Transcript Saving_Babies

Saving Babies: The Efficacy and Cost of
Recent Changes in the Medicaid
Eligibility of Pregnant Women
The Journal of Political Economy, Vol. 104, No.6
(Dec., 1996), 1263-1296
Authors: Janet Currie; Jonathan Gruber
Presented by (Jason) Chia-cheng Liao
April, 2004
• The paper try to answer:
• the key question for health care reform in the U.S. is
whether expanded health insurance eligibility will lead to
improvements in health outcome.
• A nine infant death per 1000 births, he U.S. infant
mortality rate is among the highest in the industrialized
world. The high rate is thought to reflect large numbers of
unhealthy newborns.
– How can this be improved? Then, where? 1. the poor or 2. more
coverage?
– Do women use the adequate prenatal care?
• How much is the Take-out?
• How much is the “Cost”?
• Is this cost effectiveness?
• In effort to increase the use of prenatal care, the past
decade has seen a rapid expansion in the eligibility of
pregnant women for Medicaid.
• The extensions of eligibility groups provide a natural
experiment and case study of whether changes in eligibility
can actually improve infant health and related topics.
• The authors use several data sets to investigate the
questions. The data sets include the Aid to Families with
Dependent Children (AFDC) program data, Current
Population Survey (CPS), Vital Statistics data, Medicaid
expenditure from the Health Care Financing
Administration (HCFA), and National Longitudinal Survey
of Youth (NLSY).
• The backbone of the methodology is a simulation model
of each state’s Medicaid eligibility for pregnant women
over the 1979-92 period.
• The authors divide the eligible people of Medicaid into two
groups:
– 1. “Targeted changes” applied to specific low-income groups;
They included: a) changes in eligibility for cash welfare under
AFDC program and b) changes that allowed pregnant women with
income below AFDC cutoff to receive Medicaid regardless of
family structure ( i.e. the single parent).
– 2. “Broad changes” extended Medicaid coverage to all women
with income less than specified level (e.g., 185 percent of the
federal poverty level) minus the people in group 1.
• The women in “broad changes” group have much higher income than
group “targeted changes”. For example, in Texas, the cutoff of AFDC
for a family of four was only 24 percent of the poverty line in 1979.
• By April 1990, a uniform minimum threshold had established: all
states were required to cover pregnant women with incomes up to 133
percent of the property line. States have the matching option to cover
up to 185 percent. Some states had even cover beyond that by stateonly funds.
15-44-year-old
Women’s
Medicaid
Eligibility By
State Over
Time
Methodology
• The effect of the eligibility changes on birth outcomes
– Using the Vital Statistics Data, the y (dependant) variables are: 1.
the incidence of low birth weight (less than 2,500 grams) and 2.
the infant mortality rate in each state and year.
• Regress these two (state/ year) outcomes on the fraction of
15-44-year-old women in the event of pregnancy with
Medicaid coverage.
• Problem of this strategy is that the actual fraction eligible
depends on the economic and demographic characteristic
of the state, which may also be correlated with birth
outcomes.
• For example, a state recession is associated with both
increases in eligibility and a higher incidence of low birth
weight. Then this situation could induce a spurious effect.
(biased estimation, due to omitted variables. )
Solution: simulation sampling
• The authors instrument the actual fraction eligible of
Medicaid in a state and year that depends only on the
state’s eligibility rules.
• To create the instrument, the authors first take 3,000
women from CPS in each year. Then, they calculated the
fraction of eligible women for Medicaid in this sample.
• Then, they estimated the instrumental fraction that depends
only on the legislative environment and is independent of
other characteristics of states.
• The final analysis also employed the state fixed effects
regression to control the state characteristics and Medicaid
policy.
•
•
•
The point estimation of instrument regression suggests that a 30-percentage-point in
eligibility increase (rough rate of actually happens over this time period) would lead to a
reduction of 1.9 percent of the incidence of low birth weight. ( .3x4.347/68.12=.019)
The instrument regression of the target group indicates that the 30% point in eligibility
rise was associated with an 7.8 percent decline in the incidence of low birth weight.
The result for the broad group is only 1.5 percent reduction.
*In column 4 part B, the instrumental variables regression of ‘the target’ indicates that
the 30-percentage-point rise was associated with an 8.5 percent decline in the infant
mortality rate.
*the result for ‘the broad’ is only 2.2 percent decrease of the infant mortality rate in part
C.
*The column 5 and 6 are the y incidence combined both the low birth weight and the
infant mortality rate.
Robustness Issue:
• The author also concerned the robustness of the results.
• This is caused by those significant results of previous table
could be driven by some outlying observations.
• To reduce the risk, the authors also made the robust regression
analysis. (M or R robust?)
• The results are the following:
The results are similar to those in the table 3, but the coefficients are slightly
smaller.
The only significant changes is the overall results for low birth weight are no
longer statistical significant.
How much is the take-up?
• The March CPS asks individuals whether they were
covered by Medicaid in the previous year.
• The authors estimated the marginal take-up rate for these
Medicaid policy change;
• that is, for every 100 women made eligible for the
coverage of pregnancy, how many women additional
report coverage?
• Of women 15-44 years old, about 11.4 percent of all
women were pregnant at some point during one year of the
authors’ sample years (the years of changing policy); this is
the baseline rate (i.e. if the full take-up, the rate is 0.114.)
How much is the take-up?
This table was regressed by Linear Probability Models included state and time
dummies. The CPS data set consists of 526,830 observations over 14 years.
The Medicaid policy makes an extra woman covered will raise the odds that she
will be covered by 3.9 percent.
Relative to the baseline, this is a take-up rate of 34 percent (34% = 3.9/11.4).
The take-up
The overall take-up rate is 34%.
The ‘targeted’ take-up rate is 49%, and the ‘broad’ take-up rate is
only 16%.
The authors give two reasons why the ‘broad’ take-up rate is low:
1st, the ‘broad’ population was less needy.
2nd, the broader policy changes may have been less effective.
It may be difficult to bring women who have never received
any social assistance into the Medicaid program, either they do
not know about it or the stigma effects. Previous research
report that many low-income families and their physician are
unaware that they can qualify for Medicaid.
The authors also ran the cost regression analysis as the
following:
The Cost Effectiveness:
the authors normalize and deflate health expenditure by Consumer Price Index and
use the dollars in 1986.
For example, the coefficient in the col. 2 refers to the the actual fraction made
eligible of all expenditure under the targeted change. (notice that: .301= .092+
.171+ .038)
The Cost Effectiveness
From the previous table, the majority of the spending comes
through inpatient hospital costs.
The most striking finding (from the simulated models) is
that spending per ‘broad’ eligible is actually higher than
spending per ‘target’ eligible. (the richer use more than the
poorer)
Among targeted eligible, only about half of the spending is on
inpatient hospital services, whereas among broad eligible,
over 90 percent of spending is on these services.
The Cost Effectiveness
The cost of saving a infant’s life:
The authors also estimated the increase of actual expenditure
is $202 per year per additional eligible women. $224 per
targeted eligible per year.
This leads the cost of saving a life through targeted eligibility
changes was $840,000. (if interested, check with the
footnote 22 for details)
Is this worthy?
According to the other research (Tengs et al, 1995) , the child
restraint system in cars costs about $5.5 million per child
life saved.
The similar prenatal care program also costs $1.06 million to
save a life according the Institute of Medicine.
Parental Care Utilization analysis by NLSY: Instrumental eligibility is using the
actual and simulated fractions by the CPS. Linear probability models is
employed in this analysis.
OLS in col. 1 says the targeted Medicaid-eligible women are likely to delay
prenatal care, but the instrumental models say otherwise.
Actually, the targeted group reduced the probability of delay by almost the half
when the simulated instrument is used (in col. 3).
Conclusion
Overall, the Medicaid is quite efficient especially in the
‘targeted’ group. The efficiency includes the higher take-up
rate, more cost-effectiveness, and good utilization of the
prenatal care.
The authors conclude the targeted expansions were clearly
more cost-effective than the broad eligibility changes,
these results could be good references for the improvement
of future health insurance policies.
Simulation Method: J. Cook and L. A. Stefanski. A simulation
extrapolation method for parametric measurement error models.
Journal of the American Statistical Association, 89: 1314-1328, 1995