Upcoming Presentations by Staff and Affiliates of CBO’s Health Analysis Division

Moderator
Chris Adams (CBO)

Presenters
Rena Conti (Boston University Questrom School of Business), Luca Maini (Harvard Medical School), Darshak Sanghavi (Machinify), and Sujith Ramachandran (University of Mississippi School of Pharmacy)

Description
The panel will discuss issues related to current and proposed payment policies for cell and gene therapies (CGTs). CGTs offer the potential for curative treatments for many diseases that currently have no treatment. However, CGTs could have significant up-front costs, which could pose a challenge for families, commercial insurers, and state Medicaid programs. Those challenges could make it difficult for individuals and families to access lifesaving medicines and might affect the potential returns from developing new CGTs. Are there alternative payment policies that would alleviate those problems? Are there alternative business models that could solve those problems? What issues should policymakers consider when assessing policy proposals?

Presenter
Margaret Kallus (University of California at Berkeley, dissertation fellow at CBO)

Authors
Daria Pelech (CBO) and Margaret Kallus (University of California at Berkeley, dissertation fellow at CBO)

Abstract
Medigap Access and Medicare Advantage Disenrollment Among Disabled Beneficiaries at Age 65. Medicare provided health care coverage for 7 million people with disabilities in 2024. The basic Medicare fee-for-service benefit has high cost sharing and no cap on out-of-pocket spending, which can leave enrollees exposed to substantial financial risk. To cover that cost sharing, many older beneficiaries purchase medigap plans, which supplement Medicare. But disabled beneficiaries under age 65 may find medigap unavailable or unaffordable. In some states, insurers are not required to sell medigap policies to such beneficiaries; in other states, premiums are many times higher than those charged to people over age 65.

When disabled beneficiaries turn 65, they are offered a guaranteed issue period in all states, during which insurers are required to sell them medigap plans. Depending on state policy, beneficiaries may experience a substantial drop in medigap premiums at age 65.

We combine data on medigap premiums and beneficiary-level enrollment in Medicare Advantage (MA) to test whether disabled individuals are more likely to disenroll from their MA plans at age 65. We take advantage of variation in states’ medigap regulations to examine how medigap access and premiums shape patterns of disenrollment from MA at age 65.

We find that among disabled beneficiaries who have been enrolled in MA for at least a year, the probability of disenrolling from MA is roughly three times as high at age 65 as at any other age. That increase coincides with an open enrollment period for beneficiaries and a 50 percent drop in medigap premiums, on average. The likelihood of disenrollment also varies with state policy: Disenrollment increases fourfold at age 65 in states that do not have a medigap guaranteed issue period for people under 65; doubles in states that require guaranteed issue for disabled people but allow higher premiums; and rises by only 57 percent in states that require guaranteed issue for disabled people and impose limits on variation in premiums. Lower medigap premiums and higher risk scores for beneficiaries (which project how expensive beneficiaries may be to treat) are also both associated with higher disenrollment at age 65.

Overall, the findings show that differences in medigap access and pricing before age 65 are associated with substantial shifts in enrollment behavior at age 65. The findings also point to trade-offs in medigap access for disabled beneficiaries: Expanding access could affect whether disabled beneficiaries choose MA or medigap plans, and changes in the risk pool could influence premiums for all medigap enrollees. Those dynamics are relevant for understanding how rules about supplemental coverage shape beneficiaries’ choices and plans’ costs.

Presenter
Daria Pelech (CBO)

Authors
Daria Pelech, Ru Ding, Rebecca Sachs, and Michael Cohen (all of CBO)

Abstract
This study examines whether changes in Medicare’s payments for hospital outpatient department (HOPD) services affect what commercial insurers pay providers. Specifically, we test whether the adoption of a site-neutral policy in Medicare—which aligned Medicare’s HOPD payments with its payments for similar services in physicians’ offices—affected what commercial insurers pay for those same services. Although commercial payers often use Medicare’s physician fee schedule as a benchmark for their payments, there is conflicting evidence about whether changes in Medicare’s payments for inpatient hospital services spill over to commercial insurers and limited or no evidence of such spillovers for outpatient services. We provide the first national evidence about whether Medicare’s site-neutral payment reforms spill over into commercial HOPD payments.

We analyze Medicare’s 2018 site-neutrality policy, which reduced facility fees for most off-campus hospital clinic visits by 60 percent but left payments for on-campus clinic visits unchanged. That policy led to a 15 percent reduction in facility payments for hospital outpatient clinic visits, on average, across the zip-code areas included in the study. We combine Medicare data on the percentage of claims affected by the policy in each zip-code area with commercial claims data for 49 states from the Health Care Cost Institute. We then test whether commercial insurers’ facility fees declined more in areas that were more exposed to Medicare’s payment reduction.

We find limited evidence that commercial insurers shadowed Medicare’s site-neutral policies. Commercial payments to off-campus facilities decreased more in areas with greater Medicare payment cuts, but those effects were driven by payment changes in one large state. Changes in other states and changes for claims that were not clearly billed as off-campus were minimal. Moreover, insurers who adopted the site-neutral policy seem to have implemented it differently than Medicare, eliminating facility fees for clinic visits altogether rather than reducing them by 60 percent. Unlike Medicare, they also appear to have extended payment cuts to at least some on-campus providers.

Those findings suggest that although Medicare’s payments can influence commercial insurers’ HOPD payments, adoption of Medicare’s payment policies for HOPD care is limited and mixed. Insurers’ ability to adopt payment changes may depend on the information they have about providers, such as whether a provider is an off-campus hospital department. Making that information more transparent might result in greater adoption of site-neutral policies by commercial insurers, which in turn could lower commercial prices.

Presenter
Eric Schulman (CBO)

Authors
Eric Schulman, Chris Adams, and Anthony Montano (all of CBO)

Abstract
Bad Controls and Double Machine Learning: Estimates From Drug Utilization After the Medicaid Expansion. Machine learning is becoming more common in applied policy analysis, but it is often unclear when machine learning improves causal inference beyond transparent fixed-effects methods and difference-in-differences (DiD) methods. We address that question using double machine learning (DML) in three steps, combining a Monte Carlo simulation with two motivating policy examples to clarify when DML helps and when it can hurt.

First, the Monte Carlo simulation provides a controlled setting for comparing approaches: DML can outperform fixed-effects methods when unobserved determinants of outcomes drift over time and when rich pretreatment covariates serve as proxies for that drift. But DML can fail when contemporaneous controls are affected by treatment. Second, we illustrate such a bad-controls problem in a real-world application, the Affordable Care Act’s Medicaid expansion. In that context, DML with contemporaneous controls is distorted and can perform worse than standard DiD and fixed-effects methods, whereas DML with pretreatment controls carried forward can produce estimates closer to DiD benchmarks and prior evidence. Third, we describe an early-stage Medicare Part D application where DML may be more useful than panel fixed effects because secular changes in the mix of prescriptions, especially generic substitution, interfered with predicted spending.

(The authors and text of this abstract have been updated since it was posted to ASHEcon’s website