Working Paper Series
Congressional Budget Office
Washington, D.C.

"Lesser-of" Payment Policies and the Use of
Physicians' Services Among Dual-Eligible
Beneficiaries

Tamara B. Hayford
Congressional Budget Office
Tamara.Hayford@cbo.gov

Xiaotong Niu
Congressional Budget Office
Xiaotong.Niu@cbo.gov

Sandra L. Decker
Agency for Healthcare Research and Quality

Working Paper 2023-01

January 2023

To enhance the transparency of the work of the Congressional Budget Office and to encourage external
review of that work, CBO's working paper series includes papers that provide technical descriptions of
official CBO analyses as well as papers that represent independent research by CBO analysts. Papers in this
series are available at http://go.usa.gov/xUzd7.

The findings and conclusions in this paper are those of the authors and should not be interpreted as those of
the Agency for Healthcare Research and Quality, the Department of Health and Human Services, or the
Congressional Budget Office. This paper has not been subject to the Congressional Budget Office's regular
review and editing process.

For helpful comments and suggestions, we are grateful to Lyle Nelson (formerly of CBO), Thomas M.
Selden of the Agency for Healthcare Research and Quality, David Weaver (formerly of CBO), and session
participants at the 7th Annual Conference of the American Society of Health Economists, AcademyHealth's
2018 Annual Research Meeting, and the Association for Public Policy Analysis & Management's 2018 Fall
Research Conference. We are also grateful to Joshua Varcie for fact-checking assistance and Loretta Lettner
for editorial assistance.

www.cbo.gov/publication/57843
Abstract

Most dual-eligible beneficiaries-people enrolled in both Medicare and Medicaid-are eligible
for Medicaid through their enrollment in the Qualified Medicare Beneficiary (QMB) program,
which requires that states pay for Medicare cost sharing. Since 1997, states have gradually
implemented policies under which they pay the lesser of Medicare cost sharing and the amount,
if any, by which Medicaid's payment rate for the service exceeds Medicare's payment rate.
Consequently, physicians in most states receive roughly 80 percent of the Medicare rate for
primary care services provided to QMBs.

For this analysis, we used difference-in-difference-in-differences models and administrative data
from 1999 to 2012 to assess how the implementation of "lesser-of" payment policies over that
period affected QMBs' access to care in comparison with that of Medicare-only beneficiaries.
We found that lesser-of policies were associated with a 5 percent reduction in the number of new
primary care visits and a 7 percent reduction in the likelihood of such visits among QMBs. Both
the number and the likelihood of total and established patients' primary care visits fell by about
3 percent. However, we were unable to discern spillover effects on the use of acute care,
including emergency room visits or hospitalizations. Those results suggest that continued
monitoring of quality-of-care measures among dual-eligible beneficiaries may be warranted.

Keywords: dual-eligible beneficiaries, Medicaid payment policies, qualified Medicare
beneficiaries

JEL Classification: 113, 118

il
Contents

Introduction. ............:ccsecssesesceescesecesssenceenceensesessneseaeeenceseseneseneeneesceesesenesaecanenseseneeaaeeaceceseneseneneneneeens 1
Background on Medicare Crossover Policies .............:ssseescsesesceseteseseteneeecsaseaneneseneacesesseenatenenenees 2
Methods and Data .00.... ee eeeesceseecseeencessseneecsetsnersseeseccnencnersnsesecenensnerssenseesesonereseeseesessnenenensneneeees 5
Statistical Analysis... eeesseseeeeceeesseesersseseteaeesesseeseseceseteaeesssasesesaseeteasesesaneseteaeeseeasesasenseeteas 5
Data and Sample Construction...............c.eccsesssscesesecessceseceecceacessneneceeceescessneneceneeeeeeeaeeneneneseeeseuees 7
SUMMALY Statistics...........ceeseesceseesssessseeesneesneesenecsneesneseseneeseuersuerseuseeeesseeeseeeeeneesnessnersnesesees 10
Results... eeeeeecesesesesercneesseeseneneseseesseecaseesenetcnesesceesenescseneseeseeeseesneessesseeeeseneecseseseesnseensnereneees 11
Main Results 0.0... eeeecssesecesecesecceeesseseeeseccseesseesesesecssessseeseceseesseseseesaceseseneseeeseaeeeseneneseseneeaes 11
Robustness Checks............cccesesecessceseseseeseceseceseeesceseeeeceseesseeceeeceseseseeeeeeaeeeseeeaeseaeeaeseeeeseeeaeeass 13
Discussion and Conclusions .............::csccesccseseseceseteeceseceseceseecnceenceseseseeeseseneeseseseeeaesenennseeseaeenneeans 14
References ...........cceesesecesesesccencesecesessccenecenseseseseseneeeaceseseseseneseneeneesesenseeneseceesesenesenesecennseeseneeeaeeats 16
Table 00... eeccesescceeceseceseceecccnceseeesssecseneceasensesssseaeseneeesessesuesenecsacesssenesenececeesesenaseaesesecaeseeneneseneeaes 19

ill
Introduction

Health care spending has grown faster than the economy in recent decades. In 2017, spending on
health care totaled more than $3.5 trillion, with Medicare spending accounting for about one-
fifth of that total (Martin and others, 2018). During the half century that Medicare has existed,
much attention has been paid to various demand- and supply-side policies that may contribute to
controlling costs without affecting access to care or lowering the quality of care. Much of the
supply-side focus has been on analyzing the effects of providers' payment policies on the
provision of care.

For physicians' services, the introduction in the late 1980s of a system that uses the resource-
based relative value scale was intended to reduce the costs associated with the previous cost-
based method of reimbursement. Considerable research has focused on analyzing the effects of
changes in physician reimbursement systems or in specific payment rates on the provision of
care. Standard economic theory would suggest a decline in the provision of physician care when
payment rates are reduced. A study by McGuire and Pauly (1991) indicates this is most likely to
occur when the income effects associated with a payment-rate change for physicians are small,
such as when the payment-rate change applies to patients that represent a small portion of a
physician's practice. If the income effects are strong, the effect of a fee reduction in reducing the
supply of care could be tempered, or the effect could even be positive.

Considerable research has been devoted to testing those theories, though most changes in
Medicare payment rates available for study have been relatively small. For example, a study by
Escarce (1993) examined the effects of reductions in Medicare's surgical fees that resulted from
the Omnibus Budget Reconciliation Act of 1987, though the largest payment-rate change (for
urological procedures) was only about 4 percent. One of the largest payment-rate changes was
the subject of a study by Jacobson and others (2010), which evaluated the effects of Medicare
payment cuts in the range of 10 percent to 13 percent for a set of specific services (such as
administering chemotherapy drugs that treat cancer). Thus far, evidence on the direction and
magnitude of physicians' responses to payment cuts has been mixed. For example, Escarce
found a reduction in service volume following a fee cut, and Jacobson and others found an
increase. There has been little opportunity for study of the effects of broad-based changes in
physician payment rates of substantial magnitude because concerns about access to care have
generally limited the size of changes. For example, in 2010, Medicare's sustainable growth rate
(SGR) system for updating payment rates would have led to expected cuts of 21.3 percent in
conversion factors for the physician fee schedule. Concerns about the effects of that change led
to a series of postponements that lasted until 2015 and ultimately led to the replacement of the
SGR system.
However, an often-overlooked fact is that a 20 percent decrease in physicians' fees has taken
place for a subset of Medicare beneficiaries, namely dual-eligible beneficiaries! Although some
states pay full Medicare cost sharing on behalf of dual-eligible beneficiaries who are entitled to
such assistance, most states have shifted to paying only enough cost sharing to ensure that
providers are paid Medicaid rates for services they deliver to dual-eligible beneficiaries. In most
cases, that results in providers' receiving payment only from Medicare, or 80 percent of
Medicare rates for most physicians' services. That subset of Medicare beneficiaries has a wide
variety of complex health care needs and high rates of poverty. Individuals dually enrolled in
Medicare and Medicaid account for a relatively small share of Medicare and Medicaid enrollees
but a significant share of expenditures. For example, in 2013, dual-eligible beneficiaries
accounted for one-fifth of Medicare enrollees but at least one-third of spending (MedPAC and
MACPAC, 2018).

This paper examines the effects of physician payment reductions on the use of physicians'
services among dual-eligible beneficiaries over the 1999-2012 period.

Background on Medicare Crossover Policies

Most services provided by Medicare Parts A and B require cost sharing. For Part B (physicians'
services), the cost-sharing amount is set at 20 percent of the Medicare payment rate after the
beneficiary reaches the deductible, which ranged from $100 per year in 1999 to $140 in 2012.
Many Medicare beneficiaries have supplemental health care plans that cover all or part of
Medicare coinsurance, such as Medigap plans for those who purchase them and Medicaid for
those with limited means. Medicare is the primary payer for all Medicare-covered services, but
states are required to pay Medicare cost sharing for Medicare beneficiaries enrolled in the
Qualified Medicare Beneficiary (QMB) program. A majority of dually eligible Medicare
beneficiaries are enrolled in that program: In December 2016, 10.6 million people were enrolled
in both Medicare and Medicaid, of which 7.0 million were enrolled in the QMB program (CMS,
2020)." Over our sample period, the share of Medicare beneficiaries who were dually eligible
and the share of dual-eligible beneficiaries enrolled in the QMB program remained stable.*

Since the Balanced Budget Act of 1997 was enacted, states have had the explicit authority to set
the payment policy for their Medicare cost-sharing responsibilities, and that type of policy is

 

1 We use the term "dual-eligible beneficiaries" in this paper rather than "dual-enrolled beneficiaries" to follow the
literature; however, we recognize that not all individuals who are eligible for both Medicare and Medicaid
participate in both programs.

? The QMB program is one of the four means-tested Medicare Savings Programs, which are described in more detail
at https://tinyurl.com/2w3vefny.

3 Those estimates are based on summary statistics published by the Centers for Medicare & Medicaid Services. See
https://tinyurl.com/mju354vc.
referred to as the Medicare crossover policy. Some states pay providers the full amount of cost
sharing-that is, they have implemented a "full" payment policy. Other states pay providers the
lesser of Medicare cost sharing and the amount, if any, by which Medicaid's payment rate for the
service exceeds Medicare's payment for that service-that is, they have instituted a "lesser-of"
payment policy. In the case of most physicians' services, cost sharing is equal to 20 percent of
the Medicare payment rate, and Medicare pays the remaining 80 percent. Providers are not
allowed to bill patients for the remaining balance.

Between 1997 and 1999, the number of states with a lesser-of payment policy increased sharply
from 12 states to 32 states (Nemore, 1999). By 2012, 39 states had implemented a lesser-of
payment policy for physicians' services (MACPAC, 2013). Table 1 summarizes those lesser-of
payment policies over the 1999-2012 period.* As can be seen in the table, from 1999 to 2012,
10 states retained a full payment policy over the entire period, whereas 29 states (including the
District of Columbia) had a lesser-of payment policy in effect for the entire period. Of the

12 states that changed policy during this period, all but 2 (Oklahoma and Nebraska) changed
from a full payment policy to a lesser-of payment policy.

In practice, physicians in most states with a lesser-of policy receive roughly 80 percent of the
Medicare payment rate for services provided to dual-eligible beneficiaries who are enrolled in
the QMB program. That is because Medicaid payment rates for physicians' services are
generally much lower than Medicare payment rates. A study conducted by Zuckerman and Goin
(2012) indicates that Medicaid primary care payment rates averaged 59 percent of Medicare rates
in 2012, and only a few states with lesser-of policies in effect that year had Medicaid payment
rates substantially above 80 percent. So changing from a full payment policy to a lesser-of
payment policy for Part B generally means that, rather than receiving 100 percent of Medicare's
payment amount, a physician receives 80 percent or so of the payment amount.° In the analysis
described below, we therefore chose to assess the effects of lesser-of payment policies on the use
of care among dual-eligible beneficiaries enrolled in the QMB program rather than the effects of
the precise percentage of Medicare fees paid in states with a lesser-of payment policy.

Although many states changed their lesser-of policies between 1997 and 1999, we were not able
to begin our analysis until 1999 for two reasons: The timing of the policy changes during that
period was hard to ascertain, and Medicaid data on institutional status (see the discussion below)

 

4 We focused on lesser-of payment policies that are applicable to Medicare Part B because it is well known that
Medicaid's physician fees are considerably lower than Medicare's fees (Zuckerman and Goin); the same is generally
not true for hospital payment rates (Selden and others, 2015).

5 This is true throughout our study period. Norton (1999) reports that the average Medicaid-to-Medicare primary
care fee ratio was 62 percent in 1998, the year before our study period begins. At that time, only 7 states had fee
ratios over 80 percent: Alaska (118 percent), Idaho (84 percent), Nevada (81 percent), North Carolina (83 percent),
North Dakota (87 percent), Texas (146 percent), and Washington (83 percent).
were not available before 1999. We chose to end our analysis in 2012 because Medicaid data for
many states were not available after 2012 and because the Affordable Care Act introduced a "fee
bump" for the 2013-2014 period, during which time states with Medicaid-to-Medicare fee ratios
below 100 percent were required to increase them to 100 percent (at the expense of the federal
government). Although the effect of that change on dual-eligible beneficiaries is of substantial
policy interest, we viewed that as a separate investigation. Especially because the fee-bump
policy was explicitly announced as temporary, expiring at the end of 2014, any response to that
policy, either for dual-eligible beneficiaries or for Medicaid-only beneficiaries, could be
weakened because of the time-limited nature of the change. In fact, only 7 states fully maintained
the fee bump after 2014 (Zuckerman, Skopec, and Epstein, 2017). Anticipating that outcome,
physicians may not have reacted in the same way to the temporary fee bump as they would have
to a permanent one.

Because we doubted that many physicians' practices were wholly dependent on dual-eligible
beneficiaries over the 2009-2012 period, we expected that the income effects of a reduction in
physicians' payment rates for dual-eligible beneficiaries might not be large. Therefore, in our
analysis, we hypothesized that fee reductions associated with the implementation of a lesser-of
payment policy for dual-eligible beneficiaries should decrease the supply (both the quantity and
the intensity) of physicians' services provided. It might have been easier for a physician to
reduce the quantity of services by not taking on new patients with a particular payment source
rather than by reducing care to existing patients. In addition, studying the effect of managed care
on practice patterns, Glied and Zivin (2002) found that although patients in managed care plans
tended to receive different care from patients with fee-for-service (FFS) coverage, the
differences occurred mostly across practices rather than within a practice. They found evidence
consistent with the notion that practice patterns are set on the basis of a modal payer. Within a
practice, patients with different insurance sources generally received similar care. Patients in
managed care might have received different care because they were being treated in practices
with more managed care patients and therefore a different modal payer. The study by Glied and
Zivin provides further reasoning for our hypothesis that the effects of lesser-of payment policies
on dual-eligible beneficiaries occurred more for new patients than for established patients. New
patients might have ended up seeing physicians with different practice patterns than they would
have if the state had a full payment policy. However, an established patient living in a state that
changed policy might have experienced fewer changes in care provided within previously visited
practices.

Despite the fact that most states have changed their payment policies so that physicians receive
80 percent of Medicare fees-rather than 100 percent-for services provided to dual-eligible
beneficiaries, we know of only three studies that analyzed the effects of this change. Zheng and
others (2017) used a single year of data (from 2009) for 20 states to examine the effects of lesser-
of payment policies on the use of evaluation and management (E&M) services for dual-eligible
beneficiaries enrolled in the QMB program compared with use by Medicare-only beneficiaries.
They found that full payment policies were associated with a 6.4 percent increase in the
likelihood of using E&M services (relative to no payment for cost sharing). Although they used a
control group of nondual-eligible Medicare beneficiaries, their use of a single year of data
presents challenges to internal validity if other characteristics of states are correlated with both
crossover policies and differences in the characteristics of dual-eligible beneficiaries relative to
Medicare-only beneficiaries. In addition, although dual-eligible beneficiaries have been shown in
other studies (for example, MedPAC and MACPAC) to have high use of care compared with
Medicare-only beneficiaries, Zheng and others showed low mean use of care for dual-eligible
beneficiaries relative to Medicare-only beneficiaries. That could be attributable to the authors'
inclusion of people in long-term care who were disproportionately dual-eligible beneficiaries
(CBO, 2013) and who may have had low use of physician care outside of the long-term care
facility. Their inclusion of people in the End Stage Renal Disease (ESRD) program, who were
disproportionately Medicare-only beneficiaries, may have accentuated their picture of low use of
care among dual-eligible beneficiaries relative to Medicare-only beneficiaries and may have
contributed to partly obscuring the effects of crossover policies. CMS (2015) and Mitchell and
Haber (2004) used very similar approaches to Zheng and others for 17 states in 2009 (CMS) and
for 9 states from 1996 to 1998 (Mitchell and Haber). Our study eliminated people in long-term
care and in the ESRD program to yield a more meaningful comparison of care received by dual-
eligible and Medicare-only beneficiaries. In addition, we used data for all but 4 states over

14 years, a period during which 12 states changed crossover policy. That allowed us to examine
the effect of changes in crossover policy on the care of dual-eligible beneficiaries separate from
other state characteristics that may have been related to the characteristics of dual-eligible
beneficiaries or Medicare-only beneficiaries.

Methods and Data

Using Medicare and Medicaid administrative data, we estimated the effects of Medicare
crossover policy on the utilization of Medicare services over the 1999-2012 period.

Statistical Analysis

To estimate the effects of lesser-of payment policies on the use of health care services, we used
both a difference-in-differences (DD) approach and a difference-in-difference-in-differences
(DDD) approach. Conceptually, the treatment group is the group affected by the policy-that is,
dual-eligible beneficiaries in states that changed their crossover payment policy. In our

DD approach, we compared the use of health care services in the treatment group before and
after a change in lesser-of payment policy to changes in the use of health care services among
dual-eligible beneficiaries in states that did not change lesser-of policies. We began with a
simple DD model as specified in equation (1):

Yis=a. +B LESSERs + Bo Xist TYs + O + sist, (1)
where Yis:is a measure of the use of health care services for person i in state s and year ¢;
LESSERs: is a dummy variable measuring whether a lesser-of policy has been in effect for at least
a year in state s and year t; and Xis¢ represents personal characteristics and local-level demand-
and supply-side characteristics. Equation (1) also includes state fixed effects (ys) and year fixed
effects (0:). The inclusion of state fixed effects accounts for differences across states, including
differences in Medicaid policies other than the crossover policy, that may affect the use of care
among QMBs. Estimation of equation (1) compares pre- and post-policy changes in outcomes,
netting out changes in states that did not change their policy.

To account for the possibility that health care use may be changing over time for all patients-
including dual-eligible beneficiaries and other patients in a way that could be correlated with
changes in lesser-of policies affecting dual-eligible beneficiaries-we then moved to a triple
difference approach that compared changes in outcomes following the policy change for dual-
eligible beneficiaries in states that did and did not change their policy to changes among
Medicare-only beneficiaries. As specified in equation (2):

Yis=0. +B LESSERs + Bo DUAList + 83 DUAList X LESSERst + BaXise
+ 0 YEAR; * DUAList +s STATEs X DUAList+ jist STATEs * YEAR: + 8ist, (2)

where DUAL indicates an individual in the treatment group and B3 is the DDD estimator of the
policy effect-that is, the effect of a lesser-of payment policy on dual-eligible beneficiaries
relative to Medicare-only beneficiaries. Equation (2) includes interaction terms between the state
fixed effects and the DUAL indicator, interactions between the YEAR fixed effects and the DUAL
indicator, and full interaction terms between the state fixed effects and the YEAR fixed effects.
The use of Medicare-only beneficiaries as a control group allowed us to better capture potentially
confounding, unmeasured events occurring at about the same time that a lesser-of policy went
into effect. The assumption underlying this model is that no unmeasured state-specific event
occurred at the same time that a lesser-of policy was implemented-that is, a policy affecting
"pre-post" differences in outcomes differentially for dual-eligible beneficiaries relative to
Medicare-only beneficiaries.

As discussed below, we examined two types of utilization measures: the number of visits (or
inpatient stays) and whether a person had any visit (or stay). We used a negative binomial model
for the first set of measures and linear probability models for the second. For negative binomial
models, we reported marginal effects consisting of the mean change in the predicted outcome
following the policy change for the treatment group (that is, QMBs in states with a lesser-of
policy), while holding all other covariates at their observed values. All models allowed for
correlation of errors within each state (Bertrand, Duflo, and Mullainathan, 2004).
Data and Sample Construction

We classified state crossover policies on the basis of two published reports and our own research.
Nemore provided the crossover policy by state for 1999, and MACPAC provided the crossover
policy for physicians' services for 2012. For state crossover policies in the intervening years, we
conducted Internet searches and contacted states' Medicaid agencies and the Centers for
Medicare & Medicaid Services. Between 1999 and 2012, 10 states changed from a full to a
lesser-of payment policy and 2 states changed from a lesser-of to a full payment policy (see
Table 1).

We used a 5 percent random sample of Medicare claims data for 1999 through 2012 to construct
a 14-year panel of Medicare service use for both QMBs and Medicare-only beneficiaries.
Because Medicaid programs are not required to cover Medicare cost sharing on behalf of dual-
eligible beneficiaries who are not enrolled in the QMB program, we focused our analysis only on
dual-eligible beneficiaries enrolled in the QMB program.® We identified dual and QMB status in
one of two ways. For the 1999-2005 period, we used indicators in the Medicaid Analytic eXtract
(MAX) data, which are administrative data files on Medicaid enrollees that include information
on eligibility and service utilization. For those years, we linked Medicare records to MAX data
in order to identify dual and QMB status, limiting our sample to QMBs with consistent
demographic records in both files. For the 2006-2012 period, we used the Master Beneficiary
Summary File (MBSF), because the MBSF began reporting monthly indicators of dual
enrollment in 2006. To clearly differentiate service use between dual-eligible beneficiaries who
were enrolled in the QMB program and Medicare-only beneficiaries, we included only Medicare
enrollees with either zero months of Medicaid enrollment in a year ("Medicare-only
beneficiaries") or 12 months of QMB enrollment in a year (""QMBs'').

We constructed a sample of people for whom we could observe all use of physicians' services.
That means we excluded enrollees in Medicare Advantage, those who were not enrolled in both
Medicare Parts A or B for a full year, and those residing in long-term care facilities (as indicated
in the MAX data). We excluded 4 states from the sample because their data were missing from
the MAX data in 2012.7 In addition, our sample excluded Medicare enrollees with end-stage
renal disease. Our final sample size consisted of 20.3 million person-year observations for the
full study population of QMBs and Medicare-only beneficiaries, of which about 2.4 million were
QMBs. Our primary interest was in the effect of lesser-of policies on the use of physicians'
services. We examined the effect of those policies on both the total number of visits to

 

6 Although Medicaid programs are not required to cover Medicare cost sharing for non-QMB dual-eligible
beneficiaries, some states may choose to do so. Therefore, we excluded non-QMB dual-eligible beneficiaries in
order to conduct a clean analysis of lesser-of policies because we did not know whether and which Medicaid
programs covered cost sharing for those beneficiaries during our study period.

7 Those states are Colorado, Idaho, Kansas, and Rhode Island.
physicians per person and on the probability that a person had at least one visit. We examined
those measures for all visits and specifically for primary care visits.? Because we hypothesized
stronger effects on visits by new patients compared with visits by established patients, we
examined all measures separately for each of those categories of visits. In addition to examining
the effects of lesser-of policies on the number of visits, we also measured effects on the intensity
of visits. We did this in two ways. First, we examined the effects on the number of visits
weighted by the number of relative value units (RVUs). Second, we examined average visit
intensity (or RVU-weighted visits divided by the number of visits) as a dependent variable
among those with at least one visit. Because relative value units are adjusted each year, we
estimated one set of regressions using only RVUs from 1999 in addition to our main analyses
using the annual RVUs.

In addition to the analyses described above, we considered whether lesser-of policies may have
particular effects on QMBs with chronic conditions by examining several measures of
recommended service use among people with diabetes (Parashuram, 2013). The first four are
measures used by the National Committee for Quality Assurance to assess the quality of care
delivered by health care plans to patients with diagnosed diabetes. Those measures consist of
whether or not patients with diabetes have received four tests that the American Diabetes
Association (ADA) recommends should be obtained at least annually: an HbA 1c test (to test the
severity of diabetes); a low-density lipoprotein (LDL) cholesterol test (to test for coronary
complications); a dilated eye exam (to test for vision-related complications); and a nephrology
test, specifically an ADA-recommended screening for microalbuminuria to prevent diabetic
neuropathy. (Physicians recommend that diabetics with micro- or macroalbuminuria be treated
with ACE inhibitors or ARBs). We examined a fifth possible measure of the quality of care for
patients with diabetes consisting of the total number of diabetes-related E&M visits. We
analyzed the effects of lesser-of policies on the chance that patients had any of those tests
because all patients with diabetes should have at least one of those tests per year. We also
analyzed the effects on the number of those tests because lesser-of policies may affect the
probability of follow-up testing. Finally, to analyze possible spillover effects, we also examined
the effects of crossover policies on the use of emergency department services, acute inpatient
care, and preventable hospitalizations. We identified preventable hospitalizations using software
from the Agency for Healthcare Research and Quality.'°

 

8 Specifically, a visit to a physician is indicated by a code from the Healthcare Common procedure Coding System
that is between 99201 and 99215 (inclusive). Only claims with allowed charges were included in our estimates. For
reprocessed charges or payments to secondary payers, a claim was included if the allowed charge was greater than
zero.

° Primary care visits are those to physicians in general practice, family practice, internal medicine, and geriatric
medicine.

10 In particular, we used version 6.0 ICD-9-CM from this site: https://tinyurl.com/7zvrmt8h.
AS indicated above, we controlled for a variety of personal and local-level characteristics, Xist,
including an indicator for female, a set of indicator variables for age group, a set of indicator
variables for race and ethnicity, a set of indicator variables for the number of chronic conditions,
the share of our sample of Medicare beneficiaries enrolled in Medicare Advantage plans in the
state (and its interaction with an indicator of QMB status), county-level average income per
person in 2012 dollars, and the number of primary care physicians per 1,000 residents.'! The
number of chronic conditions came from the Chronic Conditions Data Warehouse file within the
MBSF data, which identifies 27 chronic conditions on the basis of FFS claims. We collapsed
those into 22 categories that met two requirements within a condition-specific reference period:
sufficient coverage and sufficient FFS claims. '*

Finally, we performed a variety of robustness checks. First, we tested robustness to alternative
restrictions to the sample definition. In particular, we limited the analysis to one of two
subsamples-people in the bottom third of the county distribution of per capita income and those
age 65 or older-in order to perhaps increase the comparativeness of the treatment and control
groups. Second, we stratified the sample according to the number of chronic conditions (at least
three versus less than three) to see if results were stronger for those with more chronic
conditions. Third, to test the validity of the parallel-trends assumption that underpins our
analysis, we ran event study specifications for the number of visits. In particular, we modified
the approach discussed above by replacing the interaction between the indicator variable for
QMB status and the indicator for a lesser-of policy with a set of variables that create interactions
between the indicator variable for QMB status and a set of indicators for the number of years
before or since the policy's implementation. If the change to a lesser-of policy had a negative
effect on the use of physicians' services, the estimated coefficients of the interaction terms would
be negative after the change of policy. More important, the coefficients on the interaction terms
before the policy change should be statistically indistinguishable from zero.

 

1 The categories are as follows: under 65, 65 to 74, 75 to 84, and 85 or older for age group; non-Hispanic White,
non-Hispanic Black, Hispanic, non-Hispanic other, and missing for race and ethnicity; missing, zero, one, two,
three, four, and five or more for the number of chronic conditions; under $30,000, $30,000 to $40,000, and $40,000
or over for the county-level per capita income; and under 0.6, 0.6 to 0.8, and 0.8 or over for the county-level number
of primary care physicians per 1,000 people. County characteristics came from HRSA (2018). Medicare Advantage
penetration by state and year were constructed from the MBSF data, using indicators of individual-level Medicare
Advantage enrollment and state of residence for a given year.

12 Five cancer conditions (breast, colorectal, endometrial, lung, and prostate) were combined into one condition of
cancer, and we excluded the condition of "Alzheimer's Disease" because of the redundancy with the condition
"Alzheimer's Disease and Related Disorders or Senile Dementia." We used the indicators for chronic conditions
based on claims for services received through December 31 of the data year.

The reference period for determining sufficient coverage and claims is one year for most conditions and up to three
years for a few conditions. See the following website for additional details:
https:/Awww.ccwdata.org/web/guest/user-documentation.
Our approach has some limitations. One key limitation is that the composition of our study
population might have changed because of our exclusion of those enrolled in Medicare
Advantage. Medicare Advantage penetration rates fell from 18 percent in 1999 to 13 percent in
2003, but those rates have risen steadily since then. By 2012, 27 percent of Medicare
beneficiaries were enrolled in Medicare Advantage (Jacobson and others, 2015). Although recent
evidence shows that dual-eligible beneficiaries and Medicare-only beneficiaries have similar
propensities to enroll in Medicare Advantage, if the two groups had different patterns of
enrollment over time that were correlated with changes to states' lesser-of policies, that could
have affected our results (Levinson, 2018). Also, although the majority of our QMBs received
full Medicaid benefits, the incentive to enroll in Medicaid and become a dual-eligible beneficiary
might have changed after 2006 when Medicare introduced Part D, which offers drug benefits to
those enrolled only in Medicare. On the one hand, Part D provides access to prescription drug
coverage that might not have been affordable outside of Medicaid enrollment before 2006. On
the other hand, partial Medicaid enrollment (for instance, enrollment in the QMB program
without full Medicaid coverage) now granted access to deeply discounted prescription drug
coverage through the low-income subsidy program, whereas partial Medicaid enrollment did not
include any prescription drug benefits before 2006. Again, that might have affected our results if
any compositional changes were correlated with the timing of changes in a lesser-of payment
policy.

Summary Statistics

Table 2a shows that QMBs were not necessarily more likely than Medicare-only beneficiaries to
use any physicians' services over our study period, though they were much more likely to use
acute care services. On average, nearly 39 percent of QMBs, compared with 25 percent of
Medicare-only beneficiaries, visited an emergency room, a difference of nearly 56 percent. The
difference in the chance of any inpatient care was smaller, though dual-eligible beneficiaries
were 62 percent more likely than Medicare-only beneficiaries to experience a preventable
hospitalization.

Although Table 2a shows that QMBs were not necessarily more likely than Medicare-only
beneficiaries to receive care from a physician, Table 2b shows that, compared with Medicare-
only beneficiaries, they had more total visits.!? The largest difference was for primary care visits.
On average, QMBs had 3.4 primary care visits per year compared with 2.7 for Medicare-only
beneficiaries, a 24 percent difference. QMBs had more total and primary care visits compared
with Medicare-only beneficiaries as both new and established patients. In all cases, the
percentage difference in RVU-weighted measures was slightly larger than the unweighted

 

13 For summary statistics and all subsequent analysis, the counts of visits and stays are top-coded at the 99th
percentile. In other words, among those with a positive count of utilization, we estimated the 99th percentile, and we
assigned the value of the 99th percentile for those with values above the 99th percentile.

10
differences, implying that there were differences in intensity in addition to the number of visits,
although most of the difference was attributable to the number of visits. Table 2c confirms that
the differences in intensity of care received were small. Table 2d shows that, compared with the
volume of use by Medicare-only beneficiaries, the volume of use of acute care was also larger
for QMBs, especially for emergency room visits. QMBs had more than double the number of
emergency room visits compared with Medicare-only beneficiaries.

The demographic characteristics of QMBs also differed from those of Medicare-only enrollees
(see Table 3). QMBs were more likely to be female, younger, and in minority groups. The
greater share of QMBs under age 65 indicates a greater share with disabilities. The difference in
the presence of chronic conditions was small, though QMBs were more likely to be missing
complete chronic condition data. That difference suggests that QMBs may have been more likely
to have recently enrolled in Medicare or to have recently disenrolled from Medicare Advantage
relative to Medicare-only beneficiaries; either factor would contribute to an incomplete look-
back period for all of the chronic conditions included in the data.

Over the 2009-2012 period, QMBs lived in locations with different characteristics (see Table 3).
Compared with Medicare-only beneficiaries, QMBs were more likely to live in areas with lower
per capita income and fewer primary care physicians. QMBs were also more likely to reside in
states with a substantially lower share of Medicare Advantage enrollees.

Results

The results of our analysis suggest that, over the 2009-2012 period, lesser-of payment policies
reduced the number and intensity of visits to physicians by QMBs. The magnitude of those
results was generally larger for new patients than for established patients.

Main Results

Table 4 summarizes the effects of lesser-of payment policies on the use of physicians' services
over the 2009-2012 period. Results from both the DD model and the DDD model suggest a
negative effect of lesser-of policies on the number of visits to physicians. However, the estimated
magnitude of the effects and the precision of the estimates varied across models. In most cases,
estimated effects in percentage terms were larger for new patients than for established patients,
as hypothesized. For example, the DDD model suggested that the introduction of lesser-of
payment policies decreased the unweighted number of physician visits among new patients by
0.020 visits per dual-eligible beneficiary. That represents a 4.8 percent reduction compared with
the mean number of visits among QMBs before the introduction of lesser-of policies. '4 In

 

14 Percentage effects were calculated by dividing the change in the number of visits by the mean number of visits in
the year before the policy change. Specifically, the mean was calculated among QMB dual-eligible enrollees in a
given cohort in states that switched payment policies in the year that the new policy was implemented.

11
comparison, the DDD model suggested that the number of established patient visits fell by an
imprecisely estimated 1.6 percent. Results from the two specifications differ in that the DD
model produced a more precise estimate for the reduction in overall patient visits and overall
established patient visits, whereas the DDD model produced larger and more precisely estimated
reductions in primary care visits. The magnitude of estimated effects on the number of physician
visits weighted by RVUs was uniformly larger than the estimated effect on the unweighted
number of visits, suggesting that the introduction of lesser-of payment policies affected both the
number and intensity of visits over the study period.

Table 5 explicitly examines the effects of lesser-of payment policies on the intensity of visits
(among beneficiaries with at least one physician visit). This table confirms that the introduction
of lesser-of policies tended to decrease the intensity of visits. Point estimates were negative in all
cases, and the DDD estimates were relatively precise. As in Table 4, using the DDD
specification, the magnitude of the effects was larger for visits by new patients compared with
visits by established patients. For example, the DDD estimates for outcomes calculated with
annual RVUs imply that the introduction of lesser-of policies was associated with a 1.4 percent
decrease in the intensity of primary care visits for new patients compared with a 0.7 percent
decrease for established patient visits. Comparing Table 4 and Table 5, it seems that lesser-of
policies tended to decrease both the quantity and intensity of visits, especially for QMBs seeking
care from a new physician, although the effect on quantity was larger.

Table 6 examines the effects of lesser-of policies on the likelihood of any visit to a physician.°

The patterns of results across specifications in this table are very similar to the patterns of results
for the number of visits that are reported in Table 4. In general, the results suggest that, over the
2009-2012 period, the policies affected the likelihood of any visit in addition to the number of
visits. Findings from both types of specifications indicate that visits by new patients were more
responsive to the implementation of lesser-of policies. However, the DD specifications produced
results that were more precisely estimated for the overall likelihood of visits and the likelihood
of any established patient visits, whereas the DDD specifications produced results that were
larger in magnitude and more precisely estimated for primary care visits. For example, results
from the DDD model suggest that the introduction of lesser-of policies was associated with a

3.1 percent reduction in the likelihood that QMBs visited a primary care physician, compared
with about 1.6 percent for any visits-indicating that primary care providers may have been
particularly affected by the payment reduction. The estimated effect on the likelihood of any

 

15 The supplemental table also shows estimated effects for control variables for the DDD model on any primary care
visits. Estimated effects were not surprising. For example, over the study period, the chance of any primary care
visit increased with the number of recorded chronic conditions. The chance of any visit was negatively associated
with the managed care penetration rate though less so for dual-eligible beneficiaries than for Medicare-only
beneficiaries. The chance of any primary care visit was positively associated with county per capita income and
physician supply.

12
visits by new patients was also particularly strong. Results from the DDD model also suggest
that the introduction of lesser-of policies decreased the chance of any physician visit among new
patients by 4 percent and decreased the likelihood of any primary care visit among new patients
by about 7 percent.

Table 7 considers the effect of lesser-of payment policies on the receipt of certain tests or visits
among patients with diabetes. An estimated negative effect of lesser-of policies on nephrology
tests was consistent across both models, although the magnitude of the estimated effects was
smaller for the DDD model than for the DD model. Results from both models suggest a negative
effect on the chance of any nephrology test and on the number of tests. The DDD model
indicated that the introduction of lesser-of policies reduced the chance of any nephrology test by
about 3 percent and reduced the number of tests by about 5 percent. The DD model suggested
negative effects on HbA Ic tests and E&M visits, though the outcomes produced by the DDD
model were less precisely estimated and had smaller point estimates. The DDD model suggested
that lesser-of policies reduced the number of LDL tests by 7 percent over the study period. In
general, coefficients were larger in magnitude and more precisely estimated for the number of
visits or tests than the overall likelihood of those tests or visits happening, suggesting that
physicians may have been less likely to provide follow-up care in response to lower payment
rates.

Table 8 illustrates the effects of lesser-of policies on the use of acute care, including emergency
room care and inpatient care. Estimated effects were all small and imprecise, suggesting that
reductions in access to physician care did not necessarily lead to adverse health outcomes
requiring emergency room use or hospital admissions.

Robustness Checks

Table 9 reports results from robustness tests. For this table, we focused on the number of primary
care visits-the outcome most affected by a change to lesser-of payment policies. We first
examined whether the results were robust to limiting the analysis only to people in the lowest
third of a county's per capita income distribution in both the treatment and the control groups.
Applying that restriction either left the estimate unchanged (as was the case for the number of
new primary care visits) or increased the strength of the estimates (as was the case for the other
three outcomes pictured). Limiting the analysis to the elderly also either left the estimates
unchanged or strengthened the estimates. Stratifying by the number of chronic conditions, we
found that estimated effects were significant for both groups, though they were stronger among
the group with at least three chronic conditions.

Table 10 presents results from the event study analysis. Except in one case, the policy-lead
interactions were statistically indistinguishable from zero, as expected. The effects on any
primary care visits seemed to occur quickly, though they grew over time, whereas the effects on
the number of primary care visits mostly occurred with a lag, as expected.

13
Discussion and Conclusions

This paper finds that, over the 2009-2012 period, a movement toward lesser-of payment policies
for Part B services for QMBs-which reduced physician payment rates by 20 percent in most
cases-was associated with a reduction in both the number and the likelihood of physician office
visits among QMBs. In many cases, the DD estimates exhibited somewhat different patterns of
magnitudes and statistical precision across outcomes as compared with the DDD estimates.
However, the DD estimates might have been less robust because they could not account for
underlying trends or shocks that affected all Medicare beneficiaries over our study period. Our
DDD specifications suggested that, over the study period, new patient visits and primary care
visits were particularly sensitive to the implementation of lesser-of payment policies. In
particular, our DDD specifications indicated that lesser-of payment policies were associated with
a 5 percent reduction in the number of new primary care visits for QMBs and a 7 percent
reduction in the probability that QMBs visited a primary care physician for the first time. The
DDD specifications also suggested that, following the introduction of lesser-of payment policies,
both the number and the likelihood of overall and established patients' primary care visits fell by
about 3 percent.

Although the results for primary care visits by new patients, in particular, seem like a sizable
effect, we were unable to discern spillover effects on the use of acute care, including emergency
room visits or hospitalizations. Further work may be useful to assess the effects of lesser-of
payment policies on measures of quality of care. For example, we did find evidence that testing
among QMBs with diabetes is reduced with lesser-of payment policies. Whether that represents a
reduction in the overuse of testing or a reduction in needed testing requires further investigation.
We also found a larger reduction in office visits among QMBs with at least three chronic
conditions. Further work comparing quality of care among people with Medicare-only coverage
versus QMBs may be fruitful, especially in states with lesser-of payment policies.

The results of our study may have implications for the likely effects of any proposed changes in
Medicare physician payment rates for Medicare-only beneficiaries. It should be noted, however,
that effects on Medicare-only beneficiaries could differ if the income effects of a payment
reduction were stronger for changes that affect the entire Medicare population. At present,
Medicare payment rates for physician office visits are not much lower than those paid by private
insurance. Recent studies have estimated that Medicare payment rates are roughly 85 percent to
90 percent of private insurance payment rates (Biener and Selden, 2017; Pelech, 2020). This
contrasts with the situation for inpatient care. Although private insurance payment rates to
hospitals were roughly 10 percent more than Medicare's payment rates over the 1996-2001
period, that difference has grown dramatically. From 2012 to 2013, private rates for hospital care
were at least 75 percent higher than Medicare rates (Maeda and Nelson, 2018; Selden and others,
2015). So, even though Medicare payment rates for inpatient care have become much lower than
private rates, the same is much less true of physician payment rates, except in the case of QMBs.

14
In many states, physicians may be getting only 65 percent of what they would get for private
patients for treating a dual-eligible beneficiary. (Medicare pays 15 percent less than private
insurance, and providers typically get 20 percent less than Medicare rates for treating QMBs in
states with lesser-of payment policies.) The potential impact of this payment disparity on quality
of care suggests that continued monitoring of quality-of-care measures may be warranted.

15
References

Bertrand, Marianne; Esther Duflo; and Sendhil Mullainathan. 2004. "How Much Should We
Trust Differences-in-Differences Estimates?" Quarterly Journal of Economics, vol. 119, no. 1
(February), pp. 249-275.

Biener, Adam I., and Thomas M. Selden. 2017. "Public and Private Payments for Physician
Office Visits," Health Affairs, vol. 36, no. 12 (December), pp. 2160-2164.

Centers for Medicare & Medicaid Services (CMS). 2015. Access to Care Issues Among
Qualified Medicare Beneficiaries (QMB) (July), https://tinyurl.com/22sduxyr (PDF).

Centers for Medicare & Medicaid Services (CMS). 2020. MMCO Statistical & Analytic Reports,
https://tinyurl.com/4kpvkfj9.

Congressional Budget Office (CBO). 2013. Dual-Eligible Beneficiaries of Medicare and
Medicaid: Characteristics, Health Care Spending, and Evolving Policies (June),
www.cbo.gov/publication/44308.

Escarce, Jose. 1993. "Effects of Lower Surgical Fees on the Use of Physician Services Under
Medicare," JAMA, vol. 269, no. 19 (May), pp. 2513-2518.

Glied, Sherry, and Joshua Graff Zivin. 2002. "How Do Doctors Behave When Some (But Not
All) of Their Patients Are in Managed Care?" Journal of Health Economics, vol. 21, no. 2
(March), pp. 337-353.

Health Resources & Services Administration (HRSA). 2018. Area Health Resources Files
(ARF), https://data.hrsa.gov/data/download//.

Jacobson, Gretchen, and others. 2015. Medicare Advantage 2015 Spotlight: Enrollment Market
Update (Kaiser Family Foundation, June), https://tinyurl.com/mrevxjx4.

Jacobson, Mireille, and others. 2010. "How Medicare's Payment Cuts for Cancer Chemotherapy
Drugs Changed Patterns of Treatment," Health Affairs, vol. 29, no. 7 (July), pp. 1391-1399.

Levinson, Zachary. 2018. "Dual Advantage or Double the Burden? The Effect on Dual Eligibles
of Providing Medicare Benefits Through Private Plans" (draft, University of Michigan).

Maeda, Jared Lane K., and Lyle Nelson. 2018. "How Do the Hospital Prices Paid by Medicare
Advantage Plans and Commercial Plans Compare With Medicare Fee-for-Service Prices?"
INQUIRY: The Journal of Health Care Organization, Provision, and Financing, vol. 55
(January), pp. 1-8, https://doi.org/10.1177/0046958018779654.

16
Martin, Anne B., and others. 2018. "National Health Care Spending in 2017: Growth Slows to
Post-Great Recession Rates; Share of GDP Stabilizes," Health Affairs, vol. 38, no. 1
(December), pp. 96-106.

McGuire, Thomas G., and Mark V. Pauly. 1991. "Physician Response to Fee Changes With
Multiple Payers," Journal of Health Economics, vol. 10, no. 4 (September), pp. 385-410.

Medicaid and CHIP Payment and Access Commission (MACPAC). 2013. March 2013 Report to
the Congress on Medicaid and CHIP, https://tinyurl.com/4d2s424w.

Medicare Payment Advisory Commission (MedPAC) and MACPAC. 2018. Data Book:
Beneficiaries Dually Eligible for Medicare and Medicaid (January),
https://tinyurl.com/39jxhmna (PDF).

Mitchell, Janet B., and Susan G. Haber. 2004. "State Payment Limitations on Medicare Cost-
Sharing: Impact on Dually Eligible Beneficiaries," INQUIRY: The Journal of Health Care
Organization, Provision, and Financing, vol. 41, no. 4 (Winter 2004/2005), pp. 391-400.

Nemore, Patricia B. 1999. Variations in State Medicaid Buy-In Practices for Low-Income
Medicare Beneficiaries: A 1999 Update (Kaiser Family Foundation, December),
https://tinyurl.com/4s735d6m.

Norton, Stephen. 1999. Recent Trends in Medicaid Physician Fees, 1993-1998, Discussion
Paper (Urban Institute, September), https://tinyurl.com/3ew766cn.

Parashuram, Shriram. 2013. "Measuring Quality of Diabetes Care for Medicare Beneficiaries"
(Ph.D. dissertation, Division of Health Policy and Management, School of Public Health,
University of Minnesota), https://tinyurl.com/y7h8z5tk (PDF).

Pelech, Daria M. 2020. "Prices for Physicians' Services in Medicare Advantage and Commercial
Plans," Medical Care Research and Review, vol. 77, no. 3 (June), pp. 236-248,
https://doi.org/10.1177/10775587 18780604.

Selden, Thomas M., and others. 2015. "The Growing Difference Between Public and Private
Payment Rates for Inpatient Hospital Care," Health Affairs, vol. 34, no. 12 (December),
pp. 2147-2150.

Zheng, Nan Tracy, and others. 2017. "Access to Care for Medicare-Medicaid Dually Eligible
Beneficiaries: The Role of State Medicaid Payment Policies," Health Services Research, vol. 52,
no. 6 (December), pp. 2219-2236.

17
Zuckerman, Stephen, and Dana Goin. 2012. "How Much Will Medicaid Physician Fees for
Primary Care Rise in 2013? Evidence From a 2012 Survey of Medicaid Physician Fees," Kaiser
Commission on Medicaid and the Uninsured (Kaiser Family Foundation, December),
https://www.kff.org/wp-content/uploads/2013/01/8398.pdf.

Zuckerman, Stephen; Laura Skopec; and Marni Epstein. 2017. Medicaid Physician Fees After
the ACA Primary Care Fee Bump (Urban Institute, March), https://tinyurl.com/3tdh8cvf (PDF).

18
Tables

Table 1. [Return to Text]

Crossover Payment Policy by State for Medicare Part B (Physicians' Services),
1999 to 2012

 

 

Policy States

Lesser-of Payment Policy AL, AZ, CO*, DC, DE, FL, GA, IL, KS*, KY, MA, MD, MI, MT, NC, NJ, NM, NV, OH, OR, PA, RI*, TN, TX,
UT, VA, WA, WI, WV

Full Payment Policy AR, HI, IA, ME, MO, MS, NY, SD, VT, WY

States That Switched Policies AK (2008), CA (2000), CT (2000), ID* (2006), IN (2003), LA (2003), MN (2012), NE (2012), ND (2003), NH
(Year) (2000), OK (2006)**, SC (2002)

 

* These states were dropped from the analysis because of missing Medicaid Analytic eXtract (MAX) data in 2012. (That is, we could not identify dual-
eligibility status for beneficiaries in those states in 2012.)

** Of the 12 states that changed policy during the study period, all but 2 (Oklahoma and Nebraska) changed from a lesser-of payment policy to a full
payment policy.

19
Table 2a. [Return to Text]
Mean Dependent Variables, 1999 to 2012

 

 

QMB Medicare-Only Difference
Any Use
Physician 0.765 0.765 0.000
(0.004) (0.004) (0.004)
New Patient 0.293 0.282 0.012 "
(0.008) (0.008) (0.006)
Established Patient 0.748 0.753 -0.005
(0.004) (0.004) (0.005)
Primary Care Physician 0.611 0.627 0.016 "
(0.007) (0.008) (0.007)
New Patient 0.105 0.079 0.026 "
(0.007) (0.004) (0.005)
Established Patient 0.596 0.615 -0.019 "
(0.007) (0.008) (0.007)
Emergency Department 0.386 0.248 0.138 "
(0.016) (0.003) (0.013)
Acute Inpatient Care 0.204 0.174 0.030 we
(0.007) (0.002) (0.005)
Avoidable Hospital Use 0.053 0.033 0.020 "
(0.003) (0.001) (0.002)
Testing and Visits Among Beneficiaries With Diabetes
(N = 4,218,526)
HbA1c 0.485 0.486 -0.001
(0.014) (0.016) (0.010)
LDL 0.459 0.456 0.003
(0.025) (0.020) (0.018)
Eye Exam 0.391 0.449 0.058 hd
(0.011) (0.009) (0.009)
Nephrology 0.399 0.410 -0.012
(0.018) (0.012) (0.014)
E&M 0.523 0.528 -0.005
(0.006) (0.007) (0.005)

 

The asterisks *, **, and *** indicate that the difference is statistically significant at the 10 percent, 5 percent, and 1 percent levels, respectively. The
standard errors are clustered at the state level and are shown in parentheses.

E&M = evaluation and management; LDL = low-density lipoprotein; QMB = qualified Medicare beneficiary.

20
Table 2b. [Return to Text]
Mean Dependent Variables, 1999 to 2012

 

 

Unweighted RVU-Weighted
QMB Medicare-Only _ Difference QMB Medicare-Only _ Difference
Volume of Use
Physician Visits 6.58 6.05 0.53 al 12.52 11.20 1.32 "
(0.221) (0.163) (0.161) (0.473) (0.345) (0.297)
New Patient 0.44 0.40 0.04 e 1.42 1.22 0.20 le
(0.020) (0.015) (0.013) (0.076) (0.053) (0.047)
Established Patient 6.14 5.65 0.49 al 11.09 9.98 1.14 aad
(0.204) (0.152) (0.153) (0.401) (0.299) (0.259)
Primary Care Visits 3.38 2.72 0.66 " 6.30 5.02 1.28 ™
(0.110) (0.048) (0.108) (0.215) (0.098) (0.182)
New Patient 0.12 0.09 0.04 ad 0.40 0.27 0.13 aad
(0.010) (0.004) (0.008) (0.037) (0.015) (0.026)
Established Patient 3.26 2.63 0.63 ad 5.90 4.75 1.15 al
(0.102) (0.046) (0.102) (0.182) (0.088) (0.160)

 

The asterisks *, **, and *** indicate that the difference is statistically significant at the 10 percent, 5 percent, and 1 percent levels, respectively. The
standard errors are clustered at the state level and are shown in parentheses.

QMB = qualified Medicare beneficiary; RVU = relative value unit.

21
Table 2c. [Return to Text]
Mean Dependent Variables, 1999 to 2012

 

 

RVU-Weighted 1999 RVU-Weighted
QMB Medicare-Only _ Difference QMB Medicare-Only _ Difference
Average Intensity
(Conditional Upon
Any Use)
Average Intensity for
Physician Visits 1.92 1.87 0.05 ial 1.49 1.47 0.02 i
(0.016) (0.012) (0.012) (0.012) (0.009) (0.006)
New Patient 3.14 2.95 0.19 a 2.57 2.44 0.13 ial
(0.038) (0.027) (0.022) (0.033) (0.021) (0.019)
Established Patient 1.82 1.79 0.03 ae 1.39 1.39 0.00
(0.013) (0.010) (0.011) (0.009) (0.007) (0.005)
Average Intensity for
Primary Care Visits 1.92 1.92 0.00 1.48 1.50 -0.02 a
(0.015) (0.015) (0.013) (0.011) (0.011) (0.006)
New Patient 3.27 3.16 0.11 ae 2.73 2.65 0.07 oe
(0.036) (0.032) (0.021) (0.031) (0.025) (0.019)
Established Patient 1.85 1.87 -0.02 1.42 1.45 -0.03 ne
(0.014) (0.013) (0.014) (0.008) (0.009) (0.007)

 

The asterisks *, **, and *** indicate that the difference is statistically significant at the 10 percent, 5 percent, and 1 percent levels, respectively. The
standard errors are clustered at the state level and are shown in parentheses.

QMB = qualified Medicare beneficiary; RVU = relative value unit.

22
Table 2d. [Return to Text]
Mean Dependent Variables, 1999 to 2012

 

 

QMB Medicare-Only Difference

Emergency Department Visits 0.89 0.41 0.47 _
(0.051) (0.007) (0.046)

Hospital Stays 0.35 0.27 0.08 il
(0.015) (0.005) (0.012)

Avoidable Hospital Use 0.07 0.04 0.03 "
(0.004) (0.001) (0.003)

Testing and Visits Among Beneficiaries With Diabetes

HbA1c 1.01 1.04 -0.04 "
(0.033) (0.041) (0.018)

LDL 0.92 0.93 -0.01
(0.065) (0.062) (0.038)

Eye Exam 0.93 0.98 -0.05
(0.069) (0.048) (0.045)

Nephrology 0.95 0.93 0.02
(0.059) (0.042) (0.041)

E&M 1.68 1.46 0.22 "
(0.048) (0.016) (0.042)

Sample Size 2,419,054 17,872,110

 

The asterisks *, **, and *** indicate that the difference is statistically significant at the 10 percent, 5 percent, and 1 percent levels, respectively. The
standard errors are clustered at the state level and are shown in parentheses.

E&M = evaluation and management; LDL = low-density lipoprotein; QMB = qualified Medicare beneficiary.

23
Table 3. [Retum to Text]
Mean Independent Variables, 1999 to 2012

 

 

QMB Medicare-Only Difference

Female 0.636 0.599 0.038 ***
(0.010) (0.003) (0.010)

Age Group

<65 0.440 0.081 0.359 ***
(0.024) (0.003) (0.024)

65-74 0.264 0.427 -0.164 **
(0.012) (0.005) (0.012)

75-84 0.212 0.352 -0.140 **
(0.011) (0.004) (0.012)

85+ 0.084 0.139 -0.055 **
(0.003) (0.003) (0.005)

Race/Ethnicity

Non-Hispanic White 0.545 0.886 -0.341 **
(0.042) (0.010) (0.035)

Non-Hispanic Black 0.211 0.066 0.145 **
(0.029) (0.007) (0.024)

Hispanic 0.158 0.028 0.130 **
(0.040) (0.007) (0.033)

Non-Hispanic Other 0.084 0.019 0.065 ***
(0.027) (0.004) (0.023)

Missing 0.002 0.001 0.000 *
(0.000) (0.000) (0.000)

Number of Chronic Conditions

None 0.145 0.120 0.024 **
(0.004) (0.004) (0.005)

1 0.118 0.115 0.003
(0.004) (0.003) (0.003)

2 0.116 0.134 -0.015 **
(0.003) (0.002) (0.002)

3 0.114 0.137 -0.022 **
(0.002) (0.001) (0.002)

4 0.102 0.119 -0.017 **
(0.002) (0.002) (0.003)

5+ 0.254 0.242 0.012
(0.009) (0.008) (0.008)

NLA. 0.151 0.136 0.015 **
(0.005) (0.003) (0.006)

County per Capita Income in 2012 Dollars

Low (<$30,000) 0.230 0.175 0.054 *
(0.035) (0.019) (0.021)

Median ($30,000-$40,000) 0.426 0.448 -0.021
(0.028) (0.025) (0.018)

High (>=$40,000) 0.341 0.374 -0.033
(0.046) (0.034) (0.026)

County Primary Care Physicians per 1,000 People

Low (<0.6) 0.380 0.367 0.014
(0.040) (0.025) (0.023)

Medium (0.6-0.8) 0.323 0.305 0.019
(0.039) (0.021) (0.025)

High (>=0.8) 0.296 0.329 -0.032 **
(0.026) (0.026) (0.009)

State-Level MA Penetration Rate 0.124 0.204 -0.080 ***
(0.013) (0.023) (0.014)

 

The asterisks *, **, and *** indicate that the difference is statistically significant at the 10 percent, 5 percent, and 1 percent levels, respectively. The
standard errors are clustered at the state level and are shown in parentheses. MA = Medicare Advantage; N.A. = not available.

24
Table 4. [Retum to Text]

Effects of Lesser-of Payment Policies on the Number of Physician Visits

 

 

 

Unweighted RVU-Weighted
DD DDD DD DDD

All 0.242" = [-3.5%] -0.127 [-1.8%] -0.682* [-6.3%] -0.302 [-2.8%]
(0.113) (0.100) (0.355) (0.186)

New Patients -0.021* = [-5.0%] -0.020"* [-4.8%] -0.072 [-6.0%] -0.066"* [-5.5%]
0.012 (0.010) (0.045) (0.027)

Established Patients 0.219" = [-3.4%] -0.105 [-1.6%] -0.600* [-6.2%] -0.223 [-2.3%]
(0.104) (0.093) (0.310) (0.168)

Primary Care -0.071 - [-2.0%] -0.110*** [-3.0%] -0.269" [-4.7%] -0.229"* [-4.0%]
(0.075) (0.041) (0.125) (0.079)

New Patients -0.003 - [-2.0%] -0.008*** [-5.3%] -0.017 [-3.5%] -0.029"" [-6.0%]
(0.003) (0.002) (0.011) (0.006)

Established Patients -0.069 = [-2.0%] -0.102** [-2.9%] -0.247" [-4.7%] -0.195" [-3.7%]
(0.074) (0.041) (0.116) (0.076)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to
Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
marginal effects as a percentage of the pre-policy average are shown in square brackets. (The pre-policy average was calculated on the basis of the
outcome values among QMBs in the year before the policy's implementation.) Controls included those for gender, age group, race/ethnicity, original
reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual status, and controls for
county annual per capita income and the number of primary care physicians per capita. Additional controls for the DD model included state and year
fixed effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year fixed effects, Standard
errors for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; QMBs = qualified Medicare beneficiaries; RVU = relative value unit.

25
Table 5. [Retum to Text]

Effects of Lesser-of Payment Policies on the Intensity of Physician Visits

 

 

 

Annual RVUs 1999 RVUs
DD DDD DD DDD

All -0.042 [-2.6%] -0.013* [-0.8%] -0.034 [-2.3%] -0.011** [-0.7%]
(0.034) (0.007) (0.026) (0.005)

New Patients -0.037 [-1.3%] -0.029** [-1.0%] -0.038 [-1.5%] -0.025** -[-1.0%]
(0.036) (0.010) (0.033) (0.008)

Established Patients -0.041 [-2.7%] -0.008* [-0.5%] -0.033 [-2.4%] -0.007** [-0.5%]
(0.031) (0.004) (0.024) (0.004)

Primary Care -0.053 [-3.3%] -0.017" [-1.0%] -0.040 [-2.7%] -0.013"* [-0.9%]
(0.044) (0.007) (0.033) (0.006)

New Patients -0.081* = [-2.6%] -0.045* [-1.4%] -0.075* [-2.7%] -0.040"* = [-1.4%]
(0.047) (0.019) (0.040) (0.015)

Established Patients -0.048 [-3.2%] -0.011** [-0.7%] -0.039 [-2.8%] -0.009* [-0.7%]
(0.039) (0.005) (0.030) (0.005)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to

Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
percentage change-the marginal effect divided by the pre-policy average-is shown in square brackets. (The pre-policy average was calculated on the
basis of the outcome values in states that switched policy in the year of the policy's implementation.) Controls included those for gender, age group,
race/ethnicity, original reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual

status, and controls for county annual per capita income and the number of primary care physicians per capita. Additional controls for the DD model
included state and year fixed effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year
fixed effects. Standard errors for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; RVU = relative value unit.

26
Table 6. [Retum to Text]

Effects of Lesser-of Payment Policies on the Likelihood of Any Physician Visit

 

 

DD DDD
All -0.018** [-2.4%] 0.012" [-1.6%]
(0.006) (0.006)
New Patients -0.012* [-4.3%] 0.011% [-3.9%]
(0.006) (0.005)
Established Patients -0.018"* [-2.4%] 0.012" [-1.6%]
(0.006) (0.006)
Primary Care -0.008 [-1.4%] -0.018** [-3.1%]
(0.010) (0.006)
New Patients -0.003 [-2.5%] -0.008** [-6.7%]
(0.003) (0.002)
Established Patients -0.008 [-1.4%] -0.018** [-3.2%]
(0.011) (0.006)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to
Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
marginal effects as a percentage of the pre-policy average are shown in square brackets, (The pre-policy average was calculated on the basis of the
outcome values among QMBs in the year before the policy's implementation.) Controls included those for gender, age group, race/ethnicity, original
reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual status, and controls for
county annual per capita income and number of primary care physicians per capita. Additional controls for the DD model included state and year fixed
effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year fixed effects. Standard errors
for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; QMBs = qualified Medicare beneficiaries.

27
Table 7. [Return to Text]

Effects of Lesser-of Payment Policies on Testing and Visits Among Patients
With Diabetes

 

 

 

 

Number Any
DD DDD DD DDD

HbA1c -0.051** [-6.0%] 0.016 [-1.9%] 0.017" [-4.0%] 0.002 [-0.5%]
(0.022) (0.034) (0.008) (0.010)

LDL -0.071 [-8.9%] ~0.057** =--[-7.1%] -0.020 [-5.0%] -0.007 [-1.8%]
(0.051) (0.020) (0.013) (0.008)

Eye Exam -0.005 [-0.6%] 0.001 [0.1%] -0.007 [-1.9%] -0.003 [-0.8%]
(0.020) (0.021) (0.004) (0.006)

Nephrology -0.088"* -[-9.6%] -0.045** [-4.9%] 0.026" ~-_[-7.0%] -0.012* [-3.2%]
(0.027) (0.019) (0.008) (0.007)

E&M ~0.095"* = [-4.6%] 0.015 [-0.7%] -~0.018** = [-3.3%] -0.004 [-0.7%]
(0.027) (0.030) (0.005) (0.005)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to
Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
marginal effects as a percentage of the pre-policy average are shown in square brackets. (The pre-policy average was calculated on the basis of the
outcome values among QMBs in the year before the policy's implementation.) Controls included those for gender, age group, race/ethnicity, original
reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual status, and controls for
county annual per capita income and the number of primary care physicians per capita. Additional controls for the DD model included state and year
fixed effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year fixed effects. Standard
errors for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; E&M = evaluation and management; LDL = low-density lipoprotein;
QMBs = qualified Medicare beneficiaries.

28
Table 8.

Effects of Lesser-of Payment Policies on the Use of Acute Care

[Return to Text]

 

 

 

 

Number Any
DD DDD DD DDD

ER Visit -0.002 [-0.1%] 0.008 [0.4%] -0.009 [-2.8%] -0.002 [-0.6%]
(0.011) (0.006) (0.007) (0.005)

Hospital Stays -0,003 [-0.2%] 0.004* [0.2%] -0,002 [-1.0%] 0.001 [0.5%]
(0.003) (0.002) (0.002) (0.001)

Preventable

Hospitalizations -0.000 [0.0%] 0.001 [1.3%] -0.002* [-4.0%] -0.001 [-2.0%]
(0.001) (0.000) (0.001) (0.001)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to
Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
marginal effects as a percentage of the pre-policy average are shown in square brackets. (The pre-policy average was calculated on the basis of the
outcome values among QMBs in the year before the policy's implementation.) Controls included those for gender, age group, race/ethnicity, original
reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual status, and controls for
county annual per capita income and the number of primary care physicians per capita. Additional controls for the DD model included state and year
fixed effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year fixed effects. Standard
errors for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; ER = emergency room; QMBs = qualified Medicare beneficiaries.

29
Table 9. [Return to Text]

Robustness Checks: Effects of Lesser-of Payment Policies on the Number of
Primary Care Visits Under the DDD Model (Unweighted)

 

 

Low- Higher-
Income Income Non- 3+ Chronic <3 Chronic
All Counties Counties Elderly elderly Conditions Conditions
All -0.110** -0.186*** -0.065 -0.132** -0.042 -0.185** -0.073**
(0.041) (0.066) (0.057) (0.050) (0.038) (0.059) (0.036)
New -0.008*** -0.008*** -0.004 -0.008*** -0.007*** -0.014** -0.003
(0.002) (0.002) (0.003) (0.002) (0.002) (0.002) (0.002)
Established -0.102** -0.173*" -0.059 -0.124** -0.035 -0.169** -0.069**
(0.041) (0.062) (0.056) (0.049) (0.038) (0.059) (0.035)
Any -0.018*** -0.032** -0.016** -0.021** -0.009 -0.024** -0.017"
-0.006 (0.012) (0.007) (0.008) (0.008) (0.007) (0.007)

 

This table shows the marginal effects of a lesser-of policy under the DD model and of a lesser-of policy on dual-eligible beneficiaries relative to
Medicare-only beneficiaries under the DDD model from generalized linear models with a negative binomial distribution and log link function. The
marginal effects as a percentage of the pre-policy average are shown in square brackets. (The pre-policy average was calculated on the basis of the
outcome values among QMBs in the year before the policy's implementation.) Controls included those for gender, age group, race/ethnicity, original
reason for Medicare eligibility, number of chronic conditions, Medicare Advantage penetration rate by state, year, and dual status, and controls for
county annual per capita income and number of primary care physicians per capita. Additional controls for the DD model included state and year fixed
effects; for the DDD model, they included dual-specific state fixed effects, dual-specific year fixed effects, and state-year fixed effects. Standard errors
for the marginal effects are shown in parentheses under marginal effects.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DD = difference-in-differences; DDD = difference-in-difference-in-differences; QMBs = qualified Medicare beneficiaries.

30
Table 10. [Return to Text]

Event Study Specifications for the Number of Primary Care Visits
Under the DDD Model

 

 

Primary Care Visits
Any Primary
All New Established Care
One-Year Lead -0.021 -0.003 -0.017 -0.011
(0.053) (0.002) (0.057) (0.009)
Two-Year Lead -0.003 -0.007** 0.010 0.000
(0.024) (0.001) (0.023) (0.004)
Three-Year Lead -0.017 -0.001 -0.015 -0.004
(0.015) (0.003) (0.012) (0.003)
One-Year Lag -0.006 -0.005*" 0.003 -0.007*
(0.034) (0.001) (0.033) (0.004)
Two-Year Lag -0.027 -0.002 -0.025 -0.015***
(0.032) (0.002) (0.033) (0.005)
Three-Year Lag -0.150** -0.008*** -0.140*** -0.025**
(0.046) (0.001) (0.047) (0.007)

 

Standard errors (clustered by state) are shown in parentheses. Covariates are as explained in the text. Regression coefficients are reported from
negative binomial models for counts and linear probability models for any visit.

The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

DDD = difference-in-difference-in-differences.

31
Supplemental Table. [Return to Text]

Effects of a Lesser-of Payment Policy on the Probability of Any Primary Care Visit
Under the DDD Model

 

Coefficient Standard Error

Dual -0.015* (0.006)
Dual* Lesser of -0.018*** (0.006)
Female -0.052*** (0.003)
Age Group (Relative to Under 65)

Age group 2 (65-74) 0.003 (0.004)

Age Group 3 (75-84) -0.041*** (0.005)

Age Group 4 (85+) -0.102** (0.006)
Race/Ethnicity (Relative to non-Hispanic White)

Non-Hispanic Black -0.064*** (0.004)

Non-Hispanic Other 0.003 (0.013)

Hispanic -0.041** (0.009)

Missing -0.039*** (0.006)
Original Reason for Eligibility Because of Age 0.042"** (0.002)
Number of Chronic Conditions (Relative to None)

One 0.286** (0.005)

Two 0.400** (0.007)

Three 0.440" (0.007)

Four 0.458*** (0.007)

Five 0.474" (0.007)

Missing 0.320°* (0.007)
MA Penetration Rate -0.128*** (0.040)
Dual* MA Penetration Rate 0.137" (0.060)
County per Capita Income (Relative to "Low')

Medium 0.049°* (0.007)

High 0.057** (0.011)
Number of Primary Care Physicians per Capita at the County Level (Relative to "Low")

Medium 0.013** (0.004)

High 0.013** (0.005)

 

Covariates not reported include dual-specific state fixed effects, dual-specific year effects, and state-year fixed effects.
The asterisks *, **, and *** indicate statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively.
MA = Medicare Advantage.

32