Pharmaceutical Innovations and Market Dynamics: Tracking Effects on Price Indexes for Antidepressant Drugs

ERNST R. BERNDT

Massachusetts Institute

of Technology and

National Bureau of

Economic Research

IAIN M. COCKBURN

University of British

Columbia

and

National Bureau of

Economic Research

ZVI GRILICHES

Harvard University

and

National Bureau of

Economic Research

Pharmaceutical

Innovations

and

Market

Dynamics:

Tracking

EJfects

on

Price

Indexes

for

Antidepressant

Drugs

THE CONSTRUCTION

AND PUBLICATION of measures

of

price

inflation are

important tasks carried

out

by governmental

statistical

agencies.

In

the

United

States

the

Department

of

Labor's

Bureau of

Labor Statistics

(BLS) publishes price

indexes

measured at

the

point of final

consumer

demand (the

consumer

price index, CPI)

and

at the initial transaction

We

gratefully

acknowledge

the

cooperation

of

officials

from the

U.S. Bureau

of

Labor

Statistics,

in

particular

Commissioner

Katharine

Abraham,

as

well

as Assistant

Commissioner John

M.

Galvin,

Irwin

Gerduk,

and

Douglas

Kanoza in the

Office

of

Industrial Prices

and

Dennis

Fixler,

Division of Price

and Index

Number Research.

We

have also benefited

from

the

timely

and able research assistance of Gillian Currie and

Mark Moore.

We

thank

Martin

Baily,

Stan

Finkelstein,

Richard

Frank,

Theodore Keeler,

Peter

Reiss,

Jack

Triplett,

and Cliff

Winston for

comments.

Research

support

from

the

National Science Foundation,

the

U.S.

Bureau

of

Economic

Analysis,

the

Alfred

P.

Sloan

Foundation,

and

Eli

Lilly

Inc.

is

gratefully

acknowledged,

as

is

the considerable

data

support

from

Stephen Chappell

and Robert Plefka at IMS

International

and from

Rhea

Mihalison, Phyllis

Rausch,

Ditas

Riad,

and Paul

Snyderman

at

Merck & Co.

The

opinions

and

conclusions

expressed

in

this

paper

are those of the authors and do

not

necessarily

reflect views

or

positions

of

any

of the

organizations

with

which the authors

are affiliated or

those of

any

of

the

research

sponsors.

133

134

Brookings Papers: Microeconomics 1996

point,

that

is, prices

received

by producers

from

whomever makes the

first purchase (the producer price index,

PPI). These price measurement

tasks are difficult ones, particularly because

new goods embody scien-

tific discoveries and technological

progress; inherent difficulties exist

in measuring the output

of

services that themselves

combine goods and

time, and dynamic structural

and

compositional changes occur in the

underlying markets

for

production,

distribution, and sale.

The

marketplace

for health

care contains all these features

and

pre-

sents

particularly

difficult

challenges

for

price

measurement.

Health

care expenditures represent

a

significant

portion

of

gross

domestic

prod-

uct

(GDP) and are likely

to become

increasingly important

as

the U.S.

population ages.

The

conceptual

foundations for

a health care-related

CPI are

clouded,

not

only

because

physicians typically

act as

agents

for

consumers, but

also because insurance

plans pay

for

many,

but not

all, health care products

and

services.

Thus,

for

example,

the

CPI

for

prescription pharmaceutical products

currently weights only cash pay-

ment transactions from drugstores and mail-order

outlets;

it

excludes

prescription drugs purchased by managed

care

plans, Medicaid,

or

other

third

parties on behalf

of an

individual.'

Here

we focus

attention

on the

measurement

of

a

health-care-related

PPI, which,

while

arguably simpler

than

a

CPI, nonetheless presents

enormous measurement

difficulties

and obstacles.2 A PPI

measures

changes

in

selling prices

that domestic

producers

receive

for

their out-

put.

It

is

frequently

used

in

deflating

current

dollar

expenditures

to

obtain a

measure

of real

output growth

by industry.

The

reliability

and

accuracy

of

PPIs

are therefore critical

to

understanding

the

substantial

growth

in

health

care

expenditures

during

the last

ten

years.

Growth

rates

in

PPIs

by industry

are also

used

to assess

inflationary pressures

and

pricing

behavior

in

the health care sectors or to

make international

comparisons.

While

the PPI is

an

output price

index for a

specific

industry, say, pharmaceuticals,

it

is also

an

input price

index for whole-

salers who in turn sell

to

retail

drugstore

chains, hospitals,

mail-order

1. For further discussion, see

Cleeton, Goepfrich,

and

Weisbrod

(1992), and

U. S.

General Accounting Office

(1996).

2. For a recent discussion

on

problems

involved in

interpreting

various

measures

of

wholesale prices such as the

average

wholesale

price (AWP, also

known

as "Ain't

What's Paid"), see Bill

Alpert, "Hooked on Drugs: Why Do Insurers Pay such Outra-

geous Prices for Pharmaceuticals?"

Barron's,

June

10, 1996, pp.

15-19.

Ernst

R.

Berndt,

Iain M.

Cockburn, and

Zvi

Griliches

135

firms, and

managed care

organizations.3

Because issues of

pharmaceu-

tical pricing and health care cost containment

are

currently

of

great

importance to

public policy

analysts, government

statisticians, con-

sumers'

groups, and

industry

officials,

it

is

particularly timely

to

audit

closely

the

accuracy

and

reliability

of

one of the BLS health

care-

related PPIs.

That is

our

purpose

in this

paper.

Although

we

focus on

the PPI, many of

the

issues we

address

are also

germane

to

concerns

cited

by

the

Advisory

Commission To

Study

the Consumer

Price

Index

in its

final

report, released

in

December 1996.

The market on which we focus

our

audit is

that for

antidepressant

prescription

pharmaceuticals

sold between

January 1980

and

February

1996. We have

chosen

this market

segment

and

time

period

for

several

reasons,

all

relating

to the

high likelihood

of

there

being substantial

challenges

here

in

tracking price

changes.'

First,

several

very

successful new

products

have been

introduced

in

the

antidepressant drug class, with well-known

brand names such

as

Prozac, Zoloft,

and

Paxil

having

combined annual sales of

more than

$3

billion

in

the mid-

I

990s

.5

Eight

of the

twenty-one

currently marketed

chemical

entities

(molecules)

are new branded

products

launched

since

1988. Thus,

issues concerning

the

incorporation

of

new

goods

into

price

measurement,

as well as

adjustments

for

quality change,

could be

very

important

in

this

market

class.

Second,

not

only

has new

product

entry

been

substantial,

but

within

the

last

ten

years, seven branded

antidepressants

lost

patent

protection,

and

each

has

subsequently

faced

competition

from

lower-priced

generic

entrants.

Those

buyers

who

regard

the branded

and

generic

versions of

a chemical

entity

as more

or less

perfect

substitutes

realize a

substantial

price decline after

generic

entry. Although

the

BLS

has

been

making

changes

in its

CPI

procedures

for several

years,

until

mid-1996

its PPI

methods did not

adequately

link

generic products

to their

patented

an-

tecedents and

instead

generally

treated

generics

as

entirely

new

goods;

3. In

the United

States, the vast

majority of

pharmaceutical

manufacturer

sales are

to

wholesalers, not to

hospitals,

drugstore chains,

or

managed

care

organizations.

4. For

the economics

of

mental

health,

see

Frank

and

Manning

(1992), and

Jonsson

and

Rosenbaum

(1993).

Keith

and Berndt

(1994)

provide

an overview of

price

measurement

issues in the

pharmaceutical

industry.

5.

Ellen Joan

Pollock, "Side

Effects: Managed

Care's Focus on

Psychiatric

Drugs

Alarms

Many Doctors,"

Wall Street

Journal, December

1, 1995,

p.

Al.

136 Brookings Papers. Microeconomics 1996

thus these older PPI methods failed to record price declines realized by

some purchasers of generic drugs.

Recently the BLS

announced that the

May 1996 pharmaceutical PPIs

would incorporate linking procedures for generic drugs that treat ge-

nerics and their branded antecedents as perfect

substitutes.

The overall

implications of this significant change are not yet clear. Our analysis

of

1980-96 data

in

the antidepressant prescription drug marketplace

provides important information

on

what BLS-measured price growth

for

antidepressants

would have been

had

these

changes

been

introduced

earlier. We also assess the

sensitivity

of

measured

aggregate price

growth to alternative linking and weighting assumptions

that the

BLS

could

have employed.

Because

we

report findings

for an entire

thera-

peutic class, namely, antidepressants,

this

research extends that of Gril-

iches and Cockburn, who provided illustrative empirical evidence con-

cerning

two

systemic

anti-infective

drugs.6

A third

reason for focusing

on

antidepressant drugs

is that

they are

but one

component

in the treatment of

depression, along

with

psycho-

therapy and medical management.

To some

extent, psychotherapy

and

antidepressant drugs

are substitutes for each

other; indeed, controversy

surrounds the extent

to which

managed

care

organizations

are substi-

tuting prescription drugs

for talk

therapy.7

The research

findings

re-

ported here compose

one element of a

larger

research effort

in

which

we

are

creating

a

price

index for the

treatment of

depression

that

incor-

porates

both

drug

and talk

therapy components.

In

this

paper

we

begin

with a

background

discussion on the

nature

of the

medical

condition called

depression

and

provide

a historical

overview on the

evolving

medical

understanding

of

psychotherapeutic

drugs

used for the treatment

of

depression.

We

then outline data sources

and describe the

changing marketplace

for

antidepressant drugs

from

1980 to 1996, particularly

new

product

introductions and

postpatent

expiration entry by generic

firms. We review BLS

procedures

for track-

ing producer prices

in

general

and

antidepressant drugs

in

particular.

We

from

economic

theory

and then

present

results

6. Griliches and Cockburn (1994).

7. See, for example, Carol Hymowitz

and Ellen Joan

Pollock, "Cost-Cutting

Firms

Monitor Couch Time as Therapists

Fret," Wall Street Journal, July 13, 1995, p.

Al;

and Pollock, "Managed Care's

Focus

on

Psychiatric Drugs

Alarms

Many Doctors,"

p.

Al. For empirical evidence, see Berndt,

Frank, and McGuire (forthcoming).

Ernst R. Berndt,

Iain

M.

Cockburn,

and

Zvi Griliches

137

on alternative

procedures

for measuring

price inflation,

including those

involving hedonic

price adjustment.

Finally

we discuss

implications

of

our results

and offer suggestions

for

further

research.

Depression:

Diagnosis

and Prevalence

Whether

depressive disorders

are discrete

and distinguishable

from

'subclinical" depressive

symptoms

is

a question

clinicians

and

re-

searchers

have long debated;

it still has no definitive answer.

Almost

everyone

at some

time or another

has

experienced

melancholy

or

been

depressed

as

a mood,

affect,

or emotion.

To be human is to know about

a

variety

of emotions, including

sadness,

disappointment,

and

despond-

ency.

Many such

affective

occurrences are

within the normal

range

of

human experience.

It is only

with

greater

degrees

of

severity

or

longer

durations

that

such affective

states come

to be viewed

clinically

as

symptomatic

of depression.

The American

Psychiatric

Association has

issued

and updated

clin-

ical guidelines

for diagnosing

depression.8

The current

guidelines,

known as

DSM-IV,

list nine symptoms

of

depression:

(1) a depressed

mood; (2)

diminished

interest

or

pleasure

in most

activities; (3) signif-

icant unintentional

weight

loss or weight

gain,

or a decrease

or

increase

in

appetite;

(4) insomnia

or hypersomnia

nearly every

day; (5) psycho-

motor agitation

or

retardation

nearly

every day; (6)

fatigue

or loss of

energy

nearly every day;

(7) feelings

of worthlessness

or

excessive

or

inappropriate

guilt; (8)

diminished

ability

to think or

concentrate,

or

indecisiveness;

and (9)

recurrent thoughts

of death or suicide.

To

be

diagnosed

as having

a

major depressive

episode,

a

person

must show

at

least five of

these

symptoms

(including

either

a

depressed

mood or

diminished interest

in

most

activities)

for

two or more

weeks.9 These

symptoms

must

also

represent

a

change

from the individual's

A

chronic

but milder form

of

depression

is known

as

dysthymia

and

is diagnosed

when the patient

has a

depressed

mood that

persists

for

at

8.

See

American

Psychiatric

Association

(1968, 1980,

1987, 1993).

9.

It must also

be the

case

that

an organic

factor cannot

be established

as

initiating

and maintaining

the

disturbance

or

that the disturbance

is not a normal reaction

to

the

death of

a loved one.

138

Brookings

Papers:

Microeconomnics 1996

least two years

and has at least

two other

symptoms.

'"

Both

forms of

depression are

serious. Even

moderate

levels of

depression

significantly

impair

functioning

in work

and

school settings and in

social

situations.

"

Survey

evidence suggests that

in a

given

year, 9 percent

of

the

employed labor force

experiences

a

depressive

episode

and

that 80

percent

of

these workers are below the

age

of

45.1

Depression

is

widely

believed to be

an

underdiagnosed

condition;

patients

suffering

from

depression

often

present

themselves to clinicians

as

having

other

med-

ical symptoms

such as lower

back pain,

gastrointestinal

disorders, and

headaches.

1'

Depression is a

treatable condition;

modern

treatment suc-

cess rates

approach

80 to

90

percent.

'

Episodes

of illness

come and

go,

last from

several weeks to several

months,

and are

followed

by

periods

of

relatively

normal mood and behavior.

Untreated,

the

average

depressive episode lasts about four to

six

months.

Between 50 and 85

percent

of

patients

who seek

treatment

for

depression

will have

at

least

one

subsequent

episode

of

depression

in

their

lifetimes, usually

within

two or

three

years.'5

The

lifetime

average

for

depressive episodes is

five

to

seven,

but

as many as

forty episodes

have been

reported."6

Although

the

reasons are

still

not

fully

understood,

women

are about

twice as

likely

to

suffer

from

depression

as

are men.

17

Alternative

Drug

Treatments for

Depression

Before

discussing

alternative

drug

treatments for

depression,

we

briefly

review

several

medical terms.

A

synapse

is

the

point

of

contact

between

adjacent

neurons,

where

nerve

impulses

are transmitted from

one

to the

other. Neurotransmitters are

the

chemical

"messengers"

in

10.

A

tenth

symptom

associated

with dysthymia

is

feelings of hopelessness.

11.

See, for example,

the

studies and clinical

trial

findings referenced by Nolen-

Hoeksema (1990, p. 5).

12. For further

discussion

and

references, see

Greenberg, Stiglin, and others (1993)

and

Greenberg, Kessler,

and others

(1996, p. 328).

13. See

Eisenberg (1992)

for

discussion and references

documenting

the

somatiza-

tion

phenomenon, and Katon

and

others (1992)

for a

discussion of the underdiagnosis

of

depression.

14.

Regier and others (1988).

15.

American Psychiatric

Association

(1993, p. 11).

16.

Papolos

and

Papolos (1992, p. 7).

17. For an

extended discussion,

see

Nolen-Hoeksema

(1990).

Ernst R.

Berndt, Iain M.

Cockburn, and

Zvi

Griliches

139

the brain

that transmit signals

across synapses,

setting

in

motion com-

plex

neural interactions that

shape behaviors,

feelings, and thoughts.

Although

there are many different

neurotransmitters,

the

vast majority

of them

monoamines, three

of

particular importance

are

norepineph-

rine,

serotonin, and dopamine.

Today

it

is

known

that low levels of

these

monoamines are associated

with

depression.

Moreover, after per-

forming

their messenger

activities,

these

monoamines are

eventually

destroyed by monoamine oxidase

(MAO),

a

liver

and brain

enzyme,

through

a bodily absorption

process

called

reuptake.

In

this reuptake

phase,

however, MAO also

destroys

another amine

called

tyramine, a

molecule that affects blood

pressure.

Modern

biological

theories

of

depression

apparently emerged

from

several

chance discoveries.

Clinicians testing

the

antituberculosis

drug

iproniazid in

the early 1950s

observed

that subjects

experienced

relief

from

any depression, and some

even experienced

euphoria. Several

years

later,

this

drug

was shown to

inhibit the

MAO

enzyme."

About

the same

time, clinicians

prescribing reserpine,

a

drug commonly

used

to treat

hypertension,

noted that about

15

percent

of

patients taking

this

medication

became seriously

depressed. Subsequent research demon-

strated

that

reserpine

led

to

the

depletion

of all three of

the

important

monoamine neurotransmitters.

In

1957 isoniazid was

introduced;

it

was a

more

effective

antituber-

culosis

drug

than

iproniazid

and did

not inhibit MAO.

Although

the

manufacturer had

planned

to

cease

production

of

the less

effective

iproniazid,

the

coincident publication

of

psychiatric

research

linking

MAO inhibitors to the treatment of

depression

resulted

in

an

unexpected

surge

in

demand for

it;

in

1957

alone,

unmet

needs were

so

large

that

physicians

prescribed iproniazid

for more than

400,000 depressed pa-

tients.19

Because the MAO

enzyme

also inhibited

tyramine, however,

it was

soon discovered that

iproniazid, by

inhibiting MAO,

could

in-

directly

increase the amount

of

tyramine present

in

the

body,

sometimes

with lethal

consequences.

Excess

tyramine

can cause a sudden

increase

in

blood

pressure

so severe

it

on

occasion

hemorrhages

blood

vessels

in

the

brain and causes death.

The

potential

frequency

with which

this

fatal

response

could

occur

for

patients taking

MAO inhibitors was

quite

18. Baldessarini

(1990, pp.

414-18); Hyman,

Arana,

and

Rosenbaum

(1995, p. 82).

19.

Turkington-Kaplan (1994, p.

49).

140

Brookings Papers: Microeconomics

1996

large, for tyramine is present

in

common foods such as chicken

liver,

aged cheese, broad-bean pods,

soy sauce, and pickled

herring. For this

reason, MAO inhibitors (MAGIs)

were taken

off the

U.S.

market for a

time. Eventually modified MAOIs

were reintroduced,

in large part be-

cause some depressed patients

did not

respond

to

any

other medication.

Today the

MAOIs

are used

most often when other antidepressant

drugs

yield unsatisfactory results

and when electroconvulsive

treatment is

inappropriate

or

refused.20

Because of these complexities,

psychiatric

specialists currently

write about 90 percent

of

MAOI prescriptions;

general practitioners

or

internist physicians

write

only

a small

portion.2'

During the 1950s much

pharmaceutical research began

to

focus on

various mental illnesses.

Although initially analyzed by Swiss

research-

ers for

use as

an

antihistamine,

a tricyclic drug called

imipramine was

tentatively hypothesized to

be

successful

in

treating

schizophrenia. Re-

searchers soon found

that

although imipramine

was

relatively

ineffec-

tive

in

quieting agitated patients,

it

apparently

bestowed

remarkable

benefits upon

certain

depressed

individuals.22

Instead of

stimulating

the

central

nervous system (which

amphetamines do)

or

inhibiting

mono-

amine

oxidase reuptake (a

property

of the

MAGIs),

imipramine

in-

creased the brain's supply

of

norepinephrine

and

serotonin; remarkably,

about

70 percent

of

depressed

patients responded

to this

drug.

The

introduction of imipramine

(brand name Tofranil)

in

1958 was

soon

followed

by

market introductions

of numerous

com-

pounds.

These

compounds

include

amitriptyline (Elavil,

1961),

nor-

triptyline (Aventyl, 1963),

protriptyline (Vivactil, 1967),

trimipramine

(Surmontil, 1969),

and

doxepin (Sinequan, 1969).

The

tricyclic antidepressant

class of

drugs

has been

enormously

suc-

cessful

in

treating depression,

and

experience

with

these

drugs

has

been

extensive.

Today

it is known

that the various members

of

this

class

of

drugs

differ

in the

extent to

which

they

affect

the

three

monoamines.

Although

on

average

there

is

no

statistically significant

difference

in

efficacy

rates

among

the

various tricyclics,

often

patients

who

do

not

respond

to

one tricyclic

do respond

to another. About two-thirds of

people

find relief

with the first

tricylic they

are

prescribed.23

20. Baldessarini (1990,

p. 414);

American.

Psychiatric

Association (1993, p.

2).

21. [MS America (1993).

22. Baldessarini (1990,

p. 405).

23. Turkington and Kaplan

(1994, p. 91).

Ernst R. Berndt, Iain M. Cockburn, and Zvi Griliches

141

Not all patients can tolerate these drugs, however. Because they

affect several neurotransmitters other than serotonin, dopamine, and

norepinephrine, as well as receptors, the tricylic drugs are often asso-

ciated with

side effects. Although

the

side-effect

profiles

of the indi-

vidual tricyclic drugs differ slightly, common side effects include an-

ticholinergic effects (dry mouth, constipation, urinary hesitance,

blurred vision), weight gain,

increased heart

rate,

drowsiness

(which

may be a beneficial side effect initially for those depressed patients

experiencing insomnia),

increased heart

rate,

decreased blood

pressure,

dizziness

when

standing up,

and sexual

dysfunction;

side-effect

profiles

are

given

in

table 1.

The

tricyclics

also differ

in their

half-lives and

in

daily dosing frequency.

Patient

compliance

in

taking

medications

is of

course

negatively affected by

adverse side

effects

and

more

frequent

required daily dosing. A significant unattractive characteristic of the

tricyclic drugs is that overdoses are potentially lethal,

a factor

quite

important for depressed patients with suicidal tendencies.24

The

most

recent major therapeutic development

is

the 1988

launch

of

fluoxetine (brand

name

Prozac),

the

first of the

selective

serotonin

reuptake inhibitors (SSRIs); subsequent

SSRI introductions include ser-

traline (Zoloft, 1992), paroxetine (Paxil, 1993),

and

fluvoxamine

(Lu-

vox, 1994). In contrast to

the

MAOIs

and

tricyclics

that affect several

neurotransmitters, the SSRIs are selective

and

specific

in

that

they

inhibit the reuptake only

of

serotonin. Thus,

side effects associated

with the

reuptake

of

norepinephrine

or

dopamine

are reduced with the

SSRIs, and serotonin

levels are increased.

The

70

percent efficacy

rates

of the

SSRIs are

not

statistically significantly

different

from

the

MAOIs

and

tricyclics,

but adverse

interactions

with

other

drugs

occur less

frequently,

and

the

consequences

of overdoses

are

much less severe.25

With the

SSRIs, anticholinergic effects, drowsiness,

dizziness

when

standing up,

interaction

with the

cardiovascular

system,

and

weight

gain

side effects are

very

rare. Nausea is still a common side effect of

the

SSRIs, as are headaches, nervousness, anxiety,

and

various forms

24. American Psychiatric Association (1993, p. 9);

as the same

article notes,

how-

ever, "the vast majority

of

studies suggest

that all available

antidepressants decrease,

rather than increase, suicidal thoughts and indicate

no

predilection on the part of a

particular agent to

either ameliorate or

aggravate

suicidal tendencies." Also see

Potter,

Rudorfer, and Manji (1991, p. 636).

25. American Psychiatric Association (1993, pp.

7-10).

Also see

Potter, Rudorfer,

and Manji (1991).

Table

1.

Characteristics

of

Drugs

Prescribed

for

the

Treatment

of

Depression

Typical

Ha

lf-

Index

of

side

effects

Chemical

daily

dose

life

Daily

FDA

(0

=

rare,

4

=

common)

entity

(milligrams)

(hours)

frequencv

OCD

AC

DR

IA

OH

CA

GI

WTG

MAOIs

isocarboxazid

20

2

1

0

1

2

0

1

2

phenelzine

45

2

1

0

1

2

0

1

2

tranylcypromine

50

2

3

0

1

2

0

1

2

TCAs

amitriptyline

75

24

1

0

4

0

4

3

0

4

amoxapine

200

10

1

0

2

3

0

1

clomipramine

100

24

1

3

4

1

2

0

3

desipramine

150

18

1

0

1

2

0

1

doxepin

100

17

1

0

3

4

0

2

0

3

imipramine

100

22

1

0

3

1

4

3

1

3

maprotiline

100

43

1

0

2

4

0

1

0

2

nortriptyline

100

26

1

0

1

0

2

0

1

protriptyline

30

76

3.5

0

2

1

2

0

trimipramine

100

12

1

0

1

4

0

2

0

3

SSRIs,

fluoxetine

20

168

1

0

2

0

3

0

paroxetine

30

24

1

0

2

0

3

0

sertraline

50

24

1

0

2

0

3

0

fluvoxamine

100

15

1

3

1

0

3

0

nefazodone

300

18

2

0

3

4

1

0

3

0

Others

bupropion

225

14

3

0

2

0

1

0

trazodone

300

8

3

0

4

0

1

venlafaxine

112.5

5

3

0

2

3

2

1

3

0

Sources:

Depression

Guideline

Panel

(1993.

tables

7,

8,

pp.

56.

59);

for

clomipramine,

fluvoxamine,

nefazodone,

and

venlafaxine,

Physicians'

Desk

Reference

Generics

(1996.

pp.

735-39.

1383-1683,

2246-50,

3071-76).

Notes:

See

text

for

discussion

of

typical

daily

dosages

in

milligrams.

Half-life

is

average

of

elimination

half-lives

in

hours.

Daily

frequency

is

that

recommended

for

maintenance

therapy

alter

titration

has

determined

daily

dosages.

FDA

OCD

=

I

if

FDA

has

approved

obsessive-compulsive

disorder

indication.

For

side

effects,

AC

=

anticholinergic

(dry

mouth.

blurred

vision.

urinary

hesitancy,

constipation);

DR

=

drowsiness,

IA

=

insomnia-agitation;

OH

=

orthostatic

hypotension

(abnormally

low

blood

pressure):

CA

=

cardiac

arrhythmia:

GI

=

gastrointestinal

disease:

and

WTG

=

weight

gain

(more

than

6

kg).

144

Brookings

Papers:

Microeconomics

1996

of sexual dysfunction;

some

patients

encounter insomnia,

while a small

portion experience drowsiness.

In addition to their use

as

antidepressants,

two of the SSRIs-Prozac

and

Luvox, along

with a

tricyclic,

Anafranil-have received Food and

Drug Administration (FDA)

approval

for use in

treating

obsessive-

compulsive disorders (OCD).

Within

the class of

SSRIs,

Prozac has

the

longest

half-life

(see

table 1);

this has

disadvantages

for

those who

experience negative side effects

but

can be beneficial

for

those who

occasionally might forget

to

take

their medication.

Three related

drugs

have

recently

been introduced

into

the antide-

pressant

market: nefazodone

(brand

name

Serzone),

a

serotonin-related

compound

that

may

cause less sexual dysfunction;

venlafaxine

(Ef-

fexor),

a

compound

that inhibits

reuptake

of

norepinephrine

and

sero-

tonin, but not dopamine,

and thus exhibits some of the features

of both

the tricyclics and SSRIs;

and

bupropion (Wellbutrin),

a

compound

whose mechanisms

of action

are

still

not

well understood. More

gen-

erally,

researchers of the central nervous system

still do not

understand

precisely

how

the SSRIs

affect

depressive

moods and

the role of

sero-

tonin

in

this

process. Although

serotonin

levels increase within

several

days

of

taking

SSRI

(and

other

antidepressant) medications,

typically

a

change

in

depressive

moods manifests itself much

later,

after

two,

four,

or

perhaps

even

six

weeks. It

is

possible

that serotonin

causes

slight effects

in

other

neurotransmitter

systems,

which

in

turn

relieve

depression. Apparently

the serotonin neurotransmitter

system

is

very

complex.

Although

much

progress

has been

made in

developing psychothera-

peutic drugs

for

treating depression,

the

causes

and

optimal

treatments

of

depression

remain unresolved.

This has lead the American

Psy-

chiatric Association

to

issue the

following

current medical

practice

guidelines:

No one

medication

can be recommended

as

optimal

for all

patients

be-

cause

of

the substantial

heterogeneity among patients

in their

likelihood

of beneficial

response

to these medications

and

the

nature,

likelihood,

and

severity

of side

effects.

Furthermore, patients vary

in the

degree

to

which

particular side

effects and other

inconveniences of

taking

medi-

cations

(e.g.,

cost and

dietary

restrictions)

affect

their

preferences.26

26. American Psychiatric

Association

(1993, p. 7).

Ernst R. Berndt, Iain M.

Cockburn, and

Zvi

Griliches

145

Finally,

it

is

widely believed that

psychotherapy, drug

therapy, or

their

combination is

an effective

treatment

for

cases of mild to

moderate

depression. Although this consensus is based on extensive

clinical ex-

perience,

and on clinical trial data

for

drugs,

evidence

concerning

the

efficacy

of

psychotherapy based

on

controlled

experiments

is not as

extensive,

in

part

because controlled

experiments

involving

uniform

and consistent

forms

of

psychotherapy

have

proved

difficult to

design

and conduct.27 For

the more

severe forms of

depression, both

drug

treatment

and electroconvulsive treatments

appear

to

be more effica-

cious

than

psychotherapy

alone.28

The Changing

Marketplace

for

Antidepressant Drugs

Our

description

of the

changing

marketplace

for

antidepressant

drugs

is

based on the

following data sources.

Monthly price and

quantity data

for

drugstore purchases

of

antidepressant

drugs

are from

IMS

America,

a

Pennsylvania

firm that

collects

and

sells

data on the

sales and mar-

keting

of

pharmaceutical

products.

The transactions monitored

by

this

data are from wholesalers and manufacturers to

drugstores

(or

their

purchasing agents)

and are based on actual

invoices;

IMS tracks

more

than

99 percent

of

manufacturer

and wholesaler

transactions and thus

provides a near-census universe

of

drugstore

purchases.29

These in-

voices

reflect slightly

imperfectly

the

prices

manufacturers receive.

The

invoice data provide a

dollar

sales amount and

quantity

number for each

type

of

transaction;

they

include

chargebacks

(credits

to

wholesalers for

any

special price agreements

negotiated

among drug stores,

manufac-

turers,

and

wholesalers),

but rebates

(direct

payments

from

manufac-

turers to health

care providers

and

others,

such as health

maintenance

organizations

and

pharmaceutical

benefit

management

firms)

are not

always included,

nor do

the dollar

purchase

amounts on the

invoices

reflect

prompt payment

cash discounts

(usually

2

percent off).30

Further

27. See, however, the seminal

study by Elkin, Parloff, and others (1985) and

Elkin,

Shea, and others (1989).

28. See Depression Guideline Panel

(1993).

29. lMS America (1996b, p.

39-6).

30. Rebates occur in part because

health maintenance organizations and

pharmaceu-

tical benefit management companies

can affect market shares, but often these organi-

zations do not

actually take possession

of

drug products.

146

Brookings Papers. Microeconomics

1996

discussion

of the IMS

price

data is

given

in

Berndt,

Griliches,

and

Rosett,

who report that from

1986 through

1991,

the

period covered in

their

study,

the IMS data and

price

data

provided

them

by

four manu-

facturers had

very

similar

growth

rates.3'

In the

paragraphs that

follow,

we

report

sales

data,

measured

in

both

dollars and

daily dosage

units.32

Frequently

a

drug

is available in

var-

ious

strengths; considerable differences

also

occur

in

the total

daily

dosage

taken

by

individuals. To

develop

a

quantity

measure

providing

some

comparability

across diverse chemical entities and

dosage

strengths, we first take

the

midpoint

of

the

normal recommended

daily

milligram

dosage range

during

the

maintenance

phase,

as

specified

for

each chemical

entity

in

the

1996

Physicians'

Desk

Reference

and then

assess

what

integer

number

of

equal-strength

tablets at

recommended

daily

frequencies

could

feasibly

make

up

the

total

daily dosage

closest

to

this

midpoint.33

In cases of

ambiguity,

we consulted

IMS data on

volume

sales by tablet

strength. The resulting

"typical daily

dosages"

are listed

in

table

1

for each

chemical

entity.

To

express quantities

in

total number of

daily

dosages,

we divide the total

number

of

milligrams

of

active

ingredient

sold over the various

presentations

of

the

drug by

this

typical daily dosage.

The

typical daily

dosage price

is

then

com-

puted

as

sales

in

dollars divided

by

total

typical

daily dosages.

The

Overall Market

for

Antidepressant

Drugs

Growth

in

the overall market for

antidepressant

drugs

since

1980 has

been sustained and substantial.

In

1980

about 452 million

daily

dosage

units

of

antidepressant

drugs

were

sold; by

1995 this number

had

in-

creased to

about

2.44

billion,

a

factor

of more

than

five;

the

implied

average

annual

growth

rate

(AAGR)

is

11.9

percent.

Growth of

dollar

revenues

has been even

stronger,

from

a

$128

million market

in

1980

to

$3

billion in

1995,

for

an AAGR

of

23.5

percent; using

the

GDP

deflator to

convert into constant

1980

dollars,

the

1995

sales are

$1.65

billion, implying

an

AAGR of

18.6

percent.

Growth has

accelerated

31. Berndt,

Griliches, and

Rosett (1993, p. 255).

32. Because

their uses are

often for very

different purposes and

because their

vol-

umes

are

relatively small,

all

liquid

forms,

such as oral

solutions and

injections,

are

excluded.

33. The

midpoint dosage was often an infeasible

number, unless

patients broke up

tablets into

smaller units. Thus we

sought

an

integer

value.

Ernst R. Berndt,

Iain

M. Cockburn, and Zvi Griliches 147

dramatically since 1988, the year

in

which

the first SSRI

was intro-

duced.

From

1980 to

1987,

for

example,

the

AAGR in

daily dosage

quantities

was about 5.3

percent,

but

from

1987

to

1995

this

AAGR

more than tripled to 18.3 percent;

in

real dollars, these AAGRs are 9.5

percent

and

26.9 percent.

Entry and Exit

There has been much

entry

and some

exit

in

the market for

antide-

pressant drugs. Two types of entry occurred,

one

involving introduc-

tions of

patented products

and

products newly approved by

the

FDA,

and the other

involving generic

introductions after

patent protection

expired.

In

some cases branded

products

left the

market,

while both

entry

and exit

occurred

for

generic products.

This

entry

and exit

behav-

ior is

summarized

in

table 2.

Of

the twenty-one antidepressant chemical

entities on the market

in

February 1996,

fifteen were either new branded

products

or

generic

versions introduced

within the

past

ten

years.

All

three

MAOI products

were introduced

in

the

1959-61 time

pe-

riod,

and

although patent protection

has

expired,

the

market for these

products

is

apparently so

small and

unattractive that generic entry has

not been

induced.

Among

the ten

tricyclics

and

chemical

entities,

the two oldest

are

imipramine

and

amitriptyline.

The

branded

pioneers,

Elavil and

Tofranil,

not

only

faced

competition

from

generic entry beginning

in the

1970s,

but from 1975 on

they

also ex-

perienced

branded

competition

from

other

major pharmaceutical

man-

ufacturers

(Endep

for

Elavil,

and

Janimine for

Tofranil).34

The

com-

petition

these

secondary

brands

encountered from the

primary

branded

products and the generics must

have been

considerable,

for

Janimine

exited

in

1985,

and

Endep

in

1988.35

The

TCA

class of

drugs

attracted considerable branded

entry, espe-

cially

in the

1960s,

but in

the

1980s

generic entry

was

predominant,

reflecting

in

part

the

reduced

costs of

generic entry made possible by

34.

Although

the distinction

is not

completely clear,

we

distinguish

branded

products

from those

generics sold primarily by

their chemical

entity name,

often under a

private

label; thus Endep is distinguished from, say, Walgreen imipramine.

35.

It

is

possible

that

these brands exited

only

from the

IMS data

base,

not from

the

market,

in

that their sales

may

have fallen below

a

minimum

reporting

threshold

imposed

by

IMS.

Table

2.

Entry

and

Exit

in

the

Antidepressant

Drug

Market

Orginator

brand

Seconidary

brand

Genieric

distributors

Generic

name

Name

Year

Name

Entry

Exit

Entry

1988

1996

MAOIs

isocarboxazid

Marplan

1959

None

0

phenelzine

Nardil

1959

None

0

tranylcypromine

Parnate

1961

None

0

TCAs

amitriptyline

Elavil

1961

Endep

1975

1988

1977

13

24

amoxapine

Asendin

1980

1989

0

14

clomipramine

Anafranil

1990

None

0

desipramine

Pertofrane

1971

Norpramin

1975

1989

1987

9

20

doxepin

Sinequan

1969

Adapin

1973

1991

1986

12

22

imipramine

Tofranil

1958

Janimine

1975

1985

1975

12

16

maprotiline

Ludiomil

1981

1988

7

11

nortriptyline

Aventyl

1963

Pamelor

1977

1992

0

17

proptriptyline

Vivactil

1967

None

0

trimipramine

Surmontil

1969

1988

5

0

SSRIs,

fluoxetine

Prozac

1988

None

0

fluvoxamine

Luvox

1994

None

0

nefazodone

Serzone

1995

None

0

paroxetine

Paxil

1993

None

0

sertraline

Zoloft

1992

None

0

Others

bupropion

Wellbutrin

1989

None

0

trazodone

Desyrel

1981

1986

22

venlafaxine

Effexor

1994

None

0

Sources:

IMS

America,

Inc.,

and

Food

and

Drua

Administration

(annual).

Ernst R. Berndt,

Iain M. Cockburn, and

Zvi Griliches

149

passage

of

the

1984

Waxman-Hatch Act.36

By

1996

eighteen

or so

distributors were

offering generic products

for

each of the

TCA drugs

facing generic

competition, up

sharply

from about ten in

1988. Not

all

generic entry has been

sustained;

although

Surmontil

faced

generic

entry in 1988, in

1992

the

generic

competition

exited,

and

none has

emerged since then.

The introduction

of Prozac

in

1988

marked

the

entry of an

entire new

class of

antidepressants,

the

highly

successful SSRIs. Other

SSRI

branded drugs were

Zoloft,

introduced

in

1992, Paxil

in

1993,

Luvox

in

1994, and

Serzone

in

1995;

Effexor,

a

introduced

in

1994.

Prices and

Market Shares

Next we

look at

market share

and

price movements,

first

among the

four classes of

antidepressant drugs

listed

in

table 2.

During 1980-88

the

MAOIs

had

only

a

very

minor unit and

revenue

market

share,

between

1.4

percent

and

2.4

percent,

and after 1988 this share

dropped

even

further;

the

1996

share was but

0.3

percent.

In

1980

the

TCAs

accounted for about

98

percent

of

both the

daily

dosage quantities

sold and

total

antidepressant

revenues.

By

1987 the

TCA

unit share

fell

slightly,

to

90

percent,

as trazodone

(from

a differ-

ent class of

drugs)

increased

its unit market share to

about

8

percent;

the

corresponding

TCA

revenue

shares

were 77

percent

and 21

percent.

Among

the

TCAs,

three dominated

in

1980:

amitriptyline

had a 50

percent

unit

share,

doxepin

22

percent,

and

imipramine

18

percent,

for

a combined share of

90

percent. By

1987

this

combined share

fell to

80

percent,

as sales of

products

such as

desipramine,

amoxapine,

and

nortriptyline

(having fewer

and less

severe

side effects-see

table

1)

increased to a combined

14

percent

unit share. The three

largest

TCAs

accounted

for

about

82

percent

of total

TCA

dollar

sales

in

1980,

but

only 49 percent

in

1987,

in

large part

because

all three

products

faced

increased

generic

competition

in

the

1980s.

The launching of Prozac was a

huge

success. Not

only

did

this first

SSRI

take market share

away

from the

TCAs,

but

it

also

expanded

enormously

the size of the overall

antidepressant

drug marketplace.

36. For

discussion of

this legislation and its

consequences, see Grabowski

and Ver-

non

(1992).

150 Brookings Papers: Microeconomics 1996

General practitioners and internists, not just psychiatrists, were now

able to prescribe antidepressants comfortably,

for

concerns about side

effects and adverse interactions

with

Prozac were much less intense

than with the TCAs. Moreover, because the daily dosage for Prozac

was

the same

for

almost everyone, specialist knowledge

and

experience

concerning optimal patient-specific dosages, typically required for

many of the TCA drugs, were

no

longer necessary.

At

the end of

its

first year on the market (1988), the Prozac daily dosage share among

all

antidepressants

was

11

percent,

and

given

its

higher price,

its dollar

market share

was 21

percent; by

1991 these shares

had

increased to

29

percent and 51 percent, respectively.

The SSRI market continued

to

grow rapidly following entry by ad-

ditional

SSRIs,

and

by 1996

the SSRI market share

among

all antide-

pressants was 63 percent

in

daily dosage

units and a

remarkable

84

percent

in

dollars;

unit market shares

for

the

TCAs

fell

from 90 percent

in 1987 to 27 percent in 1996, while revenue shares dropped even more

dramatically, from 77 percent

to 7

percent. Clearly, for many physi-

cians and patients dealing with

the

treatment

of

depression, the SSRIs

were enormously successful

in

fulfilling unmet

needs.

Within the SSRI subclass of drugs, unit

sales of

Prozac continued to

grow, from 340 million daily units

in

1991 to 645 million in 1995. But

the

great success

of

Zoloft

and Paxil in

expanding the overall SSRI

market has implied

a

loss

in Prozac's

market share;

in

1996 SSRI daily

dosage market shares for Prozac, Zoloft,

and Paxil were

41.6 percent,

41.5 percent, and 12.6 percent, respectively,

while

corresponding

dol-

lar market shares

were

48.0

percent,

29.8

percent,

and 17.8

percent.

Moreover,

the unit shares

of

Prozac, Zoloft,

and Paxil

prescriptions

written

by nonpsychiatrists

were

39

percent,

51

percent,

and

49

percent,

respectively, indicating proportionally

more

nonspecialist prescriptions

written

for Zoloft and Paxil than

for

Prozac.37

Prozac and other SSRI

entrants have been

tremendously

successful

despite

their

higher daily dosage prices.

When Prozac was launched

in

1988,

for

example,

its

daily price

was about

$1.18,

almost double the

$0.60 daily price

of

the

branded version

of the

leading selling tricyclic,

amitriptyline,

and more

than

twenty

times

the

$0.05 daily price

for

generic

versions

of

that chemical

entity; doxepin,

the second

best-

37. IMS (1996a).

Ernst R. Berndt,

lain M. Cockburn,

and Zvi

Griliches

151

selling tricyclic,

was also

much cheaper

than Prozac-$0.70

a day in

its branded

version and $0.21

in generic form.

When Zoloft,

the second

SSRI entrant,

was launched

in 1992, its

daily price was set at about

25

percent

lower

than that of

Prozac-$1.26

compared

with

$1.69.

Ser-

zone, the most

recent

SSRI, is priced

in between Prozac and Zoloft.

In

constructing

a price

index,

what happens

following entry

of

ge-

neric competition

is very

important.38 In

table

3

we summarize

price

and market

share developments

at twelve,

twenty-four,

and thirty-six

months following

initial

generic entry

for the seven

chemical entities

experiencing

initial

generic

competition

since 1980. The

top panel

shows

that

although

considerable variability

is

present,

unweighted

average generic

prices

are

about

57

percent,

43

percent,

and

35

percent

of

brand prices

after

one,

two,

and three

years.3

Substantial differences

in

market share

penetration

are

also

present.

Measured in

daily units,

generic

market

shares

vary

from 5

percent

to

68

percent

of brand shares

after

one year

and

average

about

27

percent,

while

they

average

about

44

percent

and

54

percent

after

two

and

three

years,

respectively.

There does

not

appear

to

be

any

dominant

time

trend

to

generic

penetration

rates, although

the market share

of

the most recent

generic

entrant, nortriptyline,

is the largest

after

one, two,

and

three years.

Because generic

prices

are

lower than brand

prices,

dollar shares are

smaller

than

unit

shares;

even

so,

after

just

one

year

the

nortriptyline

dollar share

is

56 percent.

The generic

price

can

fall relative to

the

brand

price

if

the

generic

price decreases,

the

brand

price

increases,

or both.

As

the second

panel

of table 3

shows,

manufacturers

have tended

to

increase

the

price

of

branded

products

following generic

entry, apparently

focusing

on the

price inelastic

market

segment

and

letting generics gain

market share

from the elastic

segment;

after

one,

two,

and

three

years,

the

average

38. For

discussion

of generic

pricing and

responses by

incumbents,

see Caves,

Whinston,

and Hurwitz

(1991); Frank

and Salkever

(1992);

Grabowski

and Vernon

(1992);

Griliches and

Cockburn (1994);

Hurwitz

and Caves

(1988); and

Masson and

Steiner (1985).

39. These trends

in

prices

of generic drugs

for treatment

of a

relatively

chronic

condition such as depression

differ

considerably

from

those

reported by

Griliches and

Cockburn

(1994)

for systemic infectives,

which tend

to be used in the treatment of

more

acute conditions.

For

generic antidepressants

(except nortriptyline),

the initial price

discount

is larger, but

after that the

relative

price

is

flatter

than that

of

generic

systemic

anti-infectives.

Table

3.

Relative

Prices

and

Market

Share

Penetration

of

Generic

Antidepressant

Drugs

Introduced

since

1986

IWelve,

TWenty-four,

and

Thirty-six

Months

after

Introduction

Percentage

Relative

price

generic

Generic

market

share

Generic

market

share

Chemical

Entry

to

brand

in

units

in

dollars

entity

year

12

24

36

12

24

36

12

24

36

doxepin

1986

38

30

25

39

54

67

20

26

34

trazodone

1986

62

42

18

23

44

70

16

25

29

desipramine

1987

61

37

31

29

61

58

20

37

30

maprotiline

1988

61

54

42

10

22

33

6

13

17

trimipramine

1988

60

53

58

5

11

3

6

7

amoxapine

1989

58

51

50

14

37

51

9

23

35

nortriptyline

1992

61

36

22

68

80

85

56

58

56

Pioneer

brand

price

after

generic

entry

(Generic

entry

date

=

1.00)

Nominal

price

per

day

Real

GDP-deflated

price

12

24

36

12

24

36

doxepin

1986

1.11

1.35

1.50

1.08

1.27

1.35

trazodone

1986

1.01

1.22

1.58

0.98

1.14

1.42

desipramine

1987

1.13

1.35

1.60

1.09

1.25

1.42

maprotiline

1988

1.14

1.21

1.28

1.08

1.11

1.12

trimipramine

1988

1.14

1.23

1.43

1.09

1.12

1.25

amoxapine

1989

0.97

1.39

1.45

0.94

1.33

1.29

nortriptyline

1992

1.04

1.06

1.11

1.01

1.05

Source:

IMS

America,

Inc.

Ernst R.

Berndt, lain M.

Cockburn, and

Zvi

Griliches

153

nominal

price increases

for

the

branded

products

are

about

1 1

percent,

26 percent, and

42

percent

(average

real

price

increases are

about 7

percent,

18

percent,

and

27

percent,

respectively).

With

this data as

background, we now

summarize

procedures

the

BLS has

used to

track and

measure

price indexes in

this

rapidly

chang-

ing

antidepressant

drug

marketplace.

BLS

Procedures

and

Samples for

Tracking

the

Antidepressant

Drug

Market

Currently

the PPI

program at the BLS

encompasses

the

construction

of

monthly

aggregate

price

indexes

for almost five

hundred

mining

and

manufacturing

industries,

including

approximately ten

thousand

in-

dexes for

specific product

categories,

based on

reports

from

approxi-

mately

twenty-five thousand

companies that

respond

voluntarily.

For

the

specific

product

categories

for

tranquilizers

and

antidepressants,

separate

price

indexes

for

these

distinct

and

more

disaggregated

subcategories were not

offi-

cially published.

Unfortunately,

the

BLS

has not

kept

files

on

which

particular

psychotherapeutic

drugs

and

presentations

made

up

the

Cycle

A

sample

and what

their

index

weights

were.

About six

years

later,

in

December

1987, the

BLS

drew

up

a

new

sample,

implementing

where

possible

a

sampling procedure

in

which

items

were

chosen

in such

a

way that

the

probability

of

selection was

proportional

to

a

product's value

of

shipments.40

A

separate

antide-

pressant

drug

subcategory

was

created,

and

specific

items

were

chosen

for

that

subcategory

in

what the

BLS

calls

its

"Cycle

B"

sample.

For

six

years

beginning

in

December

1987,

the BLS

computed

and

pub-

lished a

PPI for

antidepressant

drugs

based on

this

Cycle

B

sample.

In

0

.-

C#7

0

~0

ID

0

.-

X _E;

Cd

0

It

0_

C0

;o

'C7

'I

_E _7E

154

Brookings Papers: Microeconomics

1996

December 1993 the BLS

again updated its sample; the items

making up

this new

sample of

antidepressant drugs

are

called "Cycle C"

products.

Under

strict

confidentiality agreements,

BLS officials

have made

available to

us

information

concerning

the set

of

antidepressant

drugs,

and their

item weights, that

make

up

the

Cycle

B

and

Cycle

C

samples.

As best we can

determine, six items were originally

in

Cycle B, and

one

additional item

was linked

in

around

May 1990.

Two of

the seven

items may be

misclassified, because the FDA has not

approved them

for treatment of

depression,

nor does

the American Medical Association

list them as as

antidepressant treatments.4'

All seven

Cycle

B

items

apparently

were branded

products;

when the

Cycle

B item

sample

was

implemented, three of the six brands faced

generic competition.

Ge-

nerics as a group accounted

for

11

percent

of total

antidepressant

rev-

enues and 44

percent

of

total

daily dosage

units sold.

Prozac,

the

pi-

oneer

SSRI,

did

not

enter

the market

until

January 1988,

and thus none

of

the

new

generation

of

SSRIs

was

included

in

the

Cycle

B

sample.

During the six-year Cycle

B

period (1987-93),

an

additional two of

the

seven

branded

drugs

in

the

sample

lost

patent protection

and faced

competition

from

generic

entrants.

Thus

at the

end of the

Cycle

B

era

(December 1993),

while five

of

the

Cycle

B

items faced

generic

com-

petition,

all seven

sample

items

were branded

products.

Details

concerning procedures

used

to construct

the

Cycle

B

sample

are no

longer

available.

BLS officials

have, however,

informally

de-

scribed how the

Cycle

C

sample

was

drawn

and how its

item

weights

were

determined.

In

early

1993

the

BLS

contacted

a

private data source

to

provide

1991

and

1992

annual sales

data

by drug,

separately

for

several market

segments

such

as

drugstores

and

hospitals.

Based on

this and related

FDA

data,

the BLS

chose a

preliminary

set of thera-

peutic

classifications

and,

using

a

sampling procedure designed

to en-

sure that a

manufacturer's

probability

of

being

selected was

propor-

tional

to its

sales,

selected about

120 manufacturers for

sampling,

of

which

approximately

75

percent cooperated voluntarily.

Item

weights

were then

constructed based

on information

these manufacturers

pro-

vided to

the

BLS. The

resulting Cycle

C

sample

of

products

used to

41. American Medical Association

(1991).

Both of these

products are known to be

prescribed "off-label" infrequently by

some physicians

for

treatments occasionally

associated with depression. BLS officials

have

suggested

that these

products may have

been selected

as antidepressants by

the

responding firms,

rather

than

by

the BLS.

Ernst R. Berndt,

lain M.

Cockburn,

and Zvi Griliclies

155

construct

PPIs

for

prescription

pharmaceuticals

numbered

between

500

and 520.42

Within

this Cycle C

sample,

first

used

in December

1993,

the

BLS

retained five

of the seven

Cycle B chemical

entitities (each

with

differ-

ent

weights,

and one switched

from brand to

generic,

with

a

changed

milligram

strength),

including one

drug

not normally considered

an

antidepressant.

Two

Cycle

B

items were

dropped,

and five new items

were added.

Of the ten

Cycle

C items, three

are

generic

and

seven

are

branded. Among the

latter,

three faced competition

from

generic

entry

at the time the Cycle

C item

sample

was drawn.43

Because

Prozac

is

manufactured

in

Puerto

Rico

(along

with

many

other drugs,

in

part because

of

provisions

in

the federal

tax code)

and

because Puerto

Rico is

not

considered part

of

the United States

for

purposes

of PPI calculations,

Prozac could not

be part of the

Cycle C

sample even

though it

is the largest-selling

SSRI.

More generally,

unlike the

CPI, which

includes drugs

manufactured

in

Puerto

Rico for

use in

the

fifty states and

the District

of

Columbia,

the PPI

excludes all

Puerto Rican

production.

Government

statistical

agencies do

not all

deal with Puerto Rican economic

accounts

in

the same

way.

For ex-

ample, the

national income

and

product

accounts

from

the

Bureau of

Economic

Analysis

exclude

Puerto Rican

production

and

that

of

other

dependencies,

but

in the balance

of

payments

accounts,

Puerto

Rico is

treated as domestic.44 The Census

Bureau defines

the

United

States as

the U.S.

customs

territory,

which consists

of the

fifty states,

the

District

of

Columbia,

and Puerto

Rico, plus

U.S.

foreign

trade

zones

and the

U.S.

Virgin

Islands.45 There

appears

to be

some

ambiguity,

however,

in

determining

what constitutes

Puerto

Rican

production

from

the

view-

point

of

the

BLS. One

of the

products

in

the

current

Cycle

C

sample,

for

example,

is

produced

both

on the mainland

(45

percent

of

domestic

consumption)

and in

Puerto

Rico

(55

percent

of

domestic

consumption).

42. In

December 1995 the

BLS supplemented

the original

Cycle C sample

by intro-

ducing fifty-one

additional products,

based

in

part on

data from new drug products

introduced after

1992, as published

in the

FDA's "Orange

Book." None

of these

products is in

the antidepressant

drug class (but

see footnote 47).

Kanoza ( 1996) provides

further details.

43. In December

1993 generics

accounted

for about

8

percent

of

total

antidepressant

market revenues,

and

37 percent

of daily dosage

units.

44. Bureau

of

Economic

Analysis (1985a,

p. 2; 1985b, p.

10).

45. Bureau

of

Economic

Analysis (1987).

156

Brookings

Papers:

Microeconomics

1996

The BLS

includes this

product in its sample,

even though

most of its

domestic

consumption

emanates

from

Puerto

Rico.46 This issue

of how

one

treats

Puerto Rican production is

important,

for

Puerto Rican

pharmaceutical

production is

about 20 to 25

percent of

mainland U.S.

production .

The current

Cycle C

sample incorporates items

from

several of

the

subclasses of

antidepressant

drugs displayed

in

tables

1

and

2, but the

weight given the

SSRI subclass

item(s)

is

(are) considerably

less

than

IMS data

would indicate

appropriate

(ignoring Puerto Rico

production

complications).

Moreover,

the older

antidepressants

appear

to be over-

weighted.

Specifically,

when

one assigns

each

antidepressant

chemical

entity

in

the

IMS data base the date of its

initial market

introduction,

calculates its

age

as of

1993:12

(the

beginning

of

the

Cycle

C

sample),

and

then

sales-weights each

entity's age using

IMS

sales of

daily

units

as

weights,

one obtains a

sales-weighted

average age

for each

entity.

In

1993:12 the

sales-weighted average age

of the IMS

universe of

antidepressant

drugs was 15.18

years,

while

that of the new

BLS

Cycle

C

sample

was an older 18.50

years;

if

one excludes

Prozac from the

IMS universe,

however, the

sales-weighted

average age jumps

to 18.53

years, virtually

identical

to

that

of the BLS

Cycle

C

sample.

In

February

1996, the last month in our

data

series, the

sales-weighted

average ages

for

the IMS

universe,

BLS

Cycle

C

sample,

and

IMS

universe

excluding

Prozac were

12.97, 16.58,

and

14.78 years,

respectively.48

Based

on

the

information it

collects,

the BLS

calculates

the

PPI

according

to a modified

Laspeyres formula,

in

which the value

of base-

46. This information was

provided

to

us

by

the manufacturer of the

product.

Note

that

the

weight employed by the BLS for this

product

could reflect

only

the

mainland

production.

47. When

the BLS supplemented

its Cycle C sample in December

1995 (see footnote

42

above), it chose four additional

antidepressant drugs.

All

four of these were found

to

be

manufactured

in

Puerto Rico, and thus

they

were

not included in the

supplemental

sample.

Regarding relative

importance,

it is not clear how best to

measure the Puerto

Rican

production proportion of U.

S. pharmaceutical

consumption.

If

one simply em-

ploys

value of shipments (VOS) data

from

the

1994 Economic

Census of Outlying Areas

(table 4,

p. 32) and from the 1992 Census

of

Manufacturers, Industry Series Drugs

(table 5b,

p. 28C-14),

both

published

by

the Bureau of the

Census,

one finds that

Puerto

Rican

VOS

is 22

percent

of

"domestic"

VOS-$1

1.1

billion, compared with $50.4

billion.

48. The

IMS

universe and BLS

Cycle

B

sales-weighted

average ages

at

the

beginning

of

Cycle

B

were 21.82 and

20.42

years, respectively;

six

years

later,

at the

end of

Cycle

B, the

respective average ages were

15.18

and 28.17.

Ernst

R.

Berndt,

lain M.

Cockburn, and

Zvi

Griliches

157

Figure 1.

BLS

Pharmaceutical Price

Indexes, 1980-96

Index

(1987:12

=

100)

210

All

prescriptions

............Antidepressants

165

-

Psychotherapeutics

120-

75-

30

I

1980:1

1987:12

1996:2

Source:

BLS

producer

price

indexes.

period

quantities at

current-period

prices

is

divided

by

the value of

base-period

quantities

at

(perhaps

temporally

different)

base-period

prices, that

is,

(1)

1t,=

[

Y

Q",P,

/

Y

Qb,PO]

100=

[={I

I

Q

(P/IP,)}/

E

QaPo,]

100,

where

Qb,

represents the

quantity

shipped

during

the

weight-base

period,

P,

is

the

current

price

of the

commodity,

and

Po

is the

price

of the

commodity

in

the

comparison

period;

the summation is over

i

goods,

but

i

subscripts

are omitted. Note that this index can be

written

as

a

weighted

average

of

price

relatives

P,IP,,

where the

weights

are

fixed

within

each

cycle.

The

monthly

time series

for the BLS PPI for all

prescription

phar-

maceutical

products,

for the

aggregate

class

of

psychotherapeutics,

and

for

the

antidepressant

subcategory

of

drugs

are

displayed

in

figure

1;

selected AAGRs are

given

in table

4.

For the

period

covered

by Cycles

Table

4.

Average

Annual

Growth

Rates

of

Alternative

Price

Indexes

Percentage

Entire

period

Cycle

A*

Cycle

B

Cycle

C

Price

index

1980:1-1996:2

1981:12-1987:12

1987:12-1993:12

1993:12-1996:2

BLS

series

All

Rx

7.94

9.82

7.13

3.06

Psychotherapeutics

NA

15.64

10.04

4.23

Antidepressants

NA

10.40

4.27

Antidepressants

BLS

and

IMS

data

BLS

sample

items

BLS

fixed

weights

NA

8.69

4.61

IMS

fixed

weights

NA

8.57

3.64

Divisia

NA

10.89

3.46

IMS

data

only

Laspeyres-fixed

All

5.04

9.35

4.17

0.42

Brands

9.26

12.37

10.34

3.65

Generics

-6.09

-4.71

-6.79

-5.66

Divisia-chained

All

7.51

10.35

7.90

2.34

Brands

8.83

11.43

9.62

3.63

Generics

-8.88

-5.13

-8.99

-18.93

Paasche-all

7.11

9.87

7.45

2.29

Divisia

brands

SSRIs

NA

3.94

TCAs

9.22

11.02

10.82

4.31

MAOIs

9.63

10.96

11.72

5.63

Others

10.89

12.77

15.29

4.11

Laspeyres

brands

SSRIs

NA

3.62

TCAs

9.02

12.41

9.40

4.28

MAOIs

9.07

10.69

12.53

1.07

Others

NA

16.11

3.13

Divisia

generics

TCAs

-9.96

-4.88

-10.72

-22.30

Others

NA

-6.40

-6.88

Laspeyres

generics

TCAs

-6.08

-4.71

-6.81

-5.57

Others

NA

-6.40

-6.88

Source:

Authors'

calculations;

see

text

for

explanation.

NA:

Not

available.

*Cycle

A

is

defined

here

as

1981:12-1987:12.

160

Brookings Papers:

Microeconomics 1996

B

and C (1987:12

through

1996:2), the

PPI

for

all prescriptions

in-

creased by about

63 percent (an

AAGR

of

6.08

percent), much

less

than the

PPIs

for

psychotherapeutics (96 percent,

AAGR of 8.53

per-

cent) and for

antidepressants (101

percent,

AAGR

of 8.80

percent).

This faster growth

of

psychotherapeutics compared with

all

prescription

drug prices

continues

a trend

going

back at least to the

beginning

of

Cycle

A;

from

1981:12

through

1987:12,

the

price

index of

psycho-

therapeutics increased 139

percent

(AAGR

of 15.64

percent), compared

with a PPI for all

prescription

products

of

75 percent

(AAGR

of

9.82

percent).49

Finally, annual average

growth rates

for

all three

price

in-

dexes

(all prescription

drugs,

psychotherapeutics,

and

antidepressants)

are

greater during

the

Cycle

B

era

(7.13 percent,

10.04

percent,

and

10.40

percent,

respectively) than

during

the

Cycle

C

time

span

to

date

(3.06

percent,

4.23

percent,

and

4.27

percent).

Given its fixed

weights

in

the context

of

a

rapidly

changing market,

the

reliability

with

which the BLS

PPI

for

antidepressants

could be

expected

to track

actual

marketplace

developments

is

ambiguous

at

best,

but

whether the BLS

sampling procedures

impart

a

systematic

bias

to

the

index

is

unclear.

The

undersampling

of

generics

would likely

impart an

upward

bias, given

the

substantial

price

reductions

they

have

experienced,

but the revenue shares of

generics

in

total are small and

falling (11 percent

in

December 1987,

8 percent

in

December

1993,

and 3

percent

in

February

1996).50 The BLS

lags marketplace

developments

in

the choice of its

sample,

and the net effect of

this

lag

on

an

aggregate price

index is

therefore

an empirical issue. But

a

different

consideration unrelated to

sampling

issues-namely,

the

absence,

until

recently,

of a

link

between

generics

and their

patented

antecedents-can more

clearly

be

expected

to

result

in

an upward bias

to

the

BLS

index.

Alternative

Price

Indexes:

Theory

and

Evidence

The

price

of a

good

before its market

introduction cannot be

ob-

served.

After

a new

good

enters

the

market,

it

may

take

quite

some

49. Of the

twenty-five

seven-digit products

in

SIC

2834-1, only

two

have a

greater

rate

of

price increase than

psychotherapeutics-central nervous system

stimulants and

antiobesity preparations,

and sedatives. See

Bureau

of Labor

Statistics

(1996, p. 61).

50. The

corresponding

daily

unit

dosage

shares are 44

percent,

38

percent,

and

27

percent, while daily dosage

levels of

generics are

32

million, 56 million, and

58

million.

Ernst R. Berndt, lain M. Cockburn, and

Zvi

Griliches

161

time for statistical agencies

to track

its

price.

As

Early

and Sinclair

have discussed, the BLS periodically

revises the sample items and

"links in" new commodities.5' For example,

in December 1995 the

BLS supplemented its Cycle C prescription

pharmaceutical sample with

fifty-one items, thereby incorporating

selected

market

developments

since the original sample (based

on 1991-92

data)

was drawn for

im-

plementation

in

December

1993. Once

items are selected for an

updated

sample, the

BLS

includes

their

price

changes

in its

price

index com-

putations.

Because this

procedure

makes no

comparisons

between new

and incumbent goods, however, changes

in

the aggregate price index

reflect only changes in the prices

of

the

products and ignore any absolute

price differentials between the

new and

comparable

incumbent

prod-

ucts.

Although

such

a

procedure may

perhaps

be

appropriate

for

truly

new goods, it surely is not appropriate

for many goods such as phar-

maceuticals for which some forms of

substitute goods or services are

available.

Considerations

from

Economic

Theory

The

theoretical solution

to this "new

goods problem"

has

long

been

known: for the time period just before

the introduction

of

the

new

good,

find that price at

which

quantity

demanded

is

just equal

to

zero and

put

this "reservation price" into the price

index calculation for the time

period just

before the

new

product

is

launched.52

This theoretical

insight

is

informative, but

it is also

challenging

to implement empirically,

for

it

requires

estimation

of demand models that

may

have

burdensome

data

requirements,

it

may

entail

making strong assumptions,

the esti-

mated reservation prices might

be sensitive to

the

choice of functional

form,

and the issue

of

proper

item

weights

is left

open.

The

special

characteristics of

generic

drugs provide

an

opportunity

to

modify price

index

computations

in a

relatively simple way, thereby

taking

into

account

the

implicit price

declines

experienced by

those

consumers

who switch from

brand

to

generic

versions of a

chemical

entity. Specifically, generic drugs

can be

envisaged

as a

particularly

simple

case of the

new

goods problem,

because

a

generic

is

a

variety

of an

existing product

identical

in

almost

all

respects

to the "old"

51. Early and Sinclair (1983).

52. Hicks (1940), Rothbarth (1940-41), and

Fisher and Shell

(1971, 1972).

162

Brookings Paipers:

Microeconomics 1996

version. In the United States,

the FDA publishes

an

"Orange

Book,"

Approved Drug Products with

Therapeutic Equivalence

Evaluations,

that certifies therapeutic equivalence.

Although the generic

versions

differ from the branded product

in

packaging (including

the inert matter

enclosing the active ingredients),

labeling,

and

provenance,

the FDA

certifies that the generics are

equivalent to

the

branded product

in

two

senses: pharmaceutical equivalence,

that

is,

the active

ingredient

is

chemically identical,

has the

same

strength, dosage form,

and route of

administration, and is manufactured

in

compliance

with

Current

Good

Manufacturing Practice regulations;

and the

generic

version

is

"bio-

equivalent"

in

that it is statistically indistinguishable

from

the branded

product in key pharmacological

aspects of therapeutic use,

such as

blood concentration profiles.

The

extent to which

generics

and branded

products

are

in

fact "al-

most perfectly substitutable"

is a hotly debated topic. Therapeutically

equivalent products may

still

vary

in characteristics such

as

inert ma-

terial, shape, color, flavor, scoring,

packaging, labeling, shelf

life, and

stability under adverse storage

conditions.

Insofar

as

any

of these char-

acteristics affects patients'

ability

to

distinguish

between

different tab-

lets and dosages, their readiness

to

take

the medicine at

the

time and

in

the amounts prescribed,

or their

possible

reactions to coloring

or

pre-

servative ingredients,

these

apparently

trivial factors

may

in

fact influ-

ence

the

realized

effectiveness

of the

generic drug

relative

to the

branded

product. Moreover,

variations

in the inert

matter

encasing

the

active

ingredient

can

affect the

speed

of

absorption

of a medication.

If one takes the

FDA at its word-"a

pill

is

a

pill

is a

pill"-the

reservation

price

is the branded price just

before the

generic

enters the

market, and in this case the

appropriate price

index for

a

particular

chemical

entity

is

straightforward,

being

the

weighted average

price

of

a

tablet across all

generic

and branded manufacturers.

If,

however,

one

takes

the

opposite

extreme position-that

taken until

recently by

the

BLS

for its

PPI-then one implicitly

treats

generic

versions of a

drug

as

entirely distinct,

nonsubstitutable

commodities.

In that

case,

the

generic

launch

price

is

also

its reservation

price.

As we noted

earlier,

the BLS has

recently changed

its

policy

and is now

introducing

a

linking

procedure

consistent with

perfect

substitutability:

"...the

predecessor

brand-name

drug price

and

successor

generic drug price

will

always

be

directly compared

without

quality

adjustment.

The direct

comparison

Ernst R.

Berndt, lain M.

Cockburn, and Zvi

Griliches

163

is predicated

on the

assumption that the

two products

are of equal

quality,

because the FDA has determined

them to be

therapeutically

equivalent.

"53

Generic

prices tend to be

considerably

lower than that of

the branded

version, and

the spread

between them tends

to increase

over time. Given

the magnitude of the

price

differential,

it is

striking

that not all

con-

sumers

switch

to the cheaper

variety.

Apparently

consumers,

or rather

physicians

who prescribe for

them, differ

in

their

perceptions concern-

ing the

efficacy and quality

of generics,

despite

FDA

certification, and

some

consumers,

prescribers,

and

insurers

prefer

the

much

higher-

priced branded

versions.

Some

consumers, prescribers,

and

insurers,

however,

do switch

to the

cheaper generic

version,

either

perceiving

no difference

between

brand and generic varieties or

taking the price

differential as more than

sufficient

compensation

for

any

difference in

quality.

Clearly,

there are considerable differences

in

information

and

knowledge

and in tastes and

preferences

among

consumers, prescribers,

and

insurers

concerning

brand-generic

differences.

Alternative

Procedures for

Incorporating

Generics

We now

consider alternatives to these two

extreme

positions,

var-

iants

we believe more

accurately reflect

the

price

declines

realized by

intermediate

purchasers of

prescription

drugs such as

pharmacies.

These

alternatives

vary

in

how diverse ultimate

consumer, physician,

and

insurer choices are taken into

account when reservation

prices

are

being computed.

Fisher and

Griliches

have

shown that even

when con-

sumers are

heterogeneous,

aggregate

Paasche and

Laspeyres

price

index

computations

provide

bounds

for

a

hypothetical

social

planner's

ideal

index, giving the minimum amount needed to

keep

all

individuals on

their

base

utility

level when

prices change.54

Griliches and

Cockburn

present

formulas for such

aggregate

indexes

in

a

world

in

which

either

the

branded or the

generic

version

of a

particular

chemical

entity

is

purchased.55

Let

Pb

be the

unit

price

of a branded

drug,

and let

pg

be

the

generic

price.

In

a

simple

linear random

utility

framework, purchaser

h

chooses

53. Kanoza

(1996, p. 9). The italics

are

in

the original.

54.

Fisher and

Griliches (1995).

55.

Griliches

and

Cockburn

(1994).

164

Brookings Papers: Microeconomics 1996

the

generic

version

if

pb

>

pg

+

b,,

where

b,

is

the

subjective premium

required by purchaser h when

buying the generic

to

compensate for the

putative

loss in

security

or

quality

associated

with

the switch.

If

one

knows the reservation price p,;

for

each

purchaser,

then Griliches and

Cockburn show

that the

aggregate

Paasche

price

index between

periods

0 and

1

is

appropriately

calculated

as

H

(2)

Pl = Qipo +

Q,j, wherep- P

"=

_

where the 0 and

1

superscripts

refer to

time

periods, Q

denotes

aggre-

gate quantities

over

all

H

consumers,

and

qg,,

is the number

of

units

of

the

generic

version

bought

on behalf of consumer h.

Although elegant, this theoretical framework

requires estimation of

reservation prices, a nontrivial

task. One feasible approach involves

making

an

assumption

concerning the distribution of preferences for

brandedness. In

the linear random

utility framework,

the

probability of

any purchaser switching

from

brand

to

generic depends on

Pb

-

Pg

>

b,,

and thus the share

of

generic

users

in

the total is

s,

=

F(b,,),

where

F(b,,)

is the cumulative distribution of

reservation prices, given a fixed

Pb*

If

no

buyer

is

willing

to

pay

more for a

generic

version than

for a

branded

one

when the branded one

is

available,

then

b,,

?

0, and in this

case

the

average

reservation

price

for

switchers

must

be

bounded be-

tween

pb

and

pg,

with the

precise

location

depending

on

the shape of

F(b,). Following Griliches and

Cockburn, one can assume that unob-

served

tastes

for

brandedness

among purchasers

are

uniformly distrib-

uted,

in which case the

average

reservation

price

is

half

way between

Pb

and

p?,

thus

"splitting

the

difference" between the two extremes

of

the

old BLS

approach,

which

assigns

all of the

brand-generic price

differential to

quality

differences

(pg being

the

reservation

price), and

the FDA

approach,

which

assigns

none of it

(Pb

being

the

reservation

price).

One notable feature

of these markets

is that a new

generic product

typically

takes several months

to achieve

significant

sales. The

product

may

take time

to move

through

distribution

channels,

and it

may

take

time

for

physicians

and

purchasers

to become

aware of its

availability

or

to

acquire

other

information

germane

to

prescribing

and

buying

decisions.

Regardless

of its

causes,

the

lagged response

of demand to

Ernst

R. Berndt,

lain M. Cockburn, and Zvi Griliches

165

price changes has

important implications for price

index computations,

particularly

at

monthly frequency,

since

weights of

new generic prod-

ucts are

typically

initially low.

This "diffusion

problem" can be approached in

several ways. One

way

is

to link

in

the new generic good after

sufficient time has passed

(say,

six

months to

a year), thereby allowing much of the

early diffusion

of

generics

to

be

completed

before

evaluating

their direct

contribution.

We discuss the BLS variant

on

this

approach

below.

Alternatively, as

Griliches

and

Cockburn have proposed

and

implemented,

the

Paasche

index formula can

be adjusted

to

reflect

the

assumption that those shift-

ing later on to

generics do so from the branded

good, with an average

reservation price

that

is

half way between the prices

of the branded and

generic good.

In

such a

case,

the Paasche

equation

2

becomes

I

V~Q)p),

+

Q""p"

(3) ,

Q),P57 + Q+Q Q ,'=

(p-

+

p')/2.

Thus shifters from

the branded to the generic

version are assumed to

have experienced

a price decline equal to half of

the branded-generic

price differential

also

in

periods subsequent to the initial

appearance of

56

generics.

The new BLS

approach

to this diffusion

issue, given

its

fixed-weight

Laspeyres index,

is

considerably

more

parsimonious

in its

data

require-

ments than are

the

above alternatives

and

addresses the choice of res-

ervation

price

and item

weights simultaneously.

Based

on

a review

of

published

research

materials,

data from the

FDA

"Orange Book,"

and

consultations with various

industry experts,

the

BLS has

determined

that

in

the month when initial

generic entry

occurs for a

chemical

entity

in

its

sample,

the

quantity

weight, say, x,,

will

be

split

into

two

components,

with a 0.642

x,,

quantity weight assigned

to the

generic,

and

a 0.358

xo

weight given

the

branded

version. These

weights

are

then

fixed

until a new or

supplemental

sample

is drawn

and

are

the

same

for

all

generic

entities. The 64.2

percent

generic weight

turns out to

be

quite

close to

the

generic

quantity

share of the

seven

antidepressants

experiencing

initial

generic entry

after 1980

(see

table

3); specifically, the 64.2

percent

share falls

in

between

the

daily dosage-

56. Ibid. The procedure used here is referred

to in

Griliches and Cockburn (1994)

as "Paasche-UD."

166 Brookings Papers:

Microeconomics 1996

weighted average generic quantity share of 57.0 percent

after twenty-

four months and 68.9 percent after thirty-six months.

Finally, yet another alternative approach

to

deriving

reservation

prices, one that we plan to pursue

in

subsequent research, is to

estimate

demand curves from data on

prices

and

quantities

and then

to

project

the

estimated demand function

to

find

the

brand-generic

price

differ-

ential

that

would choke

off demand

for

the

generics

to zero. A related

research project involves using

similar data to estimate the

shape

of

F(b,,)

consistent with observed relationships between

prices and market

shares.

New Products and Hedonic

Regressions

As

noted earlier, generics

are a

special

case

of

the

new

goods prob-

lem, for with generics the FDA has certified

equivalence. In general,

however, new goods differ in significant ways

from

older products,

reservation prices are more difficult to quantify, and thus

incorporating

new goods into price indexes

is

more complex.

One

way

in

which

the

effects of new goods

could

be incorporated

into a

price

index is

simply

first

to

regress for each branded product-month, say,

the

logarithm

of

daily dosage price

on time

dummies,

and a

dummy

variable for each

distinct

brand.

One could then use the

predicted price

for

the month

prior

to

a good's introduction

as an

approximation

of

the

reservation

price, thereby linking

in

the price of the

new

good.

The

hedonic price approach employs

instead

quality

attributes as

regressors,

in

effect making parameters

on

the

brand

dummy

variables

functions

of

quality

attributes.57

The hedonic

approach

is

particularly

relevant for branded

products,

because

their

prices

are set

by

firms

having some

market

power.

For

generic products,

the

hedonic

approach

might

be

less useful, particularly

if

competition

drives

prices

down to

marginal production-distribution costs,

and when

such

marginal

costs

are

not

dependent

on

quality

attributes.

It is

important

to note that

price

indexes

linking

in

new

products

using predicted prices

from hedonic

price equations

can

grow

at

rates

less

than, the

same

as,

or

greater

than

those

that

entirely

ignore

the link

and instead

incorporate only

the

price changes

after new

product

launch,

that

is,

those indexes

that treat the reservation

price

as

equal

to the new

57. See Griliches (1971,

1990) for a discussion of this

methodology.

Ernst R. Berndt,

lain M. Cockburn,

and Zvi Griliches

167

product's launch

price. The relationship

between

the two growth rates

depends critically

on whether the

actual price of the

new product at

the

time of its launch

is above or below

that predicted

by the hedonic price

equation, given

the new product's

quality characteristics.

Any

differ-

ence between

the linked and

nonlinked series

will

emerge only

if

there

is

positive or

negative nonpriced

quality

at launch.58

If

the new product

has a launch price

that just compensates

for

its hedonic-estimated qual-

ity, then the

launch date hedonic

residual

would

be

zero, and

AAGRs

of

price indexes

based on the

nonlinked and

the

hedonic quality-

adjusted linked

procedures would

coincide.

If the

launch

price

of the

new product

were set above (below)

its estimated

hedonic quality,

however, then

hedonic residuals

would be positive

(negative), and

the

hedonic quality-adjusted

linked

price index would

grow at a greater

(lesser) rate than

the nonlinked index.

The magnitude

of any difference

would also depend

of course on

how

quickly

the quantity weights

be-

come substantial,

that

is, the speed

of diffusion.

Launch date

residuals

from

an estimated

hedonic

price equation

are

therefore quite

important, for they

purport

to measure deviations

from

reservation prices.

Whether

these residuals

in

fact

reflect price

dis-

counts or premiums

relative

to "true" quality

is

unknown,

however,

for

hedonic residuals

could

instead

be manifestations

of

important qual-

ity aspects that

have been omitted

or other specification

or measurement

errors. Evidence

on

unmeasured

quality

as a

possible

specification

error

could be obtained

by examining

relationships

between hedonic

equation

residuals and

realized market quantity

shares;

a

negative relationship

between

them

would be consistent

with the

notion

that residuals instead

reflect unobserved

quality differentials.

Later

we

present

findings

based on hedonic

price equations

for

branded

products.59

First, however,

we

focus on alternative

price

in-

dexes where issues

of

sample

selection and

item

weighting

are central.

Empirical

Evidence:

Sampling

and

Weighting

Issues

In

table 4

we

report AAGRs

for

several

alternative

price

index

cal-

culations. In

the

top panel

we

list official

BLS

growth

rates,

which for

58. For examples and further discussion,

see Berndt and

Griliches

(1993) and

Berndt, Griliches, and Rappaport (1995).

59. For earlier attempts to estimate

hedonic

price equations

for branded

prescription

pharmaceuticals, see Berndt and Finkelstein (1992) and

Suslow

(1996).

168

Brookings

Papers: Microeconomics

1996

the

antidepressant class of drugs is 10.40

percent (Cycle B) and

4.27

percent (Cycle

C). Based on

the

BLS

sample information provided

us,

we

employ

the

IMS price data but the BLS item

weights

to

construct

an

aggregate

index, mimicking BLS

fixed-weight Laspeyres proce-

dures, with

weights updated

at

1987:12 and

1993:12.

It is

not

possible

to obtain an exact

correspondence, because two

of

the seven items in

the BLS

Cycle

B

are not

generally

considered

antidepressants,

nor is

one

of

the ten

in

Cycle C, whereas the

IMS data

are confined to anti-

depressants.

If

these items

are

removed, we

obtain

AAGRs for the

partial BLS

item

sample

that

are

lower

than those

reported by

the BLS

during Cycle

B

(8.69 percent, compared

with 10.40

percent),

and close

but slightly

larger during Cycle

C

(4.61

percent, compared

with 4.27

percent).

The

relatively

close

correspondence

during Cycle

C

is

reas-

suring,

particularly because

the one omitted item has

a

relatively low

item weight,

whereas the two omitted

items during Cycle B have

a

larger combined

relative weight.

If

one retains

the

BLS

sample

items

but

uses

instead

of

the BLS

weights those based

on

IMS

data,

the

resulting

difference

in

AAGRs is

small

during

Cycle

B

(8.57 percent, compared

with

8.69

percent),

but

larger during

Cycle

C

(3.64 percent

using

the

December 1993 IMS

weights, compared

with 4.61

percent

using

the

necessarily

older

1992

or 1993 BLS

weights). Finally,

if

one uses the

partial

BLS item

sample

and

allows

weights

to

change monthly

with the Divisia

index,

one

obtains

rather different

results, suggesting

that

weights

do

matter.

Spe-

cifically,

as seen

in the

bottom row

of the second

panel

of table

4,

during Cycle

B

the

AAGR of the Divisia index is

10.89

percent.

It is

useful to

distinguish

between

the

choice and

accuracy

of

fixed

weights

and

the

effects

of

changing weights.

We

begin by

comparing

AAGRs

of the BLS official PPI

for

antidepressants

with those

based on

the IMS

universe, using comparable

fixed-weight Laspeyres

index

pro-

cedures. As the row marked

Laspeyres-All

shows, during Cycle

B

the

Laspeyres index based on the universe

of

antidepressants grows much

less

rapidly

than

does

the

BLS

index,

4.17

percent

a

year (compared

with

10.40

percent).

In

Cycle

C the IMS universe

data have an AAGR

of

only

0.42

percent,

while

the BLS

grew

at

ten

times this

rate,

4.27

percent.

This

large disparity

is

surprising,

given

that the

only

underly-

ing difference

is one of

weights

drawn from

a

sample

rather

than the

universe,

and not the

Laspeyres

index.

Ernst R.

Berndt, lain M.

Cockburn, and

Zvi

Griliches

169

Figure

2.

Daily Doses of Antidepressants

Percentage

Millions of dollars

100

25

90

80

i

Brand shares

20

70-

60

-15

50-

40

1

I0

30-

20 -

.0Total

daily

doses

-

5

10

1980:1

1982:1 1984:1 1986:1 1988:1 1990:1 1992:1 1994:1

1996:1

Source: IMS

America,

Inc.

Recall

that the BLS

Laspeyres fixed-weight

procedure entails a

change

of

weights only

at

six-year

intervals.

If

the sizes of the

weights

selected

at

1981:12, 1987:12,

and 1993:12

do

not

accurately portray

actual

data

trends

during

the

subsequent

six

years,

the

resulting

indexes

can

yield

very misleading

AAGRs,

with

the

sign

of

the

bias

being

generally

indeterminate.

Apparently,

that is what

has

happened.

To

see this, notice

in

figure

2

that the time

trend

of

the brand share

of

daily

dosages

over

the

sixteen-year period

is

approximately U-

shaped,

starting

at 77

percent

in

1980:

1, falling

to 47

percent

in

1988:8,

and then

increasing

to

73

percent

by

1996:2. When the

BLS drew its

Cycle

B

sample

in

1987,

the

generic

share was

near its

peak

at

47

percent,

but

by

the end of

Cycle

B in

late

1993,

it had

fallen to

about

37

percent,

and

by 1996,

to

27

percent;

the

corresponding

brand

shares

increased with

time,

53

percent

in

1987,

63

percent

in

1993,

and

73

percent

in

1996.

Thus,

if the

BLS

fixed-weight procedure

had

been

applied

to the

universe

of

antidepressant drugs,

over the

entire

Cycle

B

and

in

Cycle

C to

date, generic products

would have been

overweighted

170 Brookings Papers: Microeconomics

1996

(their unit share fell as brand dosages grew more rapidly, led

by new

SSRIs such as Prozac),

and brands would have been

underweighted.

Recall from our earlier discussion that prices

of

(overweighted)

generics

have been falling, while (underweighted) brand prices have

generally

been increasing. Together

these trends

imply

that had the

six-year

fixed-

weight procedure been applied to the universe

of

drugs,

the

resulting

AAGRs

would

have

severely

understated

price growth

in

both

Cycles

B

and C. The much higher AAGR

for the BLS

published

PPI

based on

its sample than for the Laspeyres

based on the universe reflects the fact

that the BLS

sample

was in

fact

nonrepresentative,

fortuitously weight-

ing generics less and brands more than

the

then-current

market condi-

tions warranted. (Recall

that none of

the

seven items

in

the

Cycle

B

sample

was a

generic.) Clearly, using six-year

fixed

weights

in

a rapidly

changing

environment

can

lead to

highly

unreliable results.

One

way

to

assess

the role of

changing weights

is

to

compute

a

Divisia

index, which weights percentage price changes by

the

average

share

in the

current and previous month,

in

contrast to the fixed-weight

BLS procedure in which the weights are changed only every

six years.

Alternatively,

one

can employ

a Paasche

index that

sequentially

updates

the

weights monthly.

As

seen

in

the

row marked

Divisia-All,

based on

the

IMS

universe, the

Divisia

grew

more

rapidly

than

the

Laspeyres

during

all

three cycles-10.35 percent compared

with 9.35

percent

in

Cycle A,

7.90

percent compared

with

4.17

percent

in

Cycle

B,

and

2.34 percent compared with 0.42 percent

in

Cycle C;

for the

entire

sixteen-year period,

the Divisia has an AAGR

of

7.51

percent,

com-

pared

with

5.04

percent

for the

Laspeyres

index.

Finally,

as the row

marked Paasche-All

shows,

the AAGR

of

the chained Paasche index

over the sixteen

years is,

at

7.

11

percent, slightly

smaller than

the

Divisia

index,

but two

percentage points larger

than the

fixed-weight

Laspeyres.

Differences between

the

Laspeyres

and

Divisia also

persist

when

antidepressants

are

disaggregated

into brands and

generics.

Although

Laspeyres-Divisia

differences are modest

for brands

(9.26 percent

com-

pared

with 8.83

percent

over the

sixteen-year

time

frame),

for

generics

the

disparity

is

larger

in

Cycle

B

(

-

6.79

percent compared

with

-

8.99

percent),

and

in

Cycle

C

it is

dramatic

(-5.66 percent

compared

with

-

18.93

percent).

Note that the

Divisia index

incorporates

new

gener-

ics,

unlike the

fixed-weight Laspeyres.

Ernst

R.

Berndt, lain M. Cockburn, and Zvi Griliches

171

If

one

disaggregates even further to

within brands and within

gener-

ics, for the

branded SSRIs, TCAs,

and others,

Laspeyres-Divisia dif-

ferences

are

modest, but for the generic

TCAs,

these

differences are

larger

in

Cycle

B

(-6.81 percent

compared

with

-

10.72

percent) and

enormous

in

Cycle

C

(-5.57 percent

compared

with

-22.30

percent).

It is also

interesting to note that

during Cycles

B

and C,

the oldest

branded

products, the MAOIs,

generally

have

larger Divisia

price

in-

creases than the

younger TCAs,

which

in

turn have

larger price

in-

creases than

the most recent SSRIs.

This

age-price pattern

is

consistent

with

the

product life cycle pricing

results reported by Berndt,

Griliches,

and

Rosett

for

all branded

prescription

products.60

Finally, no matter what index

procedure

is

employed, the

prices

of

generic products

clearly

have been

declining throughout

the

sixteen

years and,

as measured

by

Divisia

indexes,

the rate of

decline has

sharply

accelerated

over time.6'

Empirical

Evidence: Effect of

Linking Generics to Brands

We

now

consider effects

of

linking generic prices

to their

branded

versions, rather than

following the

old BLS

procedure

of

not

linking

them

at all.

There

are at

least

three

ways

to

introduce

a link

between

generics

and

their

patented

antecedents.

In

what we call

the "FDA

Average

Price

Procedure," generics

and branded versions

of the same

chemical

entity

are treated as

perfect

substitutes,

and the

average price

of the

entity

is

simply

a current-month

weighted average

of

generic

and

branded versions.

An

alternative

is

to assume that

preferences

for

brandedness are

uniformly

distributed and then to

adjust

for

diffusion

using equation

3. We call this the

Griliches-Cockburn

adjusted

"Paasche

Diffusion"

method

(GCPD). Finally,

we

mimic the new

procedure

recently adopted by

the

BLS,

in

which

the

Laspeyres fixed

branded

weight

is

split

into a 64.2

percent

generic component and a

35.8

percent

branded

component upon

initial

generic entry

and is

fixed

thereafter;

we call this the "New BLS Procedure with

Fixed

Split

60. Berndt,

Griliches,

and

Rosett (1993).

61. An

implication is that private sector price

indexes based only on brand

prices,

such

as the

PRIME index published by the National

Association of Chain Drug

Stores

(1995),

are

likely

to overstate

price

inflation

significantly.

172

Brookings

Papers:

Microeconomics

1996

Table 5. Average Annual

Growth Rates of Price

Indexes that Link

Generics to Their Patented

Antecedents

Percentage

Entire Period

Cycle

A

Cycle

B

Cycle C

1980:1- 1981:12-

1987:12-

1993:12-

Procedure

1996:2

1987:12

1993:12 1996:2

All

drugs, no link:

Paasche

7.11

9.87

7.45

2.29

Divisia

7.51 10.35 7.90

2.34

Laspeyres

5.04

9.35 4.17

0.42

FDA

Average Price

Procedure

All with

generics

-2.98

5.33

-6.49 -

17.22

All drugs linked

2.95

5.71

1.33

1.10

Griliches-Cockburn

Adjusted Paasche

Diffusion Procedure

All with

generics

0.96

6.97 0.42

-

12.68

All

drugs linked

4.73

7.08

4.44

1.69

New BLS

procedure

with fixed

split

generic/brand

weights

All with

generics

2.69 7.18

1.97

-4.15

All

drugs

linked

3.71

7.41

2.49 0.42

Source: Authors'

calculations,

see text for

explanation.

Generic-Brand

Weights.

"62 AAGRs

based

on

these three

linking pro-

cedures are

given

in

table

5. In each

panel

we

report

AAGRs of

price

indexes where generics have been

linked to

their

patented

antecedents,

first for the

subset of

drugs

that

experienced

generic entry

and then for

all

the

drugs

in

the IMS

antidepressant

universe with

generic

versions.

For

purposes

of

comparison,

at the

top

of

table

5,

we

give

AAGRs for

Paasche, Divisia, and

Laspeyres

indexes

with

generics

not

linked in at

all.

Of

particular

interest is

the

effect

that

linking

in

generics

has on

growth of the overall

antidepressant

drug price index,

and

how

this

effect varies

among

the three

alternative methods.

When

generics

are

linked

in

using

the FDA

Average

Price

Procedure,

AAGRs are affected

62. New

generics

are introduced within

cycles

with

split weights,

with the brand

portion

retaining

the base

price

of

the brand and with

the

generic portion

having

as its

base

price

the

price

of the brand in the time

period

prior

to

entry

date.

Ernst

R.

Berndt,

lain

M. Cockburn,

and

Zvi Griliches 173

dramatically, and the

overall price

index

grows

less than half as

fast as

it does when no

attempts are made to link in

generics.

In

particular,

during the entire

sixteen-year period, the generic-linked AAGR

grows

at

only 2.95

percent, compared

with

7.

11

percent

for

the

unlinked

Paasche index or

7.51 percent for the unlinked Divisia index.

This

difference is

particularly large during Cycle

B

when

considerable ge-

neric entry occurs,

with the linked AAGR being 1.33

percent, dramat-

ically lower than the

unlinked

Paasche

(7.45

percent)

or

unlinked

Div-

isia (7.90

percent); during Cycle

C

they

are somewhat

closer

at 1.

10

percent, compared

with 2.29

percent

and 2.34

percent.

Even under the

more conservative

GCPD

"split-the-difference"

as-

sumptions, the

impact

on

aggregate growth

rates of

linking

in

the

ge-

neric drugs is very

substantial.

As

seen

in

table 5,

AAGRs are about a

third lower than

they are

in the unlinked indexes.

Specifically, for the

sixteen years, the

generic

linked

price

index

grows

at

4.73

percent,

compared

with 7.11

percent

for

the unlinked

Paasche

index

and 7.51

percent

for the unlinked Divisia

index; again

the difference

is

largest

during Cycle B,

still

large

in

Cycle A,

and smallest

in

Cycle

C.

Finally, when

we mimic the new

BLS

procedure

of

splitting

the

Laspeyres

fixed

weight

into 64.2

percent generic

and

35.8

percent

brand

components

following generic entry,

we obtain

results

that

generally

yield

an

impact

in

between the

FDA and GCPD

procedures;

the

all-

drug

linked index

has an AAGR over the sixteen

years

of

3.71

percent.

Had the BLS

implemented

the

procedures

it will now

employ,

the fixed-

weight Laspeyres

AAGR would have been

only

2.49

percent

for

the

IMS

universe

of

antidepressant drugs during Cycle

B, substantially

lower than

the published

10.40

percent

based

on the

BLS

sample (which

entirely neglected

generics).

These results

dramatically

illustrate that the old

BLS

policy

of not

linking prices of

newly

introduced

generic goods

to

the

prices

of

their

branded

predecessors

exerted a

very

substantial

upward

bias

on

the

overall

price

index

for

antidepressant drugs.

The effects

of

the new

BLS

procedure,

beginning

in

January 1996, might

well be

expected

to

result

in

lower measured rates

of

price inflation,

all

else

being equal, although

the

magnitude

of

this effect

will

depend

on

which

products

will

lose

their

patent protection

and how

important they

are.

It

is worth

noting

that

in

the

antidepressant

market,

no

drugs

are

currently

scheduled to

lose U.S.

patent protection

before

the

year

2000.

174

Brookings

Papers: Microeconomics 1996

Empirial Evidence:

Hedonic

Regressions

As we

noted

earlier,

differences

in

the

efficacy rates

among the

various

antidepressant

drugs are

statistically

insignificant, but the

side-

effect and

adverse interaction

profiles

vary

considerably,

with

the

newer

generations

of

drugs

generally

having superior

characteristics;

these

side-effect

profiles

were summarized

in

table

1.

To

capture and

quantify

these

quality

improvements over

time, we have

undertaken

a hedonic

price

analysis. The

results we report here

represent

ongoing

research.

As the

dependent variable

in

the hedonic

regression,

we

compute

for

each

of

the branded

drugs

in

our

sample

the

monthly price

per daily

dose

equivalent for the

period 1980:1

to

1996:2;

this

price

is not

the

same as a

Laspeyres,

Paasche,

or

Divisia

index,

for

those indexes do

not

provide absolute

conmparisons

across

drugs.

Because

their charac-

teristics

are

so

different

from

other

antidepressant

drugs

and their

mar-

ket shares

are so small

(less

than 1.5

percent), the MAOIs

are excluded

from the

sample

of

branded

drugs.

That

leaves

up

to

nineteen branded

drugs

in

any

one month and

a

total of

2,478

observations.

The

specification

of

attributes or characteristics in

hedonic

price

equations

is

always

somewhat

problematic,

and that

is the

case here as

well.

Considerable

collinearity

frequently occurs

among

possible attri-

butes,

which is to

be

expected

in

this

case because of the

biological

and

chemical

relationships.

Another

specification

issue is that

knowledge

about

the attributes of

drugs

diffuses at

different rates even

though

these

attributes

are

relatively

constant

(indeed,

here we

have

them

fixed

over

time).

Further,

how one scales attributes

such

as

drug

side-effect

pro-

files

is

not

without

ambiguity,

because

frequency

and

severity

are not

necessarily

related.63

Given

these

difficulties,

we have chosen to

pursue

a

relatively simple

and

parsimonious

specification,

in

which

we

regress

the

logarithm

of

the

price

of the i"

antidepressant

drug

in

month

t

on a

constant,

on

193

monthly

time

dummies

(that

for

1980:1

is

omitted),

and on

several

attribute measures that

proxy

for

quality.

We

consider

six

quality

attri-

butes,

whose

values

are

given

in

table 1: HALF

(mean

half-life of

elimination,

measured in

hours),

DR

(a

drowsiness side

effect,

scaled

zero

for

rare

and

four

if

frequent),

AC

(anticholinergic

side

effects,

63. For an extended discussion

of

specification of hedonic

equations

for

antihypertensive drugs,

see Berndt

and

Finkelstein

(1992).

Ernst R.

Berndt, lain M.

Cockburn, and Zvi

Griliches

175

zero to

four), GI

(gastrointestinal side effects,

zero

to

four), and

WTG

(weight

gain greater than

six kilograms, zero

to four). When

greater

frequency of occurrence of an

attribute such

as anticholinergic

side

effects is

usually considered

as

being

undesirable,

we

expect the

cor-

responding hedonic price

coefficient to be

negative. Our

specification

also

includes dummy variables

GEN

for

whether

the

brand

faced com-

petition

from

generics,

and OCD

if

the FDA had also

approved

the

antidepressant

drug

for treatment

of

obsessive-compulsive

disorders.

Estimation by ordinary

least squares

(OLS) yielded the

following

equation, with

heteroskedasticity-robust

standard errors

in parentheses:

ln

Pi,

-

0.217 + time

dummies

+

0.245

GEN -0.005 * HALF

(0.089)

(0.014)

(0.0002)

+

0.577 OCD

+

0.056

*

DR

-

0.056 AC

-

0.069 * GI

(0.027)

(0.006) (0.006)

(0.007)

-

0.277

WTG,

R2

= 0.8489.

(0.008)

Estimation with a random effects variance

components specification

yielded

same-signed coefficient

estimates, but

generally larger

standard

errors:

64

ln

Pi,

-

0.391

+

time dummies

+

0.125 GEN

-0.002 * HALF

(0.155)

(0.010)

(0.002)

+

0.312 OCD

+

0.064

*

DR

-

0.012 AC- 0.130

*

GI

(0.169)

(0.050) (0.081)

(0.053)

-

0.191 WTG.

(0.060)

Several results are worth

noting.

Consistent

with

all

else

being equal, branded

products

have

higher prices after

facing

generic

competition;

here this effect

is

estimated at

between

12

percent

and

24

percent.

The

a

priori

expectation

on the

sign

of

HALF is am-

biguous,

for short half-life

is

beneficial

to

those

experiencing

serious

side effects

or

adverse

interactions,

but

longer

half-life

may

be

prefer-

able for those who

might

forget

to take

medication,

such as

the

elderly.

64. Note that

fixed-effect

estimation is not

feasible,

because

the

quality

attributes

are fixed over time for

each

drug.

176 Brookings Papers:

Microeconomics 1996

In

this market, the estimated impact is negative,

but it has

statistical

significance only

with

the OLS

estimates. The estimated

parameter

on

OCD is

positive

and

substantial

but of

marginal

statistical

significance

in

the random effects estimation. The a priori

expectation on the sign

of

drowsiness

is

ambiguous,

for a

considerable

number of

depressed

patients

with an

acute episode initially experience

insomnia,

and

thus

for

them the

DR

side effect

is

beneficial;

for

others, however,

it

may

be unwanted. The 0.06 positive estimates here

suggest

that DR

is,

on

balance, valued as beneficial

in

the marketplace.

The a priori sign expectation on the remaining

three attributes is

clearer, and

in

each case both OLS and random effects

parameter

esti-

mates are negative as expected. Specifically, more

frequent

anticholi-

nergic side effects (such as dry mouth, constipation,

urinary hesitance,

and

blurred

vision), gastrointestinal impacts,

and

substantial

weight

gain are each perceived as negative attributes

in

the

antidepressant

marketplace. The estimated impact

of

WTG

is

particularly large and

significant.

With

these estimated hedonic price equations

relating prices

of anti-

depressant drugs

to

their

quality

attributes and

time,

we

have the build-

ing

blocks

necessary

to

construct

a

price

index that links in

quality

change.

Changing Weights,

Hedonic

Adjustments for

Nonpriced Quality

Change,

and

Linking

Generic

Products:

Results

from

a

Merged

Analysis

As noted

earlier,

one

way

to link

nonpriced

quality improvements

into a

price

index is to use as an

approximation

to

the reservation

price

the

predicted price

of a

product just

before

its

market

introduction,

based

on

the estimated hedonic

price equation.

This

predicted price

is

then linked to the actual launch

price

of

the new

product.

When the new product

is

priced

above

(below)

its estimated

hedonic

quality

at

launch,

then the hedonic

residual

in

that month will

be

posi-

tive

(negative),

and the hedonic

quality-adjusted

linked

price

index

will

grow

at a

greater (lesser)

rate than the unlinked index. This

implies

that

residuals

from

estimated hedonic price equations

have an

important

impact,

for

they

are a measure

of

unpriced quality

change.

In

our

sample

of

ten new

products,

the launch

period

OLS hedonic residuals of six

Ernst R.

Berndt, lain

M.

Cockburn, and

Zvi

Griliches 177

Table 6. Effects of Simultaneously Linking Generics and New Products-

Average

Annual

Growth

Rates of Price Indexes

Percentage

Index number Entire period Cycle

A

Cycle

B

Cycle C

procedure 1980:1-1996:2

1981.12-1987:12

1987:12-1993:12

1993:12-1996:2

Paasche-all 7.11 9.87 7.45 2.29

(no link)

Paasche-all

4.73 7.08 4.44

1.69

(GCPD link,

no

hedonics)

Paasche-all 4.33

7.08 3.99

0.52

(GCPD

link

and

hedonics)

Source: Authors'

calculations;

see text for

explanation.

antidepressants

turned out

to

be

negative:

-0.22 for

Asendin, -0.63

for

Luvox, -0.62

for

Wellbutrin,

-0.31 for

Effexor, -0.45 for Zo-

loft,

and -0.07 for Serzone.

With

four new

products, however, the

launch

period

residuals were

positive:

0.03 for

Ludiomil,

0.05 for

Pro-

zac,

0.21 for

Paxil,

and 0.08 for Anafranil.65

Interestingly, while signs

of

residuals were

equally

mixed for

products

introduced

during Cycle

B,

in

Cycle

C all three

of

the new

products

had

negative

residuals

(that

is, positive nonpriced quality improvements).

Using

these

predicted prices

and

residuals,

we have

computed

a

Paasche

aggregate price

index over all

drugs, simultaneously linking

in

the

generics using

the GCPD method

and

accounting

for

nonpriced

quality

differentials

at launch month

for

the new

products;

as

quantity

weights

for

new goods

in this

merged index,

for

the

first

three months

in

the

market,

we

employ

the

average

over those

three

months.

Results

are summarized

in table 6.

The

net effect of our estimated

nonzero

residuals

on

the

growth of

the

aggregate

Paasche

price

index

depends,

of

course,

on

the

quantity

and

price growth paths

of all ten new

products;

note that six

new

products

had

negative

residuals

while four had

positive

ones at

launch

date,

and that all three

of

the new

products

introduced

during Cycle

C

had

negative

residuals.

As

seen

in table

6,

for

the

entire

1980:1-1996:2

time

period,

the

hedonic-adjusted

AAGR is

four-tenths

of

a

percentage

65. These

residuals

are based on the random

effects

model. Similar

findings

resulted

from OLS

estimates, as

well as from other

specifications involving

alternative

quality

attribute measures.

178 Brookings Papers: Microeconomics 1996

point

less

than one not incorporating

hedonic

quality adjustments (4.33

percent, compared

with 4.73

percent).

Because no

new

products were

introduced during Cycle A,

the

hedonic-adjusted

and

unadjusted

AAGRs

are identical

at 7.08

percent. Although the

AAGR of

the he-

donic-adjusted price

index is

about

half a

percentage point smaller

during Cycle

B

(3.99 percent, compared

with

4.44

percent), during

Cycle

C

the AAGR of

the

hedonic-adjusted price index lags more than

a

percentage point

behind the index

not

adjusted

for

quality (0.52 per-

cent, compared

with 1.69

percent). Thus,

somewhat

surprisingly,

link-

ing in generics has a much more substantial

effect on AAGRs of

price

indexes during Cycle B, when generic entry

was

substantial, than does

accounting

for

estimated nonpriced quality

differentials.

In

Cycle C,

however, the hedonic adjustment results

in

a

larger price decline than

does the

linking

in

of

generics.

Discussion

This detailed

audit

of

the IMS universe

of

antidepressant drug prices

reveals substantial differences between

the AAGRs

of the

published

BLS

price

indexes and

those

computed

in

a

variety

of

alternative

ways.

We

now

define our

"audited"

price

index with

generics

linked

in

using

the GCPD

procedure,

hedonic

nonpriced quality changes included,

chained

Paasche

weights,

and the IMS universe of

antidepressants

as

the

data

base. As the

bottom line of table 7

shows, during Cycle

B

this

audited

price

index

grows

at an

AAGR of

3.99

percent,

whereas

the

published

BLS

PPI has an

AAGR

of 10.40

percent.

For

Cycle C,

the

audited

price

index

has an AAGR of

0.52

percent,

much smaller than

the 4.27

percent

AAGR of

the

published

BLS

index.

In both

periods,

the

BLS

index overstates

price

inflation

by

a

very

substantial

amount.

It

is useful to summarize the sources

of

these differences

in

growth

rates,

which

we

do in tabular form

in

table

7.

The

first

source

of differences concerns

sample representativeness

of

the

fixed

weights.

If we

employ

the

BLS

Laspeyres fixed-weight pro-

cedures,

but instead

of

utilizing

the

BLS

item

sample,

we use

the IMS

universe

of

drugs,

we

obtain results

given

in

the second

row

of

table 7.

The

difference is

very large during

both

Cycles

B

and

C

and,

as noted

earlier,

reflects

the

fact that universe

item

weights

set

at

1987:12 and

Table

7.

Explaining

Differences

in

Average

Annual

Growth

Rates

of

Prices

for

Antidepressant

Drugs

Percentage

Entire

period

Cycle

A

Cycle

B

Cycle

C

Procedure

1980:1-1996:2

1981:12-1987:12

1987:12-1993:12

1993:12-1996:2

BLS

index

NA

10.40

4.27

Sampling

procedures

Sample

vs.

universe

5.04

9.35

4.17

0.42

weights

Weights

Divisia-chained

7.51

10.35

7.90

2.34

Paasche-chained

7.11

9.87

7.45

2.29

Laspeyres-fixed

5.04

9.35

4.17

0.42

Laspeyres-chained

7.88

10.83

8.24

2.38

Linking

generics

FDA

average

price

2.95

5.71

1.33

1.10

GCPD

4.73

7.08

4.44

1.69

New

BLS

3.71

7.41

2.49

0.42

Linking

generics

and

hedonics

GCPD

4.33

7.08

3.99

0.52

Source:

Authors'

calculations:

see

text

for

explanation.

180

Brookings

Papers:

Microeconomics 1996

1993:12

would

subsequently

overweight

generics

whose prices

were

falling,

and

underweight

brands whose

prices

were

increasing. More

frequent

supplemental

sampling could

help

to reduce this

discrepancy,

although issues

concerning

Puerto Rican

production

would still

66

remain.

The

second

source of

differences

involves whether

fixed or

changing

weights

are used.

As

shown

in

the

third

panel

in

table

7, if

one

uses

the Divisia

or

Paasche index, based

on

monthly updates instead

of

the

relatively

fixed-weight Laspeyres

(changed

only

every

six

years),

over

the

IMS universe

of drugs, the AAGR rises from

3.3 to 3.7

percentage

points

during

Cycle B,

and about 1.9

percentage points

during

Cycle

C.

Weights

are

very important.

Obtaining

data

necessary

to

update

weights

more

frequently

is

of

course

possible,

but

not without costs. To obtain

some

evidence

on

the

possible

benefits

of

updating

weights

more

frequently,

we have

calcu-

lated AAGRs

of

Laspeyres price indexes over

the IMS

universe of

products

when,

like that for the chained Divisia

and

Paasche,

the

weights

are updated

monthly.

As

seen

in

the

bottom row

of

the

middle

panel

in

table

7,

when

Laspeyres weights

are chained on

a

monthly

basis,

AAGRs are

very

close but

slightly

larger

than those of the Div-

isia-7.88

percent, compared

with 7.51

percent over the

entire period.

Thus

what

is

empirically significant

is

failure

to

update

weights more

frequently, rather than choice

of

index number

formula.67

A

third source of differences concerns the

linking

of

generics

to

their

patented

antecedents.

As shown in the

fourth

panel

of

table

7,

the

effects of

linking

are substantial. Over the

last

sixteen

years,

the

AAGR

using

the FDA

procedure

is more than four

percentage

points

less than

that

of

a

comparable

Paasche index that

does

not link in

generics

(2.95

percent, compared

with

7.11

percent).

During

the same

period,

the

effect of

the

new BLS

procedure

is

only

slightly

smaller

(3.71

percent,

compared

with

7.11

percent).

During Cycle

B

the

new linked

procedure

would

have

generated

a

price

series

that

grew

at

2.49

percent

a

year,

almost

eight

percentage points

slower than its

published

fixed-weight

Laspeyres

(10.40

percent).

Because

patents

on

SSRI

drugs

in

the

United

66. See

footnote

47.

67. IMS America and the PRIME

Institute

construct

aggregate price

indexes

that

they

distribute

to

clients;

these indexes use

Laspeyres

index

number

procedures with

weights updated each

year, but generic products

are not linked to

predecessor brands.

Ernst R.

Berndt,

Iain

M.

Cockburn, and Zvi

Griliches

181

States all

expire after the year

2000, the

new

BLS policy that links in

generics

will not

have

any

measurable

impact

in this

therapeutic

class

for

some time,

but

in

other

therapeutic

classes

where patent

expiration

and

generic

entry

is more

extensive,

the effects could be

considerable.

The fourth

and final source

of differences

involves nonpriced

quality

differentials embodied

in

new

goods.

As shown in

the bottom

panel

of

table

7, during

Cycle

B

nonpriced

quality

results

in

a modest

difference

(less than half a

percentage

point for the

GCPD), but

during Cycle C

the effect of

accounting

for

nonpriced quality becomes

quite

substantial

(the

AAGR

is 1.2

percentage points

less than

the GCPD

index).

Al-

though

accounting

for

nonpriced

quality

change plays

a

significant mod-

erating role

in

the

growth

of

prices

in

the market

for

antidepressant

drugs,

the

impact

is

smaller

than that

in

other

high-tech markets.

For

example,

in

the

personal

computer market the

price

of

a

typical model

has been about

$2,500

for

more

than a

decade,

but that

$2,500

now

buys

much more

performance

in an

ever smaller

footprint.

As has

been

shown

elsewhere, quality-adjusted

personal

computer prices decline at

almost

30

percent

a

year,

more than double that for

price

indexes not

adjusted

for

quality change.68

Conclusions and Issues for Further Research

The market

for antidepressant

drugs

is

dynamic. Of the

twenty-one

entities

on

the market

in

1996, eight

are

entirely new,

having

been

introduced to the

market

within

the last

decade,

and an

additional

seven

brands

have

experienced

new

generic

competition

following patent

expiration.

Tracking prices and then

constructing

aggregate price

indexes

for

such

a

rapidly

changing

marketplace

are

challenging tasks,

particularly

for statistical

agencies

such as the

BLS

whose resources

are

tightening.

In

this

paper

we have

audited

the

reliability

of the BLS

producer

price

index in an

admittedly dynamic

market that

presents

enormous mea-

surement

challenges.

We

find

major

differences between the

published

BLS

numbers

and

the results

we obtain from

our

audit;

in

both

Cycles

68. For further discussion, see Berndt and Griliches (1993) and Berndt, Griliches,

and Rappaport (1995).

182

Brookings

Papers:

Microeconomics 1996

B

and C,

audited growth rates

are less

than

half those

published by the

BLS.

Of the

four

sources of difference that we have

examined

in

detail,

one-nonpriced

quality

changes

embodied

in

new

goods-plays

a mod-

est

to

significant

role

depending

on the time

period,

and

another-the

linking of new

generic products

to

their patented

antecedents-is

one

on

which the BLS has

just

recently announced

a

major

policy change.

The other two

sources

of

differences

(nonrepresentative

sampling, and

use

of

fixed-weight formulas)

could be

addressed

by

the BLS

obtaining

more frequent

(and more

costly)

information,

and

using

this data to

update

its

weights

more often.

Concerning

sample

representativeness,

one

issue that merits

greater

attention is

the treatment of

production

from

Puerto Rico. In

some of

the

national

accounts, national

production

includes Puerto

Rico;

in

others

such

production

is excluded. For

pharmaceuticals,

this

issue is

significant

because Puerto Rican

pharmaceutical

production

is

roughly

a fifth

to a

quarter

of

that

from

the U.S. mainland. If

Puerto Rico is to

be

excluded,

as is now the case for the

antidepressant PPI,

then to

the

extent

public

policy analysts

and others

seek

to

track the

price

growth

emanating

from U.S.

producers

(many

of whom

choose to

produce

significantly

in

Puerto Rico),

it will

be

necessary

to

collect

and

publish

"import" price series

from

Puerto

Rico,

and then to

combine those

data with

the more

narrowly

defined "domestic"

mainland

price

series.69

In

this

paper we have examined alternative measures of

price

growth,

but

we have not addressed the reasons

underlying

this

price

growth,

nor

have

we

attempted

to

model

quantities

sold

by

manufacturer and

the remarkable

growth

of the entire

therapeutic

class. It is

worth

em-

phasizing

that the measures of

price growth

presented

here

are

not

purported

to be

closely

related to measures

of

economic

welfare

and

consumers'

surplus.

There is reason to believe that

measures of

price

growth

employing predicted

prices

from

estimated hedonic

price equa-

tions could

differ

considerably

from

exact

price

indexes

that

employ

69. One incentive for Puerto

Rican

production

has

been

Section 936 of the

Internal

Revenue

Code,

which

has

provided

tax

benefits to firms

producing

in

Puerto

Rico. It is

worth

noting

that under the 1996 federal minimum

wage bill,

these

tax incentives will

be

phased

out over the next decade.

Ernst R. Berndt, Iain M. Cockburn, and Zvi Griliches 183

reservation prices based on estimated structural demand models.70

Pakes has argued that under a plausible set of conditions, the hedonic-

based prices provide an upper bound

to

growth

of

an exact price index .

It is worth noting that innovations providing new varieties of a product

such as antidepressant drugs have

an

unambiguous

beneficial

welfare

implication in that consumers are given the choice of another product,

one that

may

"work" for

them,

while

others

have not.

Although our analysis here has been confined

to

the market

for

an-

tidepressant drugs, our results

on

the importance

of

linking generics to

their patented antecedents may

have

implications

for

nonpharmaceuti-

cal markets.

For

example,

in the consumer electronics

industries,

new

products from branded manufacturers typically embody quality

im-

provements; these manufacturers subsequently often

sell a

virtually

identical product under a private label

or "knock-off" brand at a

much

lower price.72 To the extent that such lower-priced

versions are

not

linked to branded antecedents, the price indexes

will

fail to

incorporate

these implicit price

declines realized

by

some consumers.

Our

research

could be extended

in a number

of

ways. First,

our data

is from

the IMS, and

it

would

be useful

to

compare

the

IMS

price

data

with

actual transactions data

from the

pharmaceutical

manufacturers,

although

our

earlier

research

on this issue did not reveal

any systematic

differences.73

Second,

as

implied earlier,

it would be informative to

model much more completely

the diffusion

process

of

antidepressant

drugs, both

over time and

among

different market

segments,

such as

drugstores

and

managed

care

organizations.

It

would also be useful

to

assess how the various

price

indexes

change

when the

frequency

of

the

data

is

reduced

from

monthly

to

quarterly. Third,

we

have examined

here only

the

drug component

for the treatment

of

depression, entirely

ignoring

"substitutable"

inputs

such

as talk

therapy. Assessing

the

effects

of

the new

generation

of

SSRIs, along

with

increased

pressures

of

cost

containment,

on

the

changing

mix of

drug

and talk

therapy

in

the

overall treatment

of

depression

is

a most

interesting

avenue for

future research.

70. For

discussion,

see

Triplett (1983).

71. Pakes (1996).

72.

Several other examples of

this

brand-generic phenomenon

are

discussed

in

De-

neckere-McAfee (1996).

73. Berndt, Griliches, and Rosett

(1993), and Berndt and Greenberg (1996).

184

Brookings Papers: Microeconomics

1996

References

Advisory

Commission To Study the Consumer

Price Index. 1996. Toward a

More

Accurate

Measure of the Cost of Living.

Final

report

to

the Senate

Finance

Committee. Washington, D.C. December

4.

American Medical

Association. 1991. "Drugs

Used

in

Mood

Disorders."

Drug

Evaluations, 7th ed.

American

Psychiatric Association, Committee on Nomenclature and

Statistics.

1968.

Diagnostic and Statistical Manual of Mental

Disorders,

2d

ed. Wash-

ington, D.C.

American Psychiatric

Association,

Task

Force on

Nomenclature and Statistics.

1980. Diagnostic

and Statistical Manual of Mental

Disorders,

3d ed. Wash-

ington, D.C.

American Psychiatric Association. 1987.

Diagnostic

and Statistical Manual of

Mental Disorders:

DSM-III-R,

3d

ed.

rev.

Washington, D.C.

.

1993. "Practice

Guidelines

for

Major

Depressive Disorder

in

Adults."

American Journal

of Psychiatry

150

(4, April,

Supplement):

1-26.

Baldessarini, Ross

J. 1990. "Drugs

and the

Treatment

of

Psychiatric Disor-

ders."

In

The

Pharmacological

Basis

of

Therapeutics, edited by Alfred

Goodman Gilman and

others,

383-435.

8th

ed.

New

York:

Pergamon

Press.

Berndt, Ernst R., and Paul E.

Greenberg. 1996.

"An

Updated and Extended

Study

of

the Price Growth

of

Prescription

Pharmaceutical

Preparations."

In

Competitive

Strategies

in

the

Pharmaceutical

Industry,

edited

by

Robert

B.

Helms,

35-48.

Washington, D.C.:

American

Enterprise

Institute.

Berndt,

Ernst

R.,

Richard

G. Frank,

and Thomas G.

McGuire.

forthcoming.

"Alternative Insurance

Arrangements

and the

Treatment

of

Depression."

American Journal

of Managed Care.

Berndt,

Ernst

R.,

and Stan N.

Finkelstein. 1992.

"Price

Indexes

for Anti-

Hypertensive

Drugs

that

Incorporate Quality

Change:

A

Progress Report

on

a

Feasibility

Study." Unpublished working paper

6-92. MIT

Program

on the

Pharmaceutical

Industry, Cambridge,

Mass.

December.

Berndt,

Ernst

R.,

and Zvi

Griliches. 1993. "Price Indexes for

Microcomputers:

An

Exploratory

Study."

In Price Measurements and Their

Uses,

edited

by

Murray

F.

Foss,

Marilyn

E.

Manser,

and

Allan

H.

Young,

63-93.

University

of

Chicago

Press

for

the

National

Bureau

of Economic

Research.

Berndt,

Ernst

R.,

Zvi

Griliches,

and Joshua G.

Rosett. 1993.

"Auditing

the

Producer Price Index:

Micro

Evidence

from

Prescription

Pharmaceutical

Preparations.

"

Journal

of

Business and Economic Statistics

11

(3, July):

251-64.

Berndt,

Ernst

R.,

Zvi

Griliches,

and

Neal

J.

Rappaport.

1995.

"Econometric

Estimates

of

Price Indexes

for

Personal

Computers

in

the 1990's." Journal

of

Econometrics 68

(1, July):

243-68.

Ernst

R.

Berndt, lain

M.

Cockburn, and

Zvi

Griliches 185

Bureau

of

the Census,

U.

S. Department

of

Commerce.

1992.

1992

Census of

Manufactures, Industry Series Drugs. MC92-I-28C. Washington, D.C.

. 1994.

1992

Economic Census of Outlying Areas, Industry Statistics-

Puerto Rico. Washington, D.C.

Bureau of Economic Analysis, U.S. Department of Commerce. 1985a. "Busi-

ness Situation.

"

Survey of Current

Business 65

(10, October):

1-7.

. 1985b. "Revised

Estimates of the National Income

and

Product Ac-

counts of the

United States,

1929-85:

An Introduction."

Survey of

Current

Business 65 (12, December): 1-19.

. 1987. "Foreign Transactions."

In

Methodology Papers:

U.S.

Na-

tional Income and Product Accounts.

Washington,

D.C.

May.

Bureau

of

Labor Statistics, U.S. Department

of Labor.

1996.

Producer

Price

Indexes:

Data

for January

1996.

Washington,

D.C. June.

Caves,

Richard

E.,

Michael D.

Whinston,

and Mark A. Hurwitz.

1991. "Pat-

ent

Expiration, Entry,

and

Competition

in

the

U.S.

Pharmaceutical Indus-

try." Brookings Papers

on Economic

Activity,

Microeconomics:]. 1-48.

Cleeton,

David

L., Valy

T.

Goepfrich,

and Burton

A.

Weisbrod.

1992. "What

Does

the Consumer

Price Index

for

Prescription Drugs Really

Measure?"

Health

Care Financing

Review

13

(3, Spring):

45-51.

Deneckere, Raymond J., and

R.

Preston McAfee. 1996. "Damaged Goods."

Journal of Economics and Management Strategy

5

(2, Summer): 149-74.

Depression Guideline

Panel.

1993. Depression

in

Primary

Care.

Vol.

2:

Treat-

ment

of Major Depression.

Clinical Practice

Guideline,

No.

5, Rockville,

Md.:

U.

S. Department

of Health and Human

Services,

Public

Health Ser-

vice, Agency

for Health

Care Policy

and

Research,

AHCPR Publication

93-0551.

Early,

John

F.,

and James

H.

Sinclair. 1983.

"Quality Adjustment

in

the

Producer

Price Increases."

In The

U.S.

National

Income and Product Ac-

counts: Selected

Topics,

edited

by Murray

F.

Foss. Studies

in

Income

and

Wealth,

vol. 47.

University

of

Chicago

Press

for

the

National

Bureau of

Economic Research.

Eisenberg,

Leon.

1992. "Treating Depression

and

Anxiety

in

Primary

Care:

Closing

the

Gap

Between

Knowledge

and Practice." New

England

Journal

of

Medicine 326

(April 16):

1080-84.

Elkin, Irene,

Morris

B.

Parloff,

and others.

1985.

"NIMH

Treatment of

Depression

Collaborative Research

Program."

Archives

of

General

Psychiatry

46

(Novem-

ber):

971-83.

Fisher,

Franklin

M.,

and Karl Shell. 1971.

"Taste and

Quality Change

in

the

186

Brookings

Papers:

Microeconomics

1996

Pure Theory of

the

True-Cost-of-Living Index."

In

Price

Indexes and Qual-

ity Change:

Studies

in

New Methods

of

Measurement, edited

by

Zvi Gril-

iches, 16-54.

Harvard

University Press.

. 1972.

The Economic Theory

of

Price Indices: Two

Essays on the

Effects of Taste

Quality and

Technological

Change. New

York:

Academic

Press.

Fisher, Franklin

M., and

Zvi

Griliches. 1995.

"Aggregate Price

Indices, New

Goods,

and

Generics."

Quarterly

Journal

of Economics

110

(1, February):

229-44.

Food and Drug

Administration.

annual.

Approved

Drug Products

with Thera-

peutic

Equivalence

Evaluations ("The Orange

Book").

Washington, D.C.:

U.S.

Department

of Health and Human

Services,

Public Health

Service,

Center

for

Drug

Evaluation and

Research,

Office of

Management.

Frank, Richard

G., and David S. Salkever.

1992.

"Pricing, Patent

Loss, and

the

Market for

Pharmaceuticals."

Southern Economic

Journal

59 (2, Octo-

ber):

165-79.

Frank, Richard

G., and

Willard

G. Manning, Jr.

1992. Economics

and Mental

Health.

Johns

Hopkins

University

Press.

Grabowski,

Henry G.,

and John

M.

Vernon. 1992. "Brand

Loyalty, Entry,

and

Price

Competition

in

Pharmaceuticals

after the

1984

Drug

Act.

"

Journal

of

Law and

Economics

35

(2,

October):

331-50.

Greenberg,

Paul

E.,

Laura E.

Stiglin,

and

others.

1993.

"The

Economic Bur-

den of

Depression

in

1990.'"

Journal

of

Clinical

Psychiatry

54

(11,

Novem-

ber):

405-18.

Greenberg, Paul

E.,

Ronald

C.

Kessler,

and

others. 1996.

"Depression

in

the

Workplace:

An Economic

Perspective."

In

Selective

Seratonin

Reuptake

Inhibitors:

Advances in Basic Research and Clinical

Practice, edited

by

J.

P.

Feighner

and

W. F.

Boyers,

321-58. 2d ed.

Chichester,

U.K.:

John

Wiley

and

Sons.

Griliches, Zvi,

ed. 1971.

Price

Indexes and

Quality Change:

Studies

in

New

Methods

of

Measurement. Harvard

University

Press.

.

1990. "Hedonic Price Indexes and the

Measurement

of

Capital

and

Productivity: Some

Historical

Reflections.

"

In

Fifty

Years

of

Economic

Mea-

surement: The Jubilee

of

the

Conference

on Research in

Income and

Wealth,

edited

by

Ernst R.

Berndt

and

Jack

E.

Triplett.

NBER

Studies

in

Income

and

Wealth,

vol. 54.

University

of

Chicago

Press

for

the

National

Bureau

of

Economic Research.

Griliches, Zvi,

and lain

Cockburn. 1994.

"Generics and

New Goods

in

Phar-

maceutical Price Indexes." American Economic Review 84

(5,

December):

1213-32.

Hicks,

John R. 1940.

"The Valuation

of Social

Income." Economica 7

(26,

May):

105-24.

Ernst R. Berndt,

Iain

M.

Cockburn, and

Zvi

Griliches 187

Hurwitz, Mark A., and Richard E. Caves. 1988. "Persuasion

or

Information:

Promotion and the Shares of Brand Name and Generic

Pharmaceuticals."

Journal of Law and Economics

31

(October): 299-320.

Hyman, Steven E., George W. Arana, and Jerrold

F.

Rosenbaum.

1995. Hand-

book of Psychiatric Drug

Therapy,

3d ed.

Little,

Brown and

Company.

IMS America. 1993. National Disease and Therapeutic Index-Drugs,

July

1993-June 1994. Plymouth Meeting,

Pa.

.

1996a.

National

Disease

and

Therapeutic

Index-Disease, July 1995-

June 1996. Plymouth Meeting,

Pa.

.

1996b. Information

Services Manual.

Plymouth

Meeting,

Pa.

Jonsson, B., and

J.

Rosenbaum. 1993.

Health

Economics

of

Depression. Vol.

4:

Perspectives

in

Psychiatry.

Chichester,

U.K.:

John

Wiley & Sons.

Kanoza, Douglas. 1996. "Supplemental Sampling

in

the

PPI

Pharmaceuticals

Index." Producer

Price Indexes Detailed

Price

Report (January):

8-10.

Katon, Wayne, and others. 1992. "Adequacy

and Duration of Antidepressant

Treatment

in

Primary Care." Medical Care

30

(January):

67-76.

Keith, Alison M., and Ernst R. Berndt. 1994.

"A

Primer

on Issues

in

the

Measurement of Price Change

and

Price Impact:

An

Application

to Phar-

maceuticals." Working Paper

25-94.

MIT

Program

on

the Pharmaceutical

Industry, Cambridge, Mass. September.

Masson, Alison, and Robert

L.

Steiner.

1985. Generic Substitution and Pre-

scription

Prices: Economic

Effects of

State

Drug

Product

Selection Laws.

Washington, D.C.: Federal Trade Commission,

Bureau of Economics.

National Association of Chain Drug Stores. 1995. "NACDS

PRIME

Index

Reports Increases in Prices

for

Top

500

Prescriptions

Dispensed

in

Com-

munity Pharmacies." News release, Alexandria, Va., December

7.

Nolen-Hoeksema, Susan. 1990. Sex Differences

in

Depression.

Stanford Uni-

versity Press.

Pakes, Ariel. 1996. "Hedonic

Bounds

to

Exact

Price

Indexes." Unpublished

draft

manuscript.

Yale

University, Department

of

Economics.

May.

Papolos, Demitri F.,

and Janic

Papolos.

1992.

Overcoming

Depression, rev.

ed.

Harper

Collins

Publishers.

Physicians' Desk Reference

Generics. 1996.

2d

ed. Montvale, N.J.: Medical

Economics

Company.

Potter, William Z.,

Matthew V.

Rudorfer,

and Husseini Manji. 1991. "Drug

Therapy:

The

Pharmacologic

Treatment

of

Depression."

New

England

Jour-

nal

of

Medicine

325

(9, August 29):

633-42.

Regier, Darrel A.,

and others.

1988. "The

NIMH

Depression

Awareness,

Recognition,

and Treatment

Program: Structure, Aims,

and Scientific Ba-

sis.

"

American Journal

of Psychiatry

145

(November):

1351-57.

Rothbarth, Erwin. 1940-41.

"The Measurement

of

Change

in

Real Income

under Conditions

of

Rationing."

Review

of

Economic Studies

8

(2):

100-7.

188

Brookings Papers. Microeconomics 1996

Suslow, Valerie Y.

1996. "Measuring Quality

Change

in

the

Market for

Anti-

Ulcer Drugs."

In

Competitive Strategies

in

the Pharmaceutical

Industry,

edited

by

Robert

B.

Helms.

Washington,

D.C.: American

Enterprise

Insti-

tute.

Triplett, Jack E.

1983. "Concepts of Quality

in

Input

and

Output Price Mea-

sures: A

Resolution of the User-Value Resource-Cost Debate."

In

The

U.S.

National

Income and

Product Accounts:

Selected

Topics,

edited

by Murray

F. Foss.

University of Chicago Press for the National

Bureau

of

Economic

Research.

Turkington, Carol

Ann, and Eliot

F.

Kaplan. 1994.

Making the Prozac Deci-

sion:

Your

Guide

To

Antidepressants.

Los

Angeles:

Lowell

House.

U.

S. General

Accounting Office.

1996. Consumer

Price

Index:

Cost-of-Living

Concepts

and the

Housing

and Medical

Care

Components. Report

to

the

Ranking

Minority Member, Committee

on

Banking

and

Finanical Services,

House of

Representatives, GAO/GGD-96-166.

Washington, D.C. August.

Comments

Comment

by

Theodore E. Keeler:

This

is

an

interesting paper

with

important and

worthwhile contributions.

Many complain about

rapidly

rising costs in

the

health

care sector, yet

because

of

the complexity of

the product and

changes

in

technology

and

product quality,

it is

espe-

cially difficult

to

determine whether these

rapid

cost increases

stem

from higher

prices, higher quantities,

or

quality improvements.

The

present paper

adds valuable insight into this area.

The results

are

of

special interest

in that

they indicate the

importance

of

understanding

the value

of

improved products

and

of

reduced

costs

from generic

products.

The

results

show

also

the importance

of

quality

change and represent innovative use

of

hedonic

price indexes, and

they

represent a clear

improvement

over

conventional

BLS

price indexes.

Overall, then,

I

find

much

to

agree

with

in

this

paper.

Nevertheless,

I

do have

an

area

of

concern:

specifically,

all

the index

number theory

employed by

the

authors of this

paper

assumes that

prices

of each

drug

represent

a

consumer's

marginal willingness

to

pay

for

those

drugs.

For

many people, however,

prescription drugs

are

paid

for

through

health

insurance,

and the existence

of

moral hazard

combined

with

reasonably generous health insurance

polices

can call into

strong

question

the

validity

of the

simple proposition

that

prices represent

consumers'

marginal willingness

to

pay

for the relevant

products.

It is

easy

to see

why

a

study

such as

this

would

abstract from

the matter of

moral hazard with

insurance, because

the

likely

effects are

complicated,

and the

present

study

innovates

in other

directions.

Nevertheless,

I

believe

it

is worthwhile

to

try

to

understand

some-

thing about these

effects,

because

they

are

potentially important. Any-

189

190

Brookings Papers:

Microeconomics 1996

thing

near a

complete

or

rigorous

discussion of

these

effects are

well

beyond

the

scope

of these comments and would be a

paper

in

itself; my

goal, instead,

is to sketch

very

crudely

and

intuitively

some

likely

effects

of

moral hazard

for this

analysis, suggesting,

I

hope,

both

the

importance

of

the issue

and

the need

for

further

research.

Moral Hazard and Pharmaceutical Price Indexes

Health insurance

will

likely

exaggerate

a consumer's

apparent

mar-

ginal willingness

to

pay

for newer or more expensive drugs. Take the

case

that the authors make

of

the

introduction of

a generic substitute

for an

existing

branded

drug.

Because

the two

drugs are therapeutically

the

same,

one would

expect patients

to

move in

large

numbers to

the

generic

if

the

branded

drug

costs much more. In

fact,

that

does not

always happen.

The

authors of this

paper and Griliches

and

Cockburn

in

an

earlier

(1994)

article are

puzzled

slightly by the

fact

that

even if

the

price

of a

branded

drug

is much more than

that of the

generic,

the

branded

drug

maintains much more market share than one

might expect.

I

strongly

believe that

health

insurance

should

explain

at least

some of

that:

if the

patient pays only

10-20

percent

of

the difference

in

prices

(or

even

less),

then

the

patient

might

well

prefer

the

branded

product,

if

it is thought to be

even

modestly

superior.

A

study

that

ignores

this

fact

may

tend to overstate consumers'

true

willingness

to

pay

for

more

expensive products

of

any type.

To

see this

point

from a

slightly

different

perspective,

consider

the

following grossly simplified

case. Recall

from

the

paper

that

one rela-

tively

accurate

way

to

incorporate

a

new

product

into

a

price

index

is

to

include the

reservation price

of a

"typical" consumer

for

the

product

just

before it is introduced. BLS indexes are asserted

to

overstate in-

creases

in

these

prices

in

part

because

they

fail to

incorporate

these

relatively high

reservation

prices

at

the earliest

time of

introduction of

the

product.

Moral hazard

may

cause

this

estimated reservation

price

to

be over-

stated,

however. To see

this,

consider

the effects

of the

introduction of

a

new

(expensive) drug

with the

simplest (but

most

expensive) likely

form

of

health

insurance,

that which

simply

reimburses the

provider

(doctor, drug store, hospital,

and so

on)

a

fixed

percentage (often

75 to

Ernst R. Berndt, lain M. Cockburn, and Zvi Griliches 191

80 percent) of total expenses (with no overall limitation on them), so

that the patient makes a copayment of, say, 20 percent and the insurance

company pays the rest. If, in fact, the consumer's willingness to pay

per dose is $1, then, with the

introduction

of

insurance, the observed

reservation price (including that observed in econometric studies such

as are included in this paper)

will

be $1.00/0.2

=

$5-much higher

than the real reservation price.

Any discussion

of

biases

in

an

estimated

index

would have

to

deal

with relative changes in prices over time, and real-world insurance

is

also

more complicated-involving

issues that

go beyond my

comments

here. But it would seem quite possible

that moral

hazard

could

easily

bias

upward the estimate

of a

person's reservation price

for a new

drug,

which could clearly affect calculations of price indexes.

To make our analysis more realistic, consider

the

case

of two sub-

stitute drugs, a branded one and

a

new generic. Assume also that res-

ervation

price differentials

are distributed

among patients,

so that some

will

switch to the generic at a

low

price differential, but others require

a

higher differential. This realistic situation

is the

one on

which

the

authors have built up a new

Paasche

index

of

pharmaceutical prices.

Once again, however,

moral hazard could

easily

render the use

of

direct

comparison

of

market prices

inaccurate.

Suppose

that

one

particular

consumer is observed to be willing

to switch

from

the brand to the

generic at a price difference

of

$1 per daily

dose

(that is,

if

the brand

costs $2, the consumer

will

buy

the

generic only

if

its

price

falls

to

$1

a

dose).

In

reality, though,

if

insurance

is

covering

80

percent

of the

difference, the real reservation price

differential

is

only

20

cents:

ob-

served market

data will

exaggerate

the

apparent willingness

of the con-

sumer to pay for the more expensive drug.

If

this

simplified

view

of

reality

is

correct,

it has

clear

implications:

consumer preferences

could

easily

be closer

to the FDA

approach

of

considering

branded and

generic drugs equivalent

than the observed

data

would

suggest.

This view

also

implies

that the

coefficients

in the

hedonic price equations might

well

also

overestimate

the

consumer's

marginal willingness to pay

to obtain the

positive aspects

of a

drug,

or

to

avoid the negative effects,

relative to the consumer's true

prefer-

ences.

Why

not

buy expensive drugs

with fewer side effects

if insurance

is

paying for

it?

192 Brookings Papers:

Microeconomics 1996

Theoretical and Empirical Validity of Moral Hazard

in

Pharmaceuticals Consumption

Most health economists are convinced

that moral

hazard

is

rampant

in the fee-for-service part

of

the

U.S. health care

sector,

but

not

every-

one might agree.

After

all, people

seem to want

to

buy

health insur-

ance-in some meaningful sense, maybe

what their insurance

pays

is

a

measure

of

the value they

attach to medical care. Where

is

the market

failure? There are

strong

reasons

to

question

that.

First,

the

tax

benefits

accorded

to

employer-provided

health care have

historically

tended

to

cause overprovision

of

it; second, during

the 1960s and 1970s, Blue

Cross and

Blue Shield had

a

largely

artificial

monopoly

on U.S. health

insurance, and they represented the interests

of the

providers

(doctors

and hospitals) who controlled

them.

'

For various reasons they endea-

vored to continue the existence of

this

type

of

fee-for-service

indemnity

insurance. The growth of managed

care in the 1980s and 1990s is a

sign that over the very long pull,

the

marketplace

will not

support

this

moral hazard (more about

that

below),

and

it is

likely

to

be less

impor-

tant today than

it

was in 1980,

the first

year

of the

study.

In health care as

a

whole,

there is

ample

evidence

that

American

patients

are indeed

prone

to moral hazard

in a

fee-for-service

setting.

This evidence

is

based not only

on

econometric studies,

but also

on

a

large, expensive, controlled experiment conducted by the

RAND

Cor-

poration,

in which

a

carefully

selected

sample

of the

population

was

given varying amounts

of

insurance,

and

the

amount of health

care

then

consumed was observed.2

There

is, however,

less evidence

as

to

whether

expenditures

on

phar-

maceuticals (typically covered by insurance) are prone

to moral

hazard.

One

study,

that

of

Sean Sullivan

(1992),

based

on

the

elderly

with and

without

pharmaceutical supplemental

insurance to

Medicare,

provides

strong evidence

that

patients

are indeed

responsive

to

pharmaceutical

prices, and

that moral hazard does exist

in

this

area.3

Specifically,

the

1. For a discussion of these forces

in health

insurance, see,

for

example,

Feldstein

(1993, pp. 149-69).

2. For a survey

of

the effects

of moral hazard

in health

insurance

on

the demand for

health care in general, see any good textbook

in

health economics, such as Feldstein

(1993, pp. 74-105). For results of the RAND experiment, see Manning and others

(1987).

3. Sullivan (1992).

Ernst R. Berndt, lain

M. Cockburn, and Zvi

Griliches

193

study

found that

controlling

for

many

other

variables, patients with

drug insurance spend

12.5 percent a year more on

drugs

than

do patients

without such insurance. Another recent

study, however, found that

insurance status of

patients does not

significantly affect

whether the

patient

is

prescribed

a

generic

or brand

drug, except

that

HMO

patients

are

more

likely

to be

prescribed generic

drugs (see below).4

The

data

set,

however,

did not

include basic variables

such as

drug prices.

Rika

Mortimer,

a Ph.D.

student

at

Berkeley,

is

currently

investigating

this

very issue but has no

results

as

yet.

Managed Care

Given the costs of moral

hazard,

it

is

reasonable

to

expect

that

the

market would provide

forms

of insurance that

avoid

it.

Health

mainte-

nance

organizations

(HMOs) clearly

do have an incentive to

use less

expensive

drugs

and to avoid moral hazard. That

is

because

they

collect

a

fixed

(capitation)

fee

per patient per

unit of

time,

and

the

less

they

spend

on health

care,

the

more

money

they

make.

Empirical

evidence

indeed

shows

that

patients

in

HMOs

do

indeed use

generic

drugs

more

than

do

patients using

more conventional

indemnity

insurance with fee-

for-service reimbursement.

This

fact has

implications for the authors'

empirical

work.

Specifi-

cally,

it

implies

that

in

years

in which HMOs

had

a

greater

market

penetration,

the effects

of

moral hazard on their hedonic

coefficients

are

likely

to

be weaker. The market

share

of HMOs has

indeed

grown

from

1980

to the

present.5 That means that the

observed

willingness

to

pay

for

various attributes

(and

hence

the

coefficients)

may

well

have

shifted

substantially

over

time.

It would seem to make

sense

to

test for

that

possibility,

but

to

keep

in

mind

that this

change may

represent

not

changes

in

consumer

preferences,

but

changes

in insurance

reimburse-

ment.

4. Hellerstein

(1994).

5. According to

Oberlander (forthcoming), as of

1995,

58

percent

of

insured em-

ployees of firms with

more than ten employees belonged to

indemnity or PPO plans,

whereas fully 42

percent belonged to HMOs or point-of-service

plans.

In

1994, 62

percent belonged

to

indemnity plans

or PPOs. In

contrast,

in

1990

HMO

penetration

rates were only 15

percent or so. HMO penetration of Medicare

is

currently

still mini-

mal-only 10 percent

of Medicare enrollees are enrolled

in

any

form of

managed care.

194

Brookings

Papers:

Microeconomics

1996

An

Opposite

Bias

with

Fee-for-Service

Insurance

In the

specific

case of

antidepressants,

it is

possible

to find

an ex-

ample with

fee-for-service

reimbursement

in which

the

techniques used

by the authors

might possibly

underestimate

the

marginal

willingness

of

consumers

to pay

for

drugs.

To see this,

recall that

talk

therapy is

often viewed as a substitute

for

drugs.

Furthermore,

talk

therapy

is

quite expensive relative

to

many drugs (a

once-a-week

visit to a

psy-

chiatrist can

average out

to

$15 to $20 a

day-more

than the cost of

the daily dose

of many drugs).

If

talk

therapy were

covered

by insur-

ance, patients

might prefer

a

talk-intensive,

drug-free (or

drug-exten-

sive) treatment.

Conversely,

consumers

who have to

pay

out of

their

own

pockets

might

instead

prefer

a

drug-intensive

treatment,

with

little

talk

therapy,

even at relatively

high drug prices.

So elimination

of

moral

hazard in

mental health treatment

could

actually raise the

willingness

to

pay for

drugs relative to what is

happening

with

insurance

today.

This

might seem

probable, given

that,

as the

authors

point

out,

HMOs

often prefer

to

substitute drugs

for

talk

therapy

if

they can,

and

HMOs

may

in some

ways

mimic health

care consumers would choose in

the

absence

of moral hazard. Even ten

years ago,

however,

most

indemnity

and

PPO

(preferred provider

option)

insurance

companies,

even if

they

would

reimburse

most

things

on a

fee-for-service

basis, nevertheless

tended

to

put

tight

restrictions on

reimbursement

for

psychotherapy,

because

it was known to be

especially prone

to moral hazard.

This

means that the amount

of

moral

hazard

connected to talk

therapy

is

likely to be

limited in most

standard

insurance

policies.

Obviously

these comments

only

sketch a

rough

outline

of

some

of

the

effects

of

health

insurance on

the

authors'

results.

Nevertheless,

I

hope it is clear now

why

I

believe

that

if

we are to

understand and

interpret those

results

accurately,

we need

to

better

understand the

implications of health insurance on

pharmaceutical

consumption and

why further research

in this

direction

is

justified.

Comment

by

Martin Neil

Baily:

The

general

discussion of

depression

and

its

treatments

in a

historical

setting

I found

very

interesting.

It

is

unusual to see

such a section

in

an

economics

paper,

but as a

sometime

macroeconomist,

I

found

it

refreshing

to see human

beings

described

with normal frailties and

not

just

as

maximizing

robots.

Ernst R. Berndt, lain M. Cockburn,

and

Zvi

Griliches

195

Moving to the meat

of

the

paper,

I

had trouble at the outset because

the interpretation of a producer

price

index

(PPI)

for the

ethical

drug

industry is problematic. The

issue

surfaces almost immediately when

it

is discovered that Prozac,

a principal innovation in recent years,

is

not

included in the sample because

it

is

not manufactured in the United

States. The true value

added in the actual manufacture

of

drugs

is small

for

most drugs, and

the final price

is

determined by research

and

de-

velopment costs, marketing,

and profit. The ex-factory price

of the

drug

is set by the company based

in part on taxes and does not accurately

reflect the contribution

to value by

the

manufacturing process.

Prozac

and other

drugs may

be

manufactured

in

Puerto

Rico,

but their value

was created by research and

development and marketing carried

out

in

the

United States or, possibly,

overseas.

PPIs are often used in the

calculation of real manufacturing

output

and hence productivity,

which can create significant distortions.

In

the

case of "foreign" manufacture,

the value may be incorrectly

attributed

by country,

but even when

all

of

the value-added

is

created

within

the

United

States,

there is misattribution

by industry.

This

paper

does

not

use

the PPIs for that

purpose,

so I

am

not

criticizing

what

they

have

done. But

thinking

about

this

problem

made me realize

how

difficult

it

is to use PPIs for real output

computations

in a world where the non-

manufacturing input

to

production

is

growing.

Ethical

drugs

are

at one

end of a

spectrum,

but the same

problem

arises elsewhere. The value-

added created by an

auto or a machine

tool

plant depends

heavily

on

the design and process engineering

that is done elsewhere.

In

this

paper,

the

authors are

essentially treating

the

PPI for antide-

pressants as

an

input

into

the

consumer

price

index.

The

discussion

of

generics

and the

use of hedonic

regressions

all

go

in

this

direction,

so

I will

simply accept

that framework without

further

comment. The

interesting

issue then

is how the

BLS

deals

with

generics

and new

products and

how the results

of what

they

do

compare

with

approaches

suggested by

index number theory.

The authors

deserve

a lot

of

credit,

as indeed

they

do for their

previous work in

this

area,

for pointing

out

some

of the

problems

that

exist in BLS

procedures-in

particular,

the

fixed market basket that

delays

the introduction of

major

new

products

and that

delays recog-

nition of

changes

in

market shares.

The authors

also

point

to some odd

quirks

of the BLS

sample,

which

apparently

includes

drugs

that are not

196

Brookings

Papers: Microeconornics 1996

antidepressants.

They make some important points and add

significantly

to their earlier

work.

Whinston, and Hurwitz

reported in this

journal, has shown that the

prices

of

branded drugs

increase when

generics are introduced.6

Together with the fact that

traditionally the

BLS has handled generics by

making

them

distinct new

products, that

implies that rapid entry of generics

will

result

in

an

increase in drug

prices. The Berndt, Cockburn,

and Griliches paper is

valuable in

showing

the inflation

implications

of

the

pattern

of

market

behavior

and index

number

methodology.

The fact that the

BLS has

now changed its

approach and

is

treating generics

as

identical products

to the branded

drugs makes

the

Berndt,

Cockburn, and Griliches story

less dramatic but is

very welcome in terms

of

the accuracy of price

indexes going forward. The authors' results

showing

how much

differ-

ence the alternative

approach

would have made

in

the

past

are

dramatic

enough.

The results for new

drugs

led

me

to think about

an

argument

that is

made concerning

the extent to

which

existing index number methods

may capture

innovation and new

products.

The

argument

holds that

existing products have to

compete

with the new

products

and that in a

perfectly competitive

market, therefore,

the

price

of

the old

products

has to go down to

reflect

the entrance

of the

new

products.

In

practice

there are differentiated

products

in

imperfect

competition,

and the

entry

of

new

products

may change

the

elasticity

of

demand and cause the

price

of

the

existing products

to

rise. For

example,

CDs and CD

players

have largely driven

out

LPs

and

turntables. But

there

is still

a niche

market

of buyers who believe

in the old

products.

Perhaps

the

prices

of

these

products

have

gone up

and not

down.

Discount

clothing

stores

sell

copies

of the latest

fashions-perhaps

that drives

up

the

price

of

the

high

fashion items. In

general,

if

the

pricing

behavior that is

seen

in

drugs

also

applies

to other

products,

then

existing price

indexes will

miss

much of

the

impact

of new

products

and

quality change.

We

desperately

need

more data on these issues.

The issue

of how

to value the

generics

raises

interesting questions

of

consumer

sovereignty.

The traditional

BLS method

did

assume a

full

6. Caves,

Whinston,

and Hurwitz

(1991).

Ernst R. Berndt, lain M. Cockburn, and Zvi Griliches 197

consumer sovereignty notion.

The BLS

assumed

that a

branded

anti-

depressant was twice as valuable

as

the generic

that

is

A counterargument is that the price differentials reflect inefficiency

in the market. People do not realize that a generic drug is chemically

identical to an existing branded drug.

In this

case,

as the share of

generic

drugs increases, we should

count that

legitimately

as a

price

decrease.

This is the new BLS approach.

My own reaction

is that

the

new BLS

approach

is

correct.

I

would

count fully the generic

as a reduction in

price. But, nevertheless,

one

of

the procedures that

is

suggested

in this

paper,

the

procedure

of

splitting the difference,

is

a

reasonable, practical

alternative. It

is

a

reasonable compromise between

the two

alternatives

of

saying

that

people feel

better because

they

are

buying

a

branded

drug

rather than a

generic

one

or, conversely, saying

that

people

are foolish or

lack in-

formation

when they buy

a

branded,

rather

than a

generic, product.

A

reason that

I

prefer the new BLS method

of

counting

the

full price

decline

associated

with

the introduction

of a

generic

is

that there

is

a

private market incentive

for

companies

to disseminate what

is, accord-

ing to the FDA, false information, namely that the branded drug is

superior

to

the generic.

If the

FDA is

correct

that

the

two

are

equivalent,

then it is

still

in the interest

of the manufacturers

of

the branded

product

to

persuade customers

and doctors

that is not the

case.

Turning to the hedonic approach that

is

used

in this

paper,

I

had

trouble

understanding

what

the coefficients

on the different

drug

char-

acteristics

in

their regressions actually represent.

The

right

model

for

antidepressants may

be a

matching model,

a model such as

those used

in

job search.

A

standard view

of

hedonics

is

that

consumers

trade

off

price against

some side-effect characteristics.

To

put

it

bluntly, you

would

pay

a

little bit more

for a

drug

in

order

to

have a little bit less

constipation

or

a little bit less

of

some

other side effect.

My understanding

of

the treatment

for

depression

is

that a

doctor

will

suggest

a

particular antidepressant

based

on

the

patient's history

and

the nature

of the

depression

and other

information. There is

then,

essentially,

a

trial for some

period

of

time to see

if

that

drug

works.

The information about the effect

of a

given drug

on a

patient

is

costly

to

acquire.

Doctor and

patient may

not

know for

some

period

of months

198 Brookings Papers: Microeconomics 1996

whether a drug is going to be effective

or

the nature

of the

side effects.

If the drug works, then the patient stays on it.

If it

does not, then the

doctor gives the patient a prescription

for a

different antidepressant.

Some people have to cycle through

various

drugs

for

as

long

as a

couple

of

years, trying

to find

an

appropriate

match between

a

drug

and their

particular problems.

Such a

matching

model

helps explain

in

part why

the

demands are

relatively inelastic, because a patient

who has

finally

found a

drug

that

works will

develop a great allegiance

to

it.

There

is

resistance to

chang-

ing the drug just because a new one comes

on the

market

or

just because

there

is

a price change.

How

are price differentials among drugs established? Perhaps the

simple reduced

form

hedonic regression

works

fine,

but that is not

obvious

beforehand. Presumably,

all else

being equal,

the doctor

and

patient

will start out with a

cheaper drug

first.

(All

else

being equal

presumably means where drugs

have the same initial

probability

of

creating a successful match.)

There would be some

price elasticity

in

that process.

A

further complication

is

created by third-party payment,

however. Many patients

have

part

or

all of their

drug

cost

covered

by

insurance. This point

is

stressed extensively

in Ted

Keeler's comments.

Limits set by insurance providers

mean

that many people are

on

drug

plans that specify a certain

list of

antidepressants, and

so

they

have to

choose one

of

those

or

face

a

sharp

increase

in

the incremental cost to

move off that list.

HMOs, which account

for an

increasing

share of medical

care,

if

they

are

providing drugs

as

part

of their

package,

would have an incen-

tive

to

have

their

doctors

start

with

the

cheaper drugs-assuming equiv-

alent

probability

of success.

In

general

this

is not

a market that

resembles

a

simple

textbook

utility

tradeoff.

There

are third-party payments.

Decisions

are

made when

patients

do

not

have

full information. And the

particular

market

equi-

librium

may

look more like

a

matching

model. It would

have been

helpful

if

the authors had discussed

how

we should

actually interpret

their coefficients

and

whether

they

would be the

right

ones for use

in

a

price

index

given

these

market characteristics.

My

concerns about

the

hedonics

would

have

been more

muted

except

for the fact that

the

results seem

surprising

to

me. It

may

be the method

used,

or

the

data,

or the time

period,

but somehow

the

impact

on

welfare

Ernst R.

Berndt, lain M. Cockburn, and

Zvi

Griliches

199

of the

introduction of the new

classes of

antidepressants may not

be

fully captured.

The results

reported in the paper are that when

you adjust

for

quality,

the impact is

not all that great. I would like more

explanation

of how

much of the

quality improvement is

being captured. There

is

the shift

from the old

class of drugs

to

the

new class of

drugs.

And then

there

are new entrants

with

slightly

different

profiles.

The

impact

of the

first

should be

huge. And the impact

of the second

should

also

be

fairly

large because

the class of drugs

becomes usable

for

a broader

group of

patients-more

drugs,

even when

they

look

similar

in

average

perfor-

mance, allow

for more successful matches.

In

conclusion,

this is a

paper

that is in

progress, and

I

want to

applaud

the

work

these

authors

have

done in

this

paper and

in

previous ones on

price indexes. It is

just

terrific as a set of

work, and

I

congratulate them

on it

and look

forward

to further

progress

ahead.

Commentators'

References

Caves,

Richard E., Michael

D.

Whinston,

and Mark A.

Hurwitz. 1991.

"Pat-

ent

Expiration, Entry,

and

Competition

in

the

U.S.

Pharmaceutical

Indus-

try." Brookings Papers

on Economic

Activity, Microeconomics: 1: 1-48.

Feldstein, Paul. 1993. Health Care Economics,

4th

ed. Albany: Delmar

Pub-

lishers.

Griliches, Zvi, and Ian Cockburn. 1994. "Generics

and

New Goods

in

Phar-

maceutical Price Indexes." American Economic Review 84

(December):

1213-32.

Hellerstein, Judith K. 1994. "The Demand for Post-patent Prescription

Phar-

maceuticals." NBER Working Paper 4981.

National

Bureau of Economic

Research, Cambridge, Mass. December.

Manning,

Willard

G.,

and

others. 1987.

"Health Insurance and

the Demand

for Medical Care: Evidence from

a

Randomized Experiment."

American

Economic Review 77 (June): 251-78.

Oberlander, Jonathan. forthcoming. "Managed

Care

and

Medicare

Reform."

Journal of Health Politics, Policy,

and Law.

Sullivan, Sean. 1992. "The Demand

for

Prescription Drugs among

Elderly

Americans." Ph.D. dissertation, University

of

California.