New Drugs: Health and Economic Impacts Frank R. Lichtenberg, PhD Courtney C. Brown Professor of Business, Columbia University Graduate School of Business.

New Drugs: Health and Economic

ImpactsFrank R. Lichtenberg, PhD

Courtney C. Brown Professor of Business,Columbia University Graduate School of

Business Research Associate,

National Bureau of Economic Research

Outline

I. Introduction: Innovation, Health, and Economic Growth

II. Econometric evidence1. Longevity

a. “Case study”: HIV in U.S.b. All diseases: U.S.c. All diseases: 70 countries

(preliminary estimates)2. Ability to work (All diseases:

U.S.)

Conventional (narrow) definition of economic

growth:

increase in real GDP per capita

Utility, or welfare, depends on (leisure) time as well as goodsBecker defined an individual’s “full income”

as the value of goods consumed plus the value of leisure time “consumed”. Y* = G(Y, L)where Y* = “full income” (or utility) Y = goods consumed L = leisure time

Utility, or welfare, depends on (leisure) time as well as

goodsSimple linear approximation: 

Y* = Y + pL L

pL = the shadow price of leisure time (relative to the price of goods)

pL constant Y* = Y + pL L The change in full income is the change in GDP plus the

change in the value of leisure time consumed.

Economic importance of longevity increase

• Nordhaus: “to a first approximation, the economic value of increases in longevity over the twentieth century is about as large as the value of measured growth in non-health goods and services”

• pL L Y

• reflects changes in “quantity” (length), but not “quality”, of life

United Nations’ Human Development Index

(unweighted) average of three indexes: • a life expectancy index• an education index• an index of per capita GDP

Long-run economic growth

Two components• Increased per capita GDP• Increased longevity and quality of lifeWhat are the sources of economic

growth?

Technical progress

Economic growth

Solow (1956): technical progress is necessary for there to be sustained growth in output per hour worked

Production function with technical progress

Yt = At F (Nt, Kt)

K t+1 = (1 - ) Kt + It

Y = outputA = an index of the level of technology (“stock of

ideas”)N = laborK = capitalI = investment

Exogenous vs. endogenous technical progress

• Solow assumed exogenous technical progress: A increased at a constant rate• Subsequent research has developed and confirmed the hypothesis of endogenous technical progress

Endogenous technical progress:

Knowledge-capital-stock model (Griliches)

Yt = F (Nt, Kt, Zt)

K t+1 = (1 - ) Kt + ItZ t+1 = (1 - Z) Zt + RDt

R&DTechnical progress

Economic growth

Endogenous growth models:

Disembodied vs. Embodied Technical Progress

• Suppose that agent i in the economy (e.g. a firm or government agency) engages in research and development.

• If technical progress is disembodied, another agent (j) can benefit from agent i’s R&D whether or not he purchases agent i’s products.

• If technical progress is embodied, agent j benefits from agent i’s R&D only if he purchases agent i’s products.

• Solow conjectured that most technical progress was embodied.

R&DTechnical progress• disembodied• embodied

Economic growth

Equipment-embodied technical change

• Equipment (e.g. computers) used in manufacturing embodies a lot of R&D

• Several authors have tested the equipment-embodied technical change hypothesis by estimating manufacturing production functions, including (mean) vintage of equipment as well as quantities of capital and labor

• Finding: technical progress embodied in equipment is a major source of manufacturing productivity growth.

Pharmaceutical industry is even more R&D-intensive than the equipment industry:

Industrial R&D funds as a percent of net sales in R&D-performing companies, 1997

3.40%

3.90%

6.00%

10.50%

0% 5% 10% 15%

Drugs and medicines

Machinery andequipment (includingcomputers) Manufacturing

All Industries

R&DTechnical progress• disembodied• embodied

Economic growth• conventional• health

Key hypothesis

Pharmaceutical R&D investment

Longevityincrease

Obstacles to a direct examination of the R&D-longevity relationship

• Very long lags• Divergent estimates of R&D investment

(NSF vs. PhRMA)—30% difference in 1997• Smoothness of aggregate R&D:

pharmaceutical R&D investment is very closely approximated by an exponential trend

• Lack of disaggregated R&D data • Patent data are subject to similar limitations

New drug approvals as an “intermediate good”

Pharmaceutical R&D

investment

Longevityincrease

FDA New Drug

Approvals

Impact of new drugs on longevity

1.“Case study”: HIV in U.S.2. All diseases: U.S.3. All diseases: 70 countries

(preliminary estimates)

HIV mortality

05,000

10,00015,00020,00025,00030,00035,00040,00045,000

Source: CDC Compressed Mortality file

Hypothesis:

The development, FDA approval, and use of new HIV drugs played an important role in this dramatic reduction in HIV mortality.

Drugs with HIV indication, by FDA approval dateFDA approval date Drug Indication

18-Aug-71 MEGESTROL HIV, ANOREXIA FROM 30-Jul-76 SOMATROPIN HIV, ANOREXIA FROM

19-Mar-87 ZIDOVUDINE HIV INFECTION 21-Oct-88 OCTREOTIDE AIDS-ASSOCIATED DIARRHEA 9-Oct-91 DIDANOSINE HIV INFECTION

19-Jun-92 ZALCITABINE HIV INFECTION 24-Jun-94 STAVUDINE HIV INFECTION 17-Nov-95 LAMIVUDINE HIV INFECTION 6-Dec-95 SAQUINAVIR HIV INFECTION 1-Mar-96 RITONAVIR HIV INFECTION

13-Mar-96 INDINAVIR HIV INFECTION 21-Jun-96 NEVIRAPINE HIV INFECTION 14-Mar-97 NELFINAVIR HIV INFECTION

4-Apr-97 DELAVIRADINE HIV (IN COMBO WITH NUCLEOSIDE ANALOGUES) 17-Sep-98 EFAVIRENZ HIV (IN COMBO W/ANTIRETROVIRAL AGENTS)17-Dec-98 ABACAVIR HIV (IN COMBO W/ANTIRETROVIRAL AGENTS)15-Apr-99 AMPRENAVIR HIV (IN COMBO W/ANTIRETROVIRAL AGENTS)

No. of HIV drugs approved by FDA

1 1

0 0

1 1

0

1

2

3

2 2

0

0.5

1

1.5

2

2.5

3

3.51987-1993: 0.57 drugs/year1994-1998 2.00 drugs/year

Mortality model

• Suppose that the number of HIV deaths in year t is inversely (and linearly) related to the cumulative number of HIV drugs approved up until year t-1

• DEATHSt = a – b CUM_DRUGSt-1

= a – b (FDAt-1 + FDAt-2 + …)

• FDAt-1 is the number of drugs approved by the FDA in year t-1, etc.

Mortality model

• Implies that DEATHSt - DEATHSt-1 =

– b (CUM_DRUGSt-1 - CUM_DRUGSt-2)

• - DEATHSt = b FDAt-1

• The reduction in deaths (- DEATHSt) is proportional to the number of drugs approved in the previous year.

HIV drug approvals and HIV mortality reduction

0

0.5

1

1.5

2

2.5

3

3.5

1988 1990 1992 1994 1996

-10,000

-5,000

0

5,000

10,000

15,000

20,000

Number of HIV drugs approved

Reduction in number of HIV deaths in following year

Regression analysis

• - DEATHSt = -6328 + 6093 FDAt-1

• t-stats: (3.40) (4.74)

• Probability value associated with the FDAt-1

coefficient is .0015 • R2 = .7378• The annual number of HIV deaths is

reduced by 6093, on average, by one additional HIV drug approval

All diseases: U.S.

Increase in longevity (mean age at death)

69.6

73.4

67

68

69

70

71

72

73

74

1979 1998

Year

Mea

n ag

e at

dea

th

Longevity gains vary across diseases

0

5

10

15

20

25

0 2 4 6 8 10 12 14 16 18

Increase in mean age (in years) at death

Num

ber

of d

isea

ses

Drugs for treatment of TUBERCULOSIS

• CAPREOMYCIN• ISONIAZID• CYCLOSERINE• ETHAMBUTAL• ETHIONAMIDE• AMINOSALICYATE

SODIUM

• ACETYLCYSTEINE (INH)

• PYRAZINAMIDE• RIFAMPIN• RIFAMPIN AND

ISONIAZID• RIFAPENTINE

Drugs for treatment of HYPERCHOLESTEROLEMIA

• LOVASTATIN• PRAVASTATIN• SIMVASTATIN• CHOLESTYRAMINE• COLESTIPOL• PROBUCOL• FLUVASTATIN

• ATORVASTATIN• NIACIN(SA-

LIPOTROPIC)• CERIVASTATIN• GARLIC• PSYLLIUM,BRAN*• NEOMYCIN*• CONJ. ESTROGEN,M-

PROGESTERONE*

* Unlabeled indication

There is considerable variation across diseases—even diseases in the same broad disease groups—in the extent and timing of increases in the stock of available drugs.

Number of drugs available to treat condition in year t, as % of number of drugs available to

treat condition in 1979

100%

110%

120%

130%

140%

150%

160% Disorders of thyroidglandDisorders of otherendocrine glands

Priority- vs. Standard-review drugs

• “Priority Review” drug: one that represents a “significant improvement compared to marketed products, in the treatment, diagnosis, or prevention of a disease”

• “Standard Review” drug: one that “appears to have therapeutic qualities similar to those of one or more already marketed drugs”

Number of new drugs approved, 1979-1998

5

10

114

0

2

4

6

8

10

12

14

16

18

Syphilis Lymph cancer

StandardPriority

Basic model

• ADit = CUM_DRUGit + i + t+ it 

• ADit = mean age at which deaths caused by disease i in year t occur

• CUM_DRUGit = number of drugs approved to treat disease i up until year t

• i = 1, 2, …, 110 (approximately) “2-digit” diseases

• t = 1979, 1980, …, 1998 (~2200 obs.)

• “Year effects” (i‘s) control for changes in aggregate determinants of mortality

Measurement

• ADit: 1979-1998 Compressed Mortality File

• CUM_DRUGit:Linkage of drugs to diseases: National Drug

Data File drug-disease indications tableDrug approval dates: FDAErrors in matching drugs to diseases &

determining approval dates biased towards zero

Weighted least-squares estimation

• ADit = CUM_DRUGit + i + t+ it

• Estimate via weighted least-squares, using weight N_DEATHit

• N_DEATHit = number of deaths caused by disease i in year t

Priority-review vs. standard-review drugs

• ADit = P CUM_PRIit + S CUM_STDit

+ i + t+ it

• CUM_PRIit = number of priority-review drugs approved to treat disease i up until year t

• CUM_STDit = number of standard-review drugs approved to treat disease i up until year t

• CUM_DRUGit = CUM_PRIit + CUM_STDit

Parameter estimates

• Model 1: = .013, t = 2.27, p-value = .0232

• Model 2: P = .075, t = 4.01, p-value < .0001

S = -.009, t = 1.05, p-value = .295

• Only priority-review drug approvals increase mean age at death

Exclude CUM_STDit:

P = .065, t = 4.03, p-value < .0001

Contribution of new drugs to longevity increase

• Mean age at death increased by 3.8 years from 1979 to 1998

• The increase in the stock of priority-review drugs is estimated to have increased mean age at death by 0.39 years (4.7 months) during this period.

• About 10% of the total increase in mean age at death is due to the increase in the stock of priority-review drugs.

Contribution of new drug approvals to longevity increase, 1979-1998

0.39

3.41

New drugapprovalsOther factors

Estimate is a lower bound?

• This estimate is based on an analysis of changes in AD within diseases

• About 19% of the increase in AD was due to a shift in the distribution of diseases; the remainder was due to increase in AD from given diseases

• New drug approvals affect the disease-distribution of deaths as well as age at death from given diseases

Costs vs. Longevity Benefits of New Drug Approvals

Costs• During the period 1979-1998, 508 NMEs

(about 25/year) were approved by the FDA • OTA study: the average cost of an NME

approval was $359 million• Total cost = 508 NMES * $359 m./NME =

$182 billion

Costs vs. Longevity Benefits of New Drug Approvals

Longevity Benefits• The increase in the stock of priority-review

drugs is estimated to have increased mean age at death by 0.39 years during this period.

• There are about 2 million deaths per year• Total number of life-years gained per year

is 0.39 * 2 million = 800,000 life-years/year

Costs vs. Longevity Benefits of New Drug Approvals

Longevity Benefits• A number of authors have estimated that

the value of a life-year is approximately $150,000.

• The value of the annual gain in life-years is 800,000 * $150,000 = $120 billion.

• This $120 billion may be viewed as an annuity.

“Social rate of return” to pharmaceutical R&D

investment• DiMasi estimates that, in the last two

decades, drug development has taken about 14 years.

• Suppose that the $182 billion in R&D expenditure is evenly distributed over an initial 14-year period, i.e. $13 billion/year in years 1-14.

• In year 15 and all future years, the population experiences a gain in life-years whose annual value is $120 billion.

Investment costs and returns

-20

0

20

40

60

80

100

120

Billions of dollars

1 6

11 16 21 26 31 36 41 46 51 56 61

Year

Investment costs and returns

• The internal rate of return to this series of cash flows is 18%.

• This rate of return reflects only the value of increased longevity among Americans

• Foreigners also benefit• Additional benefits of new drugs to

Americans, including reduced hospital expenditure and reduced limitations on work and other activities

International Diffusion of New Drugs and

Global Mortality Decline

Beyond the U.S.A.

• The U.S. accounts for only one fifth of the population of OECD countries, and 5% of the world population.

• The purpose of this study is to assess the impact of international diffusion of new drugs on mortality throughout the OECD and in some non-OECD countries.

Linkage of several extremely rich databases

• IMS Drug Launches database• OECD Health database• World Health Organization Mortality

database• World Bank’s Global Development Network

Growth Database

Example: tenecteplase

Launch date Country6/00 USA3/01 Finland5/01 UK9/01 Norway10/01 Canada10/01 South Africa11/01 Ireland

Future launches

• As is often the case with duration or survival data, some of the observations are right-censored: the exact time till launch is unknown—it is only known that the time till launch exceeded some value.

• The analysis methodology must correctly utilize the censored as well as the non-censored observations.

• I use the SAS LIFETEST procedure to compute nonparametric estimates of the launch-probability distribution, and to perform hypothesis tests.

Characterizing country-specific launch behavior

• During the period 1989 to 2001, about 500 NMEs were launched (about 40 per year) in at least one country.

• From the Drug Launches database, we can determine whether or not NMEi (i = 1,…, 500+) had been launched in country j (j = 1,…, 70) after m months (m = 1,…, 156).

Drug launch probability profiles: U.S. vs. Canada

0%

20%

31%37% 40% 40% 41%

0%

39%

46%50% 52% 54% 55%

0%

10%

20%

30%

40%

50%

60%

0 2 4 6 8 10 12

Years since initial world launch

CANADAUSA

Hypotheses

• Mortality from a given disease is inversely related to the number of drugs available to treat the disease

Two measures of mortality:– Life years lost before age 70– Mean age at death

• International differences in mortality are partly attributable to differences in the probability and timing of new drug launch

• International differences in the probability and timing of new drug launch are partly attributable to differences in public policy vis-à-vis pharmaceuticals

Number of drugs available to treat disease i

Mean age at death from disease i

1987 2002

Estimates to be provided

How much have new drug launches in OECD and some non-OECD countries since 1988:– reduced the number of life-years lost in those

countries?– increased life expectancy (mean age at death) in

those countries?

In which countries is the probability of new drug launch highest (and lowest)?

How much greater would the increase in longevity have been in low-launch countries if they had had high (or average) launch rates?

How does public policy (e.g., government share of pharmaceutical purchases) affect launch rates?

Effect of drug launch behavior on longevity

• Cross-sectional analysis at the country level

• Longitudinal analysis at the country level• Longitudinal analysis at the disease-

by-country level

Three ways to analyze the effect of drug launch behavior

on longevity • cross-sectional analysis at the country

level• longitudinal analysis at the country level• longitudinal analysis at the disease-by-

country level. This approach will enable me to control for an extremely large set of unobserved factors, such as country-year fixed effects.

Longitudinal analysis at the disease-by-country level

• It is conceivable that unobserved country characteristics are not fixed, i.e. that there are unobserved factors that determine a country’s longevity trend, as well as its growth rate.

• Fortunately, due to the richness of the available data, we can control for time-varying country effects.

Longitudinal analysis at the disease-by-country level

AGE_DEATHijt = CUM_DRUGijt + ij + jt + it + uijt

AGE_DEATHijt = mean age at death in country i from disease j in year t

CUM_DRUGijt = the (cumulative) number of NMEs launched in country i for disease j up until year t

OVER65ijt = CUM_DRUGijt + ij + jt + it + uijt

OVER65ijt = fraction of deaths in country i from disease j in year t that occur at age 65 or older

Statistical issues

• Model controls for a very large number of potential determinants of mortality: it includes 450 country-disease fixed effects (ij), 300 disease-year fixed effects (jt), and 600 country-year fixed effects (it).

• To maximize the odds of obtaining a consistent estimate of the effect of drug launches on longevity (), we will also include and estimate 1350 “nuisance parameters”.

• The number of statistical degrees of freedom (= no. of observations – no. of estimated parameters) is quite large precise estimate of .

Measurement issues

Censoring of drug launch dataIn principle, want to measure the number of drugs

to treat disease i launched in country j by year tIn practice, can measure the number of post-1986

* drugs to treat disease i launched in country j by year t

*Drugs whose initial world launch occurred after 1986

Assume that latter is a noisy indicator of the former

Linking drugs to diseases

WHO disease classification

Infectious and and Parasitic Diseases

Neoplasms

Endocrine, Nutritional and Metabolic Diseases

Diseases of the Blood and Blood-forming Organs

Mental Disorders

Diseases of the Nervous System and Sense Organs

Diseases of the Circulatory System

Diseases of the Respiratory System

Diseases of the Digestive System

Diseases of the Genitourinary System

Obstetric Complications

Diseases of the Skin and Subcutaneous Tissue

Diseases of the Musculoskeletal System and Connective Tissue

Congenital Anomalies

Certain Conditions originating in the Perinatal Period

Symptoms, Signs and Ill-defined Conditions

IMS Health Drug Classification

A Alimentary Tract And Metabolism

B Blood and Blood Forming Organs

C Cardiovascular SystemD DermatologicalsG Genitourinary System and

Sex HormonesH Systemic Hormonal

Preparations (Excluding Sex Hormones)

J General Anti-Infectives, Systemic

K Hospital Solutions

L Cytostatics

M Musculoskeletal System

N Central Nervous System (CNS)

P Parasitology

R Respiratory System

S Sensory Organs

T Diagnostic Agents

V Various

Mean age

Mean age at death increased 4.5 years from 1987 to 2000. Holding number of drugs constant,

mean age at death increased 3.5 years. Increase in stock of drugs increased mean age by 1.0 years.

column 1 2 3 4 5 6

covariates

country year

class1

country year

class1

country*year

class1

country class1*

yearcountry*year class1*year

country class1

year(country) year(class1)

cum -- 0.031 0.030 0.011 0.016 0.023std. err. 0.010 0.011 0.011 0.012 0.011t-stat 3.160 2.790 1.060 1.310 2.090p-value 0.002 0.005 0.290 0.190 0.037

year=1987 (2000=0) -4.508 -3.5250.628 0.700-7.170 -5.030

<.0001 <.0001

OVER65column 7 8 9 10 11 12

covariates

country year

class1

country year

class1country*year

class1country

class1*yearcountry*year class1*year

country class1

year(country) year(class1)

cum -- 0.001 0.001 0.001 0.001 0.001std. err. 0.000 0.000 0.000 0.000 0.000

t-stat 7.410 6.010 3.680 3.150 3.860p-value <.0001 <.0001 0.000 0.002 0.000

year=1987 (2000=0) -0.093 -0.0590.010 0.011-9.790 -5.530

<.0001 <.0001

Prob. of dying after 65 increased 930 basis points from 1987 to 2000. Holding number of drugs constant, prob.

of dying after 65 increased 590 basis points. Increase in stock of drugs increased prob. of dying after 65 by 440

basis points .

Is globalization beneficial?

• Estimates suggest that international diffusion (a.k.a. globalization) of new drugs during 1987-2000 increased mean age at death in the sample countries by about 1 year

• There are about 12 million deaths per year in these countries

• Hence, globalization of new drugs added 12 million life-years per year in these countries

• More rapid diffusion would have resulted in a larger increase in life-years

Effect of public policy on the probability of drug launch

• Hypothesize that variation in drug launch behavior is due, to some extent, to variation in market size (population), per capita income, and average health expenditure.

• Drug launch behavior may also be influenced by the country’s ratio of public expenditure on pharmaceuticals to total expenditure on pharmaceuticals:Government formularies may be more likely to exclude new

drugs than private formulariesEven when new drugs are not excluded from public

formularies, governments are likely to pay lower prices than private payers

Indirect effect of public policy on longevity, via

probability of drug launchPublic policy vis-à-vis pharmaceuticals (e.g.,

RX_PUBLIC%)

Availability of new drugs (e.g.,

LAUNCH6)

Longevity

Effect of public policy on the probability of drug launch

• Examine the effect of the public share of pharmaceutical expenditure on the probability of new drug launch by estimating equations, using data on a cross-section of countries.

Country pub% launch10 Country pub% launch10Australia 50% 29% Korea 2% 53%Belgium 48% 37% Luxembourg 83% 13%Canada 32% 40% Netherlands 71% 40%Czech Republic 85% 36% New Zealand 75% 32%Denmark 47% 44% Norway 63% 32%Finland 46% 45% Portugal 64% 28%France 57% 37% Spain 72% 39%Germany 73% 41% Sweden 70% 42%Greece 71% 40% Switzerland 58% 44%Hungary 70% 38% Turkey 72% 35%Ireland 70% 38% United Kingdom 65% 49%Italy 60% 54% United States 13% 54%Japan 63% 53%

Inverse correlation between public share of pharma. expend. and

10-year launch probability

LAUNCH10i = -0.24 PUB% i + 0.54

R2 = 0.27

0%

10%

20%

30%

40%

50%

60%

0% 20% 40% 60% 80% 100%

Public pharma. expend./ Total pharma. expend.

10-y

ear

launch

pro

babilit

y

Controlling for country size

LAUNCH10i = 0.11 -.14 PUB%i + .04 log(POPi)

t-stats: (0.9) (1.9) (3.7)

(GDP not significant, controlling for population)

Impact of New Drugs on Ability to Work and to

Engage in Other Activities

Health status (H)Hmin

Suppose death occurs if H < Hmin

Probability of death can fall for tworeasons.

Health status (H)Hmin

1. Hmin declines Mean H of survivors falls

Health status (H)Hmin

2. H distribution shifts to the right Mean H of survivors rises

% of People Unable to Work, by Age

2.9%4.6%

7.9%

15.2%

19.9%

0%

5%

10%

15%

20%

25%

25-34 years 35-44 years 45-54 years 55-64 years 65-69 years

Ability to work & labor supply• UNABLE_WORKit = CUM_DRUGit + i + t+ it 

• UNABLE_WORKit = fraction of people with condition i in year t who are unable to work, mainly due to condition i

• MISSED_WORKit = CUM_DRUGit + i + t+ it • MISSED_WORKit = mean number of work days missed

by currently employed people with condition i in year t

• i = 1, 2, …, 110 (approximately) “2-digit” diseases• t = 1983, 1984, …, 1996• “Year effects” (i‘s) control for changes in aggregate

determinants of ability to work

Mean number of drugs available to treat condition (prevalence-

weighted)

23.9

35.8

0

5

10

15

20

25

30

35

40

1983 1996

1983-1996 increase in number of drugs reduced all of the following by about 12% in

1996• Number of people unable to work• Work-loss days of currently

employed persons• Restricted-activity days of all

persons• Bed days of all persons

Increase in number of drugs available decrease in inability to work

-10%

-8%

-6%

-4%

-2%

0%

2%

4%

6%

8%

10%

12%

0% 20% 40% 60% 80% 100%

% increase in number of drugs for condition

Change in %

of people

unable

to

work

due t

o c

ondit

ion

20 most prevalent conditions

Estimated effects of 1983-96 new drug approvals

• reduction in number of people unable to work: 1.44 million

• value of reduction in number of people unable to work (@ $30K/year): $43.3 billion/year

• reduction in work loss days per year of currently employed persons: 98.8 million/year

• value of reduction in work loss days (@ $100/day): $9.9 billion/year

• reduction in restricted activity days of all persons: 423 million/year

• reduction in bed days of all persons: 178 million/year