Reading Notes - Randomized trial shows healthcare payment reform has equal-sized spillover effects on patients not targeted by reform
---------------------------------------------------------------------------------------
My Reading Notes
Paper: Randomized trial shows healthcare payment reform has equal-sized spillover effects on patients not targeted by reform
Paper By Liran Einav, Amy Finkelstein, Yunan Ji, and Neale Mahoney
What is this paper about?
-
Spillovers in health care
-
Medicare changes how they pay for something to one group, that affects a different group
-
Changes for traditional Medicare affect Medicare Advantage patients
-
At similar magnitude
-
-
Possible reasons:
-
High fixed cost, low marginal cost technological innovation
-
New computer system built for traditional Medicare might be easy to extend to MA
-
(this is the reason they think is most likely - since their effects are biggest in hospitals where there are a lot of directly affected patients, and their survey of hospital admin suggests this is how the admin adjusted to the changes)
-
-
Provider constraints: not knowing insurance type and/or having preferences
-
What gap is this paper filling?
- Policies that ignore spillovers miss a big part of the effect of the policy
- Any study that looks at the direct effects only may be biased since the “control group” might also be experiencing the treatment effect
- Designing payment policies in the presence of spillovers is different depending on the number of insurance providers.
What makes this paper interesting?
-
The TM to MA spillover effects are an important piece of the healthcare market
-
This shows how to do spillover studies in general and in health care
-
Spillover effects are important for general equilibrium understanding
-
The use of the national randomization of the Traditional Medicare Spending change - cool natural experiment
--------------------------------------------------------------------------------------------------------
What are the results?
Targeting traditional medicare patients with a payment reform that changes payment from fee-for-service to a bundled comprehensive payment reduces discharges to post acute care facilities for both the targeted traditional medicare patients and non-targeted Medicare Advantage patients, who are presumably only seeing any impact because of spillover effects. The impacts are roughly the same for both groups (about 3 percentage points).
What assumptions are being made?
-
Randomization worked
-
“the (admittedly strong) identifying assumption is that, conditional on the covariates, there are no hospital-specific time trends, and thus any heterogeneity in the change in outcomes across hospitals reflects heterogeneous treatment effects.”
-
“To probe the sensitivity to this assumption, we estimate an alternative specification where we include hospital-specific linear time trends as controls. These time trends are identified from the hospital-specific outcomes in 2013 and 2014. In this specification, the identifying assumption is that, conditional on covariates, there are no hospital-specific deviations from the time trend”
-
-
Non-targeted patients did not get anything directly from the program
-
Enough targeted patients to potentially affect non-targeted as well
Other Ideas
- Given more granular (at the time level) data, I think they could have shown the time trend analysis for both groups.
<!-- -->
-
Maybe could use a bordering MSA design to test what happens in non-selected MSAs that border selected MSAs
-
Causal forest to see which covariates predict bigger treatment effects
-----------------------------------------------------------------------------------------------------------------
What is the Tweet?
Spillover effects in health care are fascinating. These authors looked at the spillover effects of an MSA-level randomized Medicare program aimed at reducing spending for traditional Medicare enrollees on the non-targeted Medicare Advantage enrollees. Both groups reduced share discharged to postacute care centers by ~3.3pp.
#