Customize your JAMA Network experience by selecting one or more topics from the list below.
Aisen PS, Schafer KA, Grundman M, et al. Effects of Rofecoxib or Naproxen vs Placebo on Alzheimer Disease ProgressionA Randomized Controlled Trial. JAMA. 2003;289(21):2819–2826. doi:10.1001/jama.289.21.2819
Context Laboratory evidence that inflammatory mechanisms contribute to neuronal
injury in Alzheimer disease (AD), along with epidemiological evidence, suggests
that nonsteroidal anti-inflammatory drugs (NSAIDs) may favorably influence
the course of the disease.
Objective To determine whether treatment with a selective cyclooxygenase (COX)
-2 inhibitor (rofecoxib) or a traditional nonselective NSAID (naproxen) slows
cognitive decline in patients with mild-to-moderate AD.
Design Multicenter, randomized, double-blind, placebo-controlled, parallel
group trial, with 1-year exposure to study medications.
Setting Forty ambulatory treatment centers affiliated with the Alzheimer's Disease
Cooperative Study consortium.
Participants Participants with mild-to-moderate AD (Mini-Mental State Examination
score of 13-26) were recruited from December 1999 to November 2000 using clinic
populations, referrals from community physicians, and local advertising. Stable
use of cholinesterase inhibitors, estrogen, low-dose aspirin, and vitamin
E was allowed. Participants with inflammatory diseases that might respond
to the study medications were excluded. Of 474 participants screened, 351
Interventions Once-daily rofecoxib, 25 mg, or twice-daily naproxen sodium, 220 mg,
Main Outcome Measures The primary outcome measure was the 1-year change in the Alzheimer Disease
Assessment Scale-Cognitive (ADAS-Cog) subscale score. Secondary outcome measures
included the Clinical Dementia Rating scale sum-of-boxes, the Neuropsychiatric
Inventory, the Quality of Life-AD, and the time to attainment of significant
end points (4-point decline from baseline ADAS-Cog score, 1-step worsening
on the global Clinical Dementia Rating scale, 15-point decline on the ADCS
activities of daily living inventory, institutionalization, or death).
Results The 1-year mean (SD) change in ADAS-Cog scores in participants treated
with naproxen (5.8 [8.0]) or rofecoxib (7.6 [7.7]) was not significantly different
from the change in participants treated with placebo (5.7 [8.2]). Results
of secondary analyses showed no consistent benefit of either treatment. Fatigue,
dizziness, and hypertension were more commonly reported in the active drug
groups, and more serious adverse events were found in the active treatment
group than in the placebo group.
Conclusion The results of this study indicate that rofecoxib or low-dose naproxen
does not slow cognitive decline in patients with mild-to-moderate AD.
Alzheimer disease (AD) is among the most important health problems of
elderly persons, affecting more than 4 million people in the United States.
In the last decade, cholinesterase inhibitors have been widely used to alleviate
symptoms of cognitive dysfunction in AD. The use of anti-inflammatory drugs
is among the strategies under active investigation for the development of
effective disease-modifying treatment for AD. This approach is supported by
a wealth of laboratory evidence that inflammatory mechanisms contribute to
neuronal damage in AD.1 Furthermore, many epidemiological
studies suggest that anti-inflammatory drugs have a protective effect, reducing
the incidence of AD.2
Results of small pilot clinical trials of nonsteroidal anti-inflammatory
drug (NSAID) efficacy in the treatment of AD have been encouraging. One report
indicated a stabilizing effect of indomethacin treatment for 6 months,3 while a follow-up study showed a trend toward benefit
with diclofenac treatment for 25 weeks.4 Neither
study showed conclusive results, and the patients with AD showed poor tolerability
of both drugs. Larger studies with better tolerated treatment regimens, such
as low-dose prednisone for 1 year5 and hydroxychloroquine
for 18 months,6 failed to demonstrate benefit
in the treatment of AD.
Recent studies have provided strong support for the testing of efficacy
of NSAIDs in the prevention of AD. A large epidemiologic study using computerized
medical records to accurately indicate drug use corroborates earlier work,
with evidence that use of NSAIDs sharply reduces AD risk.7 In
transgenic mice with AD-type amyloid deposition, the NSAID ibuprofen reduced
brain inflammation and levels of soluble and insoluble amyloid peptide, and
In this trial, we tested the hypothesis that NSAID treatment would slow
the rate of cognitive decline in participants with mild-to-moderate AD. Selection
of treatment regimens was based on consideration of the appropriate target
in AD brain, as well as medication tolerability. Results of studies in cell
culture systems,10,11 rodents,12,13 and human brain10,14 indicate
that neuronal upregulation of cyclooxygenase (COX) -2 may contribute to neurodegeneration
in AD. However, the epidemiological data support the efficacy of long-term
use of low doses of nonselective NSAIDs that inhibit COX-1 and COX-2,2 and microglial COX-1 may be an important target.15 Long-term treatment with full doses of nonselective
NSAIDs has been associated with a substantial risk of serious gastrointestinal
tract toxicity. Therefore, we used a standard dose of the selective COX-2
inhibitor rofecoxib and a low-dose of the nonselective NSAID naproxen. We
chose a treatment period of 1 year based on evidence from a pilot study of
NSAID efficacy with short-term treatment3 and
the concern about the risk of long-term exposure in individuals with AD.
The study used a randomized, double-blind, 3-group parallel design to
compare rofecoxib or naproxen with placebo. The treatment period was 12 months,
followed by a 2-month washout. Forty ambulatory treatment centers affiliated
with the Alzheimer's Disease Cooperative Study (ADCS) consortium participated
in this trial after obtaining approval from their local institutional review
boards. Written informed consent was obtained from participants and/or legally
authorized representatives, according to local guidelines.
Individuals with probable AD16 recruited
from December 1999 to November 2000 using clinic populations, referrals from
community physicians, and local advertising were eligible if they did not
have comorbid conditions that increased the risk of adverse events associated
with NSAID treatment (hypersensitivity to aspirin or NSAIDs, active peptic
ulcer disease within 5 years, renal insufficiency [serum creatinine level
>1.5 mg/dL or >132.6 µmol/L], clinically significant liver disease,
poorly controlled hypertension, congestive heart failure, or bleeding disorder).
Individuals were excluded if they had comorbid conditions that might respond
to NSAIDs (eg, inflammatory arthritis). Individuals also were excluded if
within the prior 2 months they had regularly used anti-inflammatory medications
(aspirin at a daily dose ≤325 mg was allowed), neuroleptics, antidepressants,
sedatives, anti-Parkinsonian medications, or any investigational treatment
for AD. Inclusion criteria included age older than 50 years and Mini-Mental
State Examination (MMSE)17 score within the
range of 13 to 26. Stable use of cholinesterase inhibitors was allowed.
Rofecoxib tablets, 25 mg, were overencapsulated to allow preparation
of an identical placebo capsule. Naproxen sodium tablets, 220 mg, and identical
placebo tablets were supplied by Bayer Consumer, Inc (Morristown, NJ). Each
participant received 2 bottles of study medication with coded labels at baseline
and at the 3-, 6-, and 9-month visits: 1 bottle containing rofecoxib or identical
placebo capsules, to be taken once daily; and 1 bottle of naproxen or identical
placebo capsules, to be taken twice daily.
The randomization process used a permuted block design with block size
of 3 stratified by site. The randomization sequence was generated by the ADCS
data center. Scratch-off codebreakers were used so that instances of unblinding
would be documented; all codebreakers were collected at the end of the trial.
Adequacy of masking was assessed by questionnaires completed by participants,
caregivers, psychometrists, and site investigators.
Safety assessments, including vital signs, physical examination, urinalysis,
and hematology and chemistry blood tests, were performed at each visit. Cognitive
and behavioral assessments were performed at baseline and at months 1, 3,
6, 9, 12, and 14.
The primary outcome measure for this trial was the 1-year change score
on the Alzheimer Disease Assessment Scale-Cognitive (ADAS-Cog) subscale,18 an instrument that evaluates memory, attention, reasoning,
language, orientation, and praxis. We considered a significant benefit in
the active group to be a 50% reduction in cognitive decline as indicated by
change in ADAS-Cog score compared with the placebo group. Data from a small
pilot study3 have supported the hypothesis
that an effect of this magnitude might be seen, and we considered that a smaller
effect might not justify the risk of long-term NSAID therapy in the AD participants.
Using longitudinal ADAS-Cog data from an earlier multicenter trial,5 we estimated that 100 participants were required in
each treatment group (naproxen, rofecoxib, and placebo) for a power of 0.8.
An overall drop-out rate of 20% was anticipated. Since we expected more drop-outs
from adverse drug effects in the active drug groups, fewer participants were
randomly assigned to the placebo group (participants were randomly assigned
using a ratio of 1.2 [naproxen]:1.2 [rofecoxib]:1.1 [placebo]).
Secondary outcome measures included the Clinical Dementia Rating sum-of-boxes
(CDR-SOB),19 Alzheimer's Disease Cooperative
Study activities of daily living (ADCS-ADL) scale,20 the
Neuropsychiatric Inventory (NPI),21 the Quality
of Life-AD (QOL-AD),22 and the time to attainment
of significant end points (4-point decline from baseline ADAS-Cog score, 1-step
worsening on the global CDR scale, 15-point decline on the ADCS-ADL, institutionalization,
The primary analysis was a comparison of the change in ADAS-Cog score
between the naproxen and rofecoxib groups and the placebo group using analysis
of covariance (ANCOVA), with the baseline ADAS-Cog score as a covariate. The
statistical plan included assessment of the treatment groups for significant
imbalance in age, sex, or apolipoprotein E genotype that might influence outcome;
if an imbalance was present (P<.15) and the factor
influenced ADAS-Cog change scores (P<.10), it
would be included in the ANCOVA model.
The primary analysis was conducted on an intent-to-treat basis (ie,
including all randomized participants). Because AD is a disease of progressive
cognitive deterioration, the protocol specified an alternative imputation
scheme rather than last-observation-carried-forward (LOCF) analysis. (LOCF
analysis would overestimate a treatment effect if an excess of drop-outs occurred
in the active drug group due to adverse effects.) To impute a participant's
missing score at week 52, an estimate of change in ADAS-Cog score during the
unobserved period based on all individuals with complete data from the participant's
treatment group was applied to the participant's last observed score. A linear
rate of progression was not assumed. We also conducted secondary analyses
of change in ADAS-Cog scores using the LOCF imputation, and an analysis of
completers only (ie, those who completed the week 52 visit). In addition,
we conducted a longitudinal regression analysis using the method of generalized
estimating equations.23 For this analysis,
a pair-wise comparison of the naproxen group vs the placebo group and the
rofecoxib group vs the placebo group, the outcome was the ADAS-Cog score clustered
by the baseline, 3-, 6-, 9-, and 12-month visits.
A planned secondary analysis of time to reach clinically significant
end points was performed using a Cox proportional hazards model. Intent-to-treat
analysis of secondary measures were conducted using both the slope imputation
strategy described above and LOCF for missing week 52 values. No interim analysis
was performed. Statistical analyses were performed using SAS version 8.2 (SAS
Institute, Cary, NC).
The flow of participants through the study protocol is shown in Figure 1. From a total of 474 participants
screened, 351 met the study criteria and were randomly assigned to 1 of the
3 treatment groups (rofecoxib, naproxen, or placebo). Considering the likelihood
that more participants in the active groups would drop out due to adverse
effects, more participants were randomly assigned to rofecoxib (n = 122) and
naproxen (n = 118) than to the placebo group (n = 111).
The primary outcome measure (change in ADAS-Cog score at week 52) was
obtained for study completers: 90 (76%) participants in the naproxen group,
89 (73%) in the rofecoxib group, and 88 (79%) in the placebo group. The predominant
reasons for early study discontinuation were caregiver issues (rofecoxib,
n = 12; naproxen, n = 11; and placebo, n = 10) and adverse events (rofecoxib,
n = 14; naproxen, n = 11; and placebo, n = 9).
The demographic and clinical characteristics of the treatment groups
at baseline are shown in Table 1.
No significant differences were found between the groups on any of the demographic
or baseline characteristics. Eighty-seven participants (25%) were using low-dose
aspirin during the trial: 33 (27%) in the rofecoxib group, 30 (25%) in the
naproxen group, and 24 (22%) in the placebo group. Aspirin use did not show
a significant effect on ADAS-Cog scores for all groups or for the individual
Baseline characteristics were compared between participants who discontinued
early and those who completed the study. In the naproxen group, participants
who discontinued early had lower mean (SD) MMSE scores (19.6 [3.3] vs 21.0
[3.6]; P = .04; using Wilcoxon rank sum test) and
higher mean (SD) CDR-SOB scores (7.2 [3.4] vs 5.7 [2.7]; P = .03). No significant differences were found between the rofecoxib
and placebo groups.
A total of 238 (68%) participants were taking cholinesterase inhibitors
at the time of enrollment. Twenty-four participants (rofecoxib group, n =
8; naproxen group, n = 10; and placebo group, n = 6) started treatment with
cholinesterase inhibitors prior to the month 12 visit. These participants
were included in the primary intent-to-treat analyses. In a secondary analysis
of the main outcome measure excluding these 24 subjects, the results were
similar (data not shown).
Primary Outcome Measure. The effect of treatment
on the primary and secondary measures in the intent-to-treat analysis is shown
in Table 2. None of the planned
covariates (age, sex, and apolipoprotein E genotype) was included in the analysis,
because there was neither imbalance nor association with the outcome.
Neither active treatment had a beneficial effect on the primary measure,
the change in mean ADAS-Cog scores; the group treated with rofecoxib showed
a trend toward greater cognitive decline compared with the other groups. For
the comparison of naproxen with placebo, the SE of the difference between
change scores was 1.07 (95% confidence interval for the difference in the
change scores, −0.1 + /− 2.14); therefore, it is highly unlikely
that naproxen treatment reduces the 1-year decline in ADAS-Cog score by more
than 2.04/5.7, or 36%. For the comparison of rofecoxib with placebo, the SE
of the difference between change scores was 1.03; a similar calculation indicates
it is highly unlikely that rofecoxib reduces the decline in ADAS-Cog score
by more than 15%.
When LOCF imputation of the change in ADAS-Cog score was substituted
for the primary imputation scheme, the result was similar to the primary analysis
(mean [SD] ADAS-Cog change in score in the placebo group, 4.9 [8.2]; rofecoxib
group, 5.9 [7.7], P = .52; and naproxen group, 5.0
[7.9], P = .94). Analysis of participants taking
study medication who completed the 52-week visit yielded similar results,
with no significant difference in rate of decline for the mean (SD) ADAS-Cog
score (placebo, 5.3 [8.5], n = 79; rofecoxib, 6.8 [7.7], n = 78, P = .47; naproxen, 5.3 [7.9], n = 80, P =
.9). Regression analysis using the generalized estimating equations method
demonstrated no effect of rofecoxib (P = .77) or
naproxen (P = .64) on longitudinal ADAS-Cog scores.
Secondary Outcome Measures. The intent-to-treat
analysis using the primary imputation scheme for missing values revealed no
significant difference in decline for the CDR-SOB scores across all the groups
(Table 2). Other analyses, using
LOCF imputation or assessing completers only, showed similar results (data
not shown). The ADCS-ADL scores showed a trend toward a beneficial effect
with use of rofecoxib. However, none of the secondary analyses were adjusted
for multiple comparisons because this was not specified in the original protocol.
In the slope imputation analysis, use of rofecoxib reduced the decline in
ADCS-ADL score by 22% (unadjusted P = .09); in the
LOCF analysis, use of rofecoxib reduced the decline in ADCS-ADL score by 33%
(unadjusted P = .04). In an analysis of study completers
only, rofecoxib reduced the decline in ADCS-ADL score by 27% (unadjusted P = .09). Treatment with rofecoxib or naproxen had no effect
on the NPI or QOL-AD scores.
The planned survival analysis considered the time interval from the
baseline visit to the first among 5 possible end points: death, institutionalization,
increase in global CDR score, 15-point decline on the ADCS-ADL scale, or 4-point
decline on the ADAS-Cog scale. A total of 242 subjects (69%) reached at least
1 end point (placebo group, n = 78 [ 70%]; naproxen group, n = 83 [70%]; and
rofecoxib, n = 81 [66%]). Rofecoxib (P = .63) or
naproxen (P = .50) vs placebo did not differ in time
to first end point. When time to individual end points was examined, the only
significant difference was between the rofecoxib and placebo groups for time
to institutionalization, favoring the rofecoxib group (P = .03) (data not shown). However, this analysis was based on a small
number of events (7 in the placebo group and 1 in the rofecoxib group), and
the P value was not adjusted for multiple comparisons.
Treatment emergent adverse events were grouped into categories for analysis.
Naproxen and rofecoxib groups had more frequent adverse event reports (P<.10) (Table 3).
Dizziness and fatigue were more common in the active drug groups than in the
placebo group, and dry mouth was more common in the naproxen group than the
placebo group. New cases of hypertension were reported as adverse events in
the active drug groups, but none were reported in the placebo group (naproxen
[n = 6] vs placebo [n = 0], P = .03 and rofecoxib
[n = 9] vs placebo, P = .004).
In addition to adverse events reported on case report forms, study participants
were asked about a list of symptoms that might arise with NSAID treatment.
When the proportion of each treatment group reporting new symptoms (not present
at baseline) was compared, the difference was significant compared with placebo
for dry mouth (10% for naproxen vs 2% for placebo; P =
.01) and muscle pain (13% for rofecoxib vs 4% for placebo; P = .03). The percentage of participants with new reports of any gastrointestinal
tract symptom did not differ among the treatment groups (37% for naproxen,
31% for rofecoxib, and 30% for placebo).
Despite the cases of hypertension noted above, mean blood pressure did
not differ significantly by treatment group. The effects of treatment group
on mean (SD) change were minimal for hematocrit (placebo [n = 87]: −
0.3%[2.2]; naproxen [n = 88]: − 1.0% [2.2] vs placebo, P = .01; and rofecoxib [n = 87]: − 1.2% [2.2] vs placebo; P = .05) and for serum creatinine (0.02 [0.13] mg/dL or
1.8 [11.5] µmol/L, n = 88; naproxen [n = 87]: 0.02 [0.13] mg/dL or 1.8
[11.5] µmol/L, vs placebo, P = .90; and rofecoxib
[n = 84]: 0.08 [0.16] mg/dL or 7.1 [14.1] µmol/L vs placebo, P = .01).
The total number of serious adverse events in each treatment group,
along with number of events within specific categories, did not differ significantly
The randomization code was broken in 1 instance, based on clinical need
for the management of an acute medical problem. Study personnel involved in
administering assessment tools were shielded from group assignment data in
The results of questionnaires administered at the month 12 visit indicated
that the percentage of participants who believed they were taking active study
medication also did not differ significantly across the treatment groups (rofecoxib
group, 70%; naproxen group, 74%; and placebo group, 73%; P = .82, Fisher exact test), indicating that blinding was adequately
At the 12-month visit, the percentage of informants who believed that
participants were taking active study medication also did not differ across
the participants' treatment group (52% for rofecoxib group, 51% for naproxen
group, and 59% for placebo group; P = .10, Fisher
exact test). The differences also were not significant for study coordinators
(54% for rofecoxib group, 50% for naproxen group, and 43% for placebo group; P = .52, Fisher exact test) and study physicians (56% for
rofecoxib group, 40% for naproxen group, and 40% for placebo group; P = .19, Fisher exact test).
This is the first large-scale trial, to our knowledge, to determine
the efficacy of both nonselective and COX-2 selective NSAIDs in the treatment
of AD. The result of the primary intent-to-treat analysis of this trial showed
that neither naproxen nor rofecoxib slowed the rate of cognitive decline in
comparison with placebo in participants with mild-to-moderate AD. Similar
results were obtained when an LOCF analysis was substituted for the primary
slope method of imputing missing 12-month data and when the data were analyzed
using the generalized estimating equations method. The results of an analysis
of study completers (those who were able to complete the trial while still
taking study medication) also revealed no beneficial treatment effect of the
active study drugs in comparison with placebo. Instead, the rofecoxib group
showed a trend toward greater cognitive decline compared with placebo.
In the naproxen group, baseline MMSE and CDR-SOB scores differed between
individuals who discontinued study participation early and those who completed
the study. This difference raises the possibility that imputation of the data
missing from noncompleters may have influenced the outcome of the primary
analysis. However, the nonsignificant finding of the primary analysis is in
agreement with the nonsignificant findings of the LOCF analysis, the completers
analysis, the generalized estimating equations regression analysis, and the
clinical end points survival analysis, which strongly support its validity.
Secondary analyses of the effect of treatment on a global measure of
cognitive decline, behavior, ADL, and quality of life showed no consistent
advantage of either treatment over placebo. Rofecoxib delayed the time to
institutionalization, but this result was based on a small number of end points.
Adjustment for multiple comparisons would have made this comparison nonsignificant.
In some analyses, there was a trend toward smaller decline in ADL with
active treatment, particularly with rofecoxib. The trend toward greater cognitive
decline with rofecoxib treatment suggests that the trend toward protection
against decline in ADL may be related to an analgesic effect rather than an
effect on AD.
Both active treatments were reasonably well tolerated, although participants
who dropped out due to adverse events were more commonly from the active drug
groups than from the placebo group. Surprisingly, the gastrointestinal tract
symptoms were similar in the 3 groups. The small number of serious gastrointestinal
tract events (primarily bleeding) in the active drug groups was consistent
with expected risks. The occurrence of adverse events likely was reduced by
excluding individuals at high risk of NSAID toxicity.
Two important issues must be raised when interpreting the lack of active
treatment effect. The first issue concerns selection of drug, dose, and duration
of treatment. Naproxen was selected because the epidemiological evidence supports
the efficacy of nonselective COX inhibitors, and naproxen has a long half-life
(allowing for twice-daily dosing) and is relatively well tolerated. The naproxen
dose was low because of concern that full-dose therapy with a nonselective
COX inhibitor would be associated with more serious adverse events and a high
drop-out rate compared with placebo. The nonsignificant results with naproxen
could be explained by an insufficient dose; alternatively, other drugs in
this same class, such as ibuprofen, may have greater activity in AD mediated
by an effect on processing of the amyloid precursor protein.24 Rofecoxib,
1 of 2 available selective COX-2 inhibitors at the time this study was initiated,
was administered at a standard, full anti-inflammatory dose. It is possible
that a higher dose is necessary to influence neuronal survival; a high dose
of celecoxib, above the standard anti-inflammatory dose, is required to influence
neoplastic transformation in the gastrointestinal tract.25 It
is also possible that a period of exposure to anti-inflammatory treatment
greater than the 1 year of treatment in this trial is necessary to slow the
disease process. Some epidemiological studies of AD prevention suggest that
at least 2 years of exposure is necessary to optimally reduce risk of AD.26,27
The second issue concerns the selection of participants. The most compelling
epidemiological evidence suggests that long-term exposure to NSAIDs reduces
risk of subsequent AD; on the other hand, some studies suggest a stabilizing
effect of NSAID therapy in patients with clinically evident disease,3,28 supporting the design of the current
trial. While COX-2 and other inflammatory mediators are elevated in mild-to-moderate
AD brain,1 it is nonetheless possible that
when the disease is clinically apparent, the neuropathology is too advanced
to be significantly attenuated by NSAID therapy. The results of this trial
do not address the efficacy of NSAIDs in the prevention of AD. A primary prevention
trial, in which elderly persons without dementia are randomly assigned to
treatment with an NSAID or placebo, is necessary to test the utility of long-term
NSAID therapy in AD risk reduction. Such a trial is now under way.29
The results of the current study do not support the hypothesis that
rofecoxib or naproxen can slow the progression of AD. Considering the risk
of serious toxicity, such treatment should not be recommended. In view of
evidence from cell culture and animal studies suggesting reduction of β-amyloid
generation with other NSAIDs, such as ibuprofen (but not naproxen or selective
COX-2 inhibitors),24 additional treatment trials
using other NSAIDs may be warranted. Recommendations regarding risk reduction
with long-term NSAID therapy must await the results of a primary prevention