Content uploaded by Neeraj Sood
Author content
All content in this area was uploaded by Neeraj Sood
Content may be subject to copyright.
Forum for Health Economics & Policy
Volume 14, Issue 2 2011 Article 3
(HEALTH POLICY)
How Do Consumer-Directed Health Plans
Affect Vulnerable Populations?
Amelia M. Haviland
∗
Neeraj Sood
†
Roland McDevitt
‡
M Susan Marquis
∗∗
∗
RAND Corporation, haviland@rand.org
†
University of Southern California, nsood@sppd.usc.edu
‡
Towers Watson, roland.mcdevitt@towerswatson.com
∗∗
RAND Corporation, susanm@rand.org
Copyright
c
2011 The Berkeley Electronic Press. All rights reserved.
How Do Consumer-Directed Health Plans
Affect Vulnerable Populations?
∗
Amelia M. Haviland, Neeraj Sood, Roland McDevitt, and M Susan Marquis
Abstract
We use health care claims data from 59 large employers to estimate how consumer-directed
health plans (CDHPs)—plans that combine a high deductible with personal accounts—affect
health care costs and the use of preventive services by vulnerable populations. The vulnerable
populations studied are those that will have increased access to health insurance under health care
reform: families with high health care needs and low income families. A difference-in-difference
framework is used with costs and use available for a full year before and after enrolling in a CDHP
and for controls.
Our key finding is that in almost all cases, CDHP benefit designs affect lower income populations
and the chronically ill to the same extent as non-vulnerable populations. These effects include
significant reductions in overall spending that increase with the level of the deductible and greater
reductions for high deductible plans when paired with health savings accounts (HSAs) in compar-
ison to health reimbursement arrangements (HRAs). However, enrollment in CDHPs also leads
to reductions in care that is considered beneficial for all groups, and this may have greater health
consequences for lower income and chronically ill people than for others.
KEYWORDS: consumer-directed health plans, vulnerable populations, health care costs, preven-
tive care
∗
We would like to acknowledge grant support from the California HealthCare Foundation and
the Robert Wood Johnson Foundation, helpful suggestions from the advisory group members for
those grants, including Joe Newhouse, Linda Bilheimer, Paul Ginsburg, and Dennis Scanlon, and
advice from Dr. Ateev Mehrotra, M.D. on the construction of our annualized preventive care
measures. We would like to acknowledge essential technical support from Ryan Lore, M.P.P. and
Laura Laudenberger, M.S. at Towers Watson; and from Scott Ashwood, M.A. and Al Crego, B.A.
at RAND.
INTRODUCTION
Recently enacted health care reform legislation will expand access to health
insurance coverage over the next decade, especially for low income people and
those with serious health problems who currently experience the greatest barriers
to coverage. Low-income people will find coverage more affordable due to
penalties for large employers that do not offer affordable coverage, premium
subsidies for those people without an offer of affordable employer coverage, and
expansion of Medicaid eligibility to cover parents and childless adults with
incomes below 133 percent of poverty. Those in poor health will benefit from
market reforms that prohibit rating on the basis of health status and guarantee the
issue and renewability of insurance.
On the other hand, many employees whose families are vulnerable due to
low income or serious health problems may face new cost challenges as
consumer-directed health plans (CDHPs) become more prevalent. CDHP plan
designs include a high deductible paired with a tax favored personal account that
may receive contributions from the employer. The high deductible creates cost
consciousness because the member has “skin in the game.” The personal account
and employer contribution provide resources to help the employee mitigate and
manage the additional financial risk. Two kinds of personal accounts are
associated with CDHPs: Health Reimbursement Arrangements (HRAs) and
Health Savings Accounts (HSAs).
HRAs are funded by employers to reimburse employees for qualified
medical expenses up to a stated level. These reimbursements are excluded from
taxable income of the employee. Unused amounts at the end of the year may
rollover for use as reimbursements in future years, but employees generally forfeit
any account balance that remains if they leave the employer before retirement.
Federal law does not require any minimum deductible or other cost sharing
provisions to qualify for an HRA, and the deductibles for HRA plans are often
lower than those of HSA plans.
HSAs create a stronger financial incentive for the employee to manage
health care costs carefully, because the account balance is owned by the employee
and is portable when the employee changes jobs. Created by the Medicare
Modernization Act of 2003, HSA contributions are only permitted for those
enrolled in high-deductible health plans as defined in the law; though prior
contributions to the account may be used for qualified medical expenses at any
time. The requirements for 2010 include minimum deductibles for single and
family coverage of $1,200 and $2,400 respectively.
1
The law permits employer
1
The qualifying deductible for an HSA ranged from $1,000 to $1,100 during our study period.
1
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
and employee contributions subject to limits in 2010 of $3,050 for single
coverage and $6,150 for family coverage.
2
More low income and chronically ill people are likely to enroll in CDHP
plans over the next decade for several reasons. First, the Patient Protection and
Affordable Care Act (PPACA) may accelerate the move to CDHPs among
employers because of new requirements on offering coverage and the tax on
generous plans (McDevitt and Savan, 2010). HSA-based plans will also be
offered in health insurance exchanges that will be established by 2014 to manage
the individual and small group markets (Pfeiffer, 2010), and these lower-premium
plans may be attractive to people who have not previously purchased health
insurance. Even in public insurance programs, deductibles and HSA-like
accounts may arise as states look to greater use of cost-sharing in their Medicaid
programs to control cost, especially as health reform is likely to increase cost
pressures on state budgets. The Deficit Reduction Act of 2005 gives states greater
leeway to use cost-sharing and also establishes a new, 5-year demonstration
program to allow states to set up and fund Health Opportunity Accounts –
accounts that can be used to pay for medical expenses-- in combination with high
deductible health plans for certain Medicaid eligibles (Solomon, 2007).
Given the expected growth in enrollment in CDHPs by lower income and
chronically ill people, we ask what effect CDHPs will have on access to health
care for these vulnerable populations. The preponderance of evidence is that, in
the general population, CDHPs reduce health care spending (Buntin et al., 2006,
2010, American Academy of Actuaries, 2009; Parente, S.T., R. Feldman and J.B.
Christianson. 2004; Lo Sasso, A.T., T. Rice, J.R. Gabel and H. Whitmore, 2004).
The evidence is less clear concerning the effects on persons with low income and
those at risk of high health care spending. The RAND Health Insurance
Experiment, a randomized controlled trial of the effects of cost sharing on health
care, found that cost-sharing produced substantial reductions in use but the overall
effects did not differ between the rich and poor or the healthy and sick. However,
lower income people were somewhat more likely than higher income persons to
cut back on care that is considered highly effective, especially for children
(Newhouse et al, 1993). Moreover, in contrast to current high deductible plans,
the stop loss provisions of the experiment plan were lower for low-income
families than high-income families, which we expect to reduce any difference
between low and high income people in the pure response to cost-sharing.
More recent studies suggest that CDHPs may affect vulnerable population
groups to a greater extent than the general population. CDHP enrollees with low
incomes and those with chronic conditions are more likely to report cost-related
access problems and delaying care than others in these plans based upon
2
For additional information on the rules governing HRAs and HSAs see IRS, 2009.
2
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
telephone survey data (Davis, Doty and Ho, 2006; Reed et al 2009). In addition,
evidence from one firm and one insurer respectively suggests that low-income
persons who switched to CDHPs reduced their physician office visits and visits to
emergency departments for high severity conditions more than others who
switched (Hibbard et al., 2008; Wharam et al 2007). Similarly, in one firm,
people in lower socio-economic status groups who enrolled in a CDHP reduced
both low- and high-priority visits more than other enrollees (Hibbard et al, 2008).
Additional evidence from a single firm suggests patients with chronic diseases
enrolled in high deductible plans are more likely to discontinue taking their
chronic illness medications than the chronically ill in other plans (Greene et al,
2008).
On the other hand, there is some evidence from relatively small samples
that diabetics enrolled in CDHPs did not use fewer diabetic specific preventive
services than those in traditional plans (Buntin et al, 2010; Rowe et al., 2008). In
addition, another study of a single firm found that low-risk families in a CDHP
spent less than those in traditional plans, but high-risk families in CDHPs did not
spend less than comparable families in other plans (Feldman and Parente, 2010).
Understanding the effects of benefit design in general and CDHPs in
particular on service use by the low income and chronically ill is critical to the
implementation of health reform. It is important to policymakers as they evaluate
the minimum health insurance policy standards and the subsidies for low income
populations to ensure they have access to affordable health care coverage and
adequate access to services. Policies to contain health care costs must curb
spending by those with high health care costs, because a small minority of
patients account for a large share of health care expenditures (Berk and Monheit,
2001; Stanton 2006). But how costs are reduced matters critically and high
deductibles may only lead this vulnerable population to cut back or delay
initiation of care for a problem (Newhouse et al., 1993) without curbing costs
where it counts because most of the spending by high cost patients is above the
deductible.
The effects of benefit design on the use of preventive services warrant
special focus because health reform requires that cost-sharing for proven
preventive care services be eliminated in Medicare by 2011 and for most private
insurance plans with plan years that started after September 23, 2010. Most
CDHPs currently waive the deductible for preventive services as permitted by the
legislation authorizing HSAs. Several case studies suggest that eliminating the
deductible does encourage CDHP patients to continue preventive service use even
though overall service use is reduced. However, earlier analyses as part of this
study found lower use of preventive care by CDHP patients (Buntin et al 2010).
Moreover, a study of one insurer found most people were unaware of which
services are exempt from a deductible after being enrolled for one year (Reed et
3
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
al., 2009), suggesting that exemptions may not achieve high utilization of
preventive services.
The objective of this paper is to inform our understanding of how CDHPs
affect the use of health care and preventive services by vulnerable populations.
Based on previous studies, we expect CDHPs to lower use of services but whether
the effects will be greater for vulnerable populations is uncertain. HSAs provide
stronger incentives for patients to be cost conscious than other CDHP models, but
here too it is uncertain whether the effects will differ for vulnerable populations.
Our work adds to the current literature in two ways. First, it uses data from a
large number of employers and many health insurers instead of case studies of a
single employer or single insurer. Second, as a result of this diverse data and
detailed information on insurance design, we are able to disentangle the separate
effects of deductible level and the presence of HSAs and HRAs. These two
extensions provide both more detailed and more generalizable results.
Our key finding is that in almost all cases, CDHP benefit designs affect
non-vulnerable and lower income and chronically ill populations equally. These
effects include significant reductions in overall spending that increase with the
level of the deductible and greater reductions for high deductible plans when
paired with HSAs in comparison to HRAs. However, for all populations,
enrollment in CDHPs also leads to reductions in care that is considered beneficial,
which could have greater health consequences for lower income and chronically
ill people.
METHODS
Study Design and Sample. We constructed a unique data set including medical
claims information from 2003 to 2007 for employees and dependents of 59 large
US employers. The employers entered the study from two routes. Thirty-one
employers were recruited because they were known to offer a plan with an
individual deductible of $500 or more during the period; we refer to these plans as
high deductible health plans (HDHP). These employers were selected to yield a
range of geographic regions, employee income levels, proportion of employees
enrolling in higher deductible plans, and high deductible plan characteristics such
as the size of the deductible, and the type of associated personal account. The
other 28 employers are from the Thomson Reuters MarketScan™ database; some
of these employers also offered HDHPs. These employers were selected to match
the geographic, size, and industry distribution of the recruited employers to
improve the balance in the distribution of employee characteristics across HDHP
and other plans. Forty-three of our sample employers offered an HDHP as an
option at some time during the study.
4
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Our study sample for this analysis includes full time employees and
dependents who were continuously enrolled in their employer’s health benefit
program for a period of two consecutive plan years. We use the term “family” to
include families of 1 (employee only) as well as families of 2 or more (employee
plus dependents). We have two types of families. “Treatment” families selected
a traditional health plan in the first year and a high deductible plan in the second
year. We limited the treatment sample to enrollees in firms with at least 3 percent
enrollment in an HDHP. Thus, we have the first year experience in a high
deductible plan and baseline data for 4 cohorts of families, those first enrolling in
an HDHP in 2004, 2005, 2006, or 2007. Control families were enrolled in a
traditional plan for both plan years and their employer did not offer a plan with an
individual deductible of $500 or more, during that time period. We selected a 50
percent sample of the resultant control group for analysis in order to speed
processing time; the sample was stratified by plan year and employer.
High Deductible and CDHP Account Types. Our analysis classifies the
treatment families into 5 types based on the size of the individual deductible and
the presence of an HRA or HSA: (1) Moderate deductible with no account
(individual deductible greater than $500 but less than the qualifying deductible for
an HSA
3
), (2) Moderate deductible with an HRA, (3) High deductible with no
account (deductible equal to or greater than the qualifying deductible for HSA),
(4) High deductible with an HRA (5) High deductible with an HSA. The latter
two benefit designs are those that are typically thought of as CDHPs. We
examine results for these five types of benefit designs in order to assess the effects
of different levels of the deductible and the different types of personal accounts
on patient behavior.
The deductible that was used to assign families to treatment categories
was identified from survey data and payment patterns in claims data. We
included in our analysis only plans with at least 100 employees to ensure
sufficient observations to make reliable estimates of the deductible. We validated
our claims based cost-sharing provisions by comparing them with survey
responses from 27 employers about 138 plans they offer with a total enrollment of
1.1 million members in 2005. Comparing the treatment classification based on
the two sources, we found agreement for 93 percent of enrollees. In addition, all
high deductible plans identified for this analysis were confirmed by survey data or
other communication with the employer. Ninety-four percent of the CDHP plans
and 98 percent of all the HDHP plans in our study required some cost-sharing
above the deductible until an out-of-pocket spending limit.
3
The qualifying deductible for an HSA ranged from $1,000 to $1,100 during the study period.
5
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
We sampled treatment families from four firms that would have otherwise
comprised more than 20 percent of one of these 5 treatment groups to limit the
number of treatment families contributed by any single employer to 20 percent of
the group. Our study sample sizes are shown in Table 1.
Vulnerable subpopulations. We identify the low income population using a
geocoded variable. We classify a family as low income if the median income of
families in the employee’s five-digit zip code area is below 200 percent of
poverty based on the 2000 Census. One recent study indicated that for some
socio-demographic characteristics there is relatively little misclassification when
using a geocoded versus an individual measure, although the study used finer
geographic areas than are available to us (Fremont et al, 2005). In general, since
geographic areas are not completely homogeneous, the use of geocoded data
provides an imperfect measure of the family characteristic, which might be
expected to lead to an attenuated estimate of the effect of family income.
However, community factors, as well as individual characteristics, may influence
outcomes. Other health researchers have demonstrated that in this case, the use of
the aggregate measure as a proxy for the microvariable is likely to produce an
overestimate of the effect of the individual characteristic that one would obtain
from a model that includes only the microvariable, rather than an underestimate
(Geronimus, Bound and Neidert. 1996
).
Families at high health risk are those with a member who has one of the
five most costly physical chronic illnesses: heart disease, cancer, diabetes,
hypertension, or kidney disease (Cohen and Kruass, 2003). These five conditions
also accounted for as much as 20 percent of total health care spending in 1997 and
one-quarter of the total growth in health care spending from 1987-2000 (Thorpe et
al, 2004, Cohen and Krauss, 2003, Stanton, 2006). Conditions were identified
from medical claims in the baseline year using AHRQ’s Clinical Classification
Software, which aggregates ICD-9 codes, and groupings defined by Cohen and
Krauss (2003).
6
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Table 1. Study Sample Sizes
Number of Families
Moderate
Deduct.
Moderate
Deduct.
High
Deduct.
High
Deduct.
High
Deduct.
Control
Sample
No Account HRA Account No Account HRA Account HSA Account
Cohort Year
2004 28935 0 778 6309 0 70818
2005 14005 0 2405 9456 1275 76757
2006 6880 9837 1064 2113 9038 40446
2007 22569 8431 69 3254 9851 44348
Total 72389 18268 4316 21132 20164 232369
Family Income
Below 200%
poverty 21979 3712 1104 4619 3101 45246
Above 200%
poverty 50410 14556 3212 16513 17063 187123
Health Risk
Low 55482 13660 3570 17767 16152 183449
High 16907 4608 746 3365 4012 48920
Percent of Sample
Cohort Year
2004 7.8% 0.0% 0.2% 1.7% 0.0% 19.2%
2005 3.8% 0.0% 0.7% 2.6% 0.3% 20.8%
2006 1.9% 2.7% 0.3% 0.6% 2.5% 11.0%
2007 6.1% 2.3% 0.0% 0.9% 2.7% 12.0%
Total 19.6% 5.0% 1.2% 5.7% 5.5% 63.0%
Family Income
Below 200%
poverty 6.0% 1.0% 0.3% 1.3% 0.8% 12.3%
Above 200%
poverty 13.7% 3.9% 0.9% 4.5% 4.6% 50.8%
Health Risk
Low 15.1% 3.7% 1.0% 4.8% 4.4% 49.8%
High 4.6% 1.3% 0.2% 0.9% 1.1% 13.3%
The number of treatment and control families in the vulnerable
populations is shown in Table 1.
Study Variables. Medical claims were processed by Thomson Reuters into a
standardized format. From the claims data, we calculated annual family
expenditures for medical care (insurance and patient payments for care received)
7
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
and divided these by 12 to obtain average monthly expenditures. Similarly, we
measured average monthly expenditures for care in each of three health care
settings: outpatient or emergency department, inpatient, and prescription drugs.
Expenditures are measured in current dollars; our analysis includes indicators for
year (cohort) and thus controls for medical care price inflation. Medical prices
also vary regionally, but, as we discuss below, our model accounts for time-
invariant factors that affect a family’s health care spending, which will control for
regional price variation. Medical claims data also provided the ICD-9 codes used
to determine the presence of chronic conditions and to identify high risk-families
as defined above.
We created six preventive measures from the claims data based upon
HEDIS measure definitions (NCQA, 2008): three cancer screening measures
(receipt of mammography, cervical and colorectal cancer screening) and three
recommended tests for diabetic patients (HbA1c testing, lipid profile, and
microalbumin test). For each preventive procedure, we construct a dichotomous
measure indicating whether some or none of the eligible family members obtained
the recommended care during the year. Those eligible for each of the outcome
measures are: mammography--females age 40 and older: cervical cancer
screening--females age 21 and older; colorectal cancer screening--persons age 51
and older; three recommended diabetes treatments—persons having a diagnosis
code and/or a drug claim indicating diabetes.
Means on the outcome variables for our study sample at baseline are given
in Table 2.
Statistical analysis. We use a difference-in-differences regression model to
estimate treatment effects while controlling for selection. The selection concern
is that factors that effect the decision to enroll in a CDHP also affect the outcomes
we are trying to measure. But, as we have baseline and post period measures for
each family, each family acts as its own control for unmeasured characteristics
that have the same effect on the outcomes over time, such as differences in an
inherent propensity to use health care or trust in doctors and modern health care.
Further, our regression model controls for characteristics that may be related to
the decision to select a CDHP that also affect growth in health care cost outcomes
including age of the primary insured, family composition, and measures of
education, unemployment, and race based on geocoded variables. The
combination of these two features provides a strong control for selection.
However, selection effects might still confound or bias our conclusions if there
are unaccounted for differences between control and treatment families that affect
the change in outcomes over time. To address this, we only included families as
controls that were not offered an HDHP because we hypothesized that changes in
8
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
outcomes for those who were offered and declined a HDHP might differ from the
change for families who chose to enroll in HDHPs.
Table 2. Baseline Spending and Use of
Preventive Services
Measure
Health Expenditures ($ per family per month)
Total spending 635
Inpatient spending 138
Outpatient spending 361
Prescription drug spending 135
Number of cases 368638
Cancer screening procedures (% receiving)
Cervical cancer 50.4%
Number of cases 305168
Colorectal cancer 27.8%
Number of cases 105350
Mammography 46.6%
Number of cases 185145
Diabetes procedures (% receiving)
HbA1c 68.7%
Lipid Profile 60.2%
Microalbumin 53.0%
Number of cases 19250
The basic model that we fit is:
01 0
12 3
Post HDHP HDHP *Post Vulnerable
Vulnerable *Post Vulnerable *HDHP Vulnerable *HDHP* Post
it i i i
iiii
iiit
Y
X
α
ςβ β γ
γγ γ
ημε
=+ + + + +
++ +
++
In this model,
it
Y is the outcome for family i in time period 1, 2t = ; Post is
an indicator for the 2
nd
year of observation for each cohort; HDHP
i
is a vector of
indicators denoting whether the family enrolls in one of the 5 treatment groups;
Vulnerable
i
is a vector of indicators for the vulnerable subpopulations; and X
i
is a
vector of personal and health plan characteristics that are time invariant but it also
9
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
includes cohort (year) indicators and cohort-post interactions to allow for
differences in the starting level and change in outcomes among cohorts. Thus, the
overall change in outcomes for the control population from the baseline to the
post year is measured by coefficient ζ; our measure of the treatment effect is the
difference in change for the treated population and is measured by
1
β
; the
difference in the treatment effect for those in vulnerable subpopulations is
measured by
3
γ
and the total effect of treatment for the vulnerable population is
13
β
γ
+ . The term
i
μ
is the effect of all characteristics of the family not included in
the regression and is constant through time, and
it
ε
is the remaining unexplained
variation.
We estimate the parameters of the expenditure model using generalized
least squares accounting for the correlation of errors for a family over time.
4
The
models are fit to expenditures in dollars.
5
We present the expenditure treatment
effects in both dollars and as semi-elasticities—the percent change in dollars. The
semi-elasticities are evaluated at the predicted mean baseline spending for the
relevant control populations: non-vulnerable, low income, and high risk. We fit a
logistic model to estimate treatment effects for the preventive measures. In these
models we focus on the role of the deductible only, rather than look at the five
separate treatment categories, because these analyses are restricted to certain
eligible populations and sample sizes in the separate treatment categories are too
small to produce precise estimates. The treatment effects for the preventive
procedures are shown in the tables as the difference between the treatment and
control group in the change in probability of receiving the test, and are evaluated
for a person with characteristics evaluated at the overall mean for the population
eligible for the procedure. In the tables that follow, statistically significant
treatment effects are indicated in bold. Statistically significant differences
between the vulnerable and non-vulnerable populations are indicated with the
symbols * or † indicating significance at p <0.05 or p <0.10 respectively.
Statistically significant differences between CDHP plans with an HRA account
and those with an H SA account are indicated with the symbol ‡ for significance
at p <0.05.
The complete set of parameter estimates is given in the Appendix.
4
Our sample is also clustered by employer, but there is large variation within employer in
employee characteristics and in health care utilization and so we have not incorporated this
clustering in our estimation.
5
In preliminary analysis, we tested a range of specifications using transformations of the raw
dollar scale, including natural logs, and none of the other models outperformed the model we use.
10
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
RESULTS
Annual Health Care Spending. Total spending is reduced in high deductible
health plans for both vulnerable and non-vulnerable families (Table 3). High
deductible plans paired with HSAs have significantly lower levels of total
spending than other high deductible plans for the general population—almost 30
percent lower spending for families with a high deductible and an HSA compared
to about 13 percent lower spending for similar families in other high deductible
plans. There is also a statistically significant reduction in total health care
spending for those with a moderate deductible and no health account, but the
Table 3. Effect of Benefit Design on Total Health Care Expenditures:
Difference In Change in Average Monthly Family Spending From
Baseline For Treatment Groups in Comparison to Control Group
Treatment Group Benefit Designs
Moderate
Deductible
Moderate
Deductible
High
Deductible
High
Deductible
High
Deductible
No
Account
HRA
Account
No
Account
HRA
Account
HSA
Account
Non vulnerable population
Treatment effect ($ per
family per month)
-24.60
(7.67)
0.21
(13.81)
-78.19
(27.65)
-73.43
(12.87)
-164.59 ‡
(19.46)
Semi-Elasticity (%) -4.3% 0.0% -13.7% -12.9% -28.8%
Low income Population
Treatment effect -25.26
(13.89)
-40.36
(31.86)
-77.47
(38.36)
-67.60
(23.62)
-131.13
(41.99)
Semi-elasticity -4.4% -7.0% -13.5% -11.8% -22.8%
High Risk Population
Treatment effect -26.27
(25.31)
-27.00
(35.83)
-90.48
(79.97)
-156.28
(64.91)
-147.69
(59.15)
Semi-elasticity -2.5% -2.6% -8.8% -15.1% -14.3%
* Significantly different from effect for non-vulnerable population p<.05.
† Significantly different from effect for non-vulnerable population p<.10.
‡ Significant difference between the high deductible with an HRA vs. HSA at p< .05.
Table Notes: Tests above compare non-vulnerable and vulnerable families. Standard errors in
parentheses indicate significance of treatment effect within each population group and these
estimates are in bold when significant at p < .05. Treatment effect is the difference between the
treated population and the control population in the change in spending. The semi-elasticity is the
ratio of this change to the baseline level of spending for the relevant control sample: $571 per family
per month for non-vulnerable families; $575 for low income, and $1032 for high risk.
11
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
reductions are small—about 4 percent. Thus, both the size of the deductible and
the type of personal account affect use.
The results are similar for the different components of annual spending:
inpatient, outpatient (including emergency room), and prescription drug (Table 4).
High deductibles have significant effects on all service categories for non-
vulnerable families. High deductible HSA plans showed significantly greater
reductions for outpatient services and prescription drugs than high deductible
HRAs, but reductions in inpatient hospital spending for the HSA families were
similar to those for the HRA families. Moderate deductibles with no account have
significant, but small, effects on spending categories for outpatient care and
prescription drugs. All of these tests remain statistically significant when we
adjust the single comparison tests shown in the table by the Bonferroni correction
to account for the multiple outcomes that we examine. Moderate deductibles with
HRA accounts have significant but small effects only on outpatient spending
although this result loses statistical significance after adjustment for multiple
comparisons.
There are no statistically significant differences between non-vulnerable
families and low-income or high-risk families in terms of dollar reductions in total
spending that result from benefit designs (Table 3) and few differences in the
components of spending (Table 4). However, since high-risk families have higher
levels of spending, the proportional reductions in total annual spending are
generally smaller for those at high risk. Spending reductions for outpatient care
and prescription drugs by low-income families in high deductible plans with
HSAs are smaller than reductions for non-vulnerable families; however, if we
adjust these tests for multiple comparisons, these differences do not remain
statistically significant. For families at high risk, the differences are all in
prescription drug spending where there are larger effects in three of the five
benefit designs as compared to effects for non-vulnerable families. Two of these
contrasts remain statistically significant at p = 0.10 after adjusting for multiple
comparisons.
High deductible plans with HRAs do not differ from high deductible plans
with HSAs in their effects on total spending for high-risk families. However, for
both outpatient and prescription drug components of spending, we do see
significant differences for high risk families in the effects of high deductibles
paired with HSAs as compared with other high deductible plans. It is also for
these services that we find a difference in the effect of high deductibles paired
12
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Table 4. Effect of Benefit Design on Difference in Change in Components of
Average Monthly Family Health Care Spending From Baseline For
Treatment Groups in Comparison to Control Group
Treatment Group Benefit Designs
Moderate
Deduct.
Moderate
Deduct.
High
Deduct.
High
Deduct.
High
Deduct.
No Account
HRA
Account
No
Account
HRA
Account
HSA
Account
Inpatient spending
Non vulnerable population
Treatment effect ($ per
family per month)
-8.95 12.11 -43.11 -29.36 -34.26
(5.74) (10.32) (14.53) (9.85) (13.12)
Semi-Elasticity (%) -9.0% 12.1% -43.1% -29.4% -34.3%
Low income Population
Treatment effect -3.02 -7.80 -32.70 -12.67 -48.29
(10.45) (25.32) (27.60) (18.80) (34.36)
Semi-elasticity -2.96% -7.64% -32.06% -12.43% -47.35%
High Risk Population
Treatment effect -6.58 -14.89 -47.57 -108.49 -33.21
(20.19) (26.28) (54.44) (60.58) (49.17)
Semi-elasticity -2.36% -5.34% -17.05% -38.89% -11.90%
Outpatient spending
Non vulnerable population
Treatment effect ($ per
family per month)
-12.90 -15.63 -27.52 -31.51 -99.77 ‡
(4.08) (7.00) (21.76) (6.93) (12.97)
Semi-Elasticity (%) -6.9% -8.3% -14.6% -16.8% -53.1%
Low income Population
Treatment effect -19.00 -31.85 -27.63 -39.69 -60.55 †
(7.20) (14.53) (21.54) (11.10) (19.48)
Semi-elasticity -9.9% -16.7% -14.5% -20.8% -31.7%
High Risk Population
Treatment effect -9.79 -9.98 -18.79 -25.41 -83.22 ‡
(11.79) (18.44) (45.31) (20.14) (24.23)
Semi-elasticity -2.4% -2.4% -4.6% -6.2% -20.2%
Prescription drug spending
Non vulnerable population
Treatment effect ($ per
family per month)
-2.74 3.72 -7.56 -12.56 -30.55 ‡
(1.09) (2.09) (2.39) (1.51) (1.79)
Semi-Elasticity (%) -2.0% 2.8% -5.6% -9.3% -22.6%
Low income Population
Treatment effect -3.24 -0.72 -17.15* -15.23 -22.29 *
(2.21) (2.94) (3.86) (2.16) (3.90)
Semi-elasticity -2.4% -4.1% -18.6% -16.2% -21.4%
High Risk Population
Treatment effect -9.91* -2.12 -24.13† -22.38* -31.27 ‡
(3.04) (4.14) (9.01) (4.48) (5.14)
Semi-elasticity -5.1% -1.1% -12.4% -11.5% -16.1%
* Significantly different from effect for non-vulnerable population p<.05.
† Significantly different from effect for non-vulnerable population p<.10.
‡ Significant difference between the high deductible with an HRA vs. HSA at p< .05.
Table Notes: Treatment effect is the difference between the treated population and the control population in the
change in spending. Significant treatment effects within each population are in bold when significant at p <.05.
13
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
with HSAs and other high deductible plans for the non-vulnerable population. In
general, the size of the deductible matters for all populations and HSAs are
associated with greater reductions in spending for all populations.
Receipt of Recommended Care. Families in high deductible plans reduce their
receipt of recommended cancer screening procedures relative to control families,
even though the deductible is waived for preventive care in most of the high
deductible plans (Table 5). For the non-vulnerable population, the probability of
obtaining a recommended procedure in a year is about 3 to 5 percentage points
lower among those in a high deductible plan relative to controls —reductions of
about 7-10 percent in the rate of receipt depending on the procedure. In general,
however, the moderate deductible does not appear to deter receipt of these
procedures.
Table 5. Effect of Benefit Design on Receipt of Recommended
Procedures: Difference In Change in Probability of Receipt from
Baseline for Treatment Groups in Comparison to Control Group
Treatment Group Benefit Designs and Populations
Moderate Deductible High Deductible
Non- Low High Non- Low High
vulnerable Income Risk vulnerable Income Risk
Cervical Cancer
Screening
Treatment Effect -0.8% -1.5% 0.7% * -4.8% -5.8% -3.2%
†
(0.35) (0.56) (0.62) (0.43) (0.82) (0.90)
Colorectal Cancer
Screening
Treatment Effect -0.9% -1.3% 1.1% * -2.9% -4.2% -0.8%
†
(0.59) (0.84) (0.74) (0.80) (1.44) (1.09)
Mammography
Treatment Effect 0.5% -0.9% 1.0% -3.2% -3.2% -0.3%
*
(0.46) (0.72) (0.64) (0.57) (1.09) (0.99)
* Significantly different from effect for non-vulnerable population p<.05.
† Significantly different from effect for non-vulnerable population p<.10
Table Notes: Tests above compare non-vulnerable and vulnerable families. Standard errors in
parentheses indicate significance of treatment effect within each population group and these estimates
are in bold when significant at p < .05.
14
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
As with spending, there are few significant differences between low
income and non-vulnerable families regarding the effect of plan design on receipt
of the cancer screening. However, there are significant differences for those at
high risk. For them, a high deductible is not associated with reductions in receipt
of 2 of the 3 recommended procedures and the reduction for the 3
rd
is significantly
less than for the non-vulnerable population, though this latter is not significant
when we adjust for multiple comparisons. Those at high risk in the moderate
deductible plans are in fact significantly more likely to receive 2 of the 3
treatments than the general population in these plans, though those at high risk do
not significantly increase their use of these procedures relative to traditional plans.
Focusing on one high-risk group, those with diabetes, the difference in
difference estimate suggests that diabetic patients have lower rates of receipt of
recommended tests when enrolled in a high deductible plan than the controls in
traditional plans; the differences are statistically significant for two of the three
procedures (Table 6). For the diabetic population, we estimate that the probability
of receiving the recommended tests is 2 to 5 percentage points lower when
enrolled in a high deductible plan relative to a traditional plan; this translates into
a reduction of 4-7 percent in the rate of receiving the procedures. Moderate
Table 6. Effect of Benefit Design on Receipt of
Recommended Procedures for Diabetes Patients: Difference
in Change in Probability of Receipt from Baseline for
Treatment Groups in Comparison to Control Group
Treatment Group Benefit Designs and Populations
Moderate Deductible High Deductible
Not low Low Not low Low
income Income income Income
HbA1c
Treatment Effect 0.4% -1.3% -3.9% -2.9%
(0.92) (1.33) (1.50) (2.91)
Lipid Profile
Treatment Effect -0.5% 1.4% -4.5% -3.2%
(1.06) (1.52) (1.66) (3.09)
Microalbumin
Treatment Effect -0.3% 1.8% -2.3% -4.4%
(1.22) (1.75) (1.81) (3.41)
* Significantly different from effect for non-vulnerable population p<.05.
† Significantly different from effect for non-vulnerable population p<.10
Table Notes: Tests above compare non-vulnerable and vulnerable families. Standard
errors in parentheses indicate significance of treatment effect within each population
group and these estimates are in bold when significant at p < .05.
15
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
deductibles again do not deter use. There are no significant differences between
low income persons with diabetes and other diabetics in the effect of high
deductibles on the receipt of these tests. However, because of small samples we
can not reject the hypothesis of no effect of high deductible on receipt of these
tests for those with low income.
Sensitivity Analysis. If people who elect to enroll in HDHPs “crowd-in” health
care services in the year prior to the change in their insurance in anticipation of
the higher price of care in the next year, this would underestimate their service
use in the first year of the HDHP and overestimate their service use in the
baseline year. Although the literature suggests that people do not behave in this
way (Newhouse, 1993; Kilbreith et al, 1998; Long et al, 1998), to test for the
possibility, we estimated total expenditure models using only services received in
the 2
nd
and 3
rd
quarter of the baseline and treatment year. Because the opportunity
for a demand surge prior to the change in insurance is somewhat limited by timing
of information on health plan offerings for the upcoming year, insurance
restrictions on the frequency of many procedures, and uncertainty about future
health needs, we assume that any crowding in of services would occur near the
time of the insurance change and would include services that would otherwise be
obtained early at the start of the new plan year. Our conclusions about the effects
of benefit design overall and for vulnerable populations on total annual spending
were unchanged when models were fit to the restricted dataset (results reported in
Appendix).
DISCUSSION
If health reform leads to an expansion in CDHPs as seems likely, will this leave
the low-income and chronically-ill populations with inadequate access to care? In
general, our results suggest not. Although health care spending is lower for those
in high deductible plans, the evidence suggests that non-vulnerable families, low-
income families, and high-risk families are equally affected. However, equal
effects with respect to health care spending may have different consequences for
these populations. For example, high cost sharing places a greater economic
burden on those with low income and high health care spending, and similar
effects of cost sharing on utilization may produce greater health consequences for
those with high health expenditures. These potential impacts are outside the
scope of this research. Moreover, most of the employees in the CDHP plans that
we studied had a choice of a CDHP or a traditional plan, only 4 employers in our
study fully replaced their traditional offerings during the study period. If low-
income and high-risk patients who currently opt for CDHPs behave differently
16
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
than those who opt for traditional plans, different conclusions might be reached if
more and more firms offer only CDHPs.
Our conclusion that benefit design has a similar effect on low- and high-
risk patients is consistent with results from the RAND Health Insurance
Experiment, but at odds with a recent case study of enrollees in a CDHP at one
employer (Feldman and Parente, 2010). They concluded that high risk CDHP
enrollees increased their use of services relative to those in traditional plans
because, in their case study, the former faced no cost-sharing at the margin. This
was the result of a particular feature of the plan studied—there being no cost-
sharing after the deductible was met—combined with enrollees being likely to
exceed their deductible. In contrast to the benefit design structure in the employer
they studied, almost all of those enrolled in the large employer CDHPs studied
here had significant cost-sharing in the form of coinsurance above the deductible.
This would help constrain use. Our results for low income populations vary from
those found in some other studies and may be affected by our use of a geocoded
income measure. Our findings are most properly interpreted as differences
between residents of lower and higher income neighborhoods, rather than strictly
differences between low and high income families. As noted earlier, others have
concluded that geocoded income measures produce upwardly biased estimates of
the pure micro-variable effect in health related studies where community effects
are also likely to matter, so that our conclusion is not likely to be altered if we had
access to measures of family income. In addition, unique characteristics of the
CDHPs studied in case studies discussed above may explain differences in their
results for low income families.
Our findings suggest that specific higher deductible plans designs are
more effective than others in achieving the goal of cost control. We found that
high deductible plans coupled with HSAs reduce spending by a greater amount
than other high deductible plans with HRAs or no account at all, and by a greater
amount than moderate deductible plans with or without an HRA. The greater
spending reduction in HSA plans is consistent with the stronger incentive to save
in HSAs. HSA account balances are owned by the employee, can accumulate
from year to year, can be withdrawn for non-medical use (subject to penalty), and
are portable when the employee leaves to accept a job with another employer.
HRA account balances are owned by the employers and are not portable when the
employee leaves for another job. Hence, employees not confident of staying for
the long term are faced with a “use it or lose it” incentive, encouraging them to
spend down the accounts for current health care.
One goal of health reform is to promote use of preventive services by
eliminating out-of-pocket costs for this care. Our findings suggest that simply
eliminating cost-sharing for preventive services may not increase use of these
services, at least in the short-run. Although deductibles are waived for preventive
17
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
care in the CDHP plans we studied, high deductibles led to small reductions in
receipt of recommended preventive services in low-income and high-risk families
as well as non-vulnerable families. Even with the deductible waived, high
deductibles may reduce preventive care if the deductible keeps people from
seeking care for health problems that would have prompted a referral for some
preventive or screening procedure. On the other hand, our analysis examined
people in the first year of their enrollment in CDHPs, and they may not yet be
familiar with the details of coverage. One encouraging finding on this front is
that the deductible was less of a deterrent to receipt of preventive care for high-
risk patients, who might be more engaged with medical providers and more
familiar with the terms of their insurance. In general, our findings suggest that it
will take more than first dollar coverage for preventive services to ensure that
people actually obtain these services.
If benefit design is to help in controlling costs, it will need to target any
unnecessary, low value spending by high-risk populations that account for most of
spending and much of the growth in spending. We find that CDHPs reduce the
spending by those with chronic conditions that account for a large share of health
care costs and health care cost growth, although the proportional reduction in
costs is not as large as for other populations. High deductible plans also led to
reduced spending on high value care for those at high risk. Of particular concern
are findings that those at high risk in CDHPs received significantly fewer
recommended cancer screening procedure, and diabetic patients in high
deductible plans received fewer recommended procedures for diabetics. Another
potential concern is the finding that those with chronic conditions in CDHPs,
most of whom require drug maintenance, reduced spending on prescription drugs
by more than other populations in CDHPs. This highlights the need for additional
research to explore whether more aggressive case management, educational
approaches, or other programs would help ensure that patients eliminate
unnecessary care and continue with appropriate treatment under CDHPs.
Study Limitations. There are several limitations of our study. First, we focus
only on the first year experience in a CDHP. Responses may differ once families
learn more about the benefits of their new plans. In addition, policy makers are
concerned about whether the cost savings are a one-time savings or whether
CDHPs will alter the trend in cost growth.
A second limitation is the amount of information we examined about what
kind of care is reduced. The RAND HIE found that cost sharing reduced both
necessary and unnecessary care. Our results suggest that some appropriate
services, namely preventive services, are reduced in CDHPs. Further research is
needed into how CDHPs produce cost savings and whether some of these savings
may compromise access to appropriate care.
18
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Selection remains the main threat to the validity of our conclusions, as it is
with all observational studies of CDHPs. There are two sources of selection.
First, employers choose to offer or not offer CDHPs. This decision regarding
what types of health plans to offer might be correlated with unobserved trends in
health care costs and thus bias our results. For example, our results would under
estimate the effects of CDHPs on health care costs if employers offered CDHPs in
response to a trend of rapidly rising health care costs. The second source of
selection is the employee decision to take-up CDHPs once they are offered. For
example, our results might overestimate the effects of CDHPs on health care costs
if employees who expect lower health care cost growth choose to enroll in CDHP
plans.
Existing studies attempt to control for these potential selection biases at
the employer and employee levels. Some control for selection at the employee
level by focusing on "full replacement" firms where all employees who want
employer provided insurance coverage are forced to enroll in CDHP plans. The
lack of plan choice for employees certainly mitigates bias due to employee
selection but might increase bias due to selection at the employer level as full
replacement employers might have very different cost growth trends compared to
other employers. Other studies focus on the experience of a single employer
where some but not all employees chose to enroll in CDHPs. These studies
estimate the effects of CDHPs by comparing health care cost growth for
employees who chose to enroll in CDHPs (treatment group) to growth in health
care costs for employees who chose not to enroll in CDHPs (control group). This
strategy mitigates concerns about selection at the employer level as they focus on
a single employer but compound concerns about selection at the employee level
as employees who chose to enroll in CDHPs might have different cost trajectories
compared to employees who did not chose CDHPs. Finally, both of the above
strategies typically focus on the experience of a single employer that may not
generalize to other contexts.
Our study design attempts to account for these concerns in a variety of
ways. First, we use data from a large number of employers, including those that
offered CDHPs and others that only offered traditional plans. This allows us to
compare cost growth for employees who enrolled in CDHPs to cost growth for
employees in traditional plans who were not offered CDHPs by their employers.
In other words, we do not use employees who were offered and did not enroll in
CDHPs as a control group as this group is likely to have different cost growth
trajectory. Second, we include a wide variety of covariates in our model. Third,
we have baseline and post period observations on all families, so that each family
serves as its own control for the effect of unmeasured factors that affect choice of
health plan and health care use. Finally, we used geography, industry, and firm
19
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
size in selecting employers for our study in an effort to control for selection at the
employer level. Nonetheless, selection bias remains a potential limitation of our
study, and further research exploring the effects of CHDP benefit designs on low
income and high risk populations is needed to confirm our findings.
In sum, our findings suggest that CDHPs reduce spending without unduly
restricting access for lower income and chronically ill populations. However, in
all groups, there is evidence of a small reduction in receipt of high value
preventive procedures. Further research is needed to address whether these
findings also apply after the first year of experience in a CDHP. This additional
research should evaluate whether the reductions in health care spending for
vulnerable populations have greater health or financial consequences for them
than for others.
REFERENCES
American Academy of Actuaries Consumer-Driven Health Plans Work Group:
American Academy of Actuaries Emerging Data on Consumer Driven
Health Plans, May, 2009.
Ash, A. S., Ellis, R. P., Pope, G. C., et al. (2000). Using diagnoses to describe
populations and predict costs, Health Care Financing Review, 21(3); 7-27.
Berk, M. and Monheit, A. (2001). The Concentration Of Health Care
Expenditures, Revisited. Managed care has had little impact on how
resources are spent in treating high-cost illnesses. Health Affairs, 20(2): 9-
18.
Buntin, M. B., Damberg, C., Haviland, A., Kapur, K.,
et al. (2006). Consumer-
directed health care: early evidence about effects on cost and quality.
Health Affairs, 25(6): w516-530 (web exclusive).
Buntin, M. B., Haviland, A., McDevitt, R. and Sood, N. (2010). Cost Savings in
the New Generation of High-Deductible Health Plans: A Multi-Employer
Multi-Insurer Study. Working paper.
Cohen, J. and Krauss, N. (1997). Spending and Service Use Among People With
the Fifteen Most Costly Medical Conditions. Health Affairs, 22(2): 129-
138.
20
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Cohen, R., Martinez, M. and Ward, B. (2009). Centers for Disease Control and
Prevention, Health Insurance Coverage: Early Release of Estimates from
the National Health Interview Survey 2009.
http://www.cdc.gov/nchs/data/nhis/earlyrelease/insur201006.htm
.
Craver, M.L. (Dec. 27, 2009). New Cachet for Health Reimbursement Accounts:
Both employers and retirees like the flexibility that HRAs provide. The
Kiplinger Letter.
Davis, K., Doty, M. and Ho, A. (2005). How High Is Too High? Implications of
High-Deductible Health Plans. The Commonwealth Fund, No.816.
Ellis, R. P. and Ash, A. S. (1995). Refinements to Diagnostic Cost Group (DCG)
Model. Inquiry, 32(4): 418-29.
Feldman, R. and Parente, S. (2010). Enrollee Incentives in Consumer Directed
Health Plans: Spend Now or Save for Later, Forum for Health Economics
and Policy, 13(2).
Fremont, A. M., Bierman, A, Wickstrom, S. L., Bird, C., et al. (2005). Use of
Geocoding in Managed Care Settings to Identify Quality Disparities.
Health Affairs, 24(2): 516-526.
Geronimus, A. T., Bound, J. and Neidert, L. J. (1996). On the Validity of Using
Census Geocode Characteristics to Proxy Individual Socioeconomic
Characteristics. Journal of the American Statistical Association, 91(434):
529-537.
Greene, J., Hibbard, J. and Murray, J. (2008). The Impact of Consumer-Directed
Health Plans on Prescription Drug Use. Health Affairs, 27(4): 111-1119.
Hibbard, J., Greene, J. and Tusler, M. (2008). Does Enrollment in a CDHP
Stimulate Cost-Effective Utilization. Medical Care Research and Review,
65(4): 437-449.
IRS (2009). Publication 969, Health Savings Accounts and Other Tax-Favored
Health Plans. Publication 969. November 25, 2009.
http://www.irs.gov/pub/irs-pdf/p969.pdf
.
21
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
Kilbreith, E. H., Coburn, A. F., McGuire, C., et al. (1998). State-Sponsored
Programs for the Uninsured: Is There Adverse Selection? Inquiry, 35(3):
250-265.
Krieger, N., Williams, D. and Moss, N. (1997). Measuring Social Class in Us
Public Health Research: Concepts, Methodologies, and Guidelines.
Annual Review of Public Health, 18: 341-378.
Lee, T. and Zapert, K. (2005). Do High Deductible Health Plans Threaten Quality
of Care? New England Journal of Medicine, 353(12): 1202-1204,
Perspective.
Lo Sasso, A., Rice, T., Gabel, J. and Whitmore, H. (2004). Tales from the New
Frontier: Pioneers' Experiences with Consumer-Driven Health Care.
Health Services Research, 39(4p2): 1071-1090.
Long, SH., Marquis, M. S., and Rodgers, J. (1998). Do people shift their use of
health services over time to take advantage of insurance? Journal of
Health Economics, 17(1): 105-115.
McDevitt, RD and Savan, J. (2011 in press), Prospects for Account-Based Health
Plans under the Patient Protection and Affordable Care Act. Benefits
Quarterly. 2011 First Quarter. 27(1):21-25.
NCQA: HEDIS 2008. Technical Specifications.
http://www.ncqa.org/tabid/536/Default.aspx
(Accessed October 21,
2009).
Newhouse, J. and the Insurance Experiment Group. (1993). Free For All? Lessons
from the RAND Health Insurance Experiment. Harvard University Press,
Cambridge.
Parente, S., Feldman, R. and Christianson, J. (2004). Evaluation of the Effect of a
Consumer-Driven Health Plan on Medical Care Expenditures and
Utilization. Health Services Research, 39(4-2): 1189-1210.
Pfeiffer, D. (2010). Incorporating the Best Ideas from Both Parties. The White
House Blog, March 3, 2010.
http://www.whitehouse.gov/blog/2010/03/03/incorporating-best-ideas-
both-parties
22
Forum for Health Economics & Policy, Vol. 14 [2011], Iss. 2, Art. 3
http://www.bepress.com/fhep/14/2/3
DOI: 10.2202/1558-9544.1248
Reed, M., Fung, V., Price, M., Brand, R., et al. (2009). High-Deductible Health
Insurance Plans: Efforts To Sharpen A Blunt Instrument. Health Affairs,
28(4): 1145-1154.
Rowe, J., Stevenson, T. B., Downey, R. and Newhouse, J. (2008). Effect of
Consumer-Directed Health Plans on the Use of Preventive and Chronic
Illness Services. Health Affairs, 27(1): 113-120.
Solomon, J. (2007). Cost-Sharing and Premiums in Medicaid: What Rules Apply?
Center on Budget and Policy Priorities.
http://www.cbpp.org/cms/index.cfm?fa=view&id=1125
.
Stanton, M. (2006). The High Concentration of U.S. Health Care Expenditures.
Research in Action, 19: 06-0060.
Thorpe, K, Florence, C. and Joski, P. (2004). Which Medical Conditions Account
for the Rise in Health Care Spending? Health Affairs, w4: 437-445 (web
exclusive).
Wharam, J., Landon, B., Galbraith, A., Kleinman, K., et al. (2007). Emergency
Department Use and Subsequent Hospitalizations Among Members of a
High-Deductible Health Plan. Journal of the American Medical
Association, 297(10): 1093-1102.
23
Haviland et al.: Consumer-Directed Health Plans and Vulnerable Populations
Published by Berkeley Electronic Press, 2011
