Arthritis & Rheumatism (Arthritis Care & Research)

Vol. 51, No. 5, October 15, 2004, pp 746 –754

DOI 10.1002/art.20698
© 2004, American College of Rheumatology
ORIGINAL ARTICLE

A Comparison of the Efficacy and Safety of

Nonsteroidal Antiinflammatory Agents Versus
Acetaminophen in the Treatment of Osteoarthritis:
A Meta-Analysis
CHIN LEE,1 WALTER L. STRAUS,2 ROBERT BALSHAW,3 SUNA BARLAS,2 SUZANNE VOGEL2, AND
THOMAS J. SCHNITZER1

Objective. To perform a meta-analysis comparing the efficacy and safety of recommended dosages of nonsteroidal
antiinflammatory drugs (NSAIDs), including cyclooxygenase 2 inhibitors, versus acetaminophen in the treatment of
symptomatic hip and knee osteoarthritis.
Methods. Medline and EMBASE searches were performed for original clinical trials directly comparing NSAIDs with
acetaminophen. A standardized form was used to abstract all data, including outcome measures of pain at rest, walking
pain, and dropouts due to adverse effects. Inverse-variance-weighted mean differences (WMDs) and 95% confidence
intervals (95% CI) for pain measures were determined for treatment groups. Odds ratios (ORs) and 95% CIs were
calculated for withdrawals due to adverse events. Results were compared using a random effects model.
Results. Seven articles met inclusion criteria with sufficient data for analysis. Participants had a mean age of 61.1 years
and 71.1% were women. Test of heterogeneity was not significant for either rest (P ⴝ 0.73) or walking (P ⴝ 0.76) pain. The
scores for overall pain at rest (WMD – 6.33 mm on a 100-mm visual analog scale [VAS], 95% CI –9.24, –3.41) and walking pain
(WMD –5.76 mm on a 100-mm VAS, 95% CI – 8.99, –2.52) favored the NSAID-treated group. Although NSAIDs elevated the risk
of withdrawals due to adverse events, the difference was not statistically significant (OR 1.45, 95% CI 0.93, 2.27).
Conclusion. NSAIDs are statistically superior in reducing rest and walking pain compared with acetaminophen for
symptomatic osteoarthritis. Safety, measured by discontinuation due to adverse events, was not statistically different
between NSAID- and acetaminophen-treated groups.

KEY WORDS. Osteoarthritis; Acetaminophen; Nonsteroidal antiinflammatory agents; Therapy.

INTRODUCTION based on studies showing comparable efficacy of acet-

The current American College of Rheumatology (ACR)
guidelines for the management of osteoarthritis (OA) rec-
ommends using acetaminophen as first-line oral pharma-
cotherapy for treating OA pain (1). The recommendation is
aminophen to nonsteroidal antiinflammatory drugs
(NSAIDs) in short-term, symptomatic treatment of mild-to-
moderate joint pain due to OA, in addition to its more
favorable safety profile (2,3). However, recent clinical tri-
als have found NSAIDs to be more efficacious than acet-

Dr. Lee’s work is supported by grant T32-AR-07611: Na-
tional Research Service Award in Post Doctoral Training
in Rheumatology. Preparation of this manuscript was also
supported by The Northwestern Memorial Foundation,
The Goldberg Family Charitable Trust, and Merck & Co.,
Inc.
1
Chin Lee, MD, Thomas J. Schnitzer, MD, PhD: Northwest-
ern University, The Feinberg School of Medicine, Chicago,
Illinois; 2Walter L. Straus, MD, MPH, Suna Barlas, PhD,
Suzanne Vogel, MPH: Health Dialog Data Services; West
Point, Pennsylvania; 3Robert Balshaw, PhD: Simon Fraser
University, Burnaby, British Columbia, Canada.
Dr. Schnitzer is a consultant for AstraZeneca Pharmaceu-
ticals (London, UK), Glaxo SmithKline (London, UK), Mc-
Neil Consumer (Raritan, NJ), Merck & Co., Inc. (West Point,
PA), and Norvartis (East Hanover, NJ); he has received clin-
ical research support from AstraZenecea, Merck & Co., Inc.,
and Novartis; and is on the Speakers’ Bureau for Merck &
Co., Inc. and Ortho-McNeil Pharmaceutical (Raritan, NJ).
Drs. Straus and Barlas are currently employed by Merck &
Co., Inc. (West Point, PA). Ms. Vogel was formerly employed
by Merck & Co., Inc. (West Point, PA).
Address correspondence to Chin Lee, MD, Northwestern
University, The Feinberg School of Medicine, Division of
Rheumatology, McGaw Pavilion, 240 East Huron St., 2300,
Chicago, IL 60611. E-mail: c-lee17@northwestern.edu.
Submitted for publication October 17, 2003; accepted in
revised form March 26, 2004.

746
NSAIDs Versus Acetaminophen in OA
aminophen in some groups of patients with OA (4,5).
Moreover, there is evidence suggesting OA patients have
greater preference for NSAIDs compared with acetamino-
phen (6). In the past several years, a new class of NSAIDs,
cyclooxygenase 2-selective inhibitors (coxibs), have been
introduced. Coxibs offer improved gastrointestinal safety
profiles (7–13) and possess comparable efficacy to tradi-
tional nonselective NSAIDs. The emergence of coxibs has
raised debate over the most appropriate role of NSAIDs in
the OA treatment paradigm (14 –16).
A previous meta-analysis by Eccles et al compared
NSAIDs with acetaminophen using data from 3 clinical
trials (17). All included trials evaluated only traditional
nonselective NSAIDs because they were completed prior
to publication of trials assessing the clinical efficacy of
coxibs. Furthermore, in one of the included trials, diclofe-
nac was compared with acetaminophen combined with
dextropropoxyphene in subjects from a general practice
population with arthralgia, but not necessarily with OA of
the hip or knee (18). Since the publication of this meta-
analysis, additional clinical trials assessing efficacy of
NSAIDs, both nonselective NSAIDs and coxibs, as com-
pared with acetaminophen have been reported.
We conducted a meta-analysis to compare the reported
efficacy and safety of recommended dosages of NSAIDs,
including coxibs, versus acetaminophen in the treatment
of symptomatic hip and knee osteoarthritis.
METHODS
A search of Medline (1966 to February 2003) via PubMed
and EMBASE Drugs and Pharmacology database (1991 to
first quarter 2003) via Ovid was performed to identify
original clinical trials directly comparing NSAIDs with
acetaminophen in the treatment of symptomatic hip or
knee OA. PubMed is a service of the National Library of
Medicine providing access to ⬎12 million Medline cita-
tions (19). Ovid is a literature search engine that allows
search of multiple databases, including EMBASE Drugs
and Pharmacology database, a subset of Excerpta Medica
containing ⬎1,560,000 abstracts and citations of drug and
pharmacology literature (20).
Search strategy. Medline was searched using the fol-
lowing combinations of broad Major Exploded Subject
Headings (MesH) terms or text words: “Osteoarthritis” and
“anti-inflammatory agents, nonsteroidal” or “NSAIDs” and
“acetaminophen” or “paracetamol.” The EMBASE Drugs
and Pharmacology search strategy applied the following
combinations of MesH terms or key words: “Osteoarthri-
tis” and “nonsteroidal anti-inflammatory agent” or
“nsaids” and “paracetamol” or “acetaminophen.” The
search was not restricted by language or study design.
All articles in English underwent duplicate review by 2
authors (CHL and WLS). All non-English articles were
reviewed by an investigator fluent in the language of the
publication. Bibliographies from all identified full-text ar-
ticles were screened for any additional relevant articles
suitable for inclusion in the study not found by systematic
search of Medline and EMBASE Drug and Pharmacology
747
databases. A search of unpublished studies via contact
with expert informants knowledgeable in OA therapeutic
research and through review of personal files (TJS) was
completed to obtain data or results from existing unpub-
lished studies not identified from the formal search of the
literature.
Selection criteria. Exclusion criteria for titles. Titles
identified from an initial review of the literature were
excluded if the following criteria were met: 1) identified as
a review in the title, 2) article title pertained to an animal
study, 3) article title did not pertain to OA, or 4) study title
pertained to nonoral pharmacologic therapy of OA (e.g.,
intraarticular injections or topical therapies).
Exclusion criteria for abstracts. Abstracts passing title
screening were rejected from further review if any of the
following criteria were met: 1) review or summary article,
2) letter, 3) survey, 4) commentary, 5) editorial, 6) note, 7)
conference paper, 8) case report, 9) case series, 10) case
study, 11) basic science, 12) clinical practice guideline, 13)
ad-hoc analysis of previously published data, 14) no direct
comparison of an NSAID with acetaminophen or paraceta-
mol, or 15) a study comparing an NSAID with acetamino-
phen or paracetamol combined with a nonnarcotic anal-
gesic or narcotic agent.
Exclusion criteria for articles. Full-text articles for ab-
stracts not explicitly meeting abstract exclusion criteria
were reviewed. Articles found to meet any prior exclusion
criteria applicable to abstracts or fulfilling any of the fol-
lowing criteria were excluded from our analysis: 1) a meta-
analysis, 2) a single patient trial or “n-of-1” trial, 3) lack of
sufficient analyzable data, 4) trial assessing OA in joints
other than hips or knees, or 5) duration of exposure to the
NSAID in the study was ⬍7 days.
Data abstraction. A standardized form was used to ab-
stract all data, including information on study design;
study duration; number of subjects and their demographic
data (sex and age); joint sites of OA involvement; informa-
tion on prior NSAIDs or acetaminophen used by study
participants; inclusion of a washout period; incorporation
of a flare design; type of outcome measures used, including
outcome measures of pain at rest and walking pain as
measured by a visual analog scale (VAS); and dropouts
due to adverse events. To standardize abstracted data,
outcome measures of pain using a 10-cm VAS were col-
lected and converted to a 100-mm scale. All NSAID and
acetaminophen dosages utilized in the identified studies
were within the range of therapeutic dosages used in clin-
ical practice.
The authors of the identified studies for inclusion were
contacted when further clarification of original study data
or if additional information was necessary. In the study by
Pincus et al, the authors provided baseline, 6-week fol-
lowup, and mean difference in pain data for the rest and
walking pain components of the Western Ontario and Mc-
Master Universities Osteoarthritis (WOMAC) target joint
outcome measures reported in the original study (4). Case
et al supplied baseline, 6-week, and 12-week efficacy data
for WOMAC rest and walking pain scores (21). Williams et
748
al provided 6-week data on dropouts due to adverse events
(22).
Variables. Our overall measure of efficacy was the in-
verse-variance-weighted mean difference (WMD) of rest
and walking pain measures derived from pooled pain data
(in the form of pain scores as measured by or converted to
a 100-mm VAS) across included studies. Our safety mea-
sure was based on number of subject withdrawals due to
adverse events (AEs) from either the NSAID or acetamin-
ophen groups.
Study quality assessment. We applied the Jadad crite-
ria, an accepted measure of evaluating clinical trial qual-
ity, to determine the methodologic quality of the included
studies (23). The Jadad criteria were developed using ac-
tual published clinical trial data. They assess details of 3
items directly related to reduction of bias: blinding, ran-
domization, and subject dropouts or withdrawals. The
quality of a study was scored on a scale of 0 –5, with a
score of 3 or greater indicating a high-quality study. Arti-
cles having a Jadad score ⱖ 2 were included for review due
to the limited number of existing studies comparing
NSAIDs with acetaminophen.
Statistical analysis. Abstracted information on number
of subjects in each treatment group, data on pain levels at
baseline and at the end of a study period measured on a
100-mm VAS, and accompanying standard deviations
were used to determine mean change in pain level. All rest
and walking pain data were entered into Review Manager
(RevMan), Version 4.1 (The Cochrane Collaboration, Ox-
ford, England) for analysis. In studies that did not explic-
itly provide standard deviations for rest and walking pain
scores, the standard deviations using the square roots of
the simple averages of within-study variances were calcu-
lated. Difference in mean change in pain level from base-
line to the end of a study period, and the provided or
calculated standard deviations were used to determine
point estimates for rest and walking pain data for NSAID
and acetaminophen treatment groups.
The presence of heterogeneity was explored using a
random effects model (DerSimonian and Laird’s Q-test)
(24). Pain data were pooled across all NSAID groups in
studies with multiple NSAID treatment arms to derive
point estimates for pain scores in the absence of any sig-
nificant heterogeneity. An inverse variance method was
employed to combine point estimates for rest and walking
pain scores from across all studies to derive an overall
efficacy as represented by the WMD expressed using a
100-mm VAS for NSAID and acetaminophen groups. The
95% confidence intervals (95% CIs) for WMD of rest and
walking pain measures were determined for NSAID and
acetaminophen groups in each of the individual studies,
and an overall WMD was calculated using pooled pain
data from across the studies. Cohen’s effect sizes (ESs)
were determined using the mean improvement in pain
scores and the reported or calculated standard deviations
(25,26).
Odds ratios (ORs) and 95% CIs were calculated for the
Lee et al
safety measure as indicated by withdrawals due to AEs in
all identified studies with sufficient analyzable data. An
analysis comparing safety measures after excluding low-
dose NSAID treatment arms was performed to compare
high-dose NSAID with acetaminophen groups. In this
study, for trials including more than one NSAID dosage,
total daily doses of rofecoxib of 12.5 mg, celecoxib of 200
mg, and ibuprofen of ⬍2,400 mg were categorized as low-
dose NSAID arms, whereas total daily doses of ibuprofen
of 2,400 mg and rofecoxib of 25 mg were considered high-
dose NSAID arms. Analyses of available efficacy and safety
data from all included trials and subanalyses of these data
from high-quality studies (Jadad score ⱖ 3) were per-
formed.
RESULTS
A total of 25 articles were identified following application
of exclusion criteria for titles and abstracts from the search
of Medline and EMBASE Drug and Pharmacology data-
bases (Figure 1). Fourteen articles, 2 of which were dupli-
cate studies, were rejected based on the predetermined
exclusion criteria (Table 1) (17,27–37). Of the 10 remain-
ing articles, 3 were redundant publications referenced in
both literature databases, resulting in 7 nonduplicate trials
containing sufficient data for analysis (Figure 1) (4,22,38 –
41).
Study characteristics. The 7 studies included for ana-
lysis were composed of the following study designs: 2
randomized active comparator trials, which did not in-
clude placebo arms; 2 randomized parallel-group double-
blinded trials; 2 randomized crossover trials; and 1 ran-
domized placebo-controlled double-blinded trial (Table
2). A total of 1,252 subjects were randomized, with 752
subjects in the NSAID groups and 500 subjects placed in
the acetaminophen groups. All studies included subjects
with knee OA, but only the trials by Pincus et al and
Lequesne et al included individuals with hip OA (4,39).
Seven different types of NSAIDs, including 2 coxibs,
within recommended dose ranges were compared with
acetaminophen with total daily doses ranging from 2,600
mg to 4,000 mg depending on the study. The mean ⫾ SD
duration of trial period providing data for analysis was
49.4 ⫾ 24.7 days (range 24 – 84 days). Of the 7 clinical
trials included for review, 5 were of high methodologic
quality based on a Jadad score ⱖ 3 (4,21,22,38,41).
Study participants. Overall, the mean age of the study
participants was 61.1 years and most (71.1%) were women
with OA involving the hip or knee. Subjects randomized
into NSAID and acetaminophen arms were similar as re-
gard to age and sex. Information on duration of OA in
subjects prior to study enrollment was not uniformly avail-
able from all studies.
Outcome variables. Two of the identified studies pro-
vided usable pain data for either rest or walking pain
exclusively. The study by Erturk et al (40) contained out-
come measures only for walking pain scores, whereas the
NSAIDs Versus Acetaminophen in OA 749

Figure 1. Results of Medline and EMBASE Drug and Pharmacology database search.

trial performed by Lequesne et al (39) provided outcome
results for only rest pain data. A total of 726 subjects from
the NSAID groups versus 482 subjects in the acetamino-
phen groups from 5 of the original 7 studies contributed
analyzable data for rest pain. In comparison, a total of 652
subjects from the NSAID groups versus 399 subjects in the
acetaminophen groups from 6 of 7 identified studies con-
tributed analyzable data for walking pain.
Analysis of outcome variables. The test of heterogene-
ity was not statistically significant for either rest pain (␹2 ⫽
2.81, degrees of freedom [df] ⫽ 5, P ⫽ 0.73) or walking pain

Table 1. Excluded studies or articles identified from a computerized search of Medline

and EMBASE Drug and Pharmacology databases*

Study Reason for exclusion

Seideman, 1993† No direct comparison between NSAID and acetaminophen

Brooks, 1982† No direct comparison between NSAID and acetaminophen
Solomon, 1974† Insufficient data for analysis
Wojtulewski, 1974† Insufficient data for analysis
Wojtulewski, 1974† Insufficient data for analysis
Maneksha, 1973† Insufficient data for analysis
Eccles, 1998† Meta-analysis
Wegman, 2003‡ Single-patient trial
Rovetta, 2001‡ Trial involving OA of sites other than hip or knees
Whitehead, 1997‡ Insufficient data for analysis
Hackenthal, 1997‡ Review
Nikles, 2000§ Single-patient trial
March, 1994§ Single-patient trial

* NSAID ⫽ nonsteroidal antiinflammatory drug; OA ⫽ osteoarthritis.

† Study or article identified in Medline.
‡ Study or article identified in EMBASE Drug and Pharmacology database.
§ Duplicate articles identified in both Medline and EMBASE Drug and Pharmacology databases.
750 Lee et al

Table 2. Characteristics of trials comparing efficacy of nonsteroidal antiinflammatory drug(s) to acetaminophen in treatment
of osteoarthritis*

Acetaminophen
NSAID; total total dosage, Duration, Quality
Study, year n Study design dosage, mg/day; no. mg/day; no. days score†

Bradley, 1991‡ 184 Randomized comparator Ibuprofen; 2,400; 61 4,000; 61 28 4

without placebo
Ibuprofen; 1,200; 62
Case, 2003‡ 82 Randomized placebo Diclofenac; 150; 25 4,000; 29 84 4
controlled double
blinded
Erturk, 1998§ 35 Randomized comparator Acemetacin; 90; 20 2,000; 15 84 2
without placebo
Geba, 2002¶ 382 Randomized parallel group Celecoxib; 200; 97 4,000; 94 42 5
double blinded
Rofecoxib; 12.5; 96
Rofecoxib; 25; 95
Lesquesne, 1997# 192 Randomized crossover Floctafenin; 800; 94 3,000; 98 24** 2
Pincus, 2001†† 227 Randomized crossover Diclofenac; 150; 4,000; 115§§ 42¶¶ 4
112‡‡
Williams, 1993‡ 178 Randomized parallel group Naproxen; 750; 90 2,600; 88 730## 5
double blinded

* NSAID ⫽ nonsteroidal antiinflammatory drug.

† Quality score as per Jadad criteria.
‡ Sponsored by National Institutes of Health.
§ Sponsorship not stated.
¶ Sponsored by Merck & Co. Inc.
# Sponsored by Diamant Pharmaceuticals.
** Pre- and postcrossover data included.
†† Sponsored by Pharmacia.
‡‡ Outcome data for 106 participants available.
§§ Outcome data for 112 participants available.
¶¶ Only precrossover data included because flare status of participants was unclear prior to crossover.
## Intent-to-treat data from first 42 days analyzed.

(␹2 ⫽ 2.60, df ⫽ 5, P ⫽ 0.76), thus allowing us to pool data 32) (Table 3). Of the 7 identified studies, the ORs for 6
across included studies. Point estimates for rest pain im- trials were estimable to determine safety. The OR was not
provement as indicated by the WMD revealed improve- estimable for the study by Erturk et al (40) because there
ment favoring the NSAID-treated groups in 5 of the 6
studies (Figure 2). There was a statistically significant re-
duction in rest pain (P ⬍ 0.05) favoring the NSAID groups
in 3 of the 6 studies. Overall improvement in rest pain
using pooled data across all 6 studies showed a WMD of
– 6.33 (95% CI –9.24, –3.41; Figure 2) and an average ES of
0.23 favoring NSAID-treated groups. A subanalysis of im-
provement in rest pain for 5 of the 6 studies meeting
criteria for high methodologic quality showed similar re-
sults (WMD ⫽ – 6.01; 95% CI –9.21, –2.81; ES ⫽ 0.22).
Similarly, point estimates for walking pain (as indicated
by the WMD) showed improvement in 5 of 6 studies fa-
voring the NSAID-treated groups, however, in only 2 of the
6 studies did this difference reach statistical significance.
As in the case for rest pain, the overall improvement in
walking pain using pooled data from all 6 studies demon-
strated a WMD of –5.76 (95% CI – 8.99, –2.52; Figure 2)
and an average ES of 0.23 favoring the NSAID-treated
groups. A subanalysis of improvement in walking pain for
5 of the 6 studies meeting criteria for high methodologic
quality showed similar results (WMD ⫽ –5.67; 95% CI
– 8.97, –2.38; ES ⫽ 0.22).
Figure 2. Forest plots of weighted mean differences for rest and
Results for safety showed the NSAID-treated groups as walking pain improvement in nonsteroidal antiinflammatory drug
having a numerically higher number of dropouts due to (NSAID)– versus acetaminophen-treated groups. 95% CI ⫽ 95%
AEs (n ⫽ 63) than the acetaminophen-treated groups (n ⫽ confidence interval.
NSAIDs Versus Acetaminophen in OA 751

Table 3. Comparison of dropouts due to adverse events in NSAID-, high-dose NSAID-, and acetaminophen-treated groups*

All NSAID groups High-dose NSAID groups only* Acetaminophen

Study, year Dropouts Subjects % Dropouts Subjects % Dropouts Subjects %

Bradley, 1991 10 123 8.1 6 61 9.8† 5 61 8.2

Case, 2003 3 25 12.0 3 25 12.0 2 29 6.9
Erturk, 1998 0 20 0.0 0 20 0.0 0 15 0.0
Geba, 2002 17 288 5.9 6 95 6.3‡ 6 94 6.4
Lequesne, 1997 13 94 13.8 13 94 13.8 8 98 8.2
Pincus, 2001 12 112 10.7 12 112 10.7 6 115 5.2
Williams, 1993 8 90 8.9 8 90 8.9 5 88 5.7
Total 63 752 8.4 48 497 9.7 32 500 6.4

* High-dose nonsteroidal antiinflammatory drug (NSAID)–treated groups excluded low-dose NSAID arms (total daily doses of ibuprofen ⬍2,400 mg,
rofecoxib dose of 12.5 mg, and celecoxib dose of 200 mg).
† High-dose NSAID-treated group consisted of ibuprofen 2,400 mg/day arm.
‡ High-dose NSAID-treated group consisted of rofecoxib 25 mg/day arm.

were no withdrawals due to AEs. The 6 studies contribut-
ing AE data demonstrated a 95% CI that included 1 indi-
cating no statistically significant difference in any individ-
ual study favoring either the NSAID- or acetaminophen-
treated groups. Similarly, the overall safety measure
derived from the pooled data for dropouts due to AEs
showed no statistically significant difference between the
NSAID- and acetaminophen-treated groups (OR 1.45; 95%
CI 0.93, 2.27), as shown in Figure 3. However, the overall
trend for safety favored the acetaminophen-treated groups.
A subanalysis with exclusion of low-dose NSAID arms,
thus comparing the high-dose NSAID- versus acetamino-
phen-treated groups (Table 3), revealed no statistically
significant difference favoring either group, yet the general
trend for safety further favored the acetaminophen-treated
groups (OR 1.59; 95% CI 0.99, 2.54), as shown in Figure 4.
An analysis of dropouts due to AEs of high-dose NSAID
compared with acetaminophen among the 5 studies meet-
ing criteria for high methodologic quality showed similar
findings (OR 1.52; 95% CI 0.88, 2.62). The specific types of
AEs resulting in withdrawal for all included studies were
not discernible because of the 6 studies reporting with-
drawals due to AEs, 1 study reported AEs for all subjects
without specifying the number of withdrawals due AEs. In
the remaining 5 studies, specific types of AEs leading to
subject withdrawal were clearly discernable in only 2 of
the publications.
DISCUSSION
Our results indicate a statistically significant improve-
ment of rest and walking pain favoring NSAID- versus
acetaminophen-treated groups in studies with subjects
having symptomatic hip or knee OA. Of particular note is
the consistency of this finding across the evaluated stud-
ies, with all of the trials demonstrating point estimates
favoring NSAIDs over acetaminophen. In terms of safety,
there was an absence of any statistically significant differ-
ence in AE-associated withdrawals between NSAIDs and
acetaminophen groups across studies, but there was a
trend for improved safety favoring acetaminophen. We
observed similar efficacy and safety results in our analysis
of only those studies with Jadad scores ⱖ3, therefore,
inclusion of studies having lower Jadad scores had mini-
mal impact on the findings.
A prior meta-analysis by Eccles et al comparing NSAID
and acetaminophen consisted of 3 trials, 1 of which in-
cluded acetaminophen in combination with propoxy-
phene as a direct comparator to the diclofenac; further-

Figure 3. Dropouts due to adverse events as a measure of safety in Figure 4. Dropouts due to adverse events as a measure of safety in
nonsteroial antiinflammatory drug (NSAID)– versus acetamino- high-dose nonsteroidal antiinflammatory drug (NSAID)– versus
phen-treated groups. 95% CI ⫽ 95% confidence interval. acetaminophen-treated groups.
752
more, the trial participants had arthralgias but did not
have to meet predefined OA criteria for study enrollment
(17,18). The current meta-analysis includes 5 additional
trials not evaluated in the meta-analysis by Eccles et al.
Additionally, all trials in the present meta-analysis as-
sessed individuals having predefined OA of the hip or
knee. With the exception of 2 studies (4,38), which al-
lowed use of propoxyphene as rescue therapy during the
trial, either a traditional NSAID or coxib was compared
with acetaminophen alone, thus optimizing comparability
between NSAIDs and acetaminophen.
Meta-analysis provides a systematic approach to com-
pare results across similar trials (42). The improvements in
pain outcome reported by the WMDs are consistent with
previously reported findings in all individual trials iden-
tified for analysis, with the exception of those observed by
Bradley et al (38). The Bradley trial found statistically
similar improvement in walking pain scores in both the
acetaminophen and NSAID groups, whereas NSAIDs were
found to be superior to acetaminophen alone for rest pain
scores. In contrast, analysis of the reported Bradley data in
this study shows NSAIDs to be statistically superior to
acetaminophen in reduction of both rest and walking pain
measures. The observed discrepancy may be accounted for
by our approach of pooling data across the 2 NSAID groups
from the Bradley trial that permitted a more precise esti-
mate of the WMD between NSAIDs and acetaminophen
groups; additionally, the Bradley trial was not designed or
powered as an equivalency study. The failure to show
superiority of NSAIDs to acetaminophen for walking pain
in the original study may have been a function of the
relatively small sample size.
Despite the consistent superiority for OA symptom im-
provement favoring NSAIDs in all but 1 of the included
studies, there are limited data to correlate these results
with clinical significance. A recent study determined that
differences of 9 –12 mm on a normalized 100-mm VAS
pain scale were required to achieve minimal perceptible
clinical improvement in pain among patients with knee or
hip OA (43). Similarly, estimation of Cohen’s effect sizes
to determine the standardized mean difference for im-
provement in rest and walking pain between NSAIDs and
acetaminophen in the analyzed studies are approximately
between 0.2 and 0.3, a range generally considered to reflect
a small effect size (5,25). Thus, the additional clinical
benefit in reduction of rest or walking pain with relatively
short-term (4 of the studies were ⱕ 6 weeks in duration)
use of NSAIDs compared with acetaminophen may be
limited. However, additional study is needed to better
interpret the clinical significance of these differences.
Various factors may impact the results from individual
studies, and hence our overall analysis. First, at least 2 of
the larger studies, specifically the trials by Bradley et al
(38) and Geba et al (41) included in this meta-analysis
were evaluated under clinical trial conditions that re-
quired a flare of OA pain after discontinuation of any prior
medication. With the incorporation of a flare design, these
2 higher-weighted trials contributing to our meta-analysis
focused on relief of moderate-to-severe levels of pain. It
has been proposed that NSAIDs are more effective than
acetaminophen at higher pain levels (5), which would be
Lee et al
consistent with findings from these 2 trials. The relative
efficacy of NSAIDs and acetaminophen to relieve mild
pain may differ from what is reported in such trials. Be-
cause data were not available at an individual subject level
from most of these studies, the question of heterogeneity in
pain relief with different baseline pain levels could not be
investigated.
In addition, data were unavailable regarding patients’
prior experience with acetaminophen or NSAIDs. There is
the possibility that some of these trials were undertaken in
populations enriched with subjects who were nonre-
sponders or underresponders to acetaminophen or
NSAIDs. Such data would have been useful to address
possible selection bias. These points are important to keep
in mind when attempting to extrapolate the results to the
clinical setting.
Two of the 7 studies (4,38) clearly indicated the use of
rescue medication during the trial. Propoxyphene was em-
ployed in both studies, however, the frequency of use by
either treatment group was unavailable from either publi-
cation. One study (41) allowed acetaminophen as rescue
therapy, but only during the washout period. The remain-
ing studies included in our analysis did not indicate if
rescue medication was utilized. Individuals with greater
baseline level of OA pain or use of a limited form of
treatment, whether an NSAID or acetaminophen, would be
more likely to utilize rescue therapy. However, without
data on active use of rescue medication from any of the
studies, conclusions in regard to this issue would be spec-
ulative.
A formal test of heterogeneity based on dosage of either
NSAID or acetaminophen failed to demonstrate signifi-
cance, thus allowing comparison of the different treatment
groups. However, reasonable exploration of any underly-
ing differences in efficacy based on varying NSAID dos-
ages (i.e., high versus low) among the included treatment
groups was limited by the small number of studies with
different dosage levels of NSAIDs examined. In general,
previous investigations have failed to demonstrate clini-
cally meaningful differences between varying dosage lev-
els of NSAIDs for the treatment of OA pain (26,44).
Although the absolute number of withdrawals due to
AEs was greater in the NSAID- than acetaminophen-
treated groups, this finding did not reach statistical signif-
icance. The trend for higher withdrawals in the NSAID
arms due to AEs was accentuated with elimination of
low-dose NSAIDs, suggesting an overall tolerability/safety
advantage for the tested acetaminophen regimens. How-
ever, the significance of this finding is limited by the
relatively small number of high-dose NSAID groups as-
sessed, along with the lower daily dosages of acetamino-
phen used in 3 of the comparator groups. Finally, only 1
study employing a high-dose coxib (i.e., rofecoxib 25 mg/
day) was included in analysis of high-dose NSAID arms;
however, there were insufficient data to specifically ad-
dress the safety of coxibs versus acetaminophen.
The included studies offered limited uniform data on
specific types of treatment- associated AEs leading to with-
drawal; thus, our safety variable served as a relatively
crude means of evaluating overall drug safety. Because
these trials were primarily designed to assess efficacy,
NSAIDs Versus Acetaminophen in OA
there were less systematically recorded safety data avail-
able. It is likely that other clinical trials would be needed
to more satisfactorily address the types and frequencies of
specific AEs associated with NSAIDs versus acetamino-
phen, and in turn allow a more precise assessment of
risk– benefit ratio of either therapy in the treatment of
symptomatic OA.
There are inherent limitations in most meta-analyses.
One of these is publication bias, which refers to the ten-
dency for authors and peer-refereed journals to preferen-
tially publish studies that report positive results. Studies
that report null or nonsignificant conclusions are less
likely to be published (42), and this may be applicable to
industry-supported trials (45). In a study reviewing pub-
lished randomized controlled trials of NSAIDs for treat-
ment of osteoarthritis and rheumatoid arthritis, Rochon et
al (46) reported that industry-associated NSAIDs were of-
ten equal or superior in efficacy and safety to the compar-
ison drug in industry-supported trials (46). Not all of the
trials examined in this study had industry sponsorship.
Two of the 3 studies undertaken without industry support
(22,38) demonstrated improvement for rest pain favoring
NSAIDs, and 1 (38) showed NSAIDs to be superior com-
pared with acetaminophen for walking pain. In contrast,
industry-supported trials demonstrated less robust find-
ings favoring NSAID efficacy. Regardless of sponsorship,
there was a consistent, but not statistically significant,
trend for greater AE-associated withdrawal across all stud-
ies for the NSAID groups. These observations suggest an
absence of sponsorship influence on outcomes among the
identified trials analyzed in this meta-analysis.
Despite assiduous efforts to obtain data from known
studies relevant to the issue of publication bias, we were
unable to obtain data from investigators of at least 3 pre-
viously completed OA trials containing data directly com-
paring NSAIDs with acetaminophen. We cannot draw any
additional conclusions on the effect of not having in-
cluded these data. Additionally, due to the limited number
of identified articles and the familiarity of the studies by
the investigators, the abstraction process was not blinded,
potentially introducing further bias during data collection.
However, utilization of standardized data abstraction
forms and the use of multiple data extractors were de-
signed to minimize bias in the data abstraction process.
The findings of this study can be used to generate further
discussion regarding the current ACR guidelines recom-
mending the use of acetaminophen as first-line oral phar-
macologic treatment of symptomatic hip and knee OA.
Acetaminophen has been favored as first-line OA therapy
by the ACR because of its analgesic efficacy, favorable
safety profile, and relatively low cost. In this meta-analy-
sis, NSAID therapy was consistently associated with sta-
tistically superior efficacy in relief of resting and walking
pain. In aggregate, the efficacy advantages were relatively
modest in effect and were accompanied by a trend for
higher AE risk in the NSAID-treated subjects. The small
number of available studies did not allow us to address
several outstanding questions, including the efficacy of
high-dose nonselective NSAIDs or different coxibs versus
high-dose acetaminophen (i.e., 4,000 mg per day) nor to
753
fully characterize the associated safety and tolerability
differences.
REFERENCES
1. Altman RD, Hochberg MC, Moskowitz RW, Schnitzer, TJ.
Recommendations for the medical management of osteoar-
thritis of the hip and knee: 2000 update. Arthritis Rheum
2000;43:1905–15.
2. Hochberg MC, Altman RD, Brandt KD, Clark BM, Dieppe PA,
Griffin MR, et al. Guidelines for the medical management of
osteoarthritis. Part I. Osteoarthritis of the hip. Arthritis
Rheum 1995;38:1535– 40.
3. Hochberg MC, Altman RD, Brandt KD, Clark BM, Dieppe PA,
Griffin MR, et al. Guidelines for the medical management of
osteoarthritis. Part II. Osteoarthritis of the knee. Arthritis
Rheum 1995;38:1541– 6.
4. Pincus T, Koch G, Sokka T, Lefkowith J, Wolfe F, Jordan JM,
et al. A randomized, double-blind, crossover clinical trial of
diclofenac plus misoprostol versus acetaminophen in patients
with osteoarthritis of the hip or knee. Arthritis Rheum 2001;
44:1587–98.
5. Altman RD, The IAP Study Group. Ibuprofen, acetaminophen
and placebo in osteoarthritis of the knee: a six-day double-
blind study. Arthritis Rheum 1999;42 Suppl 9:S404.
6. Pincus T, Swearingen C, Cummins P, Callahan LF. Preference
for nonsteroidal anti-inflammatory drugs versus acetamino-
phen and concomitant use of both types of drugs in patients
with osteoarthritis. J Rheumatol 2000;27:1020 –7.
7. Lanza FL, Simon TJ, Quan H, Bolognese JA, Hoover MF,
Wilson FR, et al. Specific inhibition of cyclooxygenase-2 with
MK-0966 is associated with less gastroduodenal damage than
either aspirin or ibuprofen. Aliment Pharmacol Ther 1999;13:
761–7.
8. Simon LS, Weaver AL, Graham DY, Kivitz AJ, Lipsky PE,
Hubbard RC, et al. Anti-inflammatory and upper gastrointes-
tinal effects of celecoxib in rheumatoid arthritis: a random-
ized controlled trial. JAMA 1999;282:1921– 8.
9. Langman MJ, Jensen DM, Watsoin DJ, Harper SE, Zhao PL,
Quan H, et al. Adverse upper gastrointestinal effects of rofe-
coxib compared with NSAIDs. JAMA 1999;282:1929 –33.
10. Laine L, Harper S, Simon T, Bath R, Johanson J, Schwartz H,
et al, Rofecoxib Osteoarthritis Endoscopy Study Group. A
randomized trial comparing the effect of rofecoxib, a cyclo-
oxygenase 2-specific inhibitor, with that of ibuprofen on the
gastroduodenal mucosa of patients with osteoarthritis. Gas-
troenterology 1999;117:776 – 83.
11. Bensen WG, Zhao SZ, Burke TA, Zabinski RA, Makuch RW,
Maurath CJ, et al. Upper gastrointestinal tolerability of cele-
coxib, a COX-2 specific inhibitor, compared to naproxen and
placebo. J Rheumatol 2000;27:1876 – 83.
12. Goldstein JL, Silverstein E, Agrawal NM, Hubbard RC, Kaiser
J, Maurath CJ, et al. Reduced risk of upper gastrointestinal
ulcer complications with celecoxib, a novel COX-2 inhibitor.
Am J Gastroenterol 2000;95:1681–90.
13. Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R,
Davis B, et al. Comparison of upper gastrointestinal toxicity of
rofecoxib and naproxen in patients with rheumatoid arthritis.
N Engl J Med 2000;343:1520 – 8.
14. Felson DT. The verdict favors nonsteroidal anti-inflammatory
drugs for treatment of osteoarthritis and a plea for more evi-
dence on other treatments. Arthritis Rheum 2001;44:1477– 80.
15. Schnitzer TJ. Osteoarthritis management: the role of cycloox-
ygenase-2-selective inhibitors. Clin Ther 2001;23:313–26.
16. Schnitzer TJ. Update of ACR guidelines for osteoarthritis: role
of the coxibs. J Pain Symptom Manage 2002;23(4 Suppl):S24 –
30; discussion S31– 4.
17. Eccles M, Freemantle N, Mason J, The North of England
Non-Steroidal Anti-Inflammatory Drug Guideline Develop-
ment Group. North of England evidence based guideline de-
velopment project: summary guideline for non-steroidal anti-
inflammatory drugs versus basic analgesia in treating the pain
of degenerative arthritis. BMJ 1998;317:526 –30.
754
18. Parr G, Darekar B, Fletcher A, Bulpitt CJ. Joint pain and
quality of life; results of a randomised trial. Br J Clin Pharma-
col 1989;27:235– 42.
19. PubMed Website. URL: http://www.ncbi.nlm.nih.gov/entrez/
query/static/overview.html.
20. Ovid Website. URL: http://www.ovid.com/site/index.jsp.
21. Case JP, Baliunas AJ, Block JA. Lack of efficacy of acetamin-
ophen in treating symptomatic knee osteoarthritis: a random-
ized, double-blind, placebo-controlled comparison trial with
diclofenac sodium. Arch Intern Med 2003;163:169 –78.
22. Williams HJ, Ward JR, Egger MJ, Neuner R, Brooks RH, Clegg
DO, et al. Comparison of naproxen and acetaminophen in a
two-year study of treatment of osteoarthritis of the knee. Ar-
thritis Rheum 1993;36:1196 –206.
23. Jadad AR, Moore RA, Carroll D, Jenkinson C, Reynolds DJ,
Gavaghan DJ, et al. Assessing the quality of reports of ran-
domized clinical trials: is blinding necessary? Control Clin
Trials 1996;17:1–12.
24. DerSimonian R, Laird N. Meta-analysis in clinical trials. Con-
trol Clin Trials 1986;7:177– 88.
25. Cohen J. Statistical power analysis for the behavioral sci-
ences. 2nd ed. Hillsdale (NJ): Lawrence Erlbaum Associates;
1988.
26. Gotzsche PC. Non-steroidal anti-inflammatory drugs. BMJ
2000;320:1058 – 61.
27. Seideman P, Samuelson P, Neander G. Naproxen and parac-
etamol compared with naproxen only in coxarthrosis: in-
creased effect of the combination in 18 patients. Acta Orthop
Scand 1993;64:285– 8.
28. Brooks PM, Dougan MA, Mugford S, Meffin E. Comparative
effectiveness of 5 analgesics in patients with rheumatoid ar-
thritis and osteoarthritis. J Rheumatol 1982;9:723– 6.
29. Solomon L, Abrams G. Orudis in the management of osteo-
arthritis of the knee: a double-blind trial. S Afr Med J 1974;
48:1526 –9.
30. Wojtulewski JA, Hart FD, Huskisson EC. Fenoprofen in treat-
ment of osteoarthrosis of hip and knee. BMJ 1974;2:475– 6.
31. Wojtulewski JA. Fenoprofen in the treatment of osteoarthro-
sis. Curr Med Res Opin 1974;2:551–5.
32. Maneksha S. ‘Safapryn’ and benorylate—a comparative trial
of two new preparations of aspirin and paracetamol in the
treatment of rheumatoid arthritis and osteoarthritis. Curr Med
Res Opin 1973;1:563– 8.
33. Rovetta G, Monteforte P. Dexketoprofen-trometamol in pa-
tients with osteoarthritis of the hands. Minerva Ortopedica e
Traumatologica 2001;52:27–30.
Lee et al
34. Whitehead J, Thomas P. A sequential trial of pain killers in
arthritis: issues of multiple comparisons with control and of
interval-censored survival data. J Biopharmaceutical Stat
1997;7:333–55.
35. Hackenthal E. Paracetamol and metamizol in the treatment of
chronic pain syndromes. Schmerz 1997;11:269 –75.
36. Nikles CJ, Glasziou PP, Del Mar CB, Duggan CM, Clavarino A,
Yelland MJ. Preliminary experiences with a single-patient
trials service in general practice. Med J Aust 2000;173:100 –3.
37. March L, Irwig L, Schwarz J, Simpson J, Chock C, Brooks P. n
of 1 trials comparing a non-steroidal anti-inflammatory drug
with paracetamol in osteoarthritis. BMJ 1994;309:1041–5; dis-
cussion 1045– 6.
38. Bradley JD, Brant KD, Katz BP, Kalasinski LA, Ryan SI. Com-
parison of an anti-inflammatory dose of ibuprofen, an analge-
sic dose of ibuprofen, and acetaminophen in the treatment of
patients with osteoarthritis of the knee. N Engl J Med 1991;
325:87–91.
39. Lequesne M, Fannius J, Reginster J-Y, Verdickt W, Vincent du
Laurier M. Floctafenin versus acetaminophen for pain control
in patients with osteoarthritis in the lower limbs. Rev Rhum
[Engl Ed] 1997;64:327–33.
40. Erturk H, Celker R, Aydin M, Ugur O. Comparison of efficacy
and tolerability of acemetacin and acetaminophen in the
treatment of knee osteoarthritis. J Rheum Med Rehab 1998;9:
157– 61.
41. Geba GP, Weaver AL, Polis AB, Dixon ME, Schnitzer TJ,
Vioxx, Acetaminophen, Celecoxib Trial (VACT) Group. Effi-
cacy of rofecoxib, celecoxib, and acetaminophen in osteoar-
thritis of the knee: a randomized trial. JAMA 2002;287:64 –71.
42. Petitti DB. Information retrieval. In: Meta-analysis, decision
analysis, and cost-effectiveness analysis. 2nd ed. New York:
Oxford University Press; 2000. p. 43– 67.
43. Ehrich EW, Davies GM, Watson DJ, Bolognese JA, Seidenberg
BC, Bellamy N. Minimal perceptible clinical improvement
with the Western Ontario and McMaster Universities Osteo-
arthritis Index questionnaire and global assessments in pa-
tients with osteoarthritis. J Rheumatol 2000;27:2635– 41.
44. Gotzsche PC. Non-steroidal anti-inflammatory drugs. Clin
Evid 2002;7:1063–70.
45. Davidson RA. Source of funding and outcome of clinical
trials. J Gen Intern Med 1986;1:155– 8.
46. Rochon PA, Gurwitz JH, Simms RW, Fortin PR, Felson DT,
Minaker KL, et al. A study of manufacturer-supported trials of
nonsteroidal anti-inflammatory drugs in the treatment of ar-
thritis. Arch Intern Med 1994;154:157– 63.