Read on Twitter
Next up: @ashdgandhi on selective admissions in the nursing home industry.
A perennial problem in healthcare is selection. Nursing homes want to try to admit patients who generate a lot of revenue at lower cost.
Medicaid reimbursement rates tend to be lower than Medicare. Want to avoid Medicaid patients. Costs not adjusted by care requirements. Want to avoid high-care disabilities like obesity or mobility impairments.
Then there are dynamic considerations. Facilities are mostly capacity constrained. Generally want to avoid very long-stay patients to maintain option-value to admit more profitable patients (esp. since these are often Medicaid patients).
We really want Medicaid and disability patients to be able to get adequate care, so it's a big problem if they are typically turned away by nursing homes. This is also a potential source of inequality in health outcomes (e.g. https://www.nber.org/papers/w24133 )
@ashdgandhi will try to quantify the degree of selective admissions in practice as well as consider various policies for mitigating it. Hard part: don't observe applicants or rejections.
Identification here is within-facility relationship between occupancy and admitted patient characteristics. Facilities become pickier when full. i.e. using variation in the strength of the incentive to discriminate.
Key assumption is that occupancy doesn't effect patient preferences.
As facilities become more full, less likely to be Medicaid patients
Same thing for "long-stay patients, or patients requiring antipsychotics.
Next step: predict typical occupancy at which a patient is admitted given their observed characteristics. This is a measure of how desirable that patient is.
One big question in the background: why is occupancy varying. I think the paper is trying to isolate "transitory" variation rather than say, "Occupancy goes up because Medicaid reduces coinsurances", which would be a big problem.
This is the earlier point about transitory occupancy not varying due to patient preferences.
So we have these desirability scores. Now we're going to put them in a structural model of selection. In this 20 min presentation, @ashdgandhi will verbally summarize.
Patients of different types arrive, facilities each decide whether to offer admission (if profits > opportunity cost), patients choose their favorite facility that admitted them.
Points out that getting realistic demand estimates (e.g. for staffing levels) requires taking into account facility selection as well.
Most patients get their 1st choice facility, but only about half of very long-stay patients do. These are a small minority of patients but a majority of patient days.
Counterfactual policies:
1) First-come first-serve: leads to welfare gains, better utilization of beds at high-quality facilities (more very long-term patients there)
1) First-come first-serve: leads to welfare gains, better utilization of beds at high-quality facilities (more very long-term patients there)
2) Raising Medicaid reimbursement rate: costs $37.26 per day. Actually doesn't help much. A lot of inframarginal patients. Doesn't solve underlying problem that there aren't enough beds at the facilities Medicaid patients want to go to.
But maybe there would be GE effects of this policy? Could it induce more facilities to open?
@AmandaStarc1 discussing. First point, COVID killed lots of people in nursing homes. Maybe nursing homes don't have the right incentives to invest in people not dying.
Is t he relationship between occupancy and utilization of less profitable patients causal?
It seems very plausible that it is largely supply-side as Ashvin assumes. But seems like any demand variation would also impact occupancy. Hard to see why there would be zero such variation. But direction of bias not obvious to me.
One issue @nealemahoney raises is what if hospitals have preferred networks of nursing homes. This might cause you to underestimate the degree of selective admissions (you're letting in patients because they're from the affiliated hospitals).
@LeemoreDafny points out that only about 5% of SNFs are owned by hospitals, so explicit vertical integration may not be a major concern here.
