Sie sind auf Seite 1von 17

HKS799

Case Number 2019.0

Expanding Health Insurance to Millions: Learning from the Oregon


Health Insurance Experiment

When Wendy Parris shattered her ankle, the emer-


gency room put it in an air cast and sent her on her way. Be-
cause she had no insurance, doctors did not operate to fix it. A
mother of six, Ms. Parris hobbled around for four years,
pained by the foot, becoming less mobile and gaining weight.
1
Annie Lowrey, New York Times, June 22, 2012

Introduction

In 2010, President Barack Obama signed into law the Patient Protection and Affordable Care Act (ACA), an un-
precedented overhaul of the American health care system. The ACA was designed to extend health insurance to
millions of uninsured Americans like Wendy Parris.

As the centerpiece of President Obama’s reform efforts, the ACA (commonly known as Obamacare) promised
to dramatically increase the number of people with health insurance, restrain high medical costs, and improve the
efficiency of the health care sector. Building on the nation’s existing system of private and public health insurance,
the complex ACA used a three-pronged approach to achieve universal health coverage: First, the law required that
insurers provide insurance to any American interested in buying it, offering the same price regardless of consum-
ers’ health status. Second, to make it financially feasible for health insurers to cover people with potentially expen-
sive medical conditions, it mandated that all Americans obtain health insurance—expanding the risk pool to in-
clude both healthy and less healthy individuals. And third, to ensure that all Americans had access to affordable
insurance, the ACA both expanded government-run insurance for the poor—Medicaid—and provided income-
2
based subsidies to help individuals buy private insurance.

1
Annie Lowrey, “Oregon Study Shows Benefits, and Costs, of Insuring the Uninsured,” New York Times, June 22, 2012.
2
Adapted from Landmark: The Inside Story of America’s New Health-Care Law and What It Means for Us All, The Staff of the
Washington Post, 2011, p. 68.

This case was written by Professor of Public Policy, Amitabh Chandra, Case Writer, Anjani Datla, and Research Manager, Emily
Myers, at the John F. Kennedy School of Government (HKS), Harvard University, for use at the HKS. Funding for this case was
provided by the Joseph B. Tompkins, Jr. Fund for Case Study and Research. HKS cases are developed solely as the basis for class
discussion. Cases are not intended to serve as endorsements, sources of primary data, or illustrations of effective or ineffective
management. (January 2013; Updated January 2016)

Copyright © 2014 President and Fellows of Harvard College. No part of this publication may be reproduced, revised, translated,
stored in a retrieval system, used in a spreadsheet, or transmitted in any form or by any means without the express written
consent of the Case Program. For orders and copyright permission information, please visit our website at
http://www.case.hks.harvard.edu/ or send a written request to Case Program, John F. Kennedy School of Government, Harvard
University, 79 John F. Kennedy Street, Cambridge, MA 02138.

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
Congressional debates leading up to the passage of the ACA hinged on the usefulness of expanding health in-
surance coverage to improve the country’s health system. Democrats argued that increasing access to health in-
surance was a moral imperative that would save lives and reduce health care costs. In a 2009 joint session to Con-
gress, Obama made the case for universal insurance, “everyone in this room knows what will happen if we do
nothing. More Americans will lose their coverage when they are sick and need it the most... And more will die as a
3
result. We know these things to be true.” The President also argued that reform would lead to important cost sav-
ings in medical care. “[The bill] does things…to make sure that people are getting the care they need and the
checkups they need and the screening they need before they get sick – which will save all of us money and reduce
4
pressures on emergency rooms all across the country.”

Republicans, meanwhile, were wary of the ACA ushering in the biggest increase in entitlement spending in
more than four decades. They warned that many dangers lurked inside the ACA, which would, among other things,
raise government spending, burden small businesses, and significantly limit consumer choice.

The political debate on expanding access to health insurance centered on a key question: What effect did
5
health insurance have on health? An estimated 25 million Americans were set to gain health insurance by 2016.
But little was known about how the insurance expansion would affect the use and provision of health care, and
ultimately, influence the health and well-being of the population.

What is Health Insurance?

Like all forms of insurance, health insurance allows individuals to guard against uncertain risks. According to
health economists at Harvard University, Katherine Baicker and Amitabh Chandra, health insurance, in its simplest
form, pools risks. “Many pay a premium up front, and then those who face a bad outcome (getting sick, being in a
car accident, having their home burn down) get paid out of those collected premiums,” they explained. For Chan-
dra and Baicker, a core aspect of health insurance is that not everyone falls sick at the same time, making it possi-
ble to pay for the care of the sick even though the cost of care is more than what the sick paid in as premiums.
“This is also why it is particularly important for people to get insured when they are healthy—to protect against
the risk of needing extra resources to devote to health care if they fall ill,” Baicker and Chandra said. “Uncertainty
about when we may fall sick and need more health care is the reason we purchase insurance—not just because
health care is expensive (which it is). Lots of other things are expensive, too, including housing and college tuition,

3
Office of the Press Secretary, “Remarks by the President to a Joint Session of Congress on Health Care,” September 9, 2009.
Available at: http://www.whitehouse.gov/the_press_office/Remarks-by-the-President-to-a-Joint-Session-of-Congress-on-
Health-Care.
4
Office of the Press Secretary “Remarks by the President after Meeting with Senate Democrats,” December 15, 2009. Available
at: http://www.whitehouse.gov/the-press-office/remarks-president-after-meeting-with-senate-democrats.
5
Congressional Budget Office, “Insurance Coverage Provisions of the Affordable Care Act—CBO’s February 2014 Baseline.”
Available at: http://www.cbo.gov/sites/default/files/cbofiles/attachments/43900-2014-02-ACAtables.pdf.

HKS Case Program 2 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
but we don’t have insurance to help us purchase them because they are not uncertain in the way that potentially
6
needing very expensive medical care is.”

The Health Chain

In 2012, the majority of working age Americans received health insurance from private insurance companies,
typically via their employers. In addition, two government funded health insurance programs provided coverage to
specific groups: Medicare, for people over 65, and Medicaid, for certain poor populations. The same year, an esti-
mated 48 million Americans lacked health insurance. The rate of uninsured was higher in poorer families. Nearly a
quarter of the people living on less than $25,000 a year were without insurance, compared to 8 percent among
7
people who made $75,000 or more annually.

Private and public health insurance plans covered both routine and unexpected care for their members—from
doctors’ visits, drugs, tests, and surgical procedures to emergency department visits. Although Americans did not
need health insurance to receive care (they could pay for care directly or, as in the case of Wendy Parris, receive
care from hospitals obligated to provide emergency services), an individual’s insurance status could determine
access to health care in different ways. For example, the uninsured were less likely to have a regular primary care
physician or to obtain recommended preventive care. In addition, because they were responsible for the full cost
of care (rather than just a copayment or deductible), the uninsured were more likely to put off all but the most
8
essential treatment.

Experts noted that while health insurance could increase the quantity of care, it did not automatically lead to
quality care. A 2003 study found that “Americans received less than 60 percent of recommended care, including
preventive, acute, and chronic care—and including such low-cost interventions as flu vaccines and antibiotics for
9
surgical patients.” A vigorous body of research in medicine and economics found that in the United States, getting
high-quality health care was more a function of geography than insurance status or health care spending. For in-
stance, among recipients of Medicare across the country, there were large differences in the quality of care re-
ceived. Surprisingly, in parts of the country where most was spent on Medicare, the beneficiaries were the least
10
likely to receive high-quality health care.

Yet, in other ways, health insurance could directly affect the quality of care. When health services are paid for
by insurance companies, people with insurance are more likely than the uninsured to use services of questionable

6
Baicker, Katherine, and Amitabh Chandra, "Myths and Misconceptions about US Health Insurance." Health Affairs 27, no.,
2008, w534.
7
United States Census Bureau Press Release, “Income, Poverty and Health Insurance Coverage in the United States: 2012,”
September 17, 2013, http://www.census.gov/newsroom/releases/archives/income_wealth/cb13-165.html.
8
Insurance, by its nature, is purchased before care is required. Those without insurance seeking care after an accident or illness
are typically responsible for the full cost of care. As Baicker and Chandra note, “try purchasing insurance to cover your recent
destruction of your neighbor’s Porsche: the premium would be the cost of a new Porsche.” (Baicker and Chandra, “Myths and
Misconceptions about US Health Insurance,” w535.)
9
Baicker and Chandra, “Myths and misconceptions about US Health Insurance,” w538.
10
Ibid, w538.

HKS Case Program 3 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
benefit. Similarly, health care providers are more likely to recommend services of limited or uncertain value to
patients covered by health insurance—those who don’t directly bear the cost of health care.

Moreover, some critics argued that quality of health care varied with the type of insurance. They claimed that
Medicaid, the public insurance program for the poor, was ineffective because many doctors refused to accept pa-
tients with insurance plans that had low reimbursement rates. A famous Wall Street Journal editorial alleged that
11
“Medicaid was worse than having no coverage at all.”

And others believed that health care by itself played a relatively small role in promoting overall health and
wellbeing. The World Health Organization (WHO) determined that a broad range of social, economic, and envi-
12
ronmental factors, as well as individual characteristics and behaviors were equally important. In fact, most of the
WHO’s list of factors that influenced health lay outside the realm of health care, such as income, social status, edu-
cation, the physical environment, social support networks, genetics, and gender.

For decades, economists and policymakers tried to understand the ways in which population health was tied
to health care access and, further, how access to health care was tied to health insurance. As the U.S. was set to
massively expand health insurance coverage under the ACA, finding conclusive answers to these longstanding
questions became ever more vital.

The Missing Causal Link

An oft-cited 2002 report by the Institute of Medicine estimated that lack of insurance was the cause of 18,000
13
deaths each year. Similarly, in 2009, an American Journal of Public Health study found that nearly 45,000 people
14
died each year because they did not have insurance. Several other studies offered estimates of deaths due to lack
15
of health insurance that ranged from 18,000 to 150,000.

The research methodology underlying these studies was largely the same. Researchers compared a group of
people with health insurance to a group without. After following both groups for a period of time, and controlling
for a number of factors like age, income, and education, the researchers compared health outcomes (or mortality
rates) between the two groups to arrive at conclusions.

But according to Dr. Alan Garber at Harvard University, “those who do not have insurance tend to be different
in many ways from people who have it. They tend to be less educated and to have worse health habits and lower
incomes. No matter how carefully researchers try to correct for the differences, they cannot be completely suc-

11
Scott Gottlieb, “Medicaid Is Worse Than No Coverage at All,” Wall Street Journal, March 10, 2011.
12
The World Health Organization, “Health Impact Assessment: The Determinants of Health,”
http://www.who.int/hia/evidence/doh/en/.
13
Institute of Medicine, Care Without Coverage: Too Little, Too Late, Washington: National Academies Press, 2002, pp. 161-165.
14
Andrew P. Wilper, Steffie Woolhandler, Karen E. Lasser, Danny McCormick, David H. Bor, and David Himmelstein, “Health
Insurance and Mortality in US Adults,” American Journal of Public Health, 99, 2009, pp. 2289-2295.
15
Ezra Klein, “The 150,000-Life Health-care Plan,” Washington Post Voices, December 14, 2009,
http://voices.washingtonpost.com/ezra-klein/2009/12/the_150000_life_health-care_pl.html.

HKS Case Program 4 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
16
cessful,” he said. “There is always some doubt.” These doubts could manifest in any number of ways. For in-
stance, healthier individuals are potentially able to hold on to better jobs that tend to come with health insurance
benefits. Or perhaps, individuals who are risk-averse are more likely to buy health insurance and also more likely to
make healthy lifestyle choices like wearing a seatbelt or eating a healthy diet.

In a comprehensive review of observational studies on U.S. health insurance, the authors observed that “many
of the studies claiming to show a causal effect of health insurance on health do not do so convincingly because the
17
observed correlation between insurance and good health may be driven by other, unobservable factors.”

This could explain why other observational studies reported opposite results. A 1994 paper based on national
survey data found that, after controlling for age and income, white males on Medicaid had roughly double the risk
18
of dying compared to men with no insurance. Yet those receiving Medicaid were likely to be different from the
uninsured along a number of unmeasured dimensions related to poverty and health status.

In essence, the observational studies were unable to determine, with any certainty, if health insurance caused
particular health impacts at the population level. On the observational evidence, former reporter for The Atlantic
Megan McArdle was skeptical of the connection between health insurance and health. “Even a rough approxima-
tion of how many people die because of lack of health insurance is hard to reach. Quite possibly, lack of health
19
insurance has no more impact on your health than lack of flood insurance,” she wrote.

For more definitive answers, researchers would need to apply a method common in drug trials: Randomly as-
sign participants to receive a new drug (treatment) or be given a placebo (control). In this instance, only some indi-
viduals without health insurance would randomly be chosen to receive insurance. The random assignment process
would ensure that, on average, both treatment and control groups are largely similar, with one major exception—
only individuals in the treatment group would have health insurance—allowing researchers to isolate the effects of
health insurance on health (and any other measured outcomes). For this reason, randomized controlled trials, as
they are commonly known, are considered the gold standard of research. Randomized experiments, however, re-
mained an elusive prospect in the health insurance realm. In addition to the costs associated with conducting such
a trial, the notion of denying some individuals access to health insurance, in the name of research, is ethically
problematic.

Remarkably, a randomized experiment on health insurance was conducted in the 1970s and 1980s. For several
years, the RAND Corporation ran a large-scale randomized study on health insurance, which became one of the
most famous and influential social experiments in U.S. history. Nearly 8,000 participants were divided into four
groups. Each group was assigned a different level of insurance coverage to establish how varying degrees of cost

16
Gina Kolata, “Study Find Benefits in Health Insurance for the Poor,” New York Times, July 7, 2011.
17
Helen Levy and David Meltzer, “The Impact of Health Insurance on Health,” Annual Review of Public Health, 29, 2008, pp.
399-409.
18
Paul D. Sorlie, Norman J. Johnson, Eric Backlund, and Douglas D. Bradham, “Mortality in the Uninsured Compared With That
in Persons With Public and Private Health Insurance,” Archives of Internal Medicine, 154, 1994, p. 2413.
19
Megan McArdle, “Myth Diagnosis: Everyone Knows that People Without Health Insurance Are More Likely to Die. But Are
They?” Atlantic, March 1, 2010.

HKS Case Program 5 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
sharing affected health care utilization and health outcomes. Researchers found that people with more generous
insurance used more medical services—like visiting a doctor or staying at a hospital—than people with less cover-
20
age. But the level of insurance, in general, did not affect participants’ health. Even this gold standard experiment
was not without limitations. The study did not include individuals without health insurance, an important consid-
eration for the ACA, which was about to bring into the fold millions of uninsured Americans.

Then, in 2008, a lottery to add 10,000 low-income adults to Oregon’s Medicaid program inadvertently set in
motion a “natural” random experiment with broad implications.

The Oregon Health Insurance Experiment

Budget cuts in the early-2000s had left the state of Oregon unable to cover many low-income adults under its
Medicaid extension program known as Oregon Health Plan (OHP) Standard. OHP Standard was designed for poor
individuals who did not qualify for traditional Medicaid, and offered comprehensive coverage for health care ser-
21
vices provided through managed care organizations. In 2008—for the first time in nearly five years—OHP Stand-
ard was in a position to support 10,000 new enrollees. But state officials knew that demand for Medicaid would be
far greater. To give all poor, uninsured Oregonians a fair chance at receiving health insurance, the state established
a lottery. Some individuals were randomly selected from a waiting list to receive insurance while others were not,
enabling comparison between the two groups, and in effect creating what an expert described as “a first-rate ran-
22
domized experiment.”

A group of renowned health economists led by Amy Finkelstein (at the Massachusetts Institute of Technology)
and Katherine Baicker (at Harvard University) were drawn to the research implications of the Oregon lottery. “Amy
and I stumbled across the lottery in Oregon and thought, this is an unbelievable opportunity to actually find out
23
once and for all what expanding public health insurance does,” said Baicker. Finkelstein agreed: “it’s hard to over-
state how excited we were. We thought it was a once-in-a-lifetime opportunity to bring the gold standard of ex-
24
perimental design to this question.”

Oregon led a large promotional campaign to publicize the lottery. State officials sent brochures about OHP
Standard’s new window of enrollment to nearly 350,000 households that already received benefits from the De-
partment of Human Services (DHS). The state also distributed materials about the lottery (printed in 10 languages)
at DHS offices, county health departments and major hospitals and clinics. Individuals could apply to the lottery by
providing their names in person at a DHS office, by phone, e-mail, web, fax, or mail.

20
J.P. Newhouse and the Insurance Experiment Group, Free for All? Lessons from the RAND Health Insurance Experiment, Cam-
bridge, Mass., Harvard University Press, 1993.
21
Another program, OHP Plus, covered the standard Medicaid population, including children and pregnant women, the disa-
bled, and families receiving welfare through the Temporary Assistance to Needy Families program.
22
Raj Chetty, “Yes, Economics Is a Science,” New York Times, October 20, 2013.
23
Gina Kolkata, “First Study of Its Kind Shows Benefits of Providing Medical Insurance to Poor,” New York Times, July 7, 2011.
24
Ezra Klein, “Is the Future of American Health Care in Oregon?” Washington Post Wonkblog, May 20, 2013,
http://www.washingtonpost.com/blogs/wonkblog/wp/2013/05/20/is-the-future-of-american-health-care-in-oregon/.

HKS Case Program 6 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
25
The state began accepting names for the lottery on January 29 and closed the process on February 28, 2008.
Nearly 90,000 individuals applied for the chance to receive health insurance in that 30-day period. The state se-
lected approximately 30,000 names through random lottery draws and invited those applicants and their families
to join OHP Standard. Only 60 percent of those who were selected from the lottery sent in their forms before the
end of the 45-day application period. And half of those who sent in their applications were ineligible because they
did not satisfy income or other eligibility requirements. Ultimately, 30 percent of those who won the lottery, or
around 9,000 individuals, were successfully enrolled in Medicaid. (see Appendix I for more on study methodology).

Oregon’s 30,000 lottery winners formed Finkelstein and Baicker’s treatment group (regardless of whether or
not they ultimately enrolled in Medicaid), while the remaining individuals not selected by the lottery made up the
control group. Comparing outcomes for the lottery winners (the “treatment” group) with lottery losers (the “con-
trol” group) provides an estimate of winning the lottery (for more on the statistical technique used to estimate
impact of Medicaid see Appendix I).

Between 2009 and 2011, OHP Standard’s budget allowed the state to offer insurance to those who were de-
nied the chance during the 2008 lottery. Finkelstein, Baicker and colleagues had just two years to gather infor-
mation on a broad range of health and non-health factors. They amassed troves of administrative data, obtaining
hospital discharge and mortality data for the entire state of Oregon and matching these records to both the treat-
ment and control individuals in the study population. Similarly, they collected records of all Emergency Depart-
ment visits to the twelve hospitals serving the Portland area over a three-year period, enabling them to study
emergency department utilization among the 25,000 treatment and control individuals in the Portland area.
Uniquely, they measured financial strain by mapping data on credit histories provided by the credit union,
TransUnion to the study sample (including records of any bankruptcies, liens, medical and non-medical collections,
etc.).

To supplement this administrative data, the research team conducted mail surveys of both the treatment and
control groups shortly after the lottery assignment and then again at the one year point, collecting information on
health care access, utilization, preventive care, financial strain, and health status. Not content to rely on self-
reported survey data to measure health outcomes, two years after the lottery placement, the researchers also
conducted health checks and in-person interviews among a sub-sample of treatment and control individuals resid-
26
ing in the Portland area (6,000 in each group). The interviews included questions on health care use and expens-
es, health status, insurance coverage, health-related quality of life, medications, and past diagnoses. The health

25
The Commonwealth Fund, “Oregon Lottery Opens Enrollment after 3-Year Freeze,” Newsletter, February/March 2008.
26
The survey protocol for the one-year mail survey involved up to three mail attempts, with a subset of non-respondents as-
signed to a more intensive protocol that included additional tracking efforts, mailings and phone calls. The first of the survey
mailings included a $5 cash incentive, and responders were entered into a lottery to receive an additional $200. The in-person
interview protocol involved far more comprehensive recruitment and tracking efforts. Participating individuals were inter-
viewed in their home or at a clinic. To compensate for their time and efforts, respondents received $30 for completing the in-
terview and an additional $20 for providing blood spots through a finger stick procedure. If they traveled to a clinic for the in-
terview, they were also compensated $25 for travel expenses. As an additional benefit, respondents were presented with re-
ports of their results on the physical health measures (body mass index, weight, blood pressure, cholesterol, blood sugar) as
well as information about local health resources. The survey had an effective response rate of 50 percent; in contrast, the in-
person interview and health checks achieved an effective response rate of 73 percent.

HKS Case Program 7 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
screening included measurements of weight, blood pressure, cholesterol, blood sugar, as well as a screening for
depression. Through these measures, the researchers hoped to capture the effect of Medicaid on the major risk
factors for chronic diseases like diabetes, hypertension, and heart disease, which were commonly thought to be
affected by medical care access. The study population was predominantly white and 56 percent female. More than
half of the participants were unemployed, while a significant portion reported chronic health conditions such as
asthma, diabetes, high blood pressure and depression (see Exhibit A).

Findings

It was not surprising that the Oregon Experiment drew significant attention in both academic and policy cir-
cles. Under the ACA an estimated 12 million more Americans would be enrolled into Medicaid by 2016, and 22
27
million via health exchanges—websites on which consumers can purchase insurance.

In 2012, Finkelstein, Baicker and co-authors reported on results from the first year of data collection. There
were several key differences in health care use between the insured and uninsured. Medicaid increased the likeli-
hood of being admitted to the hospital by 30 percent, outpatient care by 35 percent, and prescription drug use by
15 percent. Self-reported access to care also improved dramatically, with 92 percent of Medicaid recipients report-
ing that they received all necessary medical care over the past six months, compared to 68 percent in the control
28
group.

Wendy Parris, who had been living in Oregon with a broken, painful ankle for four years, was one of the Medi-
caid lottery recipients. According to the New York Times, “Ms. Parris got surgery for her foot [and] additional spinal
surgery. Doctors helped address her depression, brought on by the death of one of her children. Her weight came
29
back down, and her mobility was far better. ‘Medicaid saved my life,’ [Ms. Parris] said.” In surveys, the Medicaid
recipients were more likely to report that their health had improved since receiving health insurance.

After another year of follow up, data from the in-person interviews and health checks shed more light. The
two-year results confirmed initial findings on health care use. Overall, patients who received Medicaid coverage
through the lottery spent an estimated $1,172 more annually on health care—36 percent more than the unin-
sured. They also used more prescription drugs, had more doctor visits, and were more likely to have received pre-
30 ,31
ventive care, including cholesterol screening, Pap smears, and mammograms.

27
Ibid #5.
28
Amy Finkelstein, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joseph P. Newhouse, Heidi Allen, and Kathe-
rine Baicker, “The Oregon Health Insurance Experiment: Evidence from the First Year,” The Quarterly Journal of Economics 127,
no. 3, 2012, pp. 1057-1106.
29
Annie Lowrey, “Oregon Study Shows Benefits, and Costs, of Insuring the Uninsured,” New York Times, June 22, 2012.
30
Katherine Baicker, Sarah L. Taubman, Heidi L. Allen, Mira Bernstein, Jonathan H. Gruber, Joseph P. Newhouse, Eric C. Schnei-
der, Bill J. Wright, Alan M. Zaslavsky, and Amy N. Finkelstein, "The Oregon Experiment—Effects of Medicaid on Clinical Out-
comes," New England Journal of Medicine, 368, no. 18, 2013, pp. 1713-1722.
31
In contrast to the findings using administrative data, self-reported data from the two-year interviews did not show a statisti-
cally significant increase in emergency department use. The researchers discuss possible reasons for the discrepant findings in
Taubman et al, “Medicaid Increases Emergency-Department Use: Evidence from Oregon’s Health Insurance Experiment.”

HKS Case Program 8 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
Though an increase in health care utilization was expected, the magnitude and direction of some of the find-
ings were unexpected. Proponents of the ACA had long argued that emergency department visits would decrease
when patients had access to the preventive and routine care they required. The presumption was that many peo-
ple resorted to visiting the emergency department because they lacked access to other, more cost-effective
sources of health care. The findings from Oregon did not bear out this prediction. In results released in early 2014
(based on a subsample of 25,000 individuals in the Portland area), the Oregon team found that emergency de-
partment visits increased by roughly 40 percent for those who gained coverage under the lottery. Even when the
researchers considered only visits for conditions that could have been treated through primary care, the dramatic
32,33
increase in visits to the emergency department persisted (see Exhibit B).

In contrast to health care utilization, the impacts on health were mixed. Medicaid had no significant effect on
the typical markers of physical health, such as blood pressure, cholesterol, or blood sugar levels, despite the great-
er likelihood that Medicaid recipients were diagnosed with diabetes and given treatment (see Exhibit C).

On the other hand, Medicaid demonstrated powerful effects on mental health. Insurance reduced depression
rates by 30 percent. Baicker observed that “being uninsured is incredibly stressful from a financial perspective, a
psychological perspective, a physical perspective. It is a huge relief for people not to have to worry about it day in
34
and day out.”

Medicaid also had a dramatic impact on financial security, virtually eliminating catastrophic medical expenses.
In the control group, 5.5 percent of respondents reported such expenditures, and Medicaid reduced this share to 1
percent. Medicaid also decreased by more than half the number of individuals who had to borrow money or avoid
paying other bills to cover medical expenses (see Exhibit D).

Reactions

A media maelstrom ensued in the wake of the Oregon Experiment’s report. Policy wonks and political pundits
focused on the study’s mixed results to burnish arguments either for or against the ACA. Avik Roy in Forbes maga-
zine remarked that the Oregon results called into question the “$450 billion a year we spend on Medicaid, and the
35
fact that Obamacare throws… more Americans into this broken program.” On the other hand, Jonathan Cohn at
the New Republic thought the study bolstered the case for expanding Medicaid. “The big news is that Medicaid
virtually wiped out crippling medical expenses among the poor. In addition, the people on Medicaid were about
half as likely to experience other forms of financial strain… That may sound obvious—of course people with insur-

32
Sarah L. Taubman, Heidi L. Allen, Bill J. Wright, Katherine Baicker, Amy N. Finkelstein, “Medicaid Increases Emergency-
Department Use: Evidence from Oregon’s Health Insurance Experiment,” Science, Published online January 2, 2014.
33
The increase in emergency department use for primary care treatable conditions was unlikely to reflect an especially severe
lack of access to primary care providers in Portland, as the city had a higher number of primary care physicians per 100,000
residents than the U.S. average (www.dartmouthatlas.org), and Oregon paid doctors rates close to the U.S. median for care
provided to Medicaid patients.
34
Gina Kolata, “First Study of Its Kind Shows Benefits of Providing Medical Insurance to Poor.”
35
Avik Roy, “Oregon Study: Medicaid ‘Had No Significant Effect’ On Health Outcomes vs. Being Uninsured,” The Apothecary
Blog, Forbes, May 2, 2013, http://www.forbes.com/sites/theapothecary/2013/05/02/oregon-study-medicaid-had-no-
significant-effect-on-health-outcomes-vs-being-uninsured/.

HKS Case Program 9 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
ance are less likely to struggle with medical bills. But it’s also the most under-appreciated accomplishment of
36
health insurance: Whatever its effects on health, it promotes economic security.”

Others called attention to the significant mental health gains under the program, but reporter Megan McArdle
37
countered by asking if the ACA was “a $1 trillion program to treat mild depression?”

The findings on emergency department use also sparked critical remarks about the ACA’s cost-cutting poten-
tial. Michael Cannon of the Cato Institute commented “the finding that Medicaid increases emergency-room use
deals another blow to the credibility of Obamacare supporters, from President Obama all the way down to the
38
governors and legislators in the 25 states that opted to implement the law’s Medicaid expansion.”

Some of the fiercest debates on the Oregon study turned out to be over a statistical technicality. Finkelstein,
39
Baicker and co-authors reported that their study lacked the “power to detect changes in health.” They had rigor-
ously demonstrated that Medicaid recipients got more health care and more preventive care than the uninsured.
Yet, on common health measures, like blood pressure, cholesterol and blood sugar, the authors could not say, with
statistical confidence, that there were significant effects (see Exhibit C). This inconclusive finding opened the study
to a host of interpretations. According to Drew Altman, President of the Kaiser Family Foundation, “ideally we
would like to see these clinical measures change but they are strongly influenced by behavior and nutrition, which
40
are hard to change in any population in a two year time frame.” Ross Douthat, a columnist at the New York Times,
believed the findings had far greater implications, “if the best evidence suggests that health insurance is most
helpful in protecting people’s pocketbooks from similar disasters, and that more comprehensive coverage often
just pays for doctor visits that don’t improve people’s actual health, then shouldn’t we be promoting catastrophic
41
health coverage, rather than expanding Medicaid?” Other statistical challenges included how to compare out-
comes for lottery winners to lottery losers. Several observers encouraged the use of an ‘as-treated’ analysis where
outcomes for people in the treatment group who enrolled in Medicaid are compared to those in the treatment
group who did not enroll in Medicaid, as well as everyone in the control group. Another related suggestion known
as ‘per-protocol,’ would compare people in the treatment group who received Medicaid with everyone in the con-
trol group. The researchers argued that such comparisons, while intuitive, were incorrect because they did not rely
on randomization.

36
Jonathan Cohn, “Medicaid Expansion: Oregon Study Shows Benefits, Mostly” New Republic, May 1, 2013.
37
Megan McArdle, “Study: Giving People Government Health Insurance May Not Make Them Any Healthier,” Daily Beast
Asymmetrical Information Blog, May 1, 2013, http://www.thedailybeast.com/articles/2013/05/01/shocker-oregon-health-
study-shows-no-significant-health-impacts-from-joining-medicaid.html.
38
Michael F. Cannon, “Oregon Study Exposes Another ObamaCare Falsehood: Rather Than Reduce Unnecessary ER Use, Medi-
caid Increases It,” Forbes, January 2, 2014.
39
Katherine Baicker, Sarah L. Taubman, Heidi L. Allen, Mira Bernstein, Jonathan H. Gruber, Joseph P. Newhouse, Eric C. Schnei-
der, Bill J. Wright, Alan M. Zaslavsky, and Amy N. Finkelstein, "The Oregon Experiment—Effects of Medicaid on Clinical Out-
comes," New England Journal of Medicine, 368, no. 18, 2013, pp. 1713-1722.
40
Drew Altman, “Poor People Have the Same Needs as Others,” in “Room for Debate: More Medicaid, More Health?” New York
Times, May 6, 2013.
41
Ross Douthat, “What Health Insurance Doesn’t Do,” New York Times, May 4, 2013.

HKS Case Program 10 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
Other analysts questioned how representative the Oregon study population was of uninsured Americans in
general. Medicaid recipients in Oregon chose to sign up for the lottery but, under the ACA, millions of Americans
were required to obtain health insurance. Did these two groups differ in fundamental ways? “One naturally won-
ders about the population mandated to have health insurance. Will they behave like a group who volunteered in-
terest?” asked Baicker. “Of course, we can’t say, based on our study. But it's worth noting that take-up of Medicaid
is far from 100 percent, and even with the mandate, most forecasts predict take-up of far less than 100 percent, so
the individuals taking up the newly-expanded coverage are still going to be ‘selected’ based on underlying inter-
42
ests.”

On the broader test of whether the widely anticipated Oregon Experiment added to existing knowledge, there
were similar arguments. Ezra Klein at The Washington Post thought the research raised more questions than it
answered:

The problem with the Oregon study is that it doesn’t help us figure out how to make
health care or health insurance better. We don’t know if the results speak to the health
care you get through all health insurance or just Medicaid or if they're just an artifact of
the study's timeframe and sample size. We don’t know if different ways of designing in-
surance programs would lead to radically different care outcomes. And we don’t know
whether we’re seeing a problem in Medicaid, an inconvenient truth about medical care,
43
or something else.

Finkelstein and Baicker were confident their research addressed pressing policy questions. “The study put to
rest two arguments that persisted because of an absence of evidence,” said Baicker. “The first is that Medicaid
doesn’t do anything for people, because it’s bad insurance or because the uninsured have other ways of getting
care. The second is that Medicaid coverage saves money. It’s up to society to determine whether it’s worth the
44
cost.”

42
February 6, 2014 email from Katherine Baicker to the authors.
43
Ezra Klein, “Here’s What the Oregon Medicaid Study Really Said,” Washington Post Wonkblog, May 2, 2013,
http://www.washingtonpost.com/blogs/wonkblog/wp/2013/05/02/heres-what-the-oregon-medicaid-study-really-said/.
44
Annie Lowrey, “Oregon Study Shows Benefits, and Costs, of Insuring the Uninsured.”

HKS Case Program 11 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
Appendix I

The Oregon Health Insurance Experiment: Intent to Treat and Local Average Treatment Effects

Only 60 percent of the Oregon lottery winners applied to enroll in Medicaid. Some lottery winners may have
decided against filling in the application form, while others probably intended to apply but missed the 45-day ap-
45 46
plication window. Of the lottery winners that applied for Medicaid, only half were eligible. Ultimately, 30 per-
cent of the lottery winners enrolled in Medicaid. Conversely, roughly 14 percent of those who lost the lottery end-
47
ed up gaining Medicaid coverage in other ways.

In randomized experiments, researchers label participants in the treatment group who don’t get the treat-
ment, or participants in the control group who end up receiving the treatment, as “non-compliers” or “crossovers.”
In the Oregon study, as is the case with many RCTs, there was a large degree of crossover. The Oregon research
team could not compare outcomes for the sub-group of lottery winners who enrolled in Medicaid to outcomes
among individuals who lost the lottery because enrollment in Medicaid was not random. For instance, it was likely
that the lottery winners who applied for Medicaid were sicker than those who didn’t. It was also possible that indi-
viduals who cared more about their health (and were already engaging in healthy behaviors) were the ones more
likely to apply for and obtain Medicaid, regardless of whether they won the lottery.

To obtain an unbiased estimate of the effect of enrollment in Medicaid, the researchers conducted a two-step
process. In the first step, researchers estimated the effect of winning the lottery; or the “intent to treat” (ITT). To
estimate the ITT, the Oregon research team compared the average outcome for all individuals who won the lottery
to the average outcome for all individuals who lost the lottery, whether or not they enrolled in Medicaid. Because
the lottery was randomly drawn, the ITT gave the researchers an unbiased estimate of the effect of winning the
lottery.

In practical terms, the Oregon study’s ITT estimate is an estimate of the effect of increasing access to Medi-
caid, but not the impact of the Medicaid coverage itself. For that the researchers calculated the “Local Average
Treatment Effect (LATE)” of Medicaid (with the help of an econometric technique known as instrumental varia-
48
ble). The LATE estimate gave the researchers a causal estimate of the “local” impact of Medicaid on those who

45
Some may have also realized they were ineligible (or had become ineligible since signing up for the lottery).
46
Lottery applicants had to have income below 100 percent of the federal poverty line, live in the state, and had to be unin-
sured for at least 6 months to enroll.
47
The majority of control individuals who gained access to Medicaid qualified for OHP Plus, a separate Oregon Medicaid pro-
gram available to children and pregnant women, the disabled, and families receiving welfare through the Temporary Assistance
to Needy Families program.
48
An “instrumental variable” is a variable that is correlated with the key independent variable and only impacts the outcome
through the key independent variable. Following the standard framework the Oregon research team picked “being selected by
lottery” as the instrumental variable. Being selected by lottery affects whether the participant enrolls in Medicaid—even with
the crossover, more winners enrolled—and being selected by lottery only plausibly affects health outcomes by facilitating en-
rollment in Medicaid. Then the researchers conducted the standard two-step estimation process to get the ITT and LATE esti-
mates. For more on instrumental variables see Angrist, J., & A. Krueger, “Instrumental Variables and the Search for Identifica-
tion: from Supply and Demand to Natural Experiments,” Journal of Economic Perspectives, 15, 2001, pp. 69-85.

HKS Case Program 12 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
49
won the lottery and enrolled, as compared to those who lost the lottery and did not enroll. In essence, LATE is an
estimate of the impact of Medicaid on a sub-group of lottery participants who would only enroll in Medicaid if they
won the lottery and would not enroll in Medicaid if they lost the lottery (called the compliers). The example below
50
illustrates how the Oregon team arrived at the ITT and LATE estimates:

Imagine 100 people are selected in the lottery to apply for Medicaid and 100 are not. 25
of those selected enroll in Medicaid. And none of those who lost the lottery enroll in
Medicaid. Now suppose the average score [for a specific health outcome] in the control
group is 44.4, but the score among those who won the lottery (the treatment group) is
44.9, or 0.5 higher. This 0.5 increase is the impact of being able to apply for Medicaid—
or the Intent to Treat. But if we think the only reason scores are higher is because of ac-
tual enrollment in Medicaid, then the changes come from the 25 people in the treat-
ment group who ended up enrolled, and not the 75 individuals who did not enroll. The
0.5 average for the whole group selected in the lottery is, in fact, the average of 0 effect
for the 75 people who didn’t enroll and 2.0 for the 25 who did.

[(25% of the sample)* (effect X) + (75% of the sample) *(no treatment effect) = 0.5;
X=2.0]

This 2.0 increase is the effect of enrolling in Medicaid—or the Local Average Treatment
Effect. The effect of enrolling in Medicaid thus is 4 times the effect of being able to ap-
ply for Medicaid—based on the fact that the effect is coming from the 25 percent of
people drawn in the lottery who actually gain coverage and thereby drive any health ef-
fects.

49
For the LATE estimate, the researchers first estimated the effect of randomly being selected by lottery on enrolling in Medi-
caid using an ordinary least squares regression. This regression estimated the difference in enrollment between lottery winners
and losers. The estimate from the first step was then divided by the estimate from the second step to estimate the Local Aver-
age Treatment Effect of Medicaid. Mechanically, the LATE is estimated in a statistical procedure called two-stage least squares.
For more on LATE see Angrist, J., & A. Krueger, “Instrumental Variables and the Search for Identification: from Supply and De-
mand to Natural Experiments,” Journal of Economic Perspectives, 15, 2001, pp. 69-85.
50
Adapted from Katherine Baicker, Sarah L. Taubman, Heidi L. Allen, Mira Bernstein, Jonathan H. Gruber, Joseph P. Newhouse,
Eric C. Schneider, Bill J. Wright, Alan M. Zaslavsky, and Amy N. Finkelstein, “The Oregon Experiment—Effects of Medicaid on
Clinical Outcomes, Supplementary Appendix,” New England Journal of Medicine, accessible at:
http://www.nejm.org/doi/suppl/10.1056/NEJMsa1212321/suppl_file/nejmsa1212321_appendix.pdf.

HKS Case Program 13 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
51
Exhibit A

Demographic Characteristics of Study Population (Control Group)


Control Control
Variable mean Variable mean
Panel A: Full sample
Sex Age
% Female 55.7 % 50-64 26.7
% 20-50 73.3
Language
% English preferred 92.2

Panel B: Survey responders only


Race/Ethnicity Health Status
% White 82.0 % ever diagnosed with:
% Black 3.8 Diabetes 17.5
% Spanish/Hispanic/Latino 12.3 Asthma 27.6
High blood pressure 39.9
Education Emphysema or chronic bronchitis 12.9
% Less than high school 17.7 Depression (% screen positive) 55.7
% High school or GED 49.1
% Vocational training or 2-yr degree 22.0 Income (% of federal poverty line)
% 4 year college degree or more 11.2 <50% 40.6
50-75% 13.8
Employment 75-100% 14.0
% don't currently work 55.1 100-150% 17.7
% work <20 hrs per week 9.0 Above 150% 13.9
% work 20-29 hrs per week 9.9
% work 30+ hrs per week 25.9 Insurance Coverage
Any insurance? 32.5
Average household income (2008) $ 13,053 OHP/Medicaid 11.7
Private insurance 12.8
Other 10.2
Notes: All statistics are reported for control individuals only. Panel A reports the control means
for pre-randomization demographics taken from the lottery list for the whole sample
(N=45,088). Panel B reports the control means of survey questions for survey responders
(N=11,933), weighted using survey weights.

51
Adapted from Amy Finkelstein, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joseph P. Newhouse, Heidi
Allen, and Katherine Baicker, “The Oregon Health Insurance Experiment: Evidence from the First Year.” The Quarterly Journal of
Economics, 127, no. 3, 2012, pp. 1057-1106.

HKS Case Program 14 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
52
Exhibit B

Effect of Medicaid Coverage on Emergency Department Use 1


Number of visits2
Change with Medicaid
Mean Value in Coverage
Control Group (95% Confidence Interval) P Value
Panel A: Overall
All visits 1.022 ± 2.632 0.408 (0.181 to 0.635) <0.001
Panel B: By hospital admission
Inpatient visits
0.126 ± 0.602 –0.023 (-0.077 to 0.031) 0.396
(resulted in admission)
Outpatient visits
0.897 ± 2.362 0.425 (0.216 to 0.635) <0.001
(no admission)
Panel C: By timing of visit
On-hours visits
0.574 ± 1.555 0.232 (0.090 to 0.374) 0.001
(7am-8pm Monday-Friday)
Off-hours visits 0.456 ± 1.394 0.208 (0.074 to 0.342) 0.002
Panel D: By type of visit3
Requires Immediate Care
Emergent, Not Preventable
0.213 ± 0.685 0.049 (-0.016 to 0.115) 0.138
(Requires ED care, could not have been prevented)
Emergent, Preventable
0.074 ± 0.342 0.038 (0.003 to 0.072) 0.032
(Requires ED care, could have been prevented)
Primary Care Treatable
0.343 ± 0.948 0.180 (0.090 to 0.270) <0.001
(Does not require ED care)
Does Not Require Immediate Care
Non-emergent 0.201 ± 0.688 0.118 (0.051 to 0.186) 0.001
Unclassified 0.196 ± 0.734 0.059 (-0.013 to 0.132) 0.107
1
We report the estimated effect of Medicaid on overall emergency department use and on specific categories of
emergency department visits over the study period (March 10, 2008 - September 30, 2009). For each type of visit,
we report the control mean of the dependent variable ± standard deviation, the estimated effect of Medicaid
coverage (with 95 percent confidence interval in parentheses), and the p-value of the estimated effect. Sample
consists of indviduals in Portland-area zip codes (N=24,646).
2
The number-of-visits measures are unconditional, including those with no visits.
3
We also classify visits using an algorithm developed by Billings et al. that is based on the primary diagnosis code
for the visit. Those visits that require immediate care in the emergency department and that could not have
been prevented are referred to as “emergent, not preventable” (21% of control sample visits). Visits that
require immediate care in the emergency department, but could have been prevented through timely
ambulatory care are referred to as “emergent, preventable” (7%). Those visits that require immediate care, but
that could be treated in an outpatient setting, are referred to as “primary care treatable” (34%). Visits that do
not require immediate care are classified as “non-emergent” (19%). The algorithm assigns probabilities of a visit
being each type and therefore we analyze only the number of visits (not the percent with any visits) as obtained
by summing the probabilities across all visits for an individual.

52
Adapted from Sarah L. Taubman, Heidi L. Allen, Bill J. Wright, Katherine Baicker, Amy N. Finkelstein, “Medicaid Increases
Emergency-Department Use: Evidence from Oregon’s Health Insurance Experiment,” Science, Published online January 2, 2014.

HKS Case Program 15 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
53
Exhibit C

Mean Values and Absolute Change in Clinical Measures and Health Outcomes with Medicaid Coverage 1
Change with Medicaid
Mean Value in Coverage
Variable Control Group (95% Confidence Interval) 2 P Value
Blood pressure
Systolic (mm Hg) 119.3 ± 16.9 −0.52 (−2.97 to 1.93) 0.68
Diastolic (mm Hg) 76.0 ± 12.1 −0.81 (−2.65 to 1.04) 0.39
Elevated (%) 16.3 −1.33 (−7.16 to 4.49) 0.65
Hypertension
Diagnosis after lottery (%) 5.6 1.76 (−1.89 to 5.40) 0.34
Current use of medication for hypertension (%) 13.9 0.66 (−4.48 to 5.80) 0.80
Cholesterol
Total level (mg/dl) 204.1 ± 34.0 2.20 (−3.44 to 7.84) 0.45
High total level (%) 14.1 −2.43 (−7.75 to 2.89) 0.37
HDL level (mg/dl) 47.6 ± 13.1 0.83 (−1.31 to 2.98) 0.45
Low HDL level 28.0 −2.82 (−10.28 to 4.64) 0.46
Hypercholesterolemia
Diagnosis after lottery (%) 6.1 2.39 (−1.52 to 6.29) 0.23
Current use of medication for high cholesterol level (%) 8.5 3.80 (−0.75 to 8.35) 0.10
Glycated hemoglobin
Level (%) 5.3 ± 0.6 0.01 (−0.09 to 0.11) 0.82
Level ≥6.5% (%) 5.1 −0.93 (−4.44 to 2.59) 0.61
Diabetes
Diagnosis after lottery (%) 1.1 3.83 (1.93 to 5.73) <0.001
Current use of medication for diabetes (%) 6.4 5.43 (1.39 to 9.48) 0.008
Depression
Positive screening result (%) 30.0 −9.15 (−16.70 to −1.60) 0.02
Diagnosis after lottery (%) 4.8 3.81 (0.15 to 7.46) 0.04
Current use of medication for depression (%) 16.8 5.49 (−0.46 to 11.45) 0.07
Framingham risk score 3
Overall 8.2 ± 7.5 −0.21 (−1.56 to 1.15) 0.76
High-risk diagnosis 11.6 ± 8.3 1.63 (−1.11 to 4.37) 0.24
Age of 50-64 yr 13.9 ± 8.2 −0.37 (−2.64 to 1.90) 0.75
1
Plus–minus values are weighted means ± SD. Where means are shown without standard deviations, they are weighted
means. The effect of Medicaid coverage was estimated with the use of two-stage least-squares instrumental-variable
regression. All regressions include indicators for the number of household members on the lottery list, and all standard
errors were “clustered,” or adjusted to allow for arbitrary correction of error terms within households. For the blood-
pressure measures, all regressions also included controls for age (with dummies for age decile) and sex. All analyses
were weighted with the use of survey weights. The sample size was all 12,229 survey respondents for all measures
except for the Framingham risk score. HDL denotes high-density lipoprotein.
2
For variables measured as percentages, the change is expressed as percentage points.
3
The Framingham risk score was used to predict the 10-year cardiovascular risk. Risk scores were calculated separately.
for men and women on the basis of the following variables: age, total cholesterol and HDL cholesterol levels, measured
blood pressure and use or nonuse of medication for high blood pressure, current smoking status, and status with respect
to a glycated hemoglobin level ≥6.5%. Framingham risk scores, which are calculated for persons 30 years of age or older,
range from 0.99 to 30%. Samples sizes for risk scores were 9525 participants overall, 3099 participants with high-risk
diagnoses, and 3372 participants with an age of 50 to 64 years. A high-risk diagnosis was defined as a diagnosis of
diabetes, hypertension, hypercholesterolemia, myocardial infarction, or congestive heart failure before the lottery (i.e.,
before March 2008).

53
Katherine Baicker, Sarah L. Taubman, Heidi L. Allen, Mira Bernstein, Jonathan H. Gruber, Joseph P. Newhouse, Eric C. Schnei-
der, Bill J. Wright, Alan M. Zaslavsky, and Amy N. Finkelstein, "The Oregon Experiment—Effects of Medicaid on Clinical Out-
comes," New England Journal of Medicine, 368, no. 18, 2013, pp. 1713-1722.

HKS Case Program 16 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.
54
Exhibit D
Financial Hardship and Utilization1
Change with Medicaid
Mean Value in Coverage
Variable Control Group (95% Confidence Interval) 2 P Value
Financial Hardship
Any out-of-pocket spending (%) 58.8 –15.30 (–23.28 to –7.32) <0.001
Amount of out-of-pocket spending ($) 552.8 ± 1219.5 –215.35 (–408.75 to –21.95) 0.03
Catastrophic Expenditures (%) 5.5 –4.48 (–8.26 to –0.69) 0.02
Any medical debt (%) 56.8 –13.28 (–21.59 to –4.96) 0.002
Borrowed money to pay bills or skipped payment (%) 24.4 –14.22 (–21.02 to –7.43) <0.001
Utilization (no. of visits or medications)
Current prescription drugs 1.8 ± 2.8 0.66 (0.21 to 1.11) 0.004
Office visits in past 12 months 5.5 ± 11.6 2.70 (0.91 to 4.49) 0.003
Emergency department visits in past 12 months 1.0 ± 2.0 0.09 (–0.23 to 0.42) 0.57
Hospital admissions in past 12 months 0.2 ± 0.6 0.07 (–0.03 to 0.17) 0.17
Estimate of annual health care spending ($) 3 3,257 1,172 (199 to 2,144) 0.018
Preventive care in past 12 months (%)
Cholesterol-level screening 27.2 14.57 (7.09 to 22.04) <0.001
Colonoscopy in persons over 50 10.4 4.19 (–4.25 to 12.62) 0.33
Flu shot in persons over 50 35.5 –5.74 (–19.31 to 7.83) 0.41
Papanicolaou smear in women 44.9 14.44 (2.64 to 26.24) 0.016
Mammography in women over 50 28.9 29.67 (11.96 to 47.37) 0.001
1
Plus-minus values are weighted means ± standard deviation. Where means are shown without standard deviations,
they are weighted means. The effect of Medicaid coverage was estimated with the use of two-stage least-squares
instrumental-variable regression.
2
For variables measured as percentages, the change is expressed as percentage points.
3
Annual spending was calculated by multiplying the numbers of prescription drugs, office visits, visits to the
emergency department, and hospital admissions by the estimated cost of each.

54
Adapted from Katherine Baicker, Sarah L. Taubman, Heidi L. Allen, Mira Bernstein, Jonathan H. Gruber, Joseph P. Newhouse,
Eric C. Schneider, Bill J. Wright, Alan M. Zaslavsky, and Amy N. Finkelstein, "The Oregon Experiment—Effects of Medicaid on
Clinical Outcomes," New England Journal of Medicine, 368, no. 18, 2013, pp. 1713-1722.

HKS Case Program 17 of 17 Case Number 2019.0

This document is authorized for use only in Hernan Vallejo's UNIANDES MECA ECONOMIC REGULATION (1) (1910) HV at Universidad de Los Andes - Colombia (UniAndes) from Jan 2019
to May 2019.

Das könnte Ihnen auch gefallen