On Tuesday, PA Governor Tom Corbett stated that at this time he cannot recommend accepting $38 billion in
federal funding to expand Medicaid under the Affordable Care Act, thereby denying medical
assistance to more than 700,000 Pennsylvanians. This series of posts
will consider the implications of that decision.
It is difficult to arrange a definitive
test of whether a social policy such as Medicaid is effective in
achieving its goal of better health. In order to demonstrate causality, you must run an experiment with a
randomized control group design, in which some people are randomly assigned to receive Medicaid (the experimental group), while others are randomly
assigned to not receive it (the control group). Random assignment is
critical. You can't compare Medicaid recipients to all non-recipients
because to be eligible for Medicaid, you must be poor, and poor
people have worse health outcomes. Since Medicaid is voluntary, you
can't compare people who sign up and receive Medicaid to other
eligible people who don't sign up, because people seek out health
insurance when they are ill. While these flaws may seem obvious, you
should be careful. Opponents of government health insurance will
sometimes cite these flawed comparisons to convince people
that Medicaid is counterproductive.
Assuming that a randomized control group design is not possible, there are two general ways to evaluate a social reform such as Medicaid expansion. In a time series design, you measure the outcomes of a group of people from before to after the change is implemented. The main problem with this design is that other events may occur at the same time as the reform, and they may serve as alternative explanations for the results. In a comparison group design, you compare the outcomes of a group of people who receive the treatment to a comparison group that does not receive it. Outcomes are measured at the same time. The problem with this design is that the two groups may not have been equivalent at the beginning of the study. Any irrelevant difference between the two groups can be an alternative explanation for the results. It is possible to combine the good features of both these designs in a time series design with a comparison group. However, it is still possible that some outside event that coincides with the treatment is affecting one group more than the other.
All rights reserved by forwardstl |
I will discuss two studies, both published in 2012, that evaluate
Medicaid outcomes. Since these two studies are superior to any that have
gone before, previous studies are basically irrelevant. A study by Benjamin Sommers and others, published in the New
England Journal of Medicine,
utilized a time series design with comparison groups. One of its
strengths is that it used three experimental groups and four
comparison groups. In 2001 and 2002, three states, New York, Maine and
Arizona, substantially expanded Medicaid by relaxing their
eligibility requirements. For example, in New York,
you could previously apply for Medicaid if you were below the
federal poverty level. In 2001, people were
allowed to sign up if their income was at or below 150% of the poverty
level. For each of these states, they selected geographically close and
demographically similar comparison states that did not expand Medicaid access. New York's comparison state was Pennsylvania, Maine's
was New Hampshire, and Arizona's were Nevada and New Mexico. Since
they were interested in whether Medicaid saved lives, the primary
outcome measure was the mortality rate,
which in this country is reported at the county level. All the
outcomes were measured from five years before the change until five
years after.
The results showed that prior to Medicaid expansion, there were no significant differences in mortality between the expansion and comparison states. After they implemented the expansion, these states showed a 6.1% reduction in mortality relative to the comparison states. Additional analyses showed that, as you might expect, the decline in mortality was greatest among the poor, minorities and older adults. Survey data showed that Medicaid expansion was associated with a 24.7% increase in Medicaid coverage, a 21.3% decrease in the rate of delayed care due to cost, and a 3.4% increase in number of people saying their health was “excellent” or “very good.” The authors calculated that one life per year was saved for every 176 adults that were added to the Medicaid rolls.
As
impressive as these results are, they do not prove that Medicaid caused these health improvements. A critic might argue that
these three states—especially New York, which showed the greatest
drop in the death rate—are not typical of the rest of the country, and thus the study exaggerates the benefits of Medicaid.
Fortunately, circumstances have given us a randomized control group
design with which to evaluate the effects of Medicaid. This is the
“gold standard” for social policy research. In 2008, Oregon
attempted to expand its Medicaid program, but didn't have enough
money. They invited people who were eligible to apply. Ninety
thousand people applied, and 10,000 of them were randomly selected to receive Medicaid in a
lottery. Amy Finkelstein and her colleagues are conducting an
ongoing survey comparing the lucky winners to those who applied but
were turned away. They reported some preliminary results last year.
The
main finding is that the Medicaid group is 25% more likely than the control group to report
themselves in “good” or “excellent” health, as opposed to “fair” or
“poor” health. More importantly, 40% fewer people in the
experimental group reported a decline in their health over the last
six months. (The reason this difference is so much greater than in the Sommers study is that Finkelstein only compared Medicaid recipients to those who were turned away, while Sommers' data estimated the health of everyone in these states regardless of whether they were enrolled in Medicaid.) As you would expect, the Medicaid group reported more
doctor and hospital visits, more preventive care, and fewer unpaid
medical bills.
The number of people in the Oregon study is too small to
detect meaningful differences in mortality. Nevertheless, the two
studies converge to give us the best evidence we have ever had that
Medicaid improves its recipients' health and saves some of their lives. In
the next post in this series, I will look at cost considerations.
You may also be interested in reading:
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 2)
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 3)
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 4)
You may also be interested in reading:
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 2)
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 3)
Tom Corbett to PA's Working Poor: "Drop Dead!" (Part 4)
No comments:
Post a Comment
Comments are always welcome.