Transcript the value of clinical laboratory innovation

Better information, better health:
the value of clinical laboratory innovation
Frank R. Lichtenberg
Columbia University and
National Bureau of Economic Research
Role of new goods in
economic growth
• Grossman and Helpman, Innovation and Growth in the
Global Economy: “innovative goods are better than older
products simply because they provide more ‘product
services’ in relation to their cost of production.”
• Bresnahan and Gordon, The Economics of New Goods:
“New goods are at the heart of economic progress”
• Bils: Measuring the Growth from Better and Better
Goods, “Much of economic growth occurs through
growth in quality as new models of consumer goods
replace older, sometimes inferior, models.”
2
R&D intensity in 1997:
Medical equipment and supplies
industry vs. all industries
9%
8.4%
8%
7%
6%
5%
4%
3.4%
3%
2%
1%
0%
Medical equipment and supplies
All industries
Note: Percentages are total (Federal plus company and other) funds for industrial R&D performance in the
U.S. as a percent of net sales of companies that performed industrial R&D in the U.S., 1997
Source: National Science Foundation/Division of Science Resources Statistics, Survey of Industrial
Research and Development: 2000, http://www.nsf.gov/sbe/srs/nsf03318/pdf/taba19.pdf
3
Clinical laboratory innovations
• Examine the impact of a subset of the new products
generated by this industry—clinical laboratory
products—on the longevity and quality of life of
Americans.
• FDA data indicate that, in the last decade, about 100 of
these new products have been introduced.
• I hypothesize that these new products have improved
the quality of information physicians and patients have
about patients’ medical conditions, and have therefore
enabled more appropriate and effective treatment of
those conditions.
4
Outline
1. Describe the general framework I will use to assess the
impact on health of clinical laboratory innovations.
2. Explain how disease-specific measures of laboratory
innovation can be constructed by combining FDA
regulatory data with health insurance claims data.
3. Use these measures and Vital Statistics-Mortality Detail
data to examine the impact of clinical laboratory
innovation on longevity (age at time of death).
4. Use the laboratory innovation measures and data from
the National Health Interview Survey to examine the
impact of laboratory innovation on “quality of life”
(activity limitations and disability days).
5
Difference-in-differences model
HEALTHit = blab CUM_PRODit + gi Zit + ai + dt + eit
HEALTHit= a measure of health outcomes associated with disease i in year t
CUM_PRODit= an index of the cumulative number of laboratory products related to
diagnosis i that have first appeared by year t
Zit= a vector of other attributes of disease i in year t
The fixed disease effects (ai‘s) control for any permanent between-disease differences in
health determinants.
The fixed year effects (dt‘s) control for changes over time in health determinants that are
common across diseases.
If the estimate of blab is positive and statistically significant, that indicates that there were
above-average improvements in health outcomes of diseases with above-average
increases in the cumulative number of clinical laboratory products.
6
Construction of disease-specific
measures of laboratory innovation
• A variety of clinical laboratory procedures
are used to diagnose and treat people with a
given disease.
• I constructed an index of the cumulative
number of products related to patients with
diagnosis i that have first appeared by year t.
• This index is based on all of the clinical
laboratory procedures that could be linked
(using Federal regulation numbers) to
clinical laboratory products appearing in the
FDA’s Premarket Notification Database.
7
Example: regulation 862.1170
chloride test systems
CFR Title 21 Database
PART 862 CLINICAL CHEMISTRY AND
CLINICAL TOXICOLOGY DEVICES
8
Products regulated by CFR Title 21
862.1170 (chloride test systems)
First year in
Premarket
Notification
Database
Product
code
Product
JFS
COULOMETRIC, CHLORIDE
1976
CHJ
MERCURIC THIOCYANATE, COLORIMETRY,
CHLORIDE
1977
CGZ
ELECTRODE, ION-SPECIFIC, CHLORIDE
1979
CHK
MERCURIC NITRATE AND DIPHENYL CARBAZONE
(TITRIMETRIC), CHLORIDE
1987
CHG
ACID, PHOSPHORIC-TUNGSTIC
(SPECTROPHOTOMETRIC), CHLORIDE
1990
The number of products covered by this regulation increased from 3 in 1979
9
to 5 in 1990, and has remained constant since then.
Laboratory procedure coding specialists at
the Laboratory Corporation of America
ascertained that three diagnostic lab
procedures (CPT codes)—82435
(Chloride; blood), 82436 (Chloride; urine),
and 82438 (Chloride; other source)—
involve the use of chloride test systems.
10
• Suppose, for simplicity, that chloride test system
procedures are used in connection with disease A but
not in connection with disease B.
• Then chloride test system innovation (as manifested in
new products) is likely to benefit people with disease A,
but not those with disease B.
• In reality, chloride test system procedures may be used
in connection with both diseases, but to different extents.
• If chloride test system procedures account for twice as
great a fraction of all clinical lab procedures for disease A
as they do for disease B, then chloride test system
innovation is likely to benefit people with disease A twice
as much as it benefits those with disease B.
11
Estimation of SHAREip
CUM_PRODit = ∑p SHAREip CUM_PRODpt
Estimates of SHAREip were constructed from the
MEDSTAT MarketScan Commercial Claims &
Encounters and Medicare Supplemental
Research Databases. The MarketScan
databases capture person-specific clinical
utilization, expenditures and enrollment across
inpatient, outpatient, prescription drug, and
carveout services from approximately 45 large
employers, health plans, and government and
public organizations.
12
• In the year 2000, almost 14 million outpatient
and inpatient claims were diagnostic lab claims;
each of these claims included both a procedure
code (usually a CPT code) and a diagnosis
(ICD9) code.
• I computed the frequency of diagnostic lab
procedures, by (2-digit) diagnosis code. Let
N_CLAIMip = the number of claims for lab procedure p
associated with diagnosis i, and
N_CLAIMi. = ∑p N_CLAIMip = the total number of claims
for lab procedures associated with diagnosis i.
Then SHAREip = N_CLAIMip / N_CLAIMi.
13
Leading clinical lab procedures:
diabetes vs. hypertension
Ran
Ran
k
1
%
18%
Cum %
k
procedure
18%
83036-Glycated Hemoglobin Test
2
8%
27%
82947-Assay, Glucose, Blood
Quant
3
7%
33%
80061-Lipid Panel
4
5%
39%
82962-Glucose Blood Test
5
3%
42%
80053-Comprehen Metabolic
Panel
6
3%
45%
82948-Reagent Strip/Blood
Glucose
7
2%
47%
84443-Assay Thyroid Stim
Hormone
8
2%
49%
81002-Urinalysis Nonauto wo
Scope
9
2%
52%
85025-Complete CBC w Auto Diff
WBC
10
2%
54%
80048-Basic Metabolic Panel
1
2
3
%
11%
5%
5%
cum %
procedure
11%
80061-Lipid Panel
16%
80053-Comprehen Metabolic
Panel
21%
85025-Complete CBC w Auto Diff
WBC
4
4%
25%
84443-Assay Thyroid Stim
Hormone
5
4%
29%
80048-Basic Metabolic Panel
6
3%
32%
81000-Urinalysis, Nonauto w
Scope
7
3%
34%
81002-Urinalysis Nonauto wo
Scope
8
3%
37%
82565-Assay of Creatinine
9
3%
39%
84153-Assay of Psa, Total
10
2%
42%
82270-Test for Blood, Feces
14
Relative utilization of procedures
differs across diagnoses
• The Glycated Hemoglobin Test is the procedure
most frequently associated with a diabetes Dx,
accounting for 18% of all procedures, but is not
among the top ten procedures associated with a
hypertension Dx.
• Similarly, Assay of Thyroid Stimulating Hormone
accounts for twice as large a fraction of
procedures among hypertension patients as it
does among diabetes patients.
15
Effect of laboratory innovation on
longevity (age at time of death)
AGE_MEASUREit = blab CUM_PRODit
+ ai + dt + eit
AGE_MEASUREit = a measure (e.g., the
mean) of the age distribution of deaths
caused by disease i in year t
16
Controlling for pharmaceutical
innovation
AGE_MEASUREit = blab CUM_PRODit +
bdrug CUM_DRUGit + ai + dt + eit
CUM_DRUGit = the cumulative number of
drugs launched to treat disease i in year t.
17
• I estimated these equations using three alternative
measures of the age distribution of deaths:
– mean age at death
– the fraction of deaths occurring before age 65
– the fraction of deaths occurring before age 75
• These measures were computed from the CDC’s
Multiple Cause-of-Death Mortality Data Files for the
years 1979-1998. Deaths during 1979-1998 were
classified according to the ICD9 disease classification.
Since 1999, deaths have been classified according to
the ICD10 classification, causing a discontinuity in the
data. Each file contains data on approximately 2 million
deaths.
• All equations were estimated by weighted least squares,
weighting by the number of deaths caused by that
18
disease in that year.
Implications
Both laboratory and pharmaceutical
innovation have increased longevity, but
they have done so in different
(complementary) ways.
– Laboratory innovations have primarily
reduced the risk of dying before the age of 65
– Pharmaceutical innovations have primarily
extended the lives of older people.
19
Consistency with utilization data
type of claim
number of mean
claims
age
% 65 and
over
Diagnostic lab
13,819,779
45.7
9.8%
Drug
40,513,363
54.8
33.3%
20
• We can calculate the effect of laboratory
innovation on the overall increase in mean age
at death between 1979 and 1998 by multiplying
the estimate of blab in column 2 (0.336) by the
difference between weighted (by number of
deaths) mean CUM_PROD in 1979 and 1998.
• This implies that laboratory innovation during the
period 1979-1998 increased mean age at death
by 0.44 years. This represents about 12% of the
total increase in mean age at death (3.57 years)
during the period.
21
Cost per life-year gained from
laboratory innovation
•
•
•
•
•
•
There were 2.16 million deaths in the U.S. in 1998. Hence the laboratory
innovation of the previous 20 years resulted in a gain of about 952,000 lifeyears (0.44 life-years/death * 2.16 million deaths) in 1998.
We would like to know the cost per life-year gained from laboratory
innovation. This requires information about expenditure on new laboratory
procedures (i.e. procedures that were not available in 1979). This
information is not readily available.
However, we can calculate an upper bound on this, by using data on
expenditure on all (new plus old) laboratory procedures. According to the
Census Bureau, in 1998 the total revenue of medical and diagnostic
laboratories (NAICS 6215) was $18.4 billion.
This implies that $19,340 (=$18.4 billion / 952,000 life-years) is an upperbound estimate of the cost per life-year gained from laboratory innovation.
Even this figure is well below the cost-effectiveness thresholds proposed by
most medical decision-makers. Moreover, if expenditure on new
procedures accounts for half of total laboratory expenditure, then the cost
per life-year gained from laboratory innovation is only half as great.
This is the average cost per life-year gained from all laboratory innovation.
Undoubtedly some innovations have incurred higher costs, and others lower
costs, per life-year gained.
22
Effect of laboratory innovation on
activity limitations and disability days
• Now I will analyze the effect of laboratory innovation on the quality of
life, i.e. on the extent of disability and activity limitations in the U.S.
population.
• To do this, I will use data from an additional source, the National
Health Interview Survey (NHIS).
• The NHIS is the principal source of information on the health of the
civilian noninstitutionalized population of the United States and is
one of the major data collection programs of the National Center for
Health Statistics (NCHS).
• While the NHIS has been conducted continuously since 1957, the
content of the survey has been updated about every 10-15 years.
• The survey remained the same during the period 1982-1996.
• During that period, it collected information about 1.2 million chronic
and acute medical conditions afflicting about 1.6 million Americans.
23
I used these data to construct the following variables, for each (2-digit
ICD9) condition in each of the years 1982-1996:
• N_CASES: the number of times the condition appeared in the NHIS
Condition File
• LIMITED: the number of people whose ability to perform their usual
activity was limited, mainly due to the condition
• UNABLE: the number of people who were unable to perform their
usual activity, mainly due to the condition
• RADAYS: the aggregate number of restricted activity days caused
by the condition in the 2 weeks preceding the interview
• BDDAYS: the aggregate number of bed days caused by the
condition in the 2 weeks preceding the interview
• WLDAYS: the aggregate number of work-loss days caused by the
condition in the 2 weeks preceding the interview
24
Model
log(Yit) = blab CUM_PRODit
+ bdrug CUM_DRUGit + g Zit + ai + dt + eit
where Y is one of the variables indicated
above (e.g. N_CASES or RADAYS), and Z
is a vector of demographic covariates
(mean age, mean education, and percent
male)
25
• Model was estimated via weighted least-squares, with
the weight equal to the mean (across years) value of Y
for condition i, i.e. Yi. = (1/15) ∑t Yit.
• I weight the observations by Yi. because measured
percentage changes of Y are more reliable, the larger is
Yi..
• In this model, blab may be interpreted as the percentage
response of Y to one additional clinical laboratory
product.
• Suppose, for example that when Y = UNABLE, blab = 0.2. This would imply that one additional clinical
laboratory product for a condition reduces the number of
people who were unable to perform their usual activity,
mainly due to the condition, by about 20%.
26
• The coefficients on both CUM_PROD and CUM_DRUG
in the log(N_CASES) equation are negative and
statistically significant, consistent with the hypothesis
that both laboratory and pharmaceutical innovation for a
condition reduce the number of people who report that
they have that condition at a given time.
• This could be due, in part, to innovation-induced
reductions in the duration of medical conditions. If
condition incidence (the number of new cases) remains
constant, but condition duration is reduced, then
condition prevalence (the number of people who have a
condition at a given time) is reduced.
27
• During the period 1990-1996, CUM_PROD
increased by about 0.033 per year, and
CUM_DRUG increased by about 0.99 per year.
• This implies that each kind of innovation
– reduced the number of people who were limited in
their ability to perform major activities by about 1.5%
per year
– reduced the number of people who were unable to
perform major activities by about 2.1% per year
28
Assessing the benefits in 1996
• Assess the benefits in 1996, in terms of reduced activity
limitations and disability days, of the laboratory
innovation that occurred since the beginning of the
sample period (1982)
• Compute ratio of the counterfactual value of the
dependent variable in 1996, had no laboratory innovation
occurred during 1982-1996, to the actual value of the
dependent variable in 1996. This is calculated as
exp(0.462 * blab).
• The estimates imply that, in the absence of any clinical
laboratory innovation during 1982-1996, the number of
people with limited ability to perform major activities in
1996 would have been 26% higher than it actually was,
and the number of restricted-activity days in 1996 would
have been 5% higher than it actually was.
29
• The estimates imply that, in the absence of laboratory
innovation during 1982-1996, the probability of being
limited in one’s major activity would have been 2.6
percentage points higher in 1996—12.6% rather than
10.0%.
• The probability of being unable to perform one’s major
activity would have been 1.7 percentage points higher in
1996—6.4% rather than 4.7%.
• Since the U.S. population was about 265 million in 1996,
this means that 4.4 million (= 1.7% * 265 million) more
people would have been unable to perform major
activities.
• The number of restricted-activity days per person per
year would have been 0.8 days higher—15.3 days rather
than 14.5 days.
30
• Estimate that, in 1996, per capita expenditure on medical and
diagnostic laboratory services was about $64.
• Per capita expenditure on new medical and diagnostic laboratory
services (those introduced since 1982) was probably much lower.
Suppose that half of total expenditure on medical and diagnostic
laboratory services was expenditure on new services, i.e. per capita
expenditure on new laboratory services was $32.
• By spending the hypothetical $32 per year on new laboratory
services, the average person’s probability of being unable to perform
a major activity is reduced by 1.7 percentage points. The net benefit
of laboratory innovation is positive as long as the value of being able
to perform a major activity is at least $1882 (= $32 / 1.7%).
• In 1996, average annual employee compensation (including fringe
benefits) was about $34,000/year. Moreover, reduced activity
limitations increase the value of each hour of leisure time as well as
the amount of time people can work.
31
Summary
Longevity
• Overall, both laboratory and pharmaceutical innovation have
increased longevity, but they have done so in different
(complementary) ways.
– Laboratory innovations have primarily reduced the risk of dying before
the age of 65.
– Pharmaceutical innovations have primarily extended the lives of older
people.
– Evidence about the age distribution of laboratory procedure and
pharmaceutical utilization was consistent with these findings.
• Laboratory innovation is estimated to have increased mean age at
death by 0.44 years during the period 1979-1998.
• This represents about 12% of the total increase in mean age at
death (3.57 years) during the period.
• An upper-bound estimate of the cost per life-year gained from
laboratory innovation is well below the cost-effectiveness thresholds
proposed by most medical decision-makers.
32
Summary
Activity limitations and disability days
• Laboratory and pharmaceutical innovation both reduced
activity limitations and disability days.
• Both types of innovation were estimated to reduce the
number of people who were unable to perform major
activities by about 2.1% per year.
• In the absence of laboratory innovation during 19821996, the probability of being unable to perform one’s
major activity would have been 1.7 percentage points
higher in 1996—6.4% rather than 4.7%.
• The value of the reduction in disability attributable to
laboratory innovation appears to exceed expenditure on
new laboratory procedures by a substantial margin.
33