Content uploaded by Marc Berger
Author content
All content in this area was uploaded by Marc Berger on Oct 12, 2017
Content may be subject to copyright.
HEALTH ECONOMICS
Health Econ. 13: 429–436 (2004)
Published online 4 March 2004 in Wiley InterScience (www.interscience.wiley.com). DOI:10.1002/hec.853
ECONOMIC EVALUATION
When does quality-adjusting life-years matter in
cost-e¡ectiveness analysis?
y
Richard H. Chapman
a,
*, Marc Berger
b
, Milton C. Weinstein
a
, Jane C. Weeks
c
, Sue Goldie
a
and
Peter J. Neumann
a
a
Program on the Economic Evaluation of Medical Technology, Harvard Center for Risk Analysis, Harvard School of
Public Health, Boston, Massachusetts, USA
b
Merck & Co., West Point, Pennsylvania, USA
c
Dana-Farber Cancer Institute, Boston, Massachusetts, USA
Summary
Purpose: This paper investigates the impact of quality-of-life adjustment on cost-effectiveness analyses, by
comparing ratios from published studies that have reported both incremental costs per (unadjusted) life-year and per
quality-adjusted life-year for the same intervention.
Methods: A systematic literature search identified 228 original cost–utility analyses published prior to 1998.
Sixty-three of these analyses (173 ratio pairs) reported both cost/LY and cost/QALY ratios for the same
intervention, from which we calculated medians and means, the difference between ratios (cost/LY minus cost/
QALY) and between reciprocals of the ratios, and cost/LY as a percentage of the corresponding cost/QALY ratio.
We also compared the ratios using rank-order correlation, and assessed the frequency with which quality-adjustment
resulted in a ratio crossing the widely used cost-effectiveness thresholds of $20 000, $50 000, and $100 000/QALY
or LY.
Results: The mean ratios were $69 100/LY and $103 100/QALY, with corresponding medians of $24 600/LY and
$20 400/QALY . The mean difference between ratios was approximately $34 300 (median difference: $1300), with
60% of ratio pairs differing by $10 000/year or less. Mean difference between reciprocals was 59 (QA)LYs per million
dollars (median: 2.1). The Spearman rank-order correlation between ratio types was 0.86 ðp50:001Þ. Quality-
adjustment led to a ratio moving either above or below $50 000/LY (or QALY) in 8% of ratio pairs, and across
$100 000 in 6% of cases.
Conclusions: In a sizable fraction of cost–utility analyses, quality adjusting did not substantially alter the
estimated cost-effectiveness of an intervention, suggesting that sensitivity analyses using ad hoc adjustments or
‘off-the-shelf’ utility weights may be sufficient for many analyses. The collection of preference weight data should
be subjected to the same scrutiny as other data inputs to cost-effectiveness analyses, and should only be under-
taken if the value of this information is likely to be greater than the cost of obtaining it. Copyright #2004 John
Wiley & Sons, Ltd.
Keywords quality-adjusted life-years; preference weights; utilities; cost–utility analysis; cost-effectiveness analysis
Copyright #2004 John Wiley & Sons, Ltd.
Received 27 June 2002
Accepted 16 June 2003
*Correspondence to: Harvard Center for Risk Analysis, 718 Huntington Avenue, Boston, MA 02115, USA.
E-mail: pneumann@hsph.harvard.edu
y
This paper was presented at the Annual Meeting of the Society for Medical Decision Making on September 26, 2000 in Cincinnati,
OH, USA.
Introduction and Objectives
Cost–utility analyses (CUAs) that measure
cost-effectiveness in costs per quality-adjusted
life-year (QALY) have increasingly become the
standard in cost-effectiveness analysis (CEA) [1].
The US Public Health Service’s Panel on
Cost-Effectiveness in Health and Medicine
(USPHS Panel) has recommended that, for
analyses intended to inform resource allocation,
a reference case should be included that measures
cost-effectiveness as incremental costs/QALY
(here abbreviated dC/dQALY) [2 (p. 122)]. How-
ever, many investigators still estimate and
report cost-effectiveness ratios as incremental
costs/life-year (dC/dLY), rather than as
dC/dQALY [3,4]. In addition, the collection of
utility data to quality-adjust years of life can
be expensive and resource-intensive. If quality-
adjusted and unadjusted analyses are expected to
produce similar results (that is, the addition of
quality-adjustment would not change decisions),
analysts might reasonably forego performing
complete reference case CEAs to save time and
money.
Analysts are, therefore, faced with the dilemma
of how to perform CUAs within reasonable
cost and time constraints. Without some
‘methodological triage,’ analysts run the risk of
producing theoretically elegant analyses that have
little chance to influence decisions because they
take so long to complete, or of never performing
them to begin with because they are too expensive.
The USPHS Panel explicitly recognized this
dilemma in their formulation of the ‘rule of
reason’ [2 (p. 71)]. This rule states that, in
designing CEA models, analysts should consider
the importance of each cost and effect component
in deciding whether to include an element in the
actual analysis.
While some authors have discussed the issue
of whether or not quality adjustment is
always necessary when conducting a CEA, mainly
from a theoretical standpoint [5–7], there
have been few attempts to use empiric data to
answer this question [8]. Our main objective in
this study was to explore how often decisions
might differ if made based on dC/dLY rather
than dC/dQALY, using published CEAs that
present ratios as both dC/dQALY and dC/dLY
for the same intervention. That is, how often did
the extra effort of collecting and using quality-
adjustments pay off by substantially affecting the
results of CEAs? Comparisons of these paired
ratios were used to:
(1) quantify the absolute and relative differences be-
tween the cost-effectiveness of the same inter-
ventionasmeasuredbydC/dQALY and dC/
dLY;
(2) explore how the differences between measures
might affect resource allocation decisions
made using commonly used thresholds; and,
(3) identify the factors causing dC/dQALY to be
substantially higher than dC/dLY (and vice
versa), and the factors associated with larger
differences.
Methods
Data sources
We explored the differences between dC/dLY and
dC/dQALY ratios using the Cost Utility Analysis
Database developed at the Harvard Center for
Risk Analysis [1,9], a comprehensive database of
published studies that is based on a systematic
literature search of relevant computerized data-
bases. The database contains 228 original CUAs
that present 645 dC/dQALY ratios published prior
to 1998. Details about the database have been
published previously [1,9,10] and are also available
on our Web site [11].
Of the 228 articles in the CUA database, 63
reported both dC/dQALY and dC/dLY for the
same intervention, either as baseline comparisons
or through sensitivity analyses on preference
weights (by setting health-related quality of life
adjustments to zero). The 63 studies reporting both
estimates contained 173 dC/dLY and dC/dQALY
ratio pairs. (Because cost-effectiveness analyses
often compare several possible programs or in-
tensity levels, a single article could contribute
several ratio pairs, ranging from 1 to 18 here.) To
allow cross-study comparisons on a common scale,
we converted all ratios into 1998 United States
dollars, using the appropriate foreign exchange
factors and the general Consumer Price Index [9].
All statistical analyses were performed using SPSS
for Windows, ver. 10.0.7 (SPSS, Inc., Chicago, IL).
Analytic plan
Quantifying the differences between dC/dQALY and
dC/dLY. To quantify the absolute differences
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
R. H. Chapman et al.
430
between cost-effectiveness as measured by dC/
dQALY and dC/dLY for the same intervention,
the quality-adjusted and unadjusted ratios were
tabulated in two ways, as the ratio differences
(calculated as dC/dLY minus dC/dQALY esti-
mates for the same interventions), and as recipro-
cal differences (calculated as dQALY/dCminus
dLY/dCand expressed as QALYs or LYs per
million dollars). The ratio differences provide a
measure of the change in estimated cost-effective-
ness brought about by quality-adjustment. We
also determined the number of times the differ-
ences between ratios were positive and the number
of times negative. Because the incremental costs
(dC) are identical in both ratios in each pair, the
differences between the ratios are driven by
differences between the reciprocals of the effective-
ness measures (dLY and dQALY), rather than by
differences between the effect sizes themselves.
Therefore, we also calculated an alternative
measure that is proportional to the effectiveness
difference, namely, the differences between the
reciprocals of the ratios, expressed as dLY or
dQALY per million dollars. These ‘effectiveness-
cost ratios’ emphasize the incremental health gains
per resources spent, a concept that may be more
intuitive and ethically satisfying to those who are
unfamiliar with economic evaluation than is the
cost per LY or per QALY [12]. We also compared
the relative sizes of the ratios, calculating dC/dLY
as a percentage of the corresponding dC/dQALY
ratio (dQALY/dLY).
To compare the numerical dC/dLY and dC/
dQALY ratios and their reciprocals, as well as
these measures of difference between them, we
calculated descriptive statistics, including medians
and arithmetic means. We compared the rankings
of interventions based on each ratio type by
calculating the Spearman rank-order correlation
between dC/dQALY and dC/dLY. Spearman
correlations were used rather than Pearson be-
cause the ratio distributions were heavily skewed.
We also took the log of each ratio to produce more
normal distributions, and calculated Pearson
correlations. Correlations near þ1 indicate that
the rank ordering of the quality-adjusted C=E
ratios closely matched the order of the unadjusted
ratios, while those near 0 indicate that there was
little relation between the two sets of rankings.
Threshold analysis. To explore how the differences
between dC/dLY and dC/dQALY ratios might
affect resource allocation decisions made using
each type of ratio, we determined the numbers of
cases where analyzed interventions crossed one of
several commonly used ‘decision thresholds’ [14]
when dC/dQALY is used rather than dC/dLY and
vice versa. Specifically, we examined how often the
use of quality-adjustment caused an estimated
ratio to move across key thresholds used in the
literature (such as $50 000 or $100 000/LY or/
QALY), or to the intervention being dominated
(that is, gaining fewer QALYs or LYs at higher
cost that the alternative). This allowed some
quantification of how often quality-adjustment
might ‘make a difference’ in actual resource
allocation decisions.
Identifying factors associated with differences.To
identify which aspects of a given intervention were
associated with differences across ratio pairs, we
used variables related to the natural history of the
conditions and interventions, all of which could be
specified prior to the conduct of a CEA. Three
investigators (RHC, SG, MB) classified each
condition and intervention in the data set into
the appropriate categories. (For example, is the
condition acute or chronic, or is the intervention
preventive, curative, or palliative?) Discrepancies
among responses were resolved by using the
majority response. For each variable, we recorded
the value which we hypothesized would be
associated with larger differences between the
quality-adjusted and unadjusted ratios. For ex-
ample, we hypothesized that interventions for
chronic conditions would be associated with larger
differences than those for acute diseases, because
any QoL effects would more likely be long-term
rather than short-term.
We then performed two regression analyses: one
that attempted to determine which factors were
associated with positive or negative ratio differ-
ences (logistic regression), and one for factors
associated with larger differences (linear regres-
sion). Binary logistic regressions were performed
with the sign of the ratio difference as the
dependent variable and the factors discussed
above as independent variables. This regression
indicates the types of conditions or interventions
for which quality adjustment was most likely to
make the estimated QALYs either higher (if
the sign of the ratio difference is positive) or lower
(if negative) than the corresponding estimate
of unadjusted life-years. We also performed multi-
variate regression analyses with the reci-
procal difference (dQALY/dCdLY/dC)asthe
When does Quality-Adjusting Life-Years Matter 431
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
dependent variable. We hypothesized that the
reciprocal differences variable would perform
better in the regression analyses because its
distribution was more normal than the distribu-
tion of the ratio difference (dC/dLYdC/dQALY)
variable, which was highly skewed. While trans-
formation of the dependent variable makes the
interpretation of regression results somewhat more
difficult, the use of reciprocals to normalize the
dependent variable’s distribution should lead to
more efficient regression estimates. We also
checked for autocorrelation of errors using the
Durbin–Watson statistic, because we were analyz-
ing at the ratio level and many studies reported
several ratios.
Results
Quantifying the differences between dC/
dQALY and dC/dLY
The 173 dC/dLY and dC/dQALY ratio pairs from
the 63 published CUAs are listed on our Web site
[12], along with the differences between ratio types,
the reciprocals of each ratio type and their
differences, and dQALY/dLY. In 33 of these 173
ratio pairs, the incremental cost/life-year or incre-
mental cost/QALY (or both) was estimated to be
cost-saving or dominated. Because no specific cost
per QALY could be associated with cost-saving or
dominated ratios, these ratio pairs were not used
in the quantitative analyses that follow, leaving us
with 140 valid numerical ratio pairs.
The dC/dLYdC/dQALY differences were po-
sitive in 85 cases (61%) and negative in 51 cases
(36%), with four ratio pairs showing no difference
(Table 1). That is, dC/dLY ratios were greater
than their paired dC/dQALY ratios more often
than the other way around. The median dC/dLY
and dC/dQALY ratios for the sample of identical
interventions were approximately $24 600/LY and
$20 400/QALY (Table 2), while the means, ap-
proximately $69 100/LY and $103 000/QALY
(Table 2), were much higher than their respective
medians. This contrast reflects the fact that the
distributions of dC/dLY and dC/dQALY ratios
were highly positively skewed, with the highest
ratios being approximately $2 400 000/LY and
$8 900 000/QALY (for the same intervention).
With this outlier removed, the mean dC/dQALY
equals approximately $35 000/QALY and mean
dC/dLY, approximately $44 000/LY.
The Spearman rank-order correlation was 0.86
(p50:001, Table 2). After log transformation, the
Pearson correlation between the two ratio types
was 0.84 (p50:001, Figure 1). The mean difference
between ratios (dC/dLYdC/dQALY) was ap-
proximately $34 300, with a median difference of
approximately $1300 (Table 2). As expected, the
distribution of the ratio differences was also
heavily skewed (Table 2, Figure 2). The differ-
ence between the maximum ratios mentioned-
Table 1. Number of times the ratio difference (dC/
dLYdC/dQALY) was positive, zero, or negative
ðn¼140Þ
Sign of ratio difference N%
Positive (dC/dLY>dC/dQALY) 85 60.7
Zero (dC/dLYffidC/dQALY) 4 2.9
Negative (dC/dLY5dC/dQALY) 51 36.4
Total 140 100.0
Table 2. Summary descriptive statistics (n¼63 studies)
Statistic dC/dQALY
(n=145)
dC/dLY
(n=142)
Ratio
difference
dQALY/dC(10
6
)
(n=145)
dLY/dC(10
6
)
(n=142)
Reciprocal
difference ( 10
6
)
(n=140) (n=140)
Median 20 366 24 600 1276 49 41 2.1
Mean 103 075 69 104 34 306 136 141 59
Standard
deviation
734 693 215 308 549 855 280 650 238
Skewness 11.89 9.48 11.52 4.45 11.08 4.54
(S.E.=0.20)
Spearman’s
rho
0.86
n
n
p50:001.
R. H. Chapman et al.
432
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
above leads to a negative outlier difference of
$6 400 000. When this outlier is excluded, the
mean ratio difference is approximately $11 800
(standard deviation=72 800). Sixty percent of the
ratio pairs ðn¼84Þdiffered from each other by
$10 000/year or less in absolute terms (Figure 2).
The means of the reciprocals of the ratios were
similar in value (at 141 LYs per million dollars and
136 QALYs per million dollars), as were the
medians (Table 2). The distributions of the
reciprocals of the ratios were less skewed than
those for the ratios themselves, as was that of the
reciprocal differences (Table 2). The mean reci-
procal difference was 59 (QA)LYs per million
dollars, with a median difference of only 2.1
(Table 2). When dC/dLY was expressed as a
percentage of the corresponding dC/dQALY ratio,
22% of the dC/dLY ratios were within 10% of the
corresponding dC/dQALY ratio, while 19% were
greater than 1.5 times the paired dC/dQALY ratio
(data not shown).
Threshold analysis
Quality-adjustment led to a previously unadjusted
ratio moving either above or below $50 000 in 14
of the 173 ratio pairs (8.1%; Table 3). Quality-
adjustment caused a ratio pair to cross $100 000 in
11 cases (6.4%; Table 3). Five interventions were
no longer dominated when dC/dQALY was used
rather than dC/dLY, while two cases that reported
positive dC/dLY ratios were dominated for dC/
dQALY (total=7 or 4.0%). Overall, the number
of interventions for which the use of quality
adjustment would lead to the estimated ratio
crossing any of these thresholds is 32, or approxi-
mately 18% of the 173 cases.
Identifying factors associated with differences
We performed a logistic regression analysis with
the sign of the ratio difference (positive or
negative) as the dependent variable, using the
condition- and intervention-level independent
variables in Table 4. Significant explanatory
variables in the final logistic regression model
were whether the condition is chronic and whether
negative long-term sequelae (defined as interven-
tions associated with negative side effects or
Ln (dC/dLY)
1614121086
Ln (dC/dQALY)
16
14
12
10
8
6
Figure 1. Scatter-plot of dC/dLY and dC/dQALY (on a
natural log scale), Pearson correlation=0.84, p50:001
dC/dLY - dC/dQALY
52048044040036032028024020016012080400-40
Count of ratio diferences
100
80
60
40
20
0
Figure 2. Distribution of ratio differences (dC/dLYdC/
dQALY). (Graph does not show two negative outliers of
$485 000 and $6 437 000)
Table 3. Number of times adjustment for QoL caused
an estimated cost-effectiveness ratio to cross commonly
used thresholds ðn¼173Þ
Threshold value
($/LY or $/QALY)
Ratio pairs
N%
$50 000 14 8.1
$100 000 11 6.4
Dominated 7 4.0
Total 32 18.5
When does Quality-Adjusting Life-Years Matter 433
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
unintended outcomes that last longer than 2 years)
are associated with the intervention (Table 4). If a
condition was chronic, it was associated with an
increase in the likelihood of a positive ratio
difference (dC/dLY>dC/dQALY), while the pre-
sence of long-term negative sequelae was asso-
ciated with negative ratio differences.
For the linear regression model with reciprocal
differences as the dependent variable (Table 4),
we found that chronic conditions and palliative
interventions were significantly associated with
positive increases in the reciprocal difference. The
presence of long-term negative sequelae
were associated with decreases in the reciprocal
difference. However, this model explained only
about 13% of the variation in the reciprocal
differences (adjusted R2¼0:13). Attempts to
account for more of the observed variance,
through transformations of the dependent variable
and the inclusion of additional explanatory vari-
ables, were not successful. Examination of the
Durbin–Watson statistic (1.54) reveals no defini-
tive indication that autocorrelation of errors
within studies is a problem in this model, even
though we are analyzing at the ratio rather than
study level.
Discussion
Cost-effectiveness analysis is one tool to help
inform resource allocation decisions and the
prioritization of medical interventions [14,15].
We identified 63 studies that reported both
quality-adjusted and unadjusted cost-effectiveness
ratios estimated for the same intervention, which
provided us with a unique opportunity to explore
systematically the relation between quality-ad-
justed and unadjusted ratios. The two ratio types
were highly correlated, and differences between
them appeared relatively small in over two-thirds
of the cases. Although quality-adjusting life-years
is now widely advocated for cost-effectiveness
analyses, we found that in most cases quality
adjustment had relatively little effect on the final
estimated cost-effectiveness ratio. This suggests
that in many studies, quality adjustment with
minimal data collection (for example, ad hoc
adjustments or previously published utility weights
for health states) may be adequate to obtain a
reasonable estimate of the cost-effectiveness of a
given intervention.
However, we also found that quality-adjustment
had a sizeable impact in a small but non-trivial
fraction of cases. Quality-adjustment led to a ratio
moving across potential ‘cost-effective thresholds’
(such as $50 000/LY or QALY) in almost one-fifth
of the cases in this data set. This suggests that
some form of sensitivity analysis or value of
information analysis on the importance of qual-
ity-adjustment should be undertaken before decid-
ing how much effort to put into the collection of
QoL or preference weight data. For example, if a
value of information analysis indicated that the
expected value of the optimal choice with QoL
Table 4. Independent variables used in the multivariate analyses, and results of (A) the final logistic regression
model for the sign of the ratio difference (positive or negative) ðn¼136Þand (B) the final multivariate linear
regression model with reciprocal differences ((dQALYdLY)/dC10
6
) as the dependent variable (adjusted
r2¼0:13)
Variable Categories (A) Logistic regression (B) Linear regression
B(S.E) PBp
Explanatory, condition level
Term Acute, chronic
a
+1.44 (0.44) 0.001 117.43 (46.12) 0.012
Symptomatic? Y
a
/N NS
b
NS
Explanatory,
intervention level
Term Acute, chronic
a
, intermittent NS NS
Purpose Preventive, curative, palliative NS 187.73 (46.36) [palliative] 50.001
Negative sequelae? No, short-, long-term 1.17 (0.45) 0.009 161.16 (53.11) 0.003
a
Value of variable for which we hypothesized that differences between ratios would be greater.
b
NS=not significant.
R. H. Chapman et al.
434
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
information minus the expected value without
that information is smaller than the expected
cost of obtaining that information, then the
analyst should not undertake the collection of
QoL data. If other variables are likely to have a
greater effect on the final estimate of cost-
effectiveness, analysts might focus their attention
on obtaining better estimates of those inputs
rather than on the collection of health utilities
for quality adjustment. This information may be
useful because the accelerating pipeline of new
technology has the potential to overwhelm the
resources available to evaluate their cost-effective-
ness. Of course, analysts may wish to include
quality adjustment even in studies where it is not
expected to make much difference, so that a
reference case exists for comparison with other
studies [2].
To decide a priori whether quality-adjustment is
important to a given analysis, it would be helpful
to know which types of conditions and interven-
tions are most likely to be associated with
substantial differences between quality-adjusted
and unadjusted ratios. In our regression analyses,
quality-adjustment seemed to be most important
when the condition being studied was chronic,
when the intervention was palliative, and when the
intervention included long-term negative sequelae.
Interventions that are provided for chronic condi-
tions were associated with larger positive recipro-
cal differences, indicating that these interventions
are more likely to increase the incremental QALYs
gained than are those for acute conditions. This
finding confirms our prior belief that quality-
adjustment would be more important for chronic
conditions than for acute ones, because these
would be more likely to have long-term effects on
QoL. Using cost-per-QALY as the measure of
cost-effectiveness, rather than cost-per-LY, may
therefore make interventions for chronic condi-
tions appear relatively more cost-effective than
those for acute conditions, on average. Palliative
interventions, where the main purpose is to
improve QoL rather than to extend life, were also
associated with larger positive reciprocal differ-
ences. Interventions with long-term negative se-
quelae are associated with large negative reciprocal
differences. The presence of long-term side effects
from an intervention would be expected to cause a
decrease in the incremental QALYs relative to the
LYs gained from that intervention. It should be
emphasized, however, that these results are only a
rough guide, and should not be taken to indicate
that the absence of these factors means that
quality-adjustment is not important in a given
analysis (or that their presence implies that it will
necessarily be important).
There are several limitations to these analyses.
While our literature search included several
computerized databases, we were not able to
identify analyses that were either in unlisted
publications or not published at all. Thus, any
publication bias in cost–utility analyses would be
reflected in our database. A bias might exist, for
example, if investigators are more likely to per-
form a cost-per-QALY study if they anticipate
important differences from cost-per-LY. In addi-
tion, because our search algorithm was focused on
finding CUAs, we may have missed some studies
that estimated both costs per QALY and costs per
life-year. Also, this group of CEAs provides only
an incomplete snapshot of the field, and may not
be generalizable to other settings (because of
differences in disease prevalence, medical practice
or costs). Finally, the above analyses assume that
studies performed the QoL adjustments correctly.
To be complete, we would need to assess whether
all relevant effects (e.g. non-fatal side effects and
non-fatal disease effects) were included and valued
correctly.
In summary, we found that in many individual
cost–utility analyses published before 1998, quality
adjusting has not substantially affected results,
while in others it had substantial effects. If
the intervention or condition being studied is
not expected to have much impact on QoL relative
to mortality effects, and if time and resources
are especially constrained, analysts may choose
to forgo quality-adjustment altogether, allowing
them to concentrate efforts on data inputs
for which the value of information is higher.
Further research should confirm the factors that
determine when quality adjustment will be most
important.
Acknowledgements
Financial support for this study was provided in part by
a grant from the National Science Foundation and
Merck & Co., Inc. under the joint NSF/Private Sector
Research Opportunity Initiative (SBR-9730448). RHC
received financial support from training grants from the
National Library of Medicine and Agency for Health
Care Policy and Research (now the Agency for
Healthcare Research and Quality).
When does Quality-Adjusting Life-Years Matter 435
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
References
1. Neumann PJ, Stone PW, Chapman RH, Sandberg
EA, Bell CM. The quality of reporting in published
cost–utility analyses, 1976–1997. Ann Intern Med
2000; 132: 964–972.
2. Gold MR, Siegel JE, Russell LB, Weinstein MC
(eds). Cost-Effectiveness in Health and Medicine.
Oxford University Press: Oxford, 1996.
3. Tengs TO, Adams ME, Pliskin JS, et al. Five-
hundred life-saving interventions and their cost-
effectiveness. Risk Anal 1995; 15: 369–390.
4. Graham JD, Corso PS, Morris JM, Segui-Gomez
M, Weinstein MC. Evaluating the cost-effectiveness
of clinical and public health measures. Annu Rev
Public Health 1998; 19: 125–152.
5. Garber AM, Phelps CE. Economic foundations of
cost-effectiveness analysis. J Health Econ 1997; 16:
1–31.
6. Luce BR. Cost-effectiveness analysis: obstacles to
standardization and its use in regulating pharma-
ceuticals. PharmacoEconomics 1993; 3(1): 1–9.
7. Wright JC. Investments that save lives: The norms of
environmental and medical decision making. Thesis
presented to Committee on Higher Degrees in
Public Policy, Harvard University, Cambridge,
MA; May 1997; 4–29.
8. Gerard K. Cost–utility in practice: a policy maker’s
guide to the state of the art. Health Policy 1992; 21:
249–279.
9. Chapman RH, Stone PW, Sandberg EA, Bell C,
Neumann PJ. A comprehensive league table of
cost–utility ratios and a sub-table of ‘Panel-
worthy’ studies. Med Decis Making 2000; 20:
451–467.
10. Stone PW, Chapman RH, Sandberg EA, Liljas B,
Neumann PJ. Measuring costs in cost–utility
analyses: Variations in the literature. Int J Technol
Assess Health Care 2000; 16: 111–124.
11. Harvard Center for Risk Analysis, Harvard School
of Public Health. The CUA data base: standardizing
the methods and practices of cost-effectiveness
analysis; available from: hhttp://www.hsph.harvard.
edu/cearegistry/i.
12. Eddy DM. Cost-effectiveness analysis: a conversation
with my father. J Am Med Assoc 1992; 267: 1669–1675.
13. Hirth RA, Chernew ME, Miller E, Fendrick M,
Weissert WG. Willingness to pay for a quality-
adjusted life year: in search of a standard. Med
Decis Making 2000; 20: 332–342.
14. Weinstein MC, Zeckhauser R. Critical ratios
and efficient allocation. J Public Econ 1973; 2:
147–157.
15. Russell LB, Siegel JE, Daniels N, Gold MR, Luce
BR, Mandelblatt JS. Cost-effectiveness analysis as a
guide to resource allocation in health: Roles and
limitations. In: Cost-Effectiveness in Health and
Medicine, Gold MR, Siegel JE, Russell LB, Wein-
stein MC (eds). Oxford University Press: Oxford,
1996 (Chapter 1).
R. H. Chapman et al.436
Copyright #2004 John Wiley & Sons, Ltd. Health Econ. 13: 429–436 (2004)
