Medicare Home Health Claims Policy Evaluation

The policy question

CMS introduced pre-claim review for home health services in selected states. Under this policy, a home health agency began care and then submitted documentation for review before Medicare paid the claim. The analytic question was whether the policy changed home health spending and use, and whether it had the unintended effect of increasing hospitalizations.

This is a difficult question because states were not randomly assigned to the policy. Their home health markets, spending levels, patient populations, and earlier trends differed. A credible analysis therefore needed detailed longitudinal data and a carefully constructed comparison for each participating state.

Data foundation

The published analysis assembled monthly measures from January 2014 through November 2023 using three national Medicare data sources:

• 100% Traditional Medicare home health claims to measure paid amounts, patients receiving home health care, service dates, and billing agencies.
• Master Beneficiary Summary File to identify Part A-enrolled Traditional Medicare beneficiaries by month and state of residence.
• Medicare Provider Analysis and Review file to identify inpatient and skilled nursing facility discharges and count acute hospitalizations.

The enrollment file is essential. Claims provide the numerator - spending, users, or hospital stays - while the enrollment file defines the population at risk. This allows valid measures per beneficiary or per 100 beneficiaries, even when enrollment differs across states and changes over time.

How the measures were built

The unit of analysis was the state-month. Outcomes were calculated consistently for every treated and comparison state before statistical modeling.

• Home health spending per beneficiary: claim paid amounts grouped by claim start date and provider state, divided by eligible beneficiaries and adjusted to January 2025 dollars using the Consumer Price Index.
• Home health users per 100 beneficiaries: unique patients with a home health claim overlapping the month, divided by the monthly beneficiary count.
• Spending per home health user: spending divided by the number of patients using home health services.
• Post-acute home health users: patients whose home health spell began soon after an inpatient or skilled nursing facility discharge.
• Community-initiated home health users: all home health users minus those classified as post-acute.
• Hospitalizations per 100 beneficiaries: acute hospital stays from short-term acute and critical access hospitals.

Turning claims into episodes of care

A single period of home health care can generate multiple claims, so the researchers first joined claims into clinically meaningful spells. A later episode beginning within 60 days of the prior episode's end was treated as part of the same spell.

Each spell was then linked to inpatient and skilled nursing facility records. If the spell began within 14 days after a qualifying discharge, all claims in that spell were classified as post-acute home health. The remaining home health patients were classified as community-initiated. This distinction helped test whether a policy affected care following a facility stay differently from home health that began in the community.

Intervention and comparison states

The main analysis examined Illinois, Ohio, North Carolina, and Florida, using each state's actual implementation timing. Forty-six states that did not implement pre-claim review during the analysis period formed the potential comparison pool.

Texas was excluded because its implementation coincided with the beginning of the COVID-19 pandemic. That timing made it difficult to separate the policy effect from the pandemic. Oklahoma was treated as untreated because no usable post-policy months remained in the study period. These choices illustrate an important claims-analysis rule: more data do not repair a comparison whose timing is fundamentally confounded.

How the synthetic control worked

For each policy state, the researchers created a synthetic comparison state: a weighted combination of untreated states selected to reproduce the policy state's outcome before implementation. A state could receive a large, small, or zero weight. The weights differed by policy state and by outcome.

After establishing a close pre-policy match, the analysis compared the actual state's post-policy trend with its synthetic counterpart. The gap between them represented the estimated policy effect. State-specific estimates were also pooled, and uncertainty was calculated with 500 placebo repetitions.

Mechanism and geographic analyses

Measuring total spending alone would not explain why it changed. The study separated changes in the number of home health users from changes in spending per user, and separated post-acute use from community-initiated use. It also examined the number of home health agencies billing Medicare per 10,000 beneficiaries.

The Illinois analysis was divided into Cook County and the rest of the state. This geographic decomposition tested whether a statewide average masked concentrated local changes. Claims data support this kind of analysis because services, providers, and beneficiaries can be followed consistently across place and time.

Data-quality decisions

• January 2020 was excluded because the Patient-Driven Groupings Model changed home health claim structure and produced an artificially low spending value.
• December 2023 was excluded because it was the incomplete final month of the study data.
• Spending was adjusted for inflation so changes in dollars did not simply reflect general price growth.
• Policy dates were specified at the monthly level, including a pause and restart in Illinois and phased implementation in other states.
• Provider state was used for home health activity, while beneficiary state of residence defined the enrolled population.

Robustness checks

The technical appendix did not rely on one model specification. It repeated the analysis using synthetic difference-in-differences, tested alternative implementation months, and excluded agencies in treatment states whose claims were handled by Medicare Administrative Contractors outside the demonstration. The researchers also reported outcome-specific synthetic-control weights, baseline means, confidence intervals, and state-level trends.

These checks address different risks: sensitivity to the statistical estimator, uncertainty about the exact policy start, inclusion of providers not actually subject to the intervention, and poor pre-policy fit between a state and its synthetic comparison.

What this approach can deliver

• Monthly policy-effect estimates by state, market, county, provider type, or patient subgroup.
• Per-beneficiary spending and utilization measures with valid enrollment denominators.
• Episode-based measures that distinguish care pathways rather than simply counting claims.
• Mechanism analyses showing whether changes arise from patient volume, intensity, provider supply, or site of care.
• Safety and substitution outcomes, such as hospitalizations or use of other care settings.
• Transparent comparison-group weights, confidence intervals, trend plots, and sensitivity analyses.

Published findings in context

The published study reported that effects varied substantially across states. The value of the example is not one headline estimate; it is the analytic framework used to determine where a policy appeared to change spending, what component of use changed, and whether hospitalization patterns changed at the same time.

Source publication and technical appendix

Marr J, Meyers DJ, Ryan AM. Pre-Claim Review and Traditional Medicare Home Health Spending: Evidence From 4 States. Health Affairs. 2026;45(6):692-699. doi:10.1377/hlthaff.2025.01356. The detailed measure definitions, control weights, estimates, and sensitivity analyses are reported in the article's online appendix.

Frequently asked questions

How can Medicare claims be used to evaluate a policy change?

Claims can be converted into consistent measures of spending, service use, patient counts, care pathways, and hospitalizations before and after a policy begins. A comparison group then estimates what likely would have happened without the policy.

What does the beneficiary enrollment file add?

It defines who is eligible for the analysis in each month, distinguishes Traditional Medicare enrollment, supplies beneficiary geography, and provides the denominator for per-beneficiary utilization and spending rates.

What is a synthetic control?

It is a weighted combination of untreated areas selected to match the treated area's outcome pattern before the policy. Its post-policy trend provides an estimate of the outcome expected without the intervention.

For Medicare policy evaluation, home health analysis, synthetic-control studies, or longitudinal claims analytics, please email us.

See more case studies