The Impact of Competitive Bidding in Health Care:

The Case of Medicare Durable Medical Equipment

Yunan Ji∗

October 1, 2019

Click here for the latest version.

Abstract

I study the impact of introducing competitive bidding in a health care market on theprice and utilization of health care services. I focus on the Medicare durable medicalequipment (DME) sector, an important but understudied health care sector that isused by one quarter of Medicare enrollees. Since 2011, competitive bidding has beenintroduced among Medicare DME suppliers to determine procurement prices for over300 DME items in 100 metropolitan statistical areas, while suppliers in other marketscontinue to receive administratively-set prices. Exploiting the variation in pricing rulesacross MSAs over time, I estimate that the introduction of competitive bidding reducedMedicare spending for the covered items by 46%. The reduced spending is attributableto a 36% reduction in average price, and a 11% reduction in quantity, which I measureby the share of beneficiaries using the covered items. Due to features of the auctiondesign, the market would not reach its competitive equilibrium; in fact, these resultssuggest that the market has moved from having excess supply to having excess demand.Several pieces of evidence suggest that, under this Medicare-created situation of excessdemand, the allocation of DME does not appear consistent with what one might expectfrom an efficient allocation.

∗Email address: ji@g.harvard.edu. I am deeply grateful to my advisors, Amy Finkelstein and DavidCutler, for their guidance and support throughout this project. I would also like to thank Tim Layton,Grace McCormack, Joseph Newhouse, Daniel Prinz, Mark Shepard, Zirui Song, and participants at theHarvard Department of Health Care Policy Research Seminar, Harvard Health Economics Tea, HarvardHealth Economics Working Group, and MIT Public Finance Lunch Seminar for their helpful comments.

1

1 Introduction

As health care spending reaches 18 percent of the U.S. GDP - almost twice as much per

capita as other developed countries, how to sustainably finance the health care system has

become a pressing policy questions facing the U.S. economy (Anderson et al. 2005, Emanuel

et al. 2012, Kesselheim et al. 2016, Papanicolas et al. 2018). Academics and policy makers

are increasingly pointing to prices as potential culprits of high health care spending, and

are calling for solutions that improve pricing efficiency (Cooper et al. 2018, Emanuel et al.

2012, Papanicolas et al. 2018, Sinaiko and Rosenthal 2011, Verma 2018). One widely touted

solution is the use of competitive bidding to set prices for health care services, allowing

competition among providers to drive down the prices faced by public payers and patients

(Emanuel et al. 2012, Song et al. 2012a).

The use of competitive bidding in health care has become increasingly common in recent

years. Since 2006, Medicare has been setting plan payments for its privately administered

plans (“Medicare Advantage”) based on insurer bids; in certain states, Medicaid programs

have been using competitive bidding to determine payments for their managed care plans;

competitive bidding among suppliers was introduced for Medicare durable medical equipment

in 2011; there have also been proposals to introduce similar programs for clinical lab tests,

and most recently, for physician-administered drugs (Curto et al. 2018, Layton et al. 2018,

Martin and Sharp 2018, MedPAC 2018).

In great contrast to the increasing prominence of competitive bidding in health care, com-

pelling evidence on its impact has been lacking. Perhaps unsurprisingly, competitive bidding

is generally implemented as a system-wide change — such as with the nation-wide introduc-

tion of Medicare Advantage, which makes empirical estimation of its impact challenging due

to the many potential confounding factors. This is perhaps why, to date, empirical analyses

of competitive bidding in health care have almost exclusively focused on regulatory changes

within a competitive bidding system, such as changes in bidding benchmarks (Cabral et al.

2018, Duggan et al. 2016, Song et al. 2012b, Song et al. 2013). Analysis of its overall impact

2

relative to an alternative payment regime is rare, and when done, relies heavily on modeling

assumptions (Curto et al. 2018).

In this paper, I provide empirical evidence on what happens when competitive bidding

is introduced in a health care market. I focus on the Medicare durable medial equipment

(DME) sector, which provides prescription medical devices for home use. The DME sector

is attractive for several reasons. First, changes in DME pricing and utilization are of direct

policy interest as they affect the health and welfare of the over one quarter of Medicare

beneficiaries who use DME.1 Second, DME policy could play a major role in controlling

overall health care spending — appropriate DME use allows patients to receive care at home,

which has been shown to be associated with better outcomes and significantly lower health

care spending in many clinical settings (Buhagiar et al. 2017, Doyle Jr et al. 2017, among

others); furthermore, numerous studies have listed DME among the important drivers of the

unexplained geographic variation in health care spending across the U.S., and understanding

DME pricing and utilization will directly contribute to our understanding of the causes

and consequences of geographic variation in health care spending (IOM 2013, Reschovsky

et al. 2012, among others). Finally, the staggered timing across MSAs in the introduction

of competitive bidding in DME provides a rare opportunity for empirically estimating the

impact of competitive bidding in an important health care market; experience from the DME

sector can provide valuable lessons for other sectors of health care.

Prior to the introduction of competitive bidding, Medicare DME was reimbursed un-

der administratively-set prices. Starting in 2011, for certain types of DME, the Centers

for Medicare and Medicaid Services (CMS) began setting prices based on competitive bid-

ding in what eventually became 100 MSAs, while continuing to pay administratively-set

prices in the remaining MSAs (MedPAC, 2018). Using detailed administrative data from

the 100% Medicare enrollment and claims files from 2009-2015, I estimate the effect of re-

placing administrative pricing with competitive bidding on DME prices and utilization, and

1See Table 1.

3

explore heterogeneity in impact across products and patient groups. The analysis employs

a difference-in-differences strategy that compares the price and utilization in areas where

competitive bidding replaced administrative pricing to areas where administrative pricing

remained in place until the end of the study period.

Economic theory suggests that competitive bidding could lead to a more efficient allo-

cation and reduce health care spending if administrative prices were previously too high.

Empirically, however, at least two factors could complicate the result and move us away

from the desired allocation. First, as Decarolis (2014) points out in the context of Italian

public work procurement auctions, competition on price may occur at the expense of re-

duced quality. This quality-competition trade-off may be exacerbated in health care, where

quality is notoriously hard to measure (Landon et al. 2003, Nyweide et al. 2009, Walshe

2000) and patients often lack the knowledge and ability to shop based on quality. (Kolstad

and Chernew 2009 provides a summary of this literature.) Second, even the most robust

theoretical result may fail to yield its intended impact when implemented in the real world,

which may involve political and resource constraints, as well as human error. Thus it is

important to not only understand an idea in theory, but also to examine what happens in

practice when the idea is (often imperfectly) implemented.

In this paper, I find that competitive bidding in DME reduced Medicare prices by an

average of 36% relative to administrative prices, and reduced average utilization by 11% rel-

ative to administratively set prices. The reductions in price and quantity together constitute

a 46% reduction in Medicare spending on the included items. I discuss below that these

results are consistent with the pricing regime change moving from one of excess demand

to excess supply. This makes sense since, as explained in more detail below, the design of

the DME auction meant that the market would not reach its competitive equilibrium, as

CMS paid winning suppliers the median of all winning bids, rather than the market-clearing

price. Indeed, the empirical findings are consistent with prior theoretical predictions and

laboratory experiments on the expected impact of the DME bidding system (Merlob et al.

4

2012 and Cramton et al. 2015).

The average impacts mask substantial heterogeneity across product categories – across

the five product categories analyzed in this paper, the amount of reduction in spending

ranges from 25% for wheelchairs to 41% for continuous positive airway pressure machines

(CPAP), while the amount of reduction in utilization ranges from 5% for oxygen equipment

to 23% for walkers.

Given the evidence of excess demand under competitive bidding, a natural question is

whether DME use was efficiently rationed among patients - i.e. allocated to the patients for

whom it generates the highest surplus. While a formal analysis of this question is beyond

the scope of this paper, I present two pieces of evidence that are suggestive of allocative

inefficency. First, it seems plausible to assume that the surplus from new use of a given type

of DME is greater than a replacement or upgrade to an existing equipment, yet the decline

in utilization occurs similarly among patients who are new to DME (i.e. new equipment

use) and patients who have received the same type of DME in the past (i.e. replacement or

upgrade). Second, I show that the marginal patient rationed out of DME under competitive

bidding is not healthier, but is older, less likely to be white, and more likely to be on Medicaid

(a measure of low resources).

This paper contributes to several related literatures. Most narrowly, this paper studies the

DME sector, a large but understudied part of the healthcare sector. Despite the fact that one

in four Medicare beneficiaries use DME, academic research on Medicare DME writ large is

surprisingly scarce. However, a few papers have previously analyzed the move to competitive

bidding in DME. As noted above, prior theoretical work and lab experiments both predicted

that the DME auction design would create excess demand, as my empirical results suggest;

however, these prior results were designed under numerous simplifying assumptions about

the bidding rules and institutional features, which could be important in practice.2 The

2For example, both papers model the program as a auction among firms of unit capacity, while in realitythe DME auction is a multi-unit auction conducted among suppliers of different capacity who may competein multiple markets.

5

limited existing empirical work is consistent with the results I find here; there is time series

evidence that prices declined for six DME items following the introduction of competitive

bidding (Newman et al., 2017). Cramton (2011) and Cramton (2012) provide a large amount

of descriptive data on the number of submitted claims, health measures of users and non-

users of DME, and the winning and losing suppliers in the nine MSAs assigned to competitive

bidding in 2011.

In addition, this paper contributes to the small but growing literature on competitive

bidding in health care, as discussed above (Cabral et al. 2018 Duggan et al. 2016 Song et al.

2012b Song et al. 2013).

Finally, most broadly, this paper is related to the empirical literature on procurement

auctions, particularly those using quasi-experimental design to study the impact of intro-

ducing competitive bidding in a market (e.g. Cicala 2017 and Decarolis 2014).

The rest of the paper proceeds as follows: Section 2 describes the Medicare DME sector

and its competitive bidding system; Section 3 describes the data and summary statistics,

and lays out a simple conceptual framework; Section 4 describes the empirical strategy and

identification; Section 5 presents the results; Section 6 concludes.

2 Setting

2.1 Durable Medical Equipment

In an aging population, durable medical equipment (DME), such as oxygen concentrators,

wearable defibrillators, and wheelchairs, are essential to patients who rehabilitate at home.

The Centers for Medicare and Medicaid Services (CMS) define DME as medical equipment

prescribed by a physician, for home-use, and expected to last for at least three years.3

Medicare covers a wide variety of DME products, ranging from items as small as glucose

testing strips and diabetic shoe inserts to large equipment including hospital beds and patient

3https://www.medicare.gov/coverage/durable-medical-equipment-coverage.html

6

lifts. Some types of DME are used independently (e.g. wheelchairs) while others require the

relevant supplies (e.g. oxygen used with oxygen concentrators).

Medicare reimburses for DME based on the Healthcare Common Procedure Coding Sys-

tem (HCPCS), which is a standardized coding system for identifying health care products,

supplies, and services.4 For example, a HCPCS code of E1035 refers to ”Multi-positional

patient transfer system, with integrated seat, operated by care-giver, patient weight capacity

up to and including 300 lbs.” In 2009, Medicare included over 1,800 unique HCPCS codes in

its DME fee schedule.5 Related HCPCS codes are grouped into approximately 60 categories

based on Durable Medical Equipment Coding System Product Classification.6 For example,

HCPCS code E1035 and seven other HCPCS codes fall into the “Patient Lift” category.

Throughout this paper, I will use “items” to refer to unique HCPCS codes, and “product

categories” or “types of product” to refer to product classifications.

DME is frequently prescribed to patients post-discharge from acute or post-cute care

facilities,7 but certain types of DME are also often obtained following outpatient visits.8

Not surprisingly, as I document in Section 3 below, Medicare beneficiaries who use DME

are substantially less healthy and have substantially higher healthcare use than non-users.

To receive DME under Medicare benefits, a beneficiary needs to obtain a prescription from

their physician, with which they can then obtain the relevant item from a Medicare-approved

supplier. DME is covered under Medicare Part B benefits, and patients are responsible for

a 20% copayment, which may be covered by a supplemental insurance or Medicaid.9 The

4https://www.cms.gov/Medicare/Coding/MedHCPCSGenInfo/index.html5https://www.cms.gov/medicare/medicare-fee-for-service-payment/DMEPOSFeeSched/DMEPOS-Fee-

Schedule.html6https://med.noridianmedicare.com/web/jddme/contact/pdac7https://www.cms.gov/Research-Statistics-Data-and-Systems/Monitoring-Programs/

Medicare-FFS-Compliance-Programs/Recovery-Audit-Program/Approved-RAC-Topics-Items/

0019-DME-Billed-While-Inpatient.html, https://www.medicareadvocacy.org/

delivery-and-set-up-guidelines-for-durable-medical-equipment-prosthetics-orthotics-and-supplies-dmepos/8For example, continuous positive airway pressure devices (CPAP) are often prescribed to pa-

tients diagnosed with sleep apnea following an outpatient sleep study, See https://www.cms.gov/

Regulations-and-Guidance/Guidance/Transmittals/Downloads/R96NCD.pdf9https://www.medicare.gov/coverage/durable-medical-equipment-dme-coverage

7

supplier is responsible for delivering the item to the patient in a timely manner.10

DME suppliers may be independent, or affiliated or owned by a hospital or post acute care

facility.11 In addition to suppliers that specialize in DME, pharmacies may also be considered

“suppliers” if they carry DME products.12 Most DME suppliers are local or regional, and

carry a selected set of products rather than the full spectrum of equipment. Appendix Table

A1 reports summary statistics on Medicare DME suppliers. In 2009, the average supplier sold

products from just 4.5 categories, out of a total of about 60 product categories reimbursed by

Medicare. The average supplier served 168 patients from 4.6 MSAs, and received $114,069 in

Medicare reimbursement.13 The average MSA has about 400 DME suppliers, although since

most DME suppliers only carry a limited set of DME products, there are fewer suppliers

for each given product category.14 Among the 10 most used product categories, the average

number of suppliers ranges from 29 for lenses to 193 for glucose monitor.

Despite making up only 2% of total Medicare spending, DME is used by 26% of Medicare

beneficiaries annually, more than the share of beneficiaries using acute care (17.7%) and

post-acute care services (4.8%) combined.15 Changes in DME policy could therefore have

an impact on the health and well-being of a large share of beneficiaries.

2.2 Scope of Competitive Bidding in Medicare DME

Traditionally, DME has been paid based on administrative prices that largely follows the list

prices and charges from the late 1980s (MedPAC 2018). Overtime, the use of inflated and

outdated prices has led to concerns over inappropriate utilization. In recent years, reports

have shown that Medicare has been paying significantly more for DME than private insurers

10https://www.govinfo.gov/content/pkg/CFR-2005-title42-vol2/pdf/

CFR-2005-title42-vol2-sec424-57.pdf11https://www.law.cornell.edu/cfr/text/42/424.5712https://www.cms.gov/Outreach-and-Education/Medicare-Learning-Network-

MLN/MLNProducts/Downloads/DMEPOSPharmFactsheetICN905711.pdf13Suppliers are defined as unique National Provider Identifiers (NPIs). Some suppliers could share owner-

ship, which I cannot distinguish in the claims data.14Excludes suppliers with fewer than 25 claims from a given MSA in 2009.15Author’s calculation based on the 2009 Medicare claims data.

8

in the commercial market (MedPAC 2018). To address these concerns, CMS began seeking

alternative price-setting methods and tested out two small-scale competitive bidding pilot

programs in Polk County, Florida, and San Antonio, TX, between 1999 and 2002. Savings

generated from the pilot programs prompted the adoption of competitive bidding at a larger

scale. The Medicare Prescription Drug, Improvement, and Modernization Act (MMA) of

2003 authorized CMS to implement competitive bidding programs for DME, starting with the

largest MSAs and with the intention to expand to additional areas in later years (MedPAC,

2018). On January 1, 2011, nine MSAs were assigned to competitive bidding (Round 1

MSAs).16 On July 1, 2013, suppliers in another 91 MSAs also entered competitive bidding

(Round 2 MSAs). Figure 1 shows a map of these MSAs in the continental U.S.. Among

these MSAs, CMS selected items for competitive bidding that were deemed high cost and

high volume, with the exception that Class III medical devices, the highest risk level of

device classification by the Federal Drug Administration (FDA), would not be subject to

competitive bidding. 231 items in six product categories and 196 items in eight product

categories were assigned to competitive bidding in the two sets of MSAs, respectively.17

Prices for these chosen DME items would be determined based on supplier bids, whereas

the prices for other DME items continued to follow administratively-set fee schedules. The

items placed under competitive bidding in Rounds 1 and 2 together represented 54% of DME

spending under administratively set prices in 2009.

2.3 Bidding Rules

Suppliers wishing to sell items that were subject to competitive bidding to Medicare ben-

eficiaries residing in competitive bidding MSAs are required to submit bids to CMS, and

winning suppliers are granted the right to sell for three years, at a price set by CMS based

on the bids.

16Competitive bidding for these nine MSAs was initially slated to begin on 2008, but was postponed to2011. Instead, CMS imposed a 9.5% payment cut across all MSAs in 2008, regardless of whether they wouldbe subject to competitive bidding.

17See Table 3 for examples of product categories and items in each category.

9

Suppliers bid separately for each product category in each MSA (e.g. oxygen equipment

and supplies in the Boston-Cambridge-Newton, MA-NH MSA), and the competition and

contracting both occur at the product category-MSA level. The suppliers are required to

bid for every item within a given product category, each of which is assigned a weight (the

national volume of the given item relative to other items in the same category) that is

known to the suppliers. Suppliers must bid at or below the maximum allowed bid, which is

defined as the administrative price that would have been paid absent competitive bidding.

An example of the bidding form is shown in Appendix Figure A1.

After all bids have been submitted, CMS ranks suppliers based on each supplier’s com-

posite bid — the weighted sum of bids across all items in a product category, and offers

contracts starting from the supplier with the lowest composite bid. CMS continues to offer

contracts to suppliers until it deems that there are enough suppliers to meet the market

demand.18 For a more detailed description of the bidding process, see Appendix A.

A key feature of this auction will have important implications for the empirical analysis –

the price for a given item is set to the median of the winning bids, rather than a supplier’s own

bid for that item. Prior works by Cramton et al. (2015) and Merlob et al. (2012) point out

that the median pricing design, coupled with the possibility for suppliers to withdraw from

a contract post-competitive bidding, results in below-equilibrium prices and quantity. As

illustrated in Figure 2(a), assuming that CMS has perfect information about market demand,

if we rank supplier bids for a given item from the lowest to the highest and offer contracts

to suppliers until demand is satisfied, all suppliers bidding at or below the competitive

equilibrium price, p∗, would win a contract. Setting price at p∗ would allow the market to

reach equilibrium quantity q∗. However, CMS sets the price at the median of the winning

bids, which would cause half of the suppliers to be paid below their own bid, and potentially

their reservation price, resulting in a quantity that is below the market equilibrium. Cramton

et al. (2015) and Merlob et al. (2012) predict that in addition to the mechanical effect of

18https://www.cms.gov/medicare/provider-enrollment-and-certification/

medicareprovidersupenroll/dmeposaccreditation.html

10

paying half of the suppliers less than their own bid, the auction design also induces potential

strategic bidding behavior. Specifically, since suppliers are paid the median rather than their

own bid, they may be incentivized to bid below their cost in order to increase their chances of

winning. If a large number of suppliers strategically bid below their cost, the price generated

by the auction would be unsustainably low, causing the market to unravel. However, the

concerns over market unraveling due to low bids may be mitigated in practice, as CMS

requires the lowest bidders to demonstrate that the bids were ”bona-fide” by submitting

manufacturer invoices and other financial information, and failure to submit such evidence

would void the bid.19 To the extent that suppliers strategically bid below own cost in order to

win contracts, the quantity shortage caused by “median” price setting would be exacerbated.

Figure 2(b), shows that while, ex-ante, the impact on price is expected to be weakly

negative (since suppliers are required to bid no higher than the administrative price), the sign

of any effect on quantity is a-priori ambiguous, and depends on the level of the administrative

price relative to the competitive equilibrium price, as well as the elasticity of demand. In this

figure, in the extreme case where patient demand for DME is perfectly inelastic, for example

due to supplemental insurance coverage, the reduction in price would weakly reduce quantity.

In the extreme case where supply is perfectly elastic on the margin, for example if suppliers

are able to furnish additional units of DME without incurring increased cost, the reduction

in price would weakly increase quantity.

Existing evidence suggests that administrative prices were above the competitive equi-

librium. Newman et al. (2017) shows that for six respiratory and oxygen-related items,

Medicare prices had been above commercial prices under administrative fee schedules, and

were reduced to below commercial prices post-competitive bidding, suggesting that at least

for these items, administrative prices were set above the competitive equilibrium in the

pre-period. Additionally, MedPAC (2018) shows that for nine of the ten highest spending

products that were not subject to competitive bidding, Medicare reimbursement rates ex-

19See Section A of the Appendix for more details.

11

ceed that of the median private payer rate by 18% to 57%. This phenomenon likely applies

to other DME products, all of which have traditionally been paid based on a fee schedule

derived from the list prices in the 1980s and only adjusted annually based on the consumer

price index (MedPAC, 2018).

3 Data and Summary Statistics

3.1 Sample Definition and Summary Statistics

The main data for the analysis are the 100% Medicare enrollment and claims data from

2009 to 2015, which contain the universe of Medicare beneficiaries and their health care

claims over this period. The Medicare claims data allow me to observe prices of different

items in each market, the health care utilization of each Medicare beneficiary, and from which

supplier each beneficiary purchases their DME. The enrollment file, which I link to the claims

file, contains data on patient characteristics, including age, race, sex, zip code of residence,

Medicaid eligibility (a measure of low resources), and chronic conditions. I supplement these

data with publicly available Medicare fee schedules and competitive bidding prices from the

same period, which provide a denominator of all DME items covered by Medicare and their

prices in each MSA.20

I assign beneficiaries to MSAs based on their residential zipcode and county on file with

Medicare. By CMS rule, a beneficiary’s residence is used to determine whether she faces

competitive bidding or non-competitive bidding fee schedule prices, regardless of the location

of the transaction.2122 This feature of the program means that a beneficiary cannot be

charged more or less when they travel outside their MSA of residence, although they may

20Data available at https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/

DMEPOSFeeSched/DMEPOS-Fee-Schedule.html and https://www.dmecompetitivebid.com/, accessedJuly 2018.

21https://www.cms.gov/Outreach-and-Education/Medicare-Learning-Network-MLN/MLNProducts/

Downloads/DME_Travel_Bene_Factsheet_ICN904484.pdf22Competitive bidding MSAs are determined by a set of zipcodes, rather than based on the Census Bureau

definition. See https://www.dmecompetitivebid.com/

12

face a different set of suppliers depending on which suppliers are eligible to sell in each MSA.

The baseline sample includes all Medicare enrollees residing in an MSA between 2009

and 2015. Table 1 compares the characteristics and health care utilization of Medicare ben-

eficiaries who do and do not use DME. Panel (a) of Table 1 compares the demographics and

health status of DME users and non-users. On average, compared to non-users, beneficiaries

who use DME are 2.3 years (or 3%) older, 5.3 percentage points (or 10%) more likely to

be female, and 10.6 percentage points (or 68%) more likely to be on Medicaid. DME users

are also significantly sicker than non-users, as measured by the number of chronic conditions

they have. The average DME user has about five chronic conditions, more than double the

average among non-users; 80% of DME users have at least three chronic conditions, com-

pared with 20% among non-users. Panel (b) of Table 1 compares the health care utilization

of DME-users and non-users. Notably, beneficiaries who use DME spend almost four times

as much on health care services annually than non-users ($18,205 for DME users vs. $4,828

for non-users). Looking separately across different health care settings reveals that DME-

users use health care services at a much higher rate – compared with non-users, those who

use DME are three times as likely to have an inpatient admission (35.7% vs. 11.4%), four

times as likely to use institutional post acute care services, which include skilled nursing

facilities, inpatient rehabilitation facilities, and long-term care hospitals (9.7% vs. 2.4%),

and over six times as likely to use home health services (22.3% vs. 3.5%). Panel (c) of Table

1 summarizes the utilization of DME among Medicare beneficiaries. Conditional on using

any DME, the average beneficiary uses 1.7 distinct types of products (e.g. a wheelchair and

an oxygen concentrator) or 4 distinct items, regardless of type (e.g. a wheelchair, an oxygen

concentrator, liquid oxygen used with the concentrator, and a mask used with the oxygen

concentrator.) The most common type of DME is a glucose monitor, used by 10.5% of the all

Medicare benficiaries or 38% of those who use any DME. Other common items include oxy-

gen supplies and equipment (4.3% of all beneficiaries), nebulizers and related drugs (3.8%),

and wheelchairs (3.3%).

13

Table 2 compares the 2009 characteristics of the 9 MSAs that were assigned to competitive

bidding in January 2011 (“Round 1”), the 91 MSAs that were assigned to competitive bidding

in July 2013 (“Round 2”), and the 271 MSAs that remained under administrative pricing.

Among Medicare beneficiaries who reside in an MSA, 9% are in a Round 1 MSA, 64% are in

a Round 2 MSA, and 27% are in other MSAs. Since population was the main criterion for

MSA selection, MSAs assigned to competitive bidding have significantly higher populations

than non-competitive bidding MSAs. Competitive bidding MSAs also have a lower share of

white residents, but are similar to non-competitive bidding MSAs in percent female, percent

age 65 and above, high school graduation rate, and home ownership rate. Comparing to

non-competitive bidding MSAs, Round 1 and Round 2 competitive bidding MSAs are more

similar in demographic composition.

MSAs assigned to competitive bidding have a slightly lower share of population on Medi-

care, but among enrollees, a similar share of Medicare-Medicaid dual-eligibles and similar

numbers of chronic conditions as in non-competitive bidding MSAs. Total Medicare spending

is very similar across the three groups of MSAs, and so are most sub-categories of Medicare

spending, with the exception that non-competitive bidding MSAs spend slightly more on

hospital outpatient care, and on durable medical equipment. The share of Medicare en-

rollees using any DME is 18.8% in Round 1 MSAs, 19.5% in Round 2 MSAs and 22.5% in

non-competitive bidding MSAs.

With the exception of wheelchairs, product categories assigned to competitive bidding

are designed to be comprehensive, and include all relevant equipment of a given product

type and any supplies used with the equipment.23 24 Product categories were added and

removed from the list of competitive bidding items over time and also differ across the two

sets of MSAs. For the analysis, I restrict to items that were both subject to competitive

bidding in all 100 competitive bidding MSAs, and were continuously paid under competitive

23For example, all walkers and walker accessories reimbursed by Medicare are subject to competitivebidding under the ”walkers” category.

24In Round 1 MSAs, there are two categories of wheelchair while in Round 2 MSAs, there was one category.For ease of analysis, I combine the two categories in Round 1 into one wheelchair category.

14

bidding prices from the initial introduction of the program in an MSA until the end of the

study period. These DME items fall into five product categories – oxygen, continuous airway

pressure (CPAP), wheelchairs, walkers, and hospital beds – and make up 39% of overall DME

utilization.25

3.2 Variable Definitions and Summary Statistics

I perform all analyses at the MSA - half year level. I define price as the Medicare reim-

bursement price for each DME item, including both the share paid by Medicare (80%), and

patient cost-sharing (20%). The main utilization measure is the share of beneficiaries in

each MSA who use any DME item that is included in competitive bidding within each half

year, which I obtain by dividing the number of beneficiaries with a medical claim on any

included DME within each half year by the number of beneficiaries residing in each MSA. I

also construct an alternative measure of utilization – standardized utilization per beneficiary,

which is defined as the per beneficiary spending on included DME after replacing the price

paid for each item with the mean fee schedule price for that item in non-competitive bidding

MSAs. By striping away any price differences due to geography or competitive bidding,

changes in this standardized utilization measure captures changes in the quantity of DME

used. The spending measure, average spending per beneficiary, is obtained by dividing the

sum of Medicare spending on items subject to competitive bidding in each MSA in each

half year by the the number of beneficiaries residing in each MSA. For ease of comparison

across different subsamples, I log transform all price, utilization, and spending outcomes for

analyses throughout the paper. To avoid taking the log of zero, log share of beneficiaries is

defined as log(share of beneficiaries + 0.0001), log standardized utilization per beneficiary

as log(standardized utilization per beneficiary + 0.0001), and log spending per beneficiary

25Author’s analysis of the Medicare claims data.

15

is defined as log(spending per beneficiary + 0.0001).26 27

Table 3 summarizes the price of these competitive bidding DME in the first six months

of the study period, prior to the auctions. The average price of a competitive bidding item

is $157, with little variation across MSAs, but large variations across items (Table 3 row

(1), columns (1) through (3)). Among all competitive bidding items, the cheapest is a

wheelchair bearing, costing $0.6 per piece on average, and the most expensive is a heavy

duty power operated vehicle, costing $2,138 per piece on average. Comparing across the five

product categories, wheelchair is the most expensive by average price. There is substantial

heterogeneity in price within each category — for example, the lowest and highest priced

items within the “hospital beds” category cost $3.6 and $699, respectively.

4 Empirical Strategy

I estimate the effect of introducing competitive bidding by comparing the price and utilization

of DME items in MSAs where competitive bidding was introduced during the study period

to MSAs where administrative fee schedules remained in place.

Figure 3(a) shows the raw trends in log price for items subject to competitive bidding,

separately for MSAs that were assigned to competitive bidding in January 2011, MSAs that

were assigned to competitive bidding in July 2013, and MSAs that were paid by adminis-

trative fee schedule throughout this time period. Simple averages are taken across items

and MSAs without weighting. Log price in 2009 is normalized to zero. Prior to competitive

bidding, price trends in the three sets of MSAs closely followed each other. Price decreased

by 30 to 40 percent when MSAs entered competitive bidding, and the magnitude of the de-

crease is almost identical between the two sets of MSAs. The price remains stable after the

26The only analysis in this paper where any of these measures contains zero is in the first two columns ofTable 6, where I restrict to the subset of beneficiaries with prior use. The share of MSA-years with zeros are0.03% for wheelchairs and CPAP, 0.2% for oxygen, 0.9% for walkers, and 1.6% for hospital beds.

27These quantity measures are preferable to a simple count of “number of DME units used” because someitems are designed to be used in large quantities (e.g. liquid oxygen, or disposable face mask) while othersare designed to last a long time (e.g. an oxygen concentrator). Aggregating across different items thereforeimplicitly places a large weight on disposables and supplies over equipment.

16

introduction of competitive bidding, even though a second round of bidding was conducted

in 2014 for MSAs that initially entered bidding in 2011. 28

As a sanity check that the change in price was indeed due to competitive bidding rather

than other system wide changes at the MSA level, Appendix Figure A2 replicates Figure 3

Panel (a) for DME items that were paid under administrative pricing throughout the study

period. For these items, the price trends remained flat over time for all three groups of

MSAs.

Analogous to Figure 3 Panel (a), Panel (b) plots the raw trends of the main utilization

measure – log share of beneficiaries using competitive bidding DME. The pattern of utiliza-

tion in panel (b) closely follows the pattern of price in Panel (a); the share of beneficiaries

who use competitive bidding DME declined sharply immediately after competitive bidding

was introduced.

To empirically quantify the impact of competitive bidding on price and utilization, I

combine the two sets of MSAs that entered competitive bidding by creating a relative time

measure – months since competitive bidding – denoted by θr(j,t) for MSA j in a six-month

period t. θr(j,t) = 0 in the first six months MSA j enters competitive bidding. I use a

six-month time increment because the second set of MSAs entered competitive bidding six

months into the year.

For each competitive bidding MSA j in half-year t, I estimate the following difference-

in-differences event study specification

ln(yjt) = γj + τt + ΦrCBj × θr(j,t) + εjt (1)

where γj, τt indicate MSA, and half-year fixed effects, respectively. CBj is an indicator

for MSAs subject to competitive bidding. θr(j,t) are indicators for relative months. The

coefficients Φr’s quantify the impact of competitive bidding on the outcome of interest,

28Competitive bidding prices remain in place for three years, and a new round of bidding is conducted atthe end of each three year period.

17

ln(yjt). In the analysis of prices, yjt the log average price of competitive bidding DME in

MSA j in half-year t; in the analysis of utilization, yjt is either the share of beneficiaries

residing in MSA j in half-year t who had a Medical claim on any competitive bidding DME,

or the mean standardized utilization among beneficiaries residing in MSA j in half-year t.

To summarize the impact over the post-period months, I also estimate a pre-post version

of the same specification,

ln(yjt) = γj + τt + ΦCBj × Postt + εjt (2)

where Postt is an indicator for post-competitive bidding.

The difference-in-differences regression specification relies on the identifying assumption

that absent competitive bidding, the outcome of interest would have evolved in the same

pattern across the different sets of MSAs. This assumption holds for prices by construction,

as prices are otherwise administratively set and updated over-time only by multipliers stip-

ulated by law.29 This assumption also appears to hold for utilization and spending, given

the lack of a pre-trend in the event studies, shown in the next section.

5 Results

5.1 Impact on Price, Utilization, and Spending

Figure 4 plots the estimates from equation (1). The figure uses a panel of MSA-items

that is balanced in relative months, focusing on the 24 months before and after competitive

bidding was introduced. The coefficient on relative month −6 is normalized to zero. Relative

months in MSAs that never introduced competitive bidding are also set to zero. The event

study shows a flat pre period trend, which is a mechanical result due to administratively

set prices prior to competitive bidding. Similar to the raw trends, the event study shows

an average reduction of about 40% in price when competitive bidding was introduced. The

29https://www.ssa.gov/OP_Home/ssact/title18/1834.htm

18

reduction remains roughly constant in the 24 months post-competitive bidding, which is

also mechanical due to that prices generated from auctions are in effect for three years. The

estimates are fairly precise, shown by the 95% confidence intervals in the figure. A version of

the event study using the full, unbalance panel, is shown in Appendix Figure A3 and gives

similar estimates.

A pre-post analogue of the event study estimate in Figure 4 is summarized in Table 4

row (1), which reports estimates of Φ from equation (2). Column (2) of the table reports

the implied percentage change based on the coefficient estimates reported in column (1). On

average, competitive bidding in DME led to a 36% reduction in price. Row (2) reports a

44% reduction in price when I weight each item by their utilization in the first half of 2009

(the first half year of the sample).

The decline in price is consistent with the theoretical prediction discussed in Section 2.3.

Suppliers must bid at or below the administrative fee schedule price, therefore prices would

weakly decrease by design. That prices decreased by such a large amount, however, could

be a result of either (or both) of two forces, which I cannot distinguish between empirically.

As discussed in Section 2.3, first, since administratively set prices are believed to be much

higher than cost, reduced prices could be the result of competition driving prices closer to

cost. Second, the low price could be a result of CMS’s pricing rule, both mechanically and

due to the strategic bidding behavior that it might induce.

Figure 5 plots estimates from equation (1) on the utilization of items subject to com-

petitive bidding. The lack of a pre-period trend suggests that the identifying assumption

discussed in Section 4 is likely valid. The figure shows a statistically significant reduction of

about 10% in the share of patients with any DME claims after competitive bidding was intro-

duced. The decline in utilization remained roughly stable in the 24 months post-competitive

bidding, suggesting that the reduction is likely to sustain over-time. As previously illustrated

in Figure 2, although prices are predicted to go down, the impact on utilization was ex-ante

ambiguous. The reduction in both price and quantity, however, suggests that the market

19

has excess demand and is moving further from its competitive equilibrium (illustrated by

the red arrow in Figure 2).

Rows (3) through (6) of Table 4 report the impact of competitive bidding on utilization

through different specifications, all of which show a statistically significant decline. The

baseline specification is shown in row (3), which reports estimates from the pre-post analogue

of the event study estimates in Figures 5: after competitive bidding was introduced in DME,

the share of beneficiaries using any competitive bidding item was reduced by 11%.

Row (4) of Table 4 repeats the same analysis using an alternative utilization measure

– log standardized utilization per beneficiary. This measure is constructed by computing

the Medicare reimbursement assuming that each item was paid the mean fee schedule price

among MSAs that were never subject to competitive bidding. Consistent with the primary

utilization measure, there is a statistically significant decline of 7.2% in standardized utiliza-

tion per beneficiary.30

Rows (4) and (6) of the table replicate rows (3) and (5) but weight the estimates by

the number of beneficiaries residing in each MSA; the coefficient estimates can thus be

interpreted as “per beneficiary” changes. The estimates show that 10.5% fewer beneficiaries

were using DME as a result of competitive bidding and the per beneficiary standardized

utilization declined by 11.6%.

Figure 6 plots estimates from equation (1) where the outcome is log spending per ben-

eficiary on items subject to competitive bidding. Again, the lack of a pre-trend prior to

competitive bidding supports the validity of the identifying assumption. The figure shows

a roughly 50% reduction in spending when competitive bidding was introduced, and the

amount of reduction remains stable in the 24 months post-competitive bidding.

30Two things should be kept in mind when trying to compare the two point estimates. First, the twomeasures capture different margins of utilization – the primary outcome captures the extensive marginutilization at the beneficiary level, and the latter captures the average overall utilization. Second, the twomeasures also place different implicit weights on the different products. Share of beneficiaries weights allitems equally; standardized utilization per beneficiary places more weight on items that have higher feeschedule prices. Therefore, one cannot directly back out the intensive vs. extensive margin response bycomparing the two estimates.

20

Row (6) of Table 4 reports the pre-post analogue of the event study in Figure 6, which

shows a 46% reduction in per beneficiary spending on the included DME items.

As a robustness check to the above results, in Section B.2 of the Appendix, I estimate an

alternative specification at the item-MSA-half year level, rather than at the MSA-half year

level. Unlike the model in equations (1) and (2), which captures the aggregate changes in

price and utilization, the alternative specification captures an average effect across different

items, both of which show a statistically significant reduction.

5.2 Heterogeneity and Mechanisms

5.2.1 Heterogeneity Across Products

I explore heterogeneity in changes across product categories, and across items within product

categories.

Figures 7, 8, and 9 replicate the price, quantity, and spending results (from Figures 4,

5, and 6, respectively) by product category. Table 5 summarizes the corresponding model

estimates. For each of the five product categories, price declined sharply at the introduction

of competitive bidding. The magnitude of the decline varies across products, with continuous

positive airway pressure (CPAP) and walkers showing the largest price reductions (41% and

40%, respectively), and wheelchairs and hospital beds showing the smallest price reductions

(25.2% and 28.3%, respectively). The variation in the amount of price reduction could be

due to variation in markups prior to competitive bidding, or reflect differences in market

power or the extent of strategic bidding behavior.

The largest decline in utilization is in walkers, which declined by 23%, and the smallest is

in CPAP, which declined by 6%. The heterogeneity in changes in utilization does not appear

to be explained by the heterogeneity in price reductions alone – the correlation between

changes in price and changes in utilization is 0.1531. The weak correlation between the

reduction in price and the reduction in quantity at the product category level is perhaps

31Author’s calculation based on the Medicare claims data.

21

not surprising, given that these products constitute different markets that vary in their

levels of competitiveness and the nature of demand, such as how discretionary or clinically

indispensable a certain DME would be to a patient. There is similarly substantial variation

in the impact on spending; among the five categories, walkers had the largest reduction in

spending (62%) while wheelchairs had the smallest (37%). Similar patterns are found using

alternative specifications and outcome measures, shown in Appendix Table A3.

Looking within product categories reveals a clear positive correlation between changes

in price and quantity at the item level. Figure 10 shows that the correlation coefficient

between changes in log price and log utilization is 0.42 across all competitive bidding items.

The figure shows that items that saw a smaller decrease in price also had a smaller decrease

in utilization, and in some cases, an increase in utilization. This could reflect differences

in demand elasticities across items, but could also suggest substitution toward better paid

items within product categories. At the product category level though, there appears to

be little substitution toward non-competitive bidding products (Table 5 row (6)), which

is perhaps not surprising since product categories are generally large and comprehensive.

Furthermore, given that the average supplier only carries 4.5 product categories, their ability

to steer patients toward different product categories is limited. In contrast, within a product

category, it may be possible for suppliers to respond to price reductions by moving patients

toward the better paid items.

5.2.2 Nature of Rationing

Given the evidence of excess demand under competitive bidding, a natural question is

whether DME use was efficiently rationed among patients - that is, if DME was allocated

to the patients for whom it generates the highest surplus. While a formal welfare analysis

is beyond the scope of this paper, I present two pieces of evidence that are suggestive of

allocative inefficiency.

First, it seems reasonable to assume that the surplus from new use of a given type of DME

22

is greater than a replacement or upgrade to an existing equipment. Therefore, an efficient

allocation of DME would suggest a smaller reduction in new use and a larger reduction in

replacements and upgrades. To empirically test this, I estimate equation (2) separately by

prior use, defined as whether the beneficiary received a DME of the same type in the first

three years of the sample.32 Reductions in claims among patients already in possession of the

same DME likely represent reduced equipment upgrades or replacements, while reduction

among those without a prior claim likely represents reduced new use. Table 6 reports the

regression estimates. Across all five product categories, both beneficiaries with and without

prior use are statistically significantly less likely to use DME following the introduction

of competitive bidding, and the magnitude of the reduction appears comparable between

the two groups. Assuming that new uses generate greater surplus than replacements and

upgrades, these results suggest that the rationing of DME among patients is likely inefficient.

Second, it seems plausible to assume that the surplus from DME use is greater for patients

who are sicker, but should not otherwise differ by race, gender, or Medicaid status. Table 7

estimates the impact of competitive bidding on the average characteristics of patients receiv-

ing DME. Changes in the characteristics reflect the (endogenously) changing composition of

beneficiaries who receive DME under competitive bidding with excess demand compared to

administratively set prices with excess supply. Interestingly, despite the 11% reduction in

the share of beneficiaries using DME, the average patient receiving DME does not appear

any sicker relative to the control group, as measured by the number of chronic conditions.

Given that patients who use DME are generally observably sicker, as shown in Table 1,

the result raises concerns that the allocation of DME among patients may be inefficient.

Results on other patient characteristics exacerbates such concerns. Notably, the percent of

DME recipients who are Medicare-Medicaid duals decreased by a statistically significant 1.5

percentage points, relative to a pre-competitive bidding mean of 26.9 percent among R1 and

32For this exercise, I am focusing on the 91 MSAs that entered competitive bidding and the MSAs thatnever entered competitive bidding, because the first 9 MSAs that entered in 2011 do not allow for a longenough pre-period to establish prior use.

23

R2 MSAs. This likely reflects the fact that suppliers often face a lower “de facto” price with

Medicare-Medicaid dually eligible beneficiaries, due to policies that allow Medicaid (which

is responsible for the 20% patient cost-sharing for dually eligibles) to not pay the copay-

ment.33 Medicaid patients would not be rationed out in the situation of excess supply under

administrative pricing; however, in the situation of excess demand post-competitive bidding,

Medicaid patients are disproportionately rationed out due to their lower payments relative

to non-duals. Furthermore, those who receive DME are also more likely to be white as well

as slightly younger and more likely to be male. These results suggest that the average ben-

eficiary whose utilization is restricted is not healthier than the average DME user but does

appear to come from a more disadvantaged social-economic background. The result appears

consistent with the intuition that in a market where supply is limited, all else equal, those

with fewer social resources are more likely to be excluded.

6 Discussion and Conclusion

This paper studies the impact of competitive bidding on the price and utilization of durable

medical equipment, exploiting the staggered introduction of competitive bidding across dif-

ferent metropolitan statistical areas in the U.S. Difference-in-differences estimates show that

on average, competitive bidding has led to a 36% reduction in price and a 11% reduction

in the share of beneficiaries using DME, with variation across different types of products.

Total spending on the included items was reduced by 46%.

These results suggest that Medicare moved from a situation of excess supply under ad-

ministratively set prices to excess demand under its competitive bidding system. In addition,

several pieces of evidence suggest that, under this Medicare-created excess demand for DME,

the limited DME was allocated in a manner that does not appear consistent with what we

33Many states have a “lesser of” policy under which Medicaid only pays based on the lesser of Medicaid andMedicare reimbursement rates. If 80% of Medicare reimbursement rate was higher than the Medicaid rate,then Medicaid no longer pays the copayment. https://www.cms.gov/Medicare-Medicaid-Coordination/Medicare-and-Medicaid-Coordination/Medicare-Medicaid-Coordination-Office/Downloads/

Access_to_Care_Issues_Among_Qualified_Medicare_Beneficiaries.pdf

24

suspect efficient allocation would look like. Several pieces evidence suggests potential ineffi-

ciencies in allocation post-competitive bidding. First, the decline in quantity appears equally

attributable to declines in upgrades and new use. Second, the marginal patient receiving the

product is not sicker, but are less likely to be on Medicaid and more likely to be white.

Competitive bidding has been widely touted as a solution to controlling health care costs

and enhancing competition. However, analysis of the DME competitive bidding program

shows that economic tools may not yield their intended results when implemented improperly.

The study highlights the importance of complementing theoretical models with empirical

analysis when evaluating public programs, as well as taking into consideration not only the

average impact of a policy, but also its allocative consequences.

This paper also has several caveats, some of which will be address in future iterations

of this paper or in future work. One major limitation in the current paper is its lack of

analysis on health outcomes and patient well-being, which prevents it from making a more

comprehensive assessment of the impact of DME competitive bidding on patient welfare.

Furthermore, the paper analyzes a federal program that was implemented with flaws, a

natural next step would be to understand what could potentially happen should a well-

designed and well-implemented competitive bidding program be introduced. Finally, it is

important to keep in mind that demand for health care services is likely distorted due

to health insurance. Therefore, the competitive equilibrium, which an efficient auction is

designed to achieve, does not necessarily represent the first best allocation in a health care

market. Failure to achieve the competitive equilibrium allocation is only one of the many

inefficiencies in the DME market, and as pointed out by Mahoney and Weyl (2017), fixing

one dimension of imperfection in the presence of multiple market imperfections does not

always improve welfare.

25

References

Anderson, Gerard F, Peter S Hussey, Bianca K Frogner, and Hugh R Waters.

2005. “Health spending in the United States and the rest of the industrialized world.”

Health Affairs, 24(4): 903–914.

Buhagiar, Mark A, Justine M Naylor, Ian A Harris, Wei Xuan, Friedbert Kohler,

Rachael Wright, and Renee Fortunato. 2017. “Effect of inpatient rehabilitation vs a

monitored home-based program on mobility in patients with total knee arthroplasty: the

HIHO randomized clinical trial.” Jama, 317(10): 1037–1046.

Cabral, Marika, Michael Geruso, and Neale Mahoney. 2018. “Do Larger Health

Insurance Subsidies Benefit Patients or Producers? Evidence from Medicare Advantage.”

American Economic Review, 108(8): 2048–87.

Cicala, Steve. 2017. “Imperfect Markets versus Imperfect Regulation in US Electricity

Generation.” National Bureau of Economic Research. Working Paper.

Cooper, Zack, Stuart V Craig, Martin Gaynor, and John Van Reenen. 2018. “The

Price ain’t Right? Hospital Prices and Health Spending on the Privately Insured.” The

Quarterly Journal of Economics, 134(1): 51–107.

Cramton, Peter. 2011. “Medicare Auction Failure: Early Evidence from the Round 1

Rebid.” Working Paper.

Cramton, Peter. 2012. “The Hidden Costs of a Flawed Medicare Auction.” Working Paper.

Cramton, Peter, Sean Ellermeyer, and Brett Katzman. 2015. “Designed to Fail: The

Medicare Auction for Durable Medical Equipment.” Economic Inquiry, 53(1): 469–485.

Curto, Vilsa, Liran Einav, Jonathan Levin, and Jay Bhattacharya. 2018. “Can

Health Insurance Competition Work? Evidence from Medicare Advantage.” Working Pa-

per.

26

Decarolis, Francesco. 2014. “Awarding Price, Contract Performance, and Bids Screening:

Evidence from Procurement Auctions.” American Economic Journal: Applied Economics,

6(1): 108–32.

Doyle Jr, Joseph J, John A Graves, and Jonathan Gruber. 2017. “Uncovering

waste in US healthcare: Evidence from ambulance referral patterns.” Journal of health

economics, 54: 25–39.

Duggan, Mark, Amanda Starc, and Boris Vabson. 2016. “Who Benefits When the

Government Pays More? Pass-through in the Medicare Advantage Program.” Journal of

Public Economics, 141: 50–67.

Emanuel, Ezekiel, Neera Tanden, Stuart Altman, Scott Armstrong, Don-

ald Berwick, Franacois de Brantes, Maura Calsyn, Michael Chernew, John

Colmers, David Cutler, Tom Daschle, Paul Egerman, Bob Kocher, Arnold Mil-

stein, Emily Oshima Lee, John D. Podesta, Uwe Reinhardt, Meredith Rosen-

thal, Joshua Sharfstein, Stephen Shortell, Andrew Stern, Peter R. Orszag, and

Topher Spiro. 2012. “A Systemic Approach to Containing Health Care Spending.” New

England Journal of Medicine, 367(10): 949–954.

IOM. 2013. Variation in Health Care Spending: Target Decision Making, not Geography.

Institute of Medicine. National Academies Press.

Kesselheim, Aaron S, Jerry Avorn, and Ameet Sarpatwari. 2016. “The High Cost

of Prescription Drugs in the United States: Origins and Prospects for Reform.” Journal

of the American Medical Association, 316(8): 858–871.

Kolstad, Jonathan T, and Michael E Chernew. 2009. “Quality and Consumer Decision

Making in the Market for Health Insurance and Health Care Services.” Medical Care

Research and Review, 66(1 suppl): 28S–52S.

27

Landon, Bruce E, Sharon-Lise T Normand, David Blumenthal, and Jennifer Da-

ley. 2003. “Physician Clinical Performance Assessment: Prospects and Barriers.” Journal

of the American Medical Association, 290(9): 1183–1189.

Layton, Timothy J, Alice Ndikumana, and Mark Shepard. 2018. “Health Plan

Payment in Medicaid Managed Care: A Hybrid Model of Regulated Competition.” In

Risk Adjustment, Risk Sharing and Premium Regulation in Health Insurance Markets.

523–561. Elsevier.

Mahoney, Neale, and E Glen Weyl. 2017. “Imperfect competition in selection markets.”

Review of Economics and Statistics, 99(4): 637–651.

Martin, Kristi, and Jeremy Sharp. 2018. “Old Lessons for the New Medicare Part B

Drug Payment Model.” The Commonwealth Fund.

MedPAC. 2018. “Report to the Congress: Medicare and the Health Care Delivery System.”

Medicare Payment Advisory Commission.

Merlob, Brian, Charles R Plott, and Yuanjun Zhang. 2012. “The CMS Auction:

Experimental Studies of a Median-bid Procurement Auction with Nonbinding Bids.” The

Quarterly Journal of Economics, 127(2): 793–827.

Newman, David, Eric Barrette, and Katharine McGraves-Lloyd. 2017. “Medicare

Competitive Bidding Program Realized Price Savings for Durable Medical Equipment

Purchases.” Health Affairs, 36(8): 1367–1375.

Nyweide, David J, William B Weeks, Daniel J Gottlieb, Lawrence P Casalino,

and Elliott S Fisher. 2009. “Relationship of Primary Care Physicians’ Patient Caseload

with Measurement of Quality and Cost Performance.” Journal of the American Medical

Association, 302(22): 2444–2450.

28

Papanicolas, Irene, Liana R Woskie, and Ashish K Jha. 2018. “Health Care Spending

in the United States and Other High-income Countries.” Journal of the American Medical

Association, 319(10): 1024–1039.

Reschovsky, James D, Arkadipta Ghosh, Kate A Stewart, and Deborah J Chollet.

2012. “Durable Medical Equipment and Home Health among the Largest Contributors to

Area Variations in Use of Medicare Services.” Health Affairs, 31(5): 956–964.

Sinaiko, Anna D., and Meredith B. Rosenthal. 2011. “Increased Price Transparency

in Health Care a Challenges and Potential Effects.” New England Journal of Medicine,

364(10): 891–894.

Song, Zirui, David M Cutler, and Michael E Chernew. 2012a. “Potential Con-

sequences of Reforming Medicare into a Competitive Bidding System.” Journal of the

American Medical Association, 308(5): 459–460.

Song, Zirui, Mary Beth Landrum, and Michael E Chernew. 2012b. “Competitive

Bidding in Medicare: Who Benefits from Competition?” The American journal of man-

aged care, 18(9): 546.

Song, Zirui, Mary Beth Landrum, and Michael E Chernew. 2013. “Competitive

Bidding in Medicare Advantage: Effect of Benchmark Changes on Plan Bids.” Journal of

health economics, 32(6): 1301–1312.

Verma, Seema. 2018. “You Have the Right to Know the Price.” CMS.gov Blog.

Walshe, Kieran. 2000. “Adverse Events in Health Care: Issues in Measurement.” British

Medical Journal Quality & Safety, 9(1): 47–52.

29

Figure 1. Map of MSAs Assigned to Competitive Bidding

Round 1 Competitive Bidding MSAsRound 2 Competitive Bidding MSAsNot Assigned to Competitive BiddingNot an MSA

Notes: Figure shows metropolitan statistical areas (MSAs) in the continental United States that wereassigned to competitive bidding in either January 2011 (round 1) or July 2013 (round 2), or not assigned tocompetitive bidding by the end of the study period.

30

Figure 2. Price and Quantity Under CMS Pricing Rules

(a) CMS sets price at the median of winning bids

P

Q0 q q∗

Highest bid

Lowest bid

p∗ = Highest winning bid

p = Median winning bid

S

D

(b) Moving from administrative pricing to competitive bidding may increaseor decrease quantity

P

Q0 →←

Administrative pricingyCompetitive bidding price

S

D

Notes: Panel (a) illustrates how CMS determines the price for items subject to competitive bidding. Foreach item in each competitive bidding MSA, CMS pays the median of the winning bids. Panel (b) illustratesthat given previously inflated prices under administrative pricing, competitive bidding would reduce pricesby design by may increase or decrease quantity.

31

Figure 3. Raw Trends of Price and Utilization of items subject to Competitive Bidding

(a) Average Log Price

Round 1 MSAs

Round 2 MSAs

Non-Bidding MSAs

-.5

-.4

-.3

-.2

-.1

0

.1

Log

Pric

e

2009 2010 2011 2012 2013 2014 2015Calendar Year

(b) Log Share of Beneficiaries Using Included DME

Round 1 MSAs

Round 2 MSAs

Non-Bidding MSAs

-.5

-.4

-.3

-.2

-.1

0

.1

Log

Shar

e of

Ben

efic

iarie

s

2009 2010 2011 2012 2013 2014 2015Calendar Year

Notes: Panel (a) plots the simple average of log prices across items and MSAs separately for MSAsassigned to competitive bidding in January 2011, MSAs assigned to competitive bidding in July 2013, andMSAs paid by administrative fee schedules. Log prices for each MSA group in January-June 2009 arenormalized to zero. Panel (b) plots the analogue of panel (a) for log share of beneficiaries using competitivebidding DME, defined as the number of beneficiaries in each MSA who have a Medicare claim on any DMEthat was eventually included in competitive bidding in each six-month period, divided by the total numberof Medicare beneficiaries residing in that MSA. A simple average is taken across the MSAs.

32

Figure 4. Event Study: Log Price

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure shows estimates of Φr’s from equation (1). The coefficient on the six months prior to theintroduction of competitive bidding (r(j, t) = −0.5) is set to zero. 95% confidence intervals are based onstandard errors clustered at the MSA level.

33

Figure 5. Event Study: Log Share of Patients with Any Competitive Bidding DME

-.2

-.15

-.1

-.05

0

.05

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure plots estimates of Φr’s from equation (1). The outcome is the log of share of patientswith any claim on items that were included in competitive bidding. The coefficient for the six months priorto the introduction of competitive bidding (r(j, t) = −0.5) is set to zero. 95% confidence intervals are basedon standard errors clustered at the MSA level.

34

Figure 6. Event Study: Log Spending Per Beneficiary

-1

-.8

-.6

-.4

-.2

0

.2

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure plots estimates of Φr’s from equation (1). The outcome is log spending per beneficiaryon items included in competitive bidding. The coefficient for the six months prior to the introduction ofcompetitive bidding (r(j, t) = −0.5) is set to zero. 95% confidence intervals are based on standard errorsclustered at the MSA level.

35

Figure 7. Event Study: Log Price, by Product Category

(a) Oxygen

-.6

-.5

-.4

-.3

-.2

-.1

0

.1Co

effic

ient

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(b) CPAP

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(c) Wheelchair

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(d) Walker

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(e) Hospital Bed

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure replicates Figure 4 separately by each product category.

36

Figure 8. Event Study: Log Share of Beneficiaries with Any DME Claim, by ProductCategory

(a) Oxygen

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(b) CPAP

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(c) Wheelchair

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(d) Walker

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(e) Hospital Bed

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figures replicate Figure 5 separately by types of durable medical equipment.

37

Figure 9. Event Study: Log Spending per Beneficiary, by Product Category

(a) Oxygen

-1

-.8

-.6

-.4

-.2

0

.2Co

effic

ient

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(b) CPAP

-1

-.8

-.6

-.4

-.2

0

.2

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(c) Wheelchair

-1

-.8

-.6

-.4

-.2

0

.2

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(d) Walker

-1

-.8

-.6

-.4

-.2

0

.2

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

(e) Hospital Bed

-1

-.8

-.6

-.4

-.2

0

.2

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure replicates Figure 6 separately by each product category.

38

Figure 10. Scatterplot: Change in Log Price and Change in Log Utilization

Correlation: 0.42-1

-.8

-.6

-.4

-.2

0Ch

ange

in L

og P

rice

-.6 -.4 -.2 0 .2Change in Log Utilization

CPAPWheelchairsHospital BedsOxygenWalkers

Notes: Each point in the figure represents a DME item (HCPCS code). The y-axis reports the changesin log price for that item based on estimates of equation (2). The x-axis reports the changes in log of shareof patients in each MSA receiving that particular DME item, based on the same estimating equation. Thecorrelation is an unweighted correlation across all items, regardless of their product category.

39

Table 1. Summary Statistics of Medicare Beneficiaries, 2009

(1) (2) (3)All Medicare Beneficiaries BeneficiariesBeneficiaries Using DME Not Using DME

Panel (a) Patient CharacteristicsAverage Age 71.1 72.8 70.5White 83.4% 83.2% 83.5%Female 54.7% 58.6% 53.3%Medicaid 18.5% 26.3% 15.7%Disabled 18.3% 17.8% 18.4%End-Stage Renal Disease 0.7% 1.0% 0.6%Number of Chronic Conditionsa 2.99 4.99 2.29≥ 3 Chronic Conditionsa 50.6% 80.4% 20.2%≥ 8 Chronic Conditionsa 7.7% 19.0% 3.8%Panel (b) Health Care UtilizationAverage Total Medicare Spendingb $8,284 $18,205 $4,828Percent Beneficiaries with

Inpatient Admissions 17.7% 35.7% 11.4%Institutional Post-Acute Care Usec 4.5% 9.7% 2.4%Home Health Use 8.6% 22.3% 3.5%

Panel (c) DME UtilizationProduct Types Used (S.D.)d 0.5 (1.0) 1.7 (1.1)Unique Items Used (S.D.)e 1.1 (2.6) 4.0 (3.5)Most Common Product Typesd

Glucose Monitor 10.5% 38.0%Oxygen Supplies/Equipment 4.3% 15.6%Nebulizers and Related Drugs 3.8% 13.6%Wheelchairs 3.3% 12.0%Continuous Positive Airway Pressure 3.0% 10.8%Walkers 2.6% 9.4%Diabetic Shoes 2.5% 9.1%Lower Limb Orthoses 1.8% 6.5%Lenses 1.5% 5.6%Hospital Beds/Accessories 1.5% 5.4%Number of Beneficiaries 36,861,647 9,523,409 27,338,238% of All Beneficiaries 25.8 % 74.2%

Notes: Panel (a) reports the characteristics of beneficiaries. Panel (b) reports the share of Medicare beneficiarieswho used health care services in different settings, as well as the most common conditions or services in each setting.Panel (c) reports the number of distinct DME product types used, the number of unique DME items used, as wellas the most common product types by share of beneficiaries. In all panels, column (1) reports the mean for allMedicare beneficiaries enrolled in Traditional Medicare; columns (2) and (3) report the means for beneficiaries whodid or did purchase durable medical equipment in 2009, respectively. All outcomes are based on the 100% Medicareenrollment and claims files in 2009.a Based on the Chronic Conditions Segment of the 100% 2009 Medicare Beneficiary Summary File. End-of-yearchronic condition indicators are used.b Patient cost-sharing and non-Medicare payments excluded.c Includes skilled nursing facilities, inpatient rehabilitation facilities, and long-term care hospitals.d Defined based on the Durable Medical Equipment Coding System Product Classification and product categoriesused in the Durable Medical Equipment, Porsthetics, Orthotics, and Supplies Competitive Bidding program. Theseare collections of related items.e Defined as unique Healthcare Common Procedure Coding System (HCPCS) codes, which are used for reimburse-ment.

40

Table 2. MSA Summary Statistics, 2009

Competitive Bidding Competitive Bidding Non-CompetitiveMSAs (Round 1) MSAs (Round 2) Bidding MSAs

(1) Population† 3,129,132 1,850,855 210,469(,1727,962) (2,663,610) (173,751)

(2) Percent Female† 50.9 50.9 50.7(0.8) (0.8) (1.2)

(3) Percent White† 74.8 74.5 82.0(8.2) (11.8) (11.4)

(4) Percent Age 65 and Above† 12.5 12.7 13.3(3.0) (3.3) (3.3)

(5) Percent High School Graudate†∗ 85.7 85.6 85.2(4.0) (5.2) (6.6)

(6) Percent Home Ownership† 67.1 66.8 67.5(2.4) (5.1) (6.3)

(7) Percent on Medicare 15.2 16.0 18.1(3.1) (4.0) (4.6)

(8) Percent Medicare Dual∗∗ 13.2 13.8 12.9(4.0) (5.2) (5.7)

(9) Number of Chronic Conditions 2.1 2.2 2.4(0.5) (0.5) (0.6)

(10) Total Medicare Spending 6468.8 6345.9 6250.0(1450.6) (1467.9) (1599.7)

(11) Acute Spending 2039.7 2150.5 2147.3(336.6) (561.9) (613.9)

(12) Hospital Outpatient Spend 844.1 843.9 945.2(219.8) (194.9) (279.3)

(13) Skilled Nursing Spending 527.5 504.6 481.3( 129.2) (162.5) (165.0)

(14) Home Health Spending 526.9 393.4 322.0(461.6) (359.0) (274.6)

(15) Hospice Spending 332.6 266.7 250.1(51.5) (100.3) (108.1 )

(16) DME Spending 155.7 153.6 173.8(43.9) (42.6) (52.3)

(17) Percent Patients with DME 18.8 19.5 22.5(5.1) (4.6) (5.6)

Number of MSAs 9 91 271

Notes: Table reports summary statistics from 2009, the first year of the sample period, in MSAs thatwere assigned to competitive bidding in January 2011 (column (1)), in July 2013 (column (2)) and MSAsthat were not assigned to competitive bidding during the sample period (column (3)). Unless otherwisenoted, all outcomes are constructed from the 2009 Medicare master beneficiary summary file. All spendingmeasures are Medicare spending, and does not include patient cost-sharing or third-party payment.† Outcomes constructed from the 2009 American Community Survey, 3-Year estimates.∗ High school graduation rate is computed among individuals aged 25 and above.∗∗ Medicare patients who are also eligible for Medicaid.

41

Table 3. DME Prices Across MSAs and Items, January to June 2009

Mean Across SD Across SD Across Lowest Priced Highest Priced Number ofMSAs and Items MSAs Items Item(s) Item(s) Items

(1) (2) (3) (4) (5) (6)

(1) All Competitive $157 $4 $253 $0.6 $2,139 301Bidding DME [Wheelchair bearings] [Power operated vehicle,

451-600 lbs capacity]

(2) Oxygen $98 $0 $ 60 $28.8 $176 12[Portable gaseous or liquid [Stationary compressed gaseous or

oxygen system, rental] liquid oxygen system, rental]

[oxygen concentrator, rental]

(3) CPAP $109 $4 $148 $1.8 $545 26[Replacement exhalation port] [RAD with backup invasive

inteface, rental]

(4) Wheelchairs $192 $5 $302 $0.6 $2,139 185[Wheelchair bearings] [Power operated vehicle,

451-600 lbs capacity]

(5) Walkers $77 $3 $110 $1.7 $522 42[Brake for wheeled walker] [Walker with variable

wheel resistance]

(6) Hospital Beds $126 $5 $156 $3.6 $699 33[Bed cradle] [Hospital bed, extra

heavy duty extra wide]

Notes: Table reports the distribution of prices across MSAs and items prior to competitive bidding, in the first six months of the study period.Each row is a category of durable medical equipment that were subject to competitive bidding. For each category, column (1) reports the mean priceacross all MSAs and items in that category; column (2) reports the standard deviation across MSAs; column (3) reports the standard deviation acrossitems; column (4) reports the lowest price and the lowest priced item(s) in that category; column (5) reports the highest price and the highest priceditem(s) in that category; column (6) reports the number of items in the category.

42

Table 4. Impact of Competitive Bidding on DME Price and Utilization

Change with Competitive Bidding

Estimate % Change(1) (2)

(1) Log Price -0.445(0.016) -35.9%[<0.001]

(2) Log Price -0.583(Weighted by 2009 Utilization) (0.009) -44.2%

[<0.001]

(3) Log Share of Beneficiaries Using DME -0.111(0.010) -10.5%

[< 0.001]

(4) Log Share of Beneficiaries Using DME -0.134(Weighted by Number of Beneficiaries in MSA) (0.015) -12.6%

[<0.001]

(5) Log Standardized Utilization Per Beneficiary -0.074(0.013) -7.2%[<0.001]

(6) Log Standardized Utilization per Beneficiary -0.124(Weighted by Number of Beneficiaries in MSA) (0.021) -11.6%

[<0.001]

(7) Log Spending per Beneficiary -0.619(0.011) -46.1%[<0.001]

Notes: Table reports results from estimating equation (1). Column (1) reports the coefficient estimatesof Φ; robust standard errors clustered at the MSA and the p-value are reported in the parentheses and thesquare brackets, respectively. Column (2) reports the coefficient estimate in exponentiated form to representa percentage change. The sample is all items that were subject to competitive bidding between 2009 and2015. The outcome in row (1) is the simple average of log prices across items; the outcome in row (2) is theaverage log price across items, weighted by the number of beneficiaries with a claim for that item in the firstsix months of 2019.

43

Table 5. Heterogeneity in Impact Across Product Categories

Change with Competitive BiddingLog Price Log Share of Beneficiaries Log Spending per Beneficiary

Estimate % Change Estimate % Change Estimate % Change(1) (2) (3) (4) (5) (6)

(1) Oxygen Equipment -0.408 -0.052 -0.621(0.006) -33.5% (0.011) -5.0% 0.013 -46.3%

[<0.001] [<0.001] [<0.001]

(2) CPAP -0.527 -0.091 -0.594(0.007) -40.9% (0.009) -8.7% 0.013 -44.8%

[<0.001] [<0.001] [<0.001]

(3) Wheelchairs -0.290 -0.134 -0.462(0.022) -25.2% (0.024) -12.5% 0.035 -37.0%

[<0.001] [<0.001] [<0.001]

(4) Walkers -0.514 -0.262 -0.961(0.006) -40.2% (0.018) -23.0% 0.024 -61.8%

[<0.001] [<0.001] [<0.001]

(5) Hospital Beds -0.333 -0.162 -0.637(0.008) -28.3% (0.031) -15.0% 0.034 -47.1%

[<0.001] [<0.001] [<0.001]

(6) Non-Competitive Bidding DME 0.000002 -0.017 -0.011(0.00006) 0.0002% (0.007) -1.7% (0.010) -1.1%

[0.978] [0.022] [0.267]

Notes: Table replicates the main price, quantity, and spending results separately for each productcategory. Separately for each product category, columns (1) and (2) replicate row (1) of Table 4; columns(3) and (4) replicate row (2) of Table 4; columns (5) and (6) replicate row (6) of Table 4. Row (6) Non-Competitive Bidding DME includes all DME items that were never subject to competitive bidding duringthe sample period.

44

Table 6. Impact on Share of Beneficiaries Using DME, by Prior Use

Change with Competitive BiddingSample: Prior Use Sample: No Prior Use

Estimate % Change Estimate % Change(1) (2) (3) (4)

(1) Oxygen Equipment-0.053 -0.061(0.013) -5.2% (0.016) -5.9%

[<0.001] [<0.001]

(2) CPAP-0.101 -0.090(0.011) -9.6% (0.016) -8.6%

[<0.001] [<0.001]

(3) Wheelchairs-0.120 -0.115(0.030) -11.3% (0.037) -10.8%

[<0.001] [0.002]

(4) Walkers-0.410 -0.276(0.040) -33.6% (0.029) -24.1%

[<0.001] [<0.001]

(5) Hospital Beds-0.137 -0.107(0.055) -12.8% (0.043) -10.2%[0.013] [0.013]

Notes: Table replicates row (3) of Table 4, but separately for patients who had the same category ofDME in the three year period between 2009 and 2011 (columns (1) and (2)), and those who did not have thesame category of DME in those three years (columns (3) and (4)). The sample is all beneficiaries residing inthe 91 MSAs that were assigned to competitive bidding in 2013, and all beneficiaries residing in MSAs thatnever entered competitive bidding. Beneficiaries residing in the 9 MSAs that entered competitive biddingin 2011 were excluded as the sample is not long enough to measure their prior use. The average share ofbeneficiaries with and without prior use in across MSAs are 4% and 96%, respectively, for oxygen equipment,4% and 96% for CPAP, 4% and 96% for wheelchairs, 6% and 96% for walkers, 2% and 98% for hospital beds.∗ Percent of beneficiaries in sample MSAs who have or do not have prior use.

45

Table 7. Impact of Competitive Bidding on Characteristics of Patients Using DME

Change with Competitive Bidding

Outcome: Patient Characteristics Pre-Period Mean Estimate(1) (2)

0.005Number of Chronic Conditions 5.7 .015

[0.72]

-0.58Percent Over 80 Years Old 31.8 (0.2)

[0.004]

-0.81Percent Female 55.5 (0.11)

[<0.001]

-1.0Percent NonWhite 17.4 (0.23)

[<0.001]

-1.5Percent Medicaid 26.9 (0.39)

[<0.001]

Notes: Table reports results from estimating equation (2), except that outcomes are patient charac-teristics shown in each row. Column (1) reports mean in R1 and R2 MSAs prior to the introduction ofcompetitive bidding. Columns (2) reports the coefficient estimates of Φ; robust standard errors clustered atthe MSA and the p-value are reported in the parentheses and the square brackets, respectively. The sampleis all MSA-half year combinations between 2009 and 2015 (N = 5,460 MSA-half years).

46

A DME Competitive Bidding Rules

Summary of the bidding process:34

1. To be eligible to participate in the competitive bidding program, suppliers must 1) havean active National Supplier Clearinghouse supplier number, 2) meet certain qualitystandards, and 3) be accredited or the accreditation is pending. Eligible suppliers maythen submit bids in a sixty-day bidding period.

2. Bids are submitted separately for each product category in each MSA. Winning thebid grants the previlidge to sell items in the given product category to beneficiariesresiding in the MSA. Suppliers do not have to be physically located in an MSA toparticipate in competitive bidding.

3. Suppliers are provided with a bidding worksheet, which has information on list ofHCPCS codes, the definition of a bidding unit (e.g. 1 unit = 100 calories of enteralformula), weights used to compute the composite bid, which are based on historicalnational volumes of the product relative to other products in that category, and thebid limit (maximum amount the supplier is allowed to bid), which is the administrativeprice that would have been paid absent competitive bidding. Figure A1 is an excerptfrom a worksheet for the product category ”standard power wheelchairs, scooters, andrelated accessories”.

4. Suppliers must submit a bid for each product (defined as HCPCS code and applicablecode modifiers) in the product category. CMS requires the bids to be “bona-fide”,which is determined based on the information the suppliers provide on cost to purchasethe item, overhead, and profit. (e.g. the supplier may be required to submit invoices,and signed written quotes to prove that they can supply the product at the price theybid.) CMS rejects the entire bid if it determines that the bid for any product is notbona-fide.35

5. Along with each bid, suppliers must also indicate how much volume they can provideat that price, which CMS may consider when deciding how many contracts to offer inorder to satisfy the market (more on this below).

6. CMS requires the bids to be “bona-fide”, and complies with all terms and conditionsoutline by Medicare.36

7. CMS computes a composite bid for each bidding supplier that is equal to a weightedaverage of the supplier’s bids for each item in that product category.

34Based on https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/

DMEPOSCompetitiveBid/downloads/DMEPOSRegSumm.pdf35https://www.govinfo.gov/content/pkg/FR-2014-11-06/pdf/2014-26182.pdf36See for example, https://www.dmecompetitivebid.com/Palmetto/Cbic.Nsf/files/R1RC_Fact_

Sheet_Bona_Fide_Bid.pdf/$File/R1RC_Fact_Sheet_Bona_Fide_Bid.pdf

47

8. CMS ranks all suppliers from lowest composite bid to highest and offers contracts inthat order, until it determines that demand is satisfied. To determine how many con-tracts to award, CMS makes use of the volume information submitted by the suppliersin the bidding process. However, CMS awards at least five contracts per product cat-egory. To do so, it caps the share reported by each supplier at 20% (e.g. if a supplierclaims to be able to satisfy 70% of the market demand, CMS disregards the 70% anduses 20% in its calculations) CMS also requires that small suppliers make up at least30% of the awarded contracts. Small suppliers are defined as those with a annual grossrevenue (Medicare and non-Medicare combined) of $3.5 million or below. If not enoughsmall supplies initially make the cut based on the composite bids, CMS continues downthe list to make offers to additional small suppliers until the 30% number is reached.

9. The price paid to the suppliers is the median of all winning suppliers’ bids for eachitem (HCPCS code and relevant modifiers). Note that the winning suppliers are notpaid their own bids. This price is paid out without adjustment for three years.

10. If a supplier does not enter a contract with CMS, either by failing to win the biddingprocess or rejecting the contract after winning, it cannot sell any of the products inquestion in that MSA. (For example, if a supplier failed to win a contract for “OxygenSupplies and Equipment” in Pittsburg, PA, it may not sell any item in that group toMedicare beneficiaries residing in the Pittsburg, PA MSA for the next three years. )

11. A new round of competitive bidding is conducted every three years.

B Additional Results and Robustness

B.1 Event studies using the full panel of years

Due to the staggered introduction of competitive bidding across different MSAs, the mainresults in the paper use a panel that is balanced in relative months. This section shows themain results using the full panel of months. Figures A3 and A4 show event studies for logprice and log share of beneficiaries using DME, respectively. Since the study period ends in2015, we observe a different set of relative months in different MSAs, depending on when theyentered competitive bidding. This imbalance in MSA-relative months causes compositionalchanges to show up at month 30, which is only defined for the set of MSAs that enteredcompetitive bidding in 2011. To avoid confounding the result with compositional changescaused by the limited sample years, the results in the paper are based on a balanced panelof MSA and relative months. Despite the issue with sample composition in later months,the results from the full panel are almost identical to those from the balanced panel for theperiod of interest (−24 to 24 months).

48

B.2 Event studies and model estimates at the item-MSA-half yearlevel

The main regression specification in the paper estimates a difference-in-differences model atthe MSA-half year level. This section reports results from an alternative specification at theitem-MSA-half year level.

For each competitive bidding item i in MSA j in half-year t, I estimate the followingdifference-in-differences event study specification

ln(yijt) = γj + τt + λi + ΦrCBj × θr(j,t) + εijt (3)

where ln(yijt) is log price or log share of beneficiaries using competitive DME. λi, γj, τtindicate item, MSA, and half-year fixed effects, respectively. CBj is an indicator for MSAssubject to competitive bidding, θr(j,t) are indicators for relative months. The coefficients Φr’squantify the effect of competitive bidding on price.

To summarize the impact over the post-period months, I also estimate a pre-post versionof the same specification,

ln(yijt) = γj + τt + λi + ΦCBj × Postt + εijt (4)

where Postt is an indicator for post-competitive bidding.Note that the interpretation of the result differs between the specification in equation (4)

and the main specification from equation (2) in the paper. The former is an average changeacross individual items, some of which experienced an increase in utilization and others adecrease. The latter captures the overall changes in the utilization of competitive biddingitems in aggregate.

Figure A5, Figure A6, and Table A2 report event study and model estimates using thisalternative specification. The figures and table show results that are very comparable withthe baseline estimates.

49

Figure A1. Bid Preparation Sheet Example

Notes: Excerpt from a bid preparation worksheet provided to suppliers.37

37Downloaded from https://www.dmecompetitivebid.com

50

Figure A2. Raw Price Trends of DME Items Paid Under Administrative Fee Schedule

Round 1 MSAsRound 2 MSAs

Non-Bidding MSAs

-.5

-.4

-.3

-.2

-.1

0

.1

Log

Pric

e

2009 2010 2011 2012 2013 2014 2015Calendar Year

Notes: Figure replicates Figure 3 panel (a) but for DME items that were never subject to competitivebidding throughout the study period. This exercise serves as a sanity check that the price decline was aresult of competitive bidding rather than system-wide price reductions.

51

Figure A3. Event Study: Log Price (Full Panel)

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-54 -48 -42 -36 -30 -24 -18 -12 -6 0 6 12 18 24 30 36 42 48 54Months Since Competitive Bidding

Notes: Figure replicates Figure 4 in the paper, except that it uses the full, unbalanced panel. The spikesat month 30 is caused by the change in MSA compositions (since the study period ends in 2015, relativemonth 30 and later is only defined for the set of MSAs that entered competitive bidding in 2011.

52

Figure A4. Event Study: Log Share of Beneficiaries Using DME (Full Panel)

-.25

-.2

-.15

-.1

-.05

0

.05

Coef

ficie

nt

-54 -48 -42 -36 -30 -24 -18 -12 -6 0 6 12 18 24 30 36 42 48 54Months Since Competitive Bidding

Notes: Figure replicates Figure 5 in the paper, except that it uses the full, unbalanced panel.

53

Figure A5. Event Study: Log Price (Item-MSA-Half Year Level Model)

-.6

-.5

-.4

-.3

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure plots estimates of equation (3).

54

Figure A6. Event Study: Log Share of Beneficiaries (Item-MSA-Half Year Level Model)

-.2

-.1

0

.1

Coef

ficie

nt

-24 -18 -12 -6 0 6 12 18 24Months Since Competitive Bidding

Notes: Figure plots estimates of equation (3).

55

Table A1. Summary Statistics of Medicare DME Suppliers, 2009

(1) (2)Mean S.D.

Panel (a) Supplier Level Summary Statistics

Number of Product Categories Sold 4.5 5.2Number of MSAs Served 4.6 14.8Number of Beneficiaries Served 168 2,945Annual Medicare Reimbursement $114,069 $1,008,617Percent Medicare Reimbursement from Outside of an MSA 22.8% 36.8%

Panel (b) MSA Level Summary Statistics

Number of Suppliers in MSA 402 548Number of Suppliers in MSA by Product Category

Glucose Monitor 193 286Oxygen Supplies/Equipment 59 74Nebulizers and Related Drugs 172 291Wheelchairs 74 116Continuous Positive Airway Pressure 55 61Walkers 67 116Diabetic Shoes 50 90Lower Limb Orthoses 53 107Lenses 29 47Hospital Beds/Accessories 53 84

Notes: Panel (a) reports summary statistics at the supplier level. Panel (b) reports summary statisticsat the MSA level. All measures based on the 2009 Medicare claims data. Suppliers are defined as uniqueNPIs. Panel (b) restricts to suppliers with at least 25 Medicare claims in the MSA.

56

Table A2. Impact of Competitive Bidding on DME Price (Item-MSA Level Model)

Change with Competitive Bidding

Percent Change Estimate S.E. P-value(1) (2) (3) (4)

(1) Log Price -31.3% -0.375 0.010 <0.001

(2) Log Share of Beneficiaries -5.8% -0.06 0.005 <0.001

Notes: Table reports model estimates of log price from equation (4), the item-MSA-half year level model.

57

Table A3. Heterogeneity in Impact Across Product Categories (Robustness)

Change with Competitive Bidding

Percent Change Estimate S.E. P-value(1) (2) (3) (4)

Panel A Outcome: Log Price (Item-MSA Level Model)(1) Oxygen Equipment -33.5% -0.408 0.006 <0.001(2) CPAP -40.9% -0.527 0.007 <0.001(3) Wheelchairs -25.2% -0.290 0.022 <0.001(4) Walkers -40.2% -0.514 0.006 <0.001(5) Hospital Beds -28.3% -0.333 0.008 <0.001Panel B Outcome: Log Standardized Utilization per Beneficiary(1) Oxygen Equipment -4.9% -0.050 0.011 <0.001(2) CPAP -1.2% -0.012 0.012 0.307(3) Wheelchairs -16.0% -0.174 0.048 <0.001(4) Walkers -20.4% -0.228 0.019 <0.001(5) Hospital Beds -15.0% -0.163 0.035 <0.001(6) Non-Competitive Bidding DME -1.6% -0.016 0.009 0.069

Notes: Table replicates Table 5 using alternative specifications or outcomes.

58