is monitored.

And the people affected are the least able to absorb administrative failure. A commercial insurance member who receives a confusing letter can call during working hours from a stable address. A Medicaid member may be working two jobs, moving frequently, without reliable internet, and managing a disability. Designing for that reality — text messaging rather than post, address updates from any touchpoint, plain language, mobile-first, multilingual — is not a nice-to-have; it is the difference between a system that keeps eligible people covered and one that does not.

Next: Chapter 25.31, health data itself — HL7, FHIR, and why making two hospitals' systems understand each other has taken thirty years.