MAJOR ARTICLE
Assessing Resistance Costs of Antiretroviral Therapies
via Measures of Future Drug Options
Hongyu Jiang,1 Steven G. Deeks,4 Daniel R. Kuritzkes,2 Marc Lallemant,6 David Katzenstein,5 Mary Albrecht,3
and Victor DeGruttola1
1
Department of Biostatistics, Statistical and Data Analysis Center, Harvard School of Public Health and 2Section of Retroviral Therapeutics,
Brigham and Women’s Hospital and Division of AIDS, Harvard Medical School, Harvard University, and 3Beth Israel Deaconess Medical Center,
Boston, Massachusetts; 4Department of Medicine, University of California at San Francisco and San Francisco General Hospital, San Francisco,
and 5Stanford University Medical Center, Stanford, California; 6Research Unit 054, Institut de Recherche pour le Développement, Paris, France,
and Chiang Mai, Thailand
The emergence of drug-resistant human immunodeficiency virus (HIV) type 1 in the setting of antiretroviral
therapy failure limits the number of drugs available for use in subsequent therapy regimens. To quantify the
relative HIV-1–resistance costs associated with various antiretroviral therapy strategies, we developed 2 related
measures of future drug options (FDOs) by use of rule-based genotype-interpretation systems. The FDO1
metric assesses the number of drug classes that remain useful; the FDO2 metric assesses the number of drug
classes that remain useful and the number of active drugs within each class. Application of these methods is
illustrated with data from a randomized study of 3 therapy regimens in nucleoside analog–experienced patients.
Each therapy regimen resulted in a unique pattern of drug-resistance (and cross-resistance) mutations. The
regimen with the highest virologic failure rate preserved greater future drug options. Quantification of future
drug options as an outcome of antiretroviral therapy trials may complement traditional clinical, virologic,
and immunologic end points, thereby providing novel insights.
Highly active antiretroviral therapy (HAART) fails to
result in complete virus suppression for many patients
[1, 2]. The reasons for this failure are often multifactorial and include nonadherence, use of suboptimal regimens with inadequate potency, infection with drugresistant variants, accelerated drug metabolism, and
advanced immunodeficiency [3–5]. As a consequence
of incomplete virus suppression, viral replication occurs
in the presence of a strong selective pressure, and, after
a varying period of time, resistance-associated mutations emerge [6]. Because resistance to 1 drug often
Received 21 January 2003; accepted 29 April 2003; electronically published 23
September 2003.
Presented in part: XI International HIV Drug Resistance Workshop, 2–5 July
2002, Seville, Spain (abstract 99).
Financial support: National Institutes of Health (grants AI24643, AI38855,
AI32770, AI27659, and RR16482); Adult Clinical Trials Group (Virology Support
Laboratory contract AI38858).
Reprints or correspondence: Dr. Hongyu Jiang, Dept. of Biostatistics, Harvard
School of Public Health, 655 Huntington Ave., Boston, MA 02115 (hjiang@
sdac.harvard.edu).
The Journal of Infectious Diseases 2003; 188:1001–8
2003 by the Infectious Diseases Society of America. All rights reserved.
0022-1899/2003/18807-0010$15.00
confers cross-resistance to other drugs within a class
[7] and because the virus and host factors that led to
the initial failure often persist during the second regimen, many patients for whom an initial regimen fails
often are unable to achieve a durable response to a
subsequent “salvage” regimen. High virologic failure
rates during salvage therapy are seen even when new
therapeutic classes are used [8, 9].
Although drug resistance results in loss of virologic
activity and increasing levels of viremia, many patients
continue to derive immunologic and clinical benefit
from therapy, even after highly resistant variants emerge
(as long as therapy is maintained) [2, 10, 11]. Again,
the reasons for this phenomenon are multifactorial and
include continued partial activity of therapy against the
highly resistant variant, reduced viral “fitness” of the
drug-resistant variant, and, perhaps, enhanced human
immunodeficinecy virus (HIV)–specific immune responses [12]. Many patients in clinical practice are
now continuing to receive stable therapy, despite ongoing viral replication, because of the continued benefit
conferred by partially effective therapy regimens and
the risk of rapidly rendering multiple drugs within
Resistance Costs of HAART
• JID 2003:188 (1 October) • 1001
each drug class useless when aggressive switch strategies are
used. This approach is commonly used, despite evidence that
further virus evolution is likely to occur with continued therapy
[13]. The extent of reduction in chances of achieving virus
suppression (resistance cost) of such a strategy in clinical trials
or observational studies is difficult to measure because of the
lack of a standardized methodology for quantifying the effect
of resistance mutations on future therapeutic options.
There are fewer options for management of virologic failure
in resource-poor regions. Access to antiretroviral drugs is increasing for 98% of HIV-infected individuals, who live in
resource-poor regions [14, 15]. However, in developing countries, drug options, monitoring tools, and sequential therapeutic schemes are limited. Thus, systematic identification and
adoption of simple and effective “therapy option–sparing”
strategies are vital.
On the basis of these concerns, we have developed a methodology that quantifies the effect of any given therapeutic strategy on drug resistance. Specifically, we sought to define a metric
that quantifies the number of antiretroviral drugs that are likely
to be effective, on the basis of genotypic and/or phenotypic
resistance measurements at specified time points after study
entry, in the context of a study of antiretroviral therapy. Currently available measures of resistance, such as the genotype
summary score, make use of genotype-interpretation systems
to count the number of drugs in a patient’s regimen to which
the patient’s virus is expected to be sensitive [16, 17]. Here,
we extend this concept to consider sensitivity to all possible
future drug regimens and apply our metric to a recently published randomized clinical study [18]. Our end points, denoted
“measures of future drug options” (FDOs), summarize the susceptibility of the viruses from each study regimen to all available
drugs and compare FDOs remaining between study arms.
Larger values of FDO end points correspond to greater drug
susceptibility and more drug/regimen options in the future.
METHODS
Quantification of resistance. To help in interpretation of
viral genotype, systems have been developed to assess whether
a virus strain is expected to be susceptible to US Food and
Drug Administration (FDA)–approved drugs. Here, we explore
the application of 2 FDO measures, each of which is based on
binary genotypic-sensitivity scores derived from a genotypeinterpreting system.
FDO measure 1 (FDO1). FDO1 evaluates the number of
drug classes to which a virus strain is sensitive and assigns extra
credit to full-class susceptibility to the nucleoside/nucleotide
reverse-transcriptase inhibitor (NRTI) or protease inhibitor
(PI) classes. FDO1 is defined as follows: on the basis of the
viral genotype, a patient’s virus is scored as susceptible (1 point)
1002 • JID 2003:188 (1 October) • Jiang et al.
or resistant (0 points) to each antiretroviral drug by use of a
genotype-interpretation system. The number of drug classes
that include at least 1 drug to which the patient’s virus is
susceptible (NC) is then determined. Because remaining susceptible to all drugs in the NRTI or PI class offers greater
flexibility in selecting a subsequent regimen, remaining fully
susceptible to the drugs in these classes is given extra weight
by adding 0.3 to NC for each class for which this is true; we
exclude the nonnucleoside reverse-transcriptase inhibitor
(NNRTI) class because partial susceptibility within this class is
unlikely [8]. The resulting score is the value of FDO1. Considering all 3 drug classes (NRTI, NNRTI, and PI), FDO1 has
9 possible values: 0, 1, 1.3, 2, 2.3, 2.6, 3, 3.3, or 3.6.
FDO measure 2 (FDO2). FDO2 is determined by 2 components: the number of drug classes that include at least 1 drug
to which the patient’s virus is susceptible and the total number
of effective drugs within each class. In determining the order
of FDO2, the first component is deemed to be more important
than the second, which can be used to break ties. FDO2 is
calculated as follows: the susceptibility of virus to each drug
and the number of drug classes containing at least 1 active drug
(NC) is determined as for FDO1. FDO2 is then calculated using
the following formula: NC + ND/TND , where ND is the total
number of active drugs, and TND is a constant that guarantees
that ND/TND is always a fraction between 0 and 1.
The constant TND is defined on the basis of the number of
drugs within the classes being studied. For example, there are
currently 7, 3, and 6 FDA-approved drugs in the NRTI, NNRTI,
and PI classes, respectively. Considering these 3 classes of drugs,
we might choose TND to be 17, the total number of drugs (16)
plus 1. Thus, the ratio of ND to TND will always be !1. Note
that this system ensures that the different combinations of NC
and ND achieve FDO2 values that are strictly ordered and are
determined primarily by the number of susceptible drug classes
(NC) and secondarily by the total number of susceptible drugs
(ND). In all cases, the resulting FDO2 is an ordinal variable
taking no more than 34 values between 0 and 4.
Statistical methods. Three types of scientific questions
may be addressed with the FDO metric, each of which requires
unique analytical approaches. First, therapy effects on resistance
over some fixed period of follow-up (e.g., 48 or 96 weeks from
study entry) may be compared by measuring the FDO end
points at the start and the end of the period. Resistance cost
(CP) is defined as the difference in FDO end points at those 2
time points. Rank-based tests can be used to identify the presence of therapy effect. One concern that arises is how to account
for patients for whom resistance-testing data are not available
(either because the level of viremia is low or undetectable or
because a patient prematurely discontinues therapy). For patients with low or undetectable levels of viremia, one can use
the last genotype data available and assume that no additional
mutations have arisen since that time—a reasonable assumption given the limited risk of new resistance associated mutations in patients with durable virologic control during HAART
[19]. Patients who have prematurely interrupted therapy may
no longer have detectable levels of drug resistance at the end
of the period of interest (given the observation that such mutations often wane in the absence of drug pressure). The last
genotype data available during active therapy could be used for
such patients.
Second, we can compare resistance costs of therapies over
variable periods of time—for example, time from start of therapy to virologic failure. The resistance costs associated with
virologic failure (CVF) can be quantified by calculating the
changes in FDO measures from start of therapy to virologic
failure. Because of the limited follow-up time and drug potency,
virologic failures are observed in only a fraction of patients.
Resistance cost CVF is measurable only in patients with observed
virologic failure. Our analysis estimates an average resistance
cost associated with mean time to virologic failure for patients
receiving the same therapy. The special estimation procedure
is detailed in the Appendix [20–22]. After obtaining the estimates of resistance costs and their variances, for each therapy
arm, we may conduct pairwise comparison between therapy
arms, in terms of CVF, by use of 2-sample t tests.
Third, because therapies may differ in both effective duration
and resistance costs at failure, we propose to compare a
duration-adjusted resistance cost measure (dCVF) among therapy arms, which is the ratio of resistance cost at virologic failure
(CVF) to the effective duration of therapy (T). The durationadjusted resistance cost measure may be interpreted as the average amount of FDOs expended per unit of time over the
effective duration of therapy. The estimation and testing procedures for dCVF are similar to those for CVF.
Application of FDO in clinical trials. To illustrate our
methods, we used data derived from the AIDS Clinical Trials
Group (ACTG) 364 study, a randomized, multicenter, 3-arm
clinical trial comparing PI nelfinavir and/or the NNRTI efavirenz in combination with 2 nucleoside analogs in HIVinfected patients with prolonged but exclusive NRTI exposure
(median, 5.6 years) [18]. All patients were PI and NNRTI naive
at entry. In ACTG 364, 195 patients were randomly assigned
to 1 of the 3 therapy arms, denoted as nelfinavir, efavirenz, or
nelfinavir plus efavirenz, respectively (all patients also received
nucleoside analogs). Our analysis focused on 161 patients for
whom baseline genotypic data were available. When a patient’s
virus load was confirmed to be 12000 HIV RNA copies/mL at
or after week 16 (used as the definition of virologic failure for
our purposes), a genotypic test was conducted. Eighty patients
experienced virologic failure and received a postbaseline genotypic test before study closure. The remaining 81 patients
did not experience viologic failure at the time of study closure.
Written, informed consent was obtained from all patients or
their guardians, and human-experimentation guidelines of the
US Department of Health and Human Services and the individual institutions were followed in the conduct of this research.
Both FDO1 and FDO2 are used to illustrate the comparison
of the resistance costs across therapy arms. In the first step of
computing these 2 measures, we use the Stanford Inferred Drug
Resistance Score Interpretation Rules (Beta Test) to obtain
the binary sensitivity scores to each drug (available at: http://
hivdb.stanford.edu) [23]. A cutoff point of 15 was chosen, to
distinguish between susceptible and resistant, according to the
interpreting rules of the Stanford interpretation system. Any
other interpreting system could also be used. This flexibility in
defining FDO end points ensures that the most clinically meaningful interpreting system can be used and, conversely, that one
can test or compare different interpretation systems using the
same genotypic data set.
RESULTS
Genotypic resistance and FDOs at study entry (ACTG 364).
Most of the 161 patients had evidence of significant NRTI
resistance at baseline. The most prevalent resistance mutations
at baseline (with ⭓10% prevalence) were NRTI mutations:
M41L (58.4%), D67N (36.7%), K70R (36.7%), V118I (26.7%),
M184V (57.8%), L210W (34.8%), T215Y (48.5%), T215F
(15.5%), and K219Q (23.7%). None of the NNRTI mutations
was detected, but 2 minor PI mutations (polymorphisms unrelated to PI responses) were found in the baseline genotypes
(L63P [58.4%] and V77I [29.2%]). Mutation patterns did not
differ significantly between the therapy arms, with respect to
polymorphisms.
Table 1 shows the overall baseline resistance in ACTG 364
patients, in terms of measures of future drug options. In deriving FDO2, TND was set to be a constant 20. The distributions
of baseline FDO1 and FDO2 in this study did not differ much,
with FDO2 having a slightly larger range. As expected, the
Kruskal-Wallis test results showed no difference in the distributions of these baseline measures across the 3 therapy arms.
Figure 1 shows the time to virologic failure for the 3 therapy
arms, on the basis of the 161 patients for whom baseline genotypic data were available. When the study ended, 36%, 44%,
Table 1. Summary of baseline resistance in patients enrolled
in AIDS Clinical Trials Group 364.
Median (IQR)
Pa
Resistance measure
Mean (SE)
FDO1
2.86 (0.55)
3.3 (2.3–3.3)
.23
FDO2
3.06 (0.60)
3.5 (2.45–3.55)
.36
NOTE. FDO, measures of Future drug option (see Methods for an explanation of the difference between FDO1 and FDO2); IQR, interquartile range.
a
Calculated by use of the Kruskal-Wallis rank sum test for equal median.
Resistance Costs of HAART • JID 2003:188 (1 October) • 1003
Figure 1. Curves of virologic failure free probabilities, by therapy arms
in AIDS Clinical Trials Group 364. Nos. and percentages are the sample
sizes and proportion of patients without virologic failures in each arm,
respectively. EFV, efavirenz triple therapy; NFV, nelfinavir triple therapy;
NFV + EFV, quadruple therapy.
and 72% patients in the nelfinavir, efavirenz, and nelfinavir
plus efavirenz arms, respectively, had not experienced virologic
failure (P p .0005, log rank test). The median time to virologic
failure in the nelfinavir arm was ∼8.0 months, compared with
∼26 months in the efavirenz arm and 134 months in the nelfinavir plus efavirenz arm. The pattern seen in figure 1 is consistent with the data analysis results using all 195 patients [18].
Genotypic resistance at first virologic failure (ACTG 364).
Before performing the resistance costs analyses, we examined
the viral genotypic profiles at the time of patients’ first virologic
failure during the study. Table 2 shows that viruses isolated
from 75% of patients in the efavirenz arm and from 57% of
patients in the nelfinavir plus efavirenz arm developed an
efavirenz-resistance mutation (K103N and G190A/S) at the
time of virologic failure. Viruses isolated from 45% of patients
for whom the nelfinavir regimen failed developed a major
nelfinavir-resistance mutation (D30N); 20%–30% developed a
minor nelfinavir-resistance mutation (M36I and N88D/S). In
contrast, viruses isolated from only 2 of 14 patients for whom
the nelfinavir plus efavirenz regimen failed developed the D30N
mutation (table 2).
Among the 80 patients who experienced virologic failure,
viruses isolated from 14 developed the M184V mutation,
whereas viruses isolated from 26 patients who had the M184V
mutation at baseline reverted to wild type at the first virologic
failure. Although there is evidence showing that the reversion
to wild type in the consensus pol sequence occurs in patients
who discontinue lamivudine therapy [24], it is likely that the
M184V persists below assay detection and will affect future
options. To address this issue, we did 2 sets of analysis by
assuming either that reversion represented true susceptibility
1004 • JID 2003:188 (1 October) • Jiang et al.
or that all mutations observed in past and current genotype
testing persisted when calculating the current values of FDO
end points, noted as “reversion to wild type possible” and “reversion to wild type not possible,” respectively.
Resistance-costs comparison at week 48. We compared
the resistance cost across the 3 study arms over the period from
study entry to 48 weeks after entry. If a patient’s week-48 genotype data were missing, we use the last available genotype
data to compute the FDO. Of the 161 patients, 98 experienced
no virologic failure by week 48 and had only baseline genotype
data available (27, 31, and 40 patients in the nelfinavir, efavirenz, and nelfinavir plus efavirenz arms, respectively); another
34 patients had only 1 postentry genotype test at or before week
48 (16, 12, and 6 patients in the nelfinavir, efavirenz, and nelfinavir plus efavirenz arms, respectively); the remaining 29 patients had 2 postentry genotype tests by week 48 (16, 9, and 4
patients in the nelfinavir, efavirenz, and nelfinavir plus efavirenz
arms, respectively).
Table 3 shows the mean resistance costs by therapy arms and
the P values calculated with the Kruskal-Wallis test (a rank sum
test for equivalence in the cost distribution), where CW48-1 and
Table 2. New mutations detected at the time of virologic failure
in patients enrolled in AIDS Clinical Trials Group 364.
Study arm
Outcome
NFV
(n p 59)
EFV
(n p 52)
Patients experiencing
a
virological failure
38 (64)
28 (54)
Time to failure, median months
8.2
NFV + EFV
(n p 50)
14 (28)
26
134
b
New mutation at failure
D30N
17 (44.7)c
0 (0)
2 (14.3)d
M36I
12 (31.6)
0 (0)
2 (14.3)
N88D/S
8 (21.1)
0 (0)
2 (14.3)
M46I/L
4 (10.5)
0 (0)
0 (0)
K103N
0 (0)
8 (57.1)e
21 (75)
f
G190A/S
0 (0)
3 (10.7)
0 (0)
L90M
0 (0)
0 (0)
1 (7.1)
NOTE. Data are no. (%) of patients, except where noted. EFV, efavirenz
triple therapy; NFV, nelfinavir triple therapy; NFV + EFV, quadruple therapy.
a
Percentages were calculated relative to the total no. of patients in each
therapy arm.
b
Percentage of patients with a new mutation was calculated relative to
the no. of patients in each therapy arm who experienced virological failure,
not the total no. of patients in each therapy arm.
c
Among the 17 patients, 5 had only the D30N mutation, 8 had both the
D30N and M36I mutations, 2 had the D30N and N88D mutations, 1 had the
D30N and M46I mutations, and 1 had the D30N, M46L, and N88S mutations.
d
None of the 2 patients had the K103N mutation, but both had the M36I
mutation; 1 of them also had the N88D and L90M mutations.
e
Among the 8 patients who had the K103N mutation, only 1 developed a
protease inhibitor mutation (N88S) as well.
f
Among the 3 patients, 1 had the G190S mutation without the K103N
mutation, and 2 had the G190A mutation with the K103N mutation.
Table 3. Comparison of resistance costs induced by therapies
from entry to 48 weeks after study entry.
Possibility of reversion to
wild type, resistance cost
end point
Mean resistance cost of
therapy, by study arm
NFV
EFV
NFV + EFV
Pa
CW48-1
⫺0.046
0.23
0.16
.07
CW48-2
⫺0.13
0.27
0.17
.002
CW48-1
0.18
0.37
0.22
.38
CW48-2
0.10
0.42
0.24
.20
Possible
Not possible
NOTE. CW48-1 and CW48-2, 48-week changes in resistance cost measures
future drug option (FDO) 1 and FDO2, respectively; EFV, efavirenz triple therapy; NFV, nelfinavir triple therapy; NFV + EFV, quadruple therapy.
a
Calculated by use of the Kruskal-Wallis rank sum test for testing the
equivalence in the cost distribution among the 3 therapy arms.
CW48-2 denote 48-week changes in resistance cost measures
FDO1 and FDO2, respectively. Assuming reversion to wild type,
there is a marginally insignificant therapy effect on 48-week
changes in the FDO1 measure, and a significant difference in
48-week changes in the FDO2 measure across therapy arms.
Assuming no reversion to wild type, there is no significant
therapy effect on 48-week resistance cost in terms of either
FDO1 or FDO2.
Comparison of resistance costs between study entry and
time of virologic failure. We compared the resistance costs
between study entry and the time of virologic failure across the
3 therapy arms. The result is again sensitive to the assumption
whether reversion to wild type at virologic failure represents
true drug susceptibility. If complete reversion to wild type is
assumed to be possible, the resistance cost observed at virologic
failure to the nelfinavir regimen is not significantly different
from 0, whereas the other 2 therapies have nontrivial resistance
costs (table 4). Two-sample t tests show that failure of the
nelfinavir regimen leads to significantly lower reductions in
FDOs, compared with failure of the efavirenz regimen (CVFFDO1: mean difference, ⫺0.57; P p .001 ; CVF-FDO2: mean difference, ⫺0.85; P ! .0001) and compared with the nelfinavir plus
efavirenz regimen (CVF-FDO1: mean difference, ⫺0.34; P p .03;
CVF-FDO2: mean difference, ⫺0.49; P p .004).
When complete reversion to wild type is assumed not to be
possible, resistance costs to all therapies are significantly 10.
Two-sample t tests show that failure of the nelfinavir regimen
does not lead to a significant difference in reduction of FDOs,
compared with failure of the nelfinavir plus efavirenz regimen
(CVF-FDO1: mean difference, ⫺0.06; P p .66 ; CVF-FDO2: mean difference, ⫺0.17; P p .17), but does significantly lower reduction
in FDOs, compared with failure of the efavirenz regimen (CVFFDO1: mean difference, ⫺0.49; P p .001 ; CVF-FDO2: mean difference, ⫺0.77; P ! .0001).
Comparison of duration-adjusted resistance costs between
entry and time of virologic failure. Because the 3 therapy
regimens in ACTG 364 have quite different effective duration
and resistance costs, it is of interest to consider both aspects and
compare the effective duration-adjusted resistance costs from
study entry to time of virologic failure. In table 4, dCVF-FDO1 and
dCVF-FDO2 represent the average effective duration-adjusted resistance cost in terms of FDO1 and FDO2, respectively.
When reversion to wild type is assumed to be possible, the
efavirenz and nelfinavir plus efavirenz study arms both have
significantly 10 duration-adjusted resistance cost, in terms of
both dCVF-FDO1 and dCVF-FDO2 measures. Of note, the efavirenz
arm has significantly higher rate of expending FDO over the
effective duration than does the nelfinavir arm (P p .006, for
dCVF-FDO1, and P ! .0001, for dCVF-FDO2 [2-sample t test]). When
reversion to wild type is assumed not to be possible, all therapy
arms have significantly 10 duration-adjusted resistance cost, in
terms of both dCVF-FDO1 and dCVF-FDO2 measures. There is no
Table 4. Comparison of resistance costs and duration-adjusted
resistance costs from entry to virologic failure (VF) among AIDS
Clinical Trials Group 364 study arms.
Reversion to wild
type not possible
Reversion to wild
type possible
Cost at VF, study arm
Mean (SE)
P
a
Mean (SE)
P
a
Resistance cost
CVF-FDO1
NFV
b
0.003 (0.116)
.03
0.366 (0.067) .66
EFV
0.571 (0.127)
.17
0.854 (0.133) .01
NFV + EFV
0.342 (0.107)
—
0.424 (0.113)
—
CVF-FDO2
NFV
b
⫺0.164 (0.127)
.004 0.221 (0.070) .17
EFV
0.685 (0.133)
.04
NFV + EFV
0.329 (0.117)
—
0.987 (0.142) .002
0.415 (0.123)
—
Duration-adjusted cost
dCVF-FDO1
b
NFV
0.010 (0.016)
.07
0.057 (0.010) .75
EFV
0.072 (0.016)
.36
0.118 (0.019) .04
NFV + EFV
0.051 (0.016)
—
0.064 (0.017)
—
dCVF-FDO2
b
NFV
⫺0.016 (0.018)
.01
0.035 (0.011) .24
EFV
0.091 (0.017)
.07
0.139 (0.021) .004
NFV + EFV
0.047 (0.017)
—
0.060 (0.018)
—
NOTE. CVF-FDO1 and CVF-FDO2, resistance cost at VF in terms of measures
of future drug option (FDO) 1 and FDO2, respectively; dCVF-FDO1 and dCVF-FDO2,
average effective duration-adjusted resistance cost in terms of FDO1 and
FDO2, respectively; EFV, efavirenz triple therapy; NFV, nelfinavir triple therapy;
NFV + EFV, quadruple therapy.
a
Resistance cost between each triple-drug regimen vs. the quadruple-drug
regimen (2-sample t test).
b
Mean cost is not significantly different from 0 (each P 1 .20 ); all other mean
costs are highly significantly different from 0 (each P ! .002).
Resistance Costs of HAART • JID 2003:188 (1 October) • 1005
significant difference in the effective duration-adjusted resistance cost between the nelfinavir and the nelfinavir plus efavirenz arms, whereas the efavirenz arm has significantly higher
duration-adjusted resistance cost than either of the other 2
therapy arms.
DISCUSSION
Many factors need to be considered in determining the therapy
success of HAART, primarily the ability of a therapy regimen
or strategy to prolong life and prevent disease. However, given
the clinical effectiveness of HAART [2, 25, 26], the use of
clinical end points and/or death as outcome measures is not
feasible in many settings. Instead, the use of surrogate markers
is necessary to determine optimal therapy strategies. Accepted
surrogate markers include plasma HIV RNA levels and CD4+
T cell counts. In the present report, we describe a surrogate
marker that captures the preservation of future therapy options,
as measured by the absence of drug resistance [27–29]. Our
proposed metric (FDOs) allows for easy quantification of the
resistance costs associated with any given strategy. In the present
study, we applied the FDO approach to a randomized clinical
study of 3 therapy regimens: 1 based on a PI, 1 based on an
NNRTI, and 1 based on the combination of a PI and an NNRTI.
Because each regimen resulted in a unique pattern of drugresistance (and cross-resistance) mutations, we showed that
important new insights could be provided when the FDO was
included as a therapy outcome.
The potential utility of the FDO end point is illustrated by
considering the therapy outcomes observed in ACTG 364. In
this study of heavily NRTI-experienced patients who added 1 or
2 new drug classes when initiating salvage therapy, the proportion
of patients who maintained virus load suppression throughout
the study was greatest in those receiving nelfinavir plus efavirenz,
intermediate in those receiving efavirenz, and lowest in those
receiving nelfinavir. However, as shown in the present study,
patients randomly assigned to a regimen including efavirenz had
significant reductions in their FDO at the time of virologic failure,
whereas patients randomly assigned to receive nelfinavir had a
significantly lower reduction in their FDO. Thus, although patients were more likely to experience failure with a regimen containing nelfinavir but not efavirenz, they experienced failure with
a virus that retained greater susceptibility to other drugs, including PIs other than nelfinavir and each of the drugs within
the NNRTI class. This outcome reflected the observation that
45% of the patients for whom therapy with nelfinavir failed had
dominant plasma virus containing D30N as the only primary
mutation [30]. This mutation is associated with the preservation
of susceptibility to all other PIs [31–33].
There are many clinical studies for which the FDO can be
used as an outcome measure. For example, an FDO end point
1006 • JID 2003:188 (1 October) • Jiang et al.
may be particularly useful in investigating approaches to treating patients who experience sustained virologic rebound while
receiving HAART. Current guidelines recommend switching to
a new therapy regimen soon after the detection of virus, in an
effort to avoid accumulation of resistance mutations. Nonetheless, patients who continue to receive a stable antiretroviral
regimen after the emergence of drug-resistant virus typically
maintain some degree of virus suppression, as well as immunologic and clinical benefits [19, 34]. Although patients who
continue receiving a partially suppressive regimen may, in fact,
accumulate new resistance mutations, it is also likely that many
patients who change therapies will experience virologic rebound
and develop resistance mutations while receiving their new regimen. Heretofore, there has been no formal investigation of the
costs and benefits of delaying changes in therapy, in part because of a lack of appropriate end points. The proposed FDO
measures are tools that can quantify the risk of continued drug
therapy after initial virologic failure.
FDO end points might also be useful in determining when
therapy-naive patients should start receiving an antiretroviral
regimen. Early initiation of antiretroviral therapy is likely to
delay disease progression in the short term, but some sequential
therapy strategies may exhaust therapy options sooner because
of the emergence of resistance when therapy fails. Similarly, the
FDO end point may be useful in studies evaluating the benefits
and risks of sequential therapy interruptions. Interrupting therapy may be associated with increased resistance in some patients
and decreased resistance (or prevention of further virus evolution) in others [12, 35]. The robustness of the FDO metric
would allow careful assessment of each of these strategies.
The FDO metric may also be particularly useful in anticipating the evolution of drug resistance in resource-poor
regions, particularly where virologic monitoring is unlikely because of costs and the need for sophisticated laboratory infrastructure. The current therapy and monitoring standards in
resource-rich countries emphasize frequent monitoring for virologic failure and aggressive adjustment of therapy when viremia is detected [3, 4]. This often leads to a rapid sequence
of therapeutic switches after virologic failure. A monitoring
strategy driven by CD4 cell counts or clinical signs and symptoms that demonstrate the patient’s immune restoration is
more easily implemented in resource-poor regions and may be
as effective in preventing disease progression [36]. M.L. and
Sombat Thanprasertsuk have proposed a clinical study in Thailand to compare monitoring based on virus load with that based
on CD4 cell counts (personal communication). Similarly, Rabkin et al. are initiating a study comparing minimal and moreintensive clinical and laboratory monitoring strategies [36]. Assessment of the resistance cost and the FDOs that remain after
clinical and immunologic failure of antiretroviral regimens in
resource-limited settings is essential for comparing the longterm efficacy of different therapeutic and monitoring strategies.
FDO end points may also play a vital role in international
studies of transmission of HIV infection (i.e., those conducted
among serodiscordant sex partners). The HIV Prevention Trials
Network (HPTN) is planning a randomized study to evaluate
whether immediate therapy of HIV-infected patients with CD4
cell counts of 200–350 cells/mL reduces risk of transmission to
initially seronegative sex partners, compared with deferring
therapy until CD4 cell counts decrease to !200 cells/mL. The
study will be conducted in areas of high HIV prevalence
throughout the world. Although immediate therapy may be
expected to reduce risk of transmission early in the study, evaluation of the true clinical benefit of each strategy must take
into account both its long-term impact on transmission rates
and the degree to which transmitted viruses are susceptible to
available drugs. Because only a limited number of therapies are
available in the developing world, transmission of resistant virus
is of especially grave concern. As HPTN investigators have
noted, monitoring of the study should be based not only on
rates of transmission of HIV, but also on the genetic characteristics of transmitted viruses.
There are several caveats to the FDO metric that deserve
comment. First, our approach assumes that no mutations accumulate during periods of time when virus load is below the
limit of assay detection and, therefore, that no additional resistance costs accrue during this period. This assumption may
be warranted on the basis of small studies of virus evolution
in HIV DNA during virologic suppression [37, 38]. Comparison of the FDO between early and late virologic failures can
serve to test the hypothesis that continued replication in the
presence of drug increases the resistance cost. Second, the resistance cost end points we have proposed quantify the future
drug options available to HIV-1–infected patients on the basis
of analysis of genotypic data. However, it would be straightforward to incorporate phenotypic data into the FDO metric.
Third, the FDO metric can rapidly accommodate new drugs
as they are developed. For example, for highly drug-experienced
patients, the selection of drugs in a salvage regimen that included the newly available fusion inhibitor enfuvirtide could
be optimized by FDO analysis. Finally, the FDO can incorporate
changes in our understanding of pathogenesis, drug resistance,
and susceptibility. Mutational interactions, which may include
increased susceptibility to some drugs on the basis of resistance
to others, may be incorporated into FDO strategies and evaluation [39–42]. To compare antiretroviral drug strategies and
regimens, the FDO metric is an analytical tool to define new,
additional outcomes in clinical trials. For the clinician, the FDO
approach may offer a method to synthesize resistance information, to optimize sequential drug regimens in the long-term
therapy of HIV infection.
Acknowledgment
We thank the AIDS Clinical Trials Group 364 team, for providing the data, and the reviewers, for their helpful suggestions
to improve our paper.
APPENDIX
Let T denote the random time from entry to virologic failure
of a patient, and let U denote the random censoring time,
assumed to be independent of T. Notice that T is observable
only if U 1 T. We may denote the random resistance cost accumulated during period T as CVF, which is measurable only
if the failure time T is observed. For each patient, the observed
information is the observed time X, which is the minimum of
T and U, an indicator D p 1 if X p T , 0 if otherwise, and the
resistance cost CVF if D p 1. We may assume that the cost
accumulation depends on the failure time T in the same way
for all patients receiving a particular therapy. Averaging over
all possible values of failure time T leads to a common distribution of C, whose mean m is of interest.
With a sample of size n, m can be estimated as follows [22]:
ˆ ), where G(.) is the survival function of
m̂ p Snip1D iC VFi /G(X
i
U, and Ĝ(.) is the Kaplan-Meier estimator of G(.) based on
data {(X i, 1 ⫺ D i),i p 1, … ,n}. Thus, we estimate the mean
resistance cost by computing a weighted average of the observed
resistance costs, where the weight is the inverse probability of
not being censored at the observed time.
Furthermore, let S(.) denote the survival function of T, and
let Ŝ(.) denote the corresponding Kaplan-Meier estimator based
on data {(X i,D i),i p 1, … ,n}. Define
H(V,X i) p
1
冘
n
ˆ ) jp1
nS(X
i
DV
j j I(X j ⭓ X i)
,
ˆ )
G(X
i
where the variable V can be CVF or CVF2. Then the variance of
m̂, j 2 can be estimated by the formula
{冘
冘
ˆ 2 1 (1 ⫺ D i) ˆ 2
1 1 D i(C VFi ⫺ m)
[H(C VFi,X i) ⫺ Hˆ 2(C VFi,X i)]
+
ˆ
ˆ )
n n ip1
n ip1 G(X
G(X i)
i
n
ĵ2 p
n
}
.
References
1. Lucas GM, Chaisson RE, Moore RD. Highly active antiretroviral therapy in a large urban clinic: risk factors for virologic failure and adverse
drug reactions. Ann Intern Med 1999; 131:81–7.
2. Ledergerber B, Egger M, Opravil M, et al. Clinical progression and
virological failure on highly active antiretroviral therapy in HIV-1 patients: a prospective cohort study. Swiss HIV Cohort Study. Lancet
1999; 353:863–8.
3. Yeni PG, Hammer SM, Carpenter CC, et al. Antiretroviral treatment
for adult HIV infection in 2002: updated recommendations of the
International AIDS Society—USA Panel. JAMA 2002; 288:222–35.
Resistance Costs of HAART • JID 2003:188 (1 October) • 1007
4. Dybul M, Fauci AS, Bartlett JG, Kaplan JE, Pau AK. Guidelines for
using antiretroviral agents among HIV-infected adults and adolescents.
The Panel on Clinical Practices for the Treatment of HIV. Ann Intern
Med 2002; 137(5 Pt. 2):381–433.
5. British HIV Association (BHIVA) guidelines for the treatment of HIVinfected adults with antiretroviral therapy. HIV Med 2001; 2:276–313.
6. Molla A, Korneyeva M, Gao Q, et al. Ordered accumulation of mutations
in HIV protease confers resistance to ritonavir. Nat Med 1996; 2:760–6.
7. Harrigan PR, Larder BA. Extent of cross-resistance between agents used
to treat human immunodeficiency virus type 1 infection in clinically
derived isolates. Antimicrob Agents Chemother 2002; 46:909–12.
8. Hammer SM, Vaida F, Bennett KK, et al. Dual vs. single protease
inhibitor therapy following antiretroviral treatment failure: a randomized trial. JAMA 2002; 288:169–80.
9. Gulick RM, Hu XJ, Fiscus SA, et al. Randomized study of saquinavir
with ritonavir or nelfinavir together with delavirdine, adefovir, or both
in human immunodeficiency virus–infected adults with virologic failure on indinavir: AIDS Clinical Trials Group Study 359. J Infect Dis
2000; 182:1375–84.
10. Deeks SG, Barbour JD, Martin JN, Swanson MS, Grant RM. Sustained
CD4+ T cell response after virologic failure of protease inhibitor–based
regimens in patients with human immunodeficiency virus infection. J
Infect Dis 2000; 181:946–53.
11. Le Moing V, Thiebaut R, Chene G, et al. Predictors of long-term
increase in CD4+ cell counts in human immunodeficiency virus–
infected patients receiving a protease inhibitor–containing antiretroviral regimen. J Infect Dis 2002; 185:471–80.
12. Deeks SG, Wrin T, Liegler T, et al. Virologic and immunologic consequences of discontinuing combination antiretroviral-drug therapy in
HIV-infected patients with detectable viremia. N Engl J Med 2001;
344:472–80.
13. Barbour J, Wrin T, Grant R, et al. Evolution of phenotypic drug susceptibility and viral replication capacity during long-term virologic
failure of protease inhibitor therapy in HIV-infected adults. J Virol
2002; 76:11104–12.
14. The HIV/AIDS situation in the world. ONUSIDA/01.75F WHO/CDS/
NCS/2001.2. December 2001.
15. United Nations. Declaration of Commitment on HIV/AIDS. A/RES/
S-26/2. August 2001.
16. DeGruttola V, Dix L, D’Aquila R, et al. The relation between baseline
HIV drug resistance and response to antiretroviral therapy: re-analysis
of retrospective and prospective studies using a standardized data analysis plan. Antivir Ther 2000; 5:41–8.
17. Gilbert PB, Hanna GJ, De Gruttola V, et al. Comparative analysis of
HIV type 1 genotypic resistance across antiretroviral trial treatment
regimens. AIDS Res Hum Retroviruses 2000; 16:1325–36.
18. Albrecht MA, Bosch RJ, Hammer SM, et al. Nelfinavir, efavirenz, or
both after the failure of nucleoside treatment of HIV infection. N Engl
J Med 2001; 345:398–407.
19. Hermankova M, Ray SC, Ruff C, et al. HIV-1 drug resistance profiles
in children and adults with viral load of !50 copies/mL receiving combination therapy. JAMA 2001; 286:196–207.
20. Huang Y. The two-sample problem with induced dependent censorship. Biometrics 1999; 55:1108–13.
21. Zhao H, Tsiatis AA. Estimating mean quality adjusted lifetime with censored data. Sankhya, Series B, Indian Journal of Statistics 2000; 62:175–88.
22. Bang H, Tsiatis AA. Estimating medical costs with censored data. Biometrika 2000; 87:329–43.
23. Kantor R, Machekano R, Gonzales MJ, Dupnik K, Schapiro JM, Shafer
RW. Human immunodeficiency virus reverse transcriptase and protease
sequence database: an expanded data model integrating natural language
text and sequence analysis programs. Nucleic Acids Res 2001; 29:296–9.
24. Winters MA, Bosch RJ, Albrecht MA, Katzenstein DA. Clinical impact
of the M184V mutation on switching to didanosine or maintaining
lamivudine treatment [abstract 122]. ACTG 364 Study Team. In: Pro-
1008 • JID 2003:188 (1 October) • Jiang et al.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
gram and abstracts of the XI International HIV Drug Resistance Workshop (Seville, Spain). Antivir Ther 2002; 7(Suppl 1):S134.
Hammer SM, Squires KE, Hughes MD, et al. A controlled trial of two
nucleoside analogues plus indinavir in persons with human immunodeficiency virus infection and CD4 cell counts of 200 per cubic
millimeter or less. AIDS Clinical Trials Group 320 Study Team. N Engl
J Med 1997; 337:725–33.
Egger M, Hirschel B, Francioli P, et al. Impact of new antiretroviral
combination therapies in HIV infected patients in Switzerland: prospective multicentre study. Swiss HIV Cohort Study. BMJ 1997; 315:1194–9.
Gilbert PB, Ribaudo HJ, Greenberg L, et al. Considerations in choosing
a primary endpoint that measures durability of virological suppression
in an antiretroviral trial. AIDS 2000; 14:1961–72.
Carr A. Improvement of the study, analysis, and reporting of adverse
events associated with antiretroviral therapy. Lancet 2002; 360:81–5.
Hughes MD, Johnson VA, Hirsch MS, et al. Monitoring plasma HIV1 RNA levels in addition to CD4+ lymphocyte count improves assessment of antiretroviral therapeutic response. ACTG 241 Protocol Virology Substudy Team. Ann Intern Med 1997; 126:929–38.
Katzenstein D, Downey G, Bosch R, Hellmann N, Becker M, Albrecht
M. Evolution of protease phenotype and genotype changes at initial and
continued virologic failure among nucleoside-experienced subjects receiving nelfinavir (NFV) in ACTG 364 [abstract 144]. ACTG 364 Virology
team. In: Program and abstracts of the XI International HIV Drug Resistance Workshop (Seville, Spain). Antivir Ther 2002; 7(Suppl 1):S156.
Tebas P, Patick AK, Kane EM, et al. Virologic responses to a ritonavirsaquinavir–containing regimen in patients who had previously failed
nelfinavir. AIDS 1999; 13:F23–8.
Kemper CA, Witt MD, Keiser PH, et al. Sequencing of protease inhibitor therapy: insights from an analysis of HIV phenotypic resistance
in patients failing protease inhibitors. AIDS 2001; 15:609–15.
Hirsch MS, Brun-Vezinet F, D’Aquila RT, et al. Antiretroviral drug
resistance testing in adult HIV-1 infection: recommendations of an
International AIDS Society—USA Panel. JAMA 2000; 283:2417–26.
Gulick RM, Mellors JW, Havlir D, et al. Simultaneous vs. sequential
initiation of therapy with indinavir, zidovudine, and lamivudine for
HIV-1 infection: 100-week follow-up. JAMA 1998; 280:35–41.
Martinez-Picado J, Morales-Lopetegi K, Wrin T, et al. Selection of drugresistant HIV-1 mutants in response to repeated structured treatment
interruptions. AIDS 2002; 16:895–9.
Rabkin M, El-Sadr W, Katzenstein D, et al. Antiretroviral therapy in
resource-poor settings: clinical research priorities. Lancet 2002; 360:
1503–5.
Zhang L, Ramratnam B, Tenner-Racz K, et al. Quantifying residual
HIV-1 replication in patients receiving combination antiretroviral therapy. N Engl J Med 1999; 340:1605–13.
Zhang L, Chung C, Hu BS, et al. Genetic characterization of rebounding
HIV-1 after cessation of highly active antiretroviral therapy. J Clin
Invest 2000; 106:839–45.
Ait-Khaled M, Stone C, Amphlett G, et al. M184V is associated with
a low incidence of thymidine analogue mutations and low phenotypic
resistance to zidovudine and stavudine. AIDS 2002; 16:1686–9.
Markland W, Rao BG, Parsons JD, et al. Structural and kinetic analyses
of the protease from an amprenavir-resistant human immunodeficiency virus type 1 mutant rendered resistant to saquinavir and resensitized to amprenavir. J Virol 2000; 74:7636–41.
Katzenstein DA, Bosch RJ, Hellmann N, Wang N, Bacheler L, Albrecht
MA. Phenotypic susceptibility and virological outcome in nucleosideexperienced patients receiving three or four antiretroviral drugs. AIDS
2003; 17:821–30.
Naeger LK, Margot NA, Miller MD. Increased drug susceptibility of HIV1 reverse transcriptase mutants containing M184V and zidovudineassociated mutations: analysis of enzyme processivity, chain-terminator
removal and viral replication. Antivir Ther 2001; 6:115–26.