Rheumatology 2000;39:975–981
Meta-analysis of treatment termination rates
among rheumatoid arthritis patients receiving
disease-modifying anti-rheumatic drugs
A. Maetzel, A. Wong, V. Strand1, P. Tugwell2, G. Wells and
C. Bombardier3
University Health Network, Toronto, Ontario, Canada 1Division of Immunology,
Stanford University, Stanford CA, USA, 2Department of Medicine and The
Loeb Health Research Institute, Clinical Epidemiology Unit, University of
Ottawa, Ontario, 3Arthritis & Autoimmunity Research Centre, University Health
Network, Toronto, Ontario and Clinical Epidemiology and Healthcare Research
Program, University of Toronto, Ontario, Canada
Abstract
Objective. To summarize the evidence on treatment withdrawal rates reported in
observational studies and randomized controlled trials (RCTs) of methotrexate (MTX ),
parenteral gold (GST ), sulphasalazine (SSZ) and hydroxychloroquine (HCQ) among patients
with rheumatoid arthritis (RA).
Methods. Two independent Medline searches were used to retrieve relevant studies
published between 1966 and 1997. Those which disclosed information on the number of
patients withdrawing from the drug were retained. Cumulative probabilities of survival on
treatment were then computed using actuarial survival estimates, and differences were tested
using log-rank, Wilcoxon and Cox proportional hazards tests.
Results. A total of 159 studies provided withdrawal information, and the numbers of
patients who withdrew, in general or because of inefficacy or toxicity, could be abstracted
from 110 studies contributing 142 treatment arms (MTX, 48; GST, 56; SSZ, 22; HCQ, 16).
Data for HCQ were available only up to 24 months, but combined percentages of patients
estimated to have continued MTX, GST or SSZ, respectively, for 60 months were 36, 23 and
22% when all failures were considered, 75, 73 and 53% when withdrawals due to lack of
efficacy alone were considered, and 65, 36 and 48% when only withdrawals due to toxicity
were taken into account. The Cox proportional hazards test performed on all withdrawals,
after adjusting for year of publication and type of study, revealed that patients remained on
MTX significantly longer than they did on the other three agents; however, the patients stayed
significantly longer on GST than MTX when withdrawals for inefficacy were analysed
separately. No significant differences in withdrawal rates were noted between observational
studies and RCTs.
Conclusion. Patients with RA stay significantly longer on MTX than on other diseasemodifying anti-rheumatic drugs. Higher withdrawal rates among those given GST are mainly
due to high toxicity, whereas the majority of withdrawals from SSZ and HCQ result from
lack of efficacy. Withdrawal rates in observational studies are similar to those reported in
RCTs.
Given the chronic nature of rheumatoid arthritis (RA),
it is important to obtain long-term evidence of the
success of therapy with disease modifying anti-rheumatic
drugs (DMARDs). This may be provided by the findings
of observational studies or randomized controlled trials
(RCTs).
Compared with observational studies, RCTs generate
more detailed and standardized data concerning the
therapeutic response [1], but are generally conducted
over periods that rarely exceed 1 yr. The therapeutic
potential of the study drugs must therefore often be
extrapolated beyond the period of investigation.
Furthermore, patient populations in RCTs are highly
selected and monitoring is intense. Some investigators
Submitted 22 October 1999; revised version accepted 24 February
2000.
Correspondence to: A. Maetzel, University Health Network, 610
University Avenue, Room 16-741, Toronto, Ontario M5G 2M9,
Canada.
975
© 2000 British Society for Rheumatology
976
A. Maetzel et al.
have argued that observational studies, which typically
document treatment failure as the proportion of patients
who withdraw from therapy due to lack of efficacy,
treatment toxicity or reasons unrelated to therapy, provide more realistic estimates of how patients respond to
therapy [2–7].
Observational studies of drug therapy often use survival analysis to assess treatment withdrawal rates. For
outcomes such as time to treatment failure, the product
limit method is a fundamental part of this approach,
allowing the estimation of withdrawal rates even when
patients enter a study at different times and are therefore
observed for varying periods [8]. However, because
estimates of survival become unstable at follow-up
periods reached by only a few patients, such analyses
are usually stopped when too few subjects remain on
treatment for reasonable inferences to be drawn. It is
therefore advantageous to combine data from several
observational studies in order to increase sample size,
particularly at the tail end of the survival curve.
A number of observational studies of DMARD
therapy in RA have now been published, creating an
important opportunity to summarize their findings in
combination with those of RCTs. Agents considered in
the present meta-analysis are parenteral gold (GST ),
hydroxychloroquine (HCQ), methotrexate (MTX ) and
sulphasalazine (SSZ). The primary objective was to
determine overall treatment withdrawal rates and to
look at how they break down according to lack of
efficacy and toxicity. A secondary objective was to
compare treatment withdrawal rates from observational
studies with those from RCTs.
Methods
Selection of references
The Medline database from 1966 to August 1997 was
searched by combining the keyword ‘rheumatoid arthritis’ with text words and keywords for GST, SSZ, MTX
and HCQ, including common synonyms. Letters, editorials and comments were excluded, as were articles written
in languages other than English, French, German and
Spanish. The resulting references were reduced to a set
of 1387 by excluding all those that did not address
efficacy or toxicity. A further 1035 references derived
from a broader search conducted by the RA subgroup
of the Cochrane Musculoskeletal Review Group were
added, to give a total of 2422. The Cochrane set of
references was retrieved from the Excerpta Medica and
Medline electronic databases, using key words and text
words for (1) randomized clinical trial, (2) clinical trial,
(3) comparative study, (4) evaluation study, (5) followup study, (6) prospective study, combined with all
references indexed under (7) arthritis, rheumatoid.
Screening and assessment of references
Based on information in titles and abstracts, the 2422
references resulting from both searches were classified
according to the four agents of interest and the design
of the study. Eligible designs were RCTs, observational
studies including case series, case–control studies and
cohorts, and case reports. Paper copies were obtained
of all clearly relevant papers, plus those with uninformative titles and/or abstracts. A total of 445 investigations
were found that provided original information. These
were then assessed by two authors according to the
following criteria: (1) does the study document the
experience of patients from the day therapy was initiated? (2) does the study provide data for one of the
four therapies of interest? (3) does the study provide
information on treatment withdrawal rates? (4) do all
included patients have a documented diagnosis of RA?
Investigations meeting all four criteria were retained as
the set of studies providing withdrawal information. Of
these, only studies disclosing the actual number of
patients withdrawing from the study drug were used for
the meta-analysis. Authors of four large excluded studies
were approached with a request to provide additional
information but were either unable to provide the data
or declined to collaborate.
All observational studies were further appraised
according to the following criteria: (1) was reporting of
termination rates a primary or secondary objective?
(2) was outcome ascertainment prospective or retrospective? (3) was the clinical response specifically
defined? (4) with regard to follow-up, were (a) all
patients followed over a fixed interval independent of
treatment status (no censoring), or were (b) patients
followed over intervals varying from a few months to
several years depending on study entry and treatment
termination (with censoring)? (5) were lifetable or
Kaplan–Meier survival analysis techniques applied?
(6) did the study provide information about the numbers
of patients at risk, censored and terminating treatment
at each interval?
Data extraction
Numbers of patients reported to have withdrawn from
treatment with each DMARD were abstracted independently by two of the authors, whose findings were
checked for agreement and corrected where necessary.
The data set was then constructed. Take, for example,
the treatment arm of a study of 6 months’ duration in
which five of 20 patients had withdrawn from treatment
by 6 months. Those five patients were coded as one
observation and classified as treatment failures at the
midpoint of the interval (month 3), whereas the 15
patients who continued treatment were coded as a
second observation and classified as censored at the end
of the study (month 6). Each observation was weighted
by the number of patients, such that the first counted
as contributing five patients and the second as contributing 15. Observations from a few studies conducted over
long periods without providing detailed interval
information were modified by apportioning withdrawals
to 6-month intervals assuming a constant failure rate.
Analysis
Survival analysis and statistical comparisons were then
performed. Actuarial survival curves stratified by
Meta-analysis of RA treatment termination rates
DMARD were plotted for all studies entered into the
database using four definitions of failure: (1) withdrawal
due to lack of efficacy, toxicity or other reason; (2) withdrawal due to lack of efficacy or toxicity; (3) withdrawal
due to lack of efficacy; and (4) withdrawal due to
toxicity. For the purposes of analyses 2, 3 and 4,
withdrawals not defined as treatment failure were coded
as treatment continuations up to the midpoint of the
interval.
Statistical comparisons were undertaken in two steps:
first, the cumulative survival probabilities for treatment
with each of the four DMARDs were compared pairwise
using log-rank and Wilcoxon statistics; secondly, withdrawal rates for each agent were compared using Cox’s
proportional hazards model (proc PHREG in SAS )
with adjustment for study type and year of publication.
As many treatment failures occurred at the same timepoint, the exact algorithm for tied events was used.
HCQ treatment was excluded from the comparisons
because of the limited amount of data available and
because it is generally prescribed in mild RA, although
survival curves are presented. Three statistical comparisons were performed within each of the four groups:
(1) MTX vs GST; (2) MTX vs SSZ; and (3) SSZ vs
GST. Further subgroup analyses compared the effect of
study type on treatment withdrawal rates for MTX,
SSZ and GST separately. A time limit of 24 months
was set and the same statistical approaches were used.
Results
Of the initial 445 studies, 159 satisfied all four screening
criteria (A list of all 159 references is available from the
author upon request.); 71 were RCTs and 88 observational studies. Reasons for the 286 exclusions were as
follows: 132 (46.2%) did not evaluate an inception
cohort, 220 (76.9%) did not provide information on
treatment withdrawal, 32 (11.2%) studied diagnoses
other than RA, and 45 (15.7%) evaluated drugs other
than MTX, GST, HCQ or SSZ (some studies were
excluded for more than one reason). Among the 159
studies were 110 that provided withdrawal information
detailed enough to be included in the present metaanalysis.
Further assessment of the observational studies for
methodological characteristics showed that provision of
information on treatment termination rates was a primary objective in 42 studies (48%) and outcome ascertainment was prospective in 62 (70%); treatment success or
failure was defined by explicit criteria in 21 (24%).
Follow-up over a fixed interval independent of treatment
status was documented in 38 studies (43%), and of the
remaining 50 observational studies that reported a variable follow-up, 26 presented the data using Kaplan–
Meier survival analysis techniques. Raw numbers of
patients at risk, censored, and terminating treatment at
within-study intervals were reported in only 13, four
and seven papers, respectively.
The 110 studies included in the meta-analysis contributed 142 arms, the majority of them reporting results
977
with GST or MTX administration. Only three observational study arms provided withdrawal information for
HCQ ( Table 1). The maximum observation time on
HCQ treatment was only 24 months, whereas data for
GST, MTX and SSZ were reported for as long as
72 months.
Survival on therapy curves including all types of
withdrawals indicated that 36, 23 and 22% of the patients
continued MTX, GST and SSZ, respectively, for 60
months ( Fig. 1), with median survival times of 41, 24
and 18 months, respectively. The combined number of
patients at risk at baseline for each of the three drugs
was 2875 (MTX ), 3155 (GST ) and 1418 (SSZ). When
considering only withdrawals for lack of efficacy or
toxicity (excluding administrative reasons or relocation),
the percentage of patients estimated to have continued
MTX compared with GST or SSZ increased to 51%,
indicating that a considerable proportion failed to continue MTX for reasons other than lack of efficacy or
toxicity (data not shown as a figure).
Data on withdrawals for lack of efficacy only ( Fig. 2)
indicated that 75, 73 and 53% of patients receiving
MTX, GST and SSZ, respectively, continued therapy
for 60 months. For this subgroup analysis, the combined
number of patients at risk at baseline for each of the
three drugs was 2013 (MTX ), 2233 (GST ) and 1392
(SSZ). Analysis of withdrawals due to toxicity (treating
all other withdrawals as non-informative) revealed that
65, 36 and 48% of the patients continued MTX, GST
and SSZ, respectively, for 60 months ( Fig. 3).
Combined data from all published studies indicated
that, after adjusting for study type and year of publication, patients treated with SSZ were up to 1.9 times
more likely to fail therapy than patients treated with
MTX ( Table 2). Patients treated with GST were up to
2.3 times more likely to fail therapy than patients treated
with MTX. However, when looking at failures for lack
of efficacy only, it turns out that patients treated with
GST were significantly less likely to discontinue therapy
than patients on MTX. This finding is significant only
after adjusting for type of study and year of publication.
Patients on SSZ were more likely than patients on GST
to withdraw from therapy overall, and more than twice
as likely as patients receiving GST to withdraw because
of lack of efficacy. Patients on SSZ, however, were less
likely than those taking GST to withdraw because
of toxicity.
The comparison of withdrawal rates between observational studies and RCTs involving MTX, GST and SSZ
revealed no statistically significant differences for up to
24 months (Fig. 4).
Discussion
The present study summarizes rates of treatment withdrawal in published observational studies and RCTs of
MTX, GST, SSZ and HCQ. Analysis of the combined
data shows that patients stayed longest on single
DMARD therapy with MTX. However, nearly twothirds of all subjects started another therapy within 5 yr
A. Maetzel et al.
978
T 1. Number of patients at risk and duration of treatment in study arms from observational studies and randomized clinical trials reported
in the literature providing information on treatment withdrawals of patients
Gold
Hydroxychloroquine
OBS
RCT
OBS
Methotrexate
RCT
OBS
Sulphasalazine
RCT
OBS
RCT
Length
(months)
N
Pat.
N
Pat.
N
Pat.
N
Pat.
N
Pat.
N
Pat.
N
Pat.
N
Pat.
0, 6
6, 12
12, 24
24, 36
36, 48
48, 60
60, 72
Total
3
9
5
2
1
2
–
22
647
366
349
127
49
392
–
1930
14
18
2
–
–
–
–
34
387
689
149
–
–
–
–
1225
2
–
1
–
–
–
–
3
569
–
23
–
–
–
–
592
4
8
1
–
–
–
–
13
194
327
18
–
–
–
–
539
2
7
3
3
2
2
1
20
523
191
198
171
174
247
78
1582
14
11
2
1
–
–
–
28
404
616
88
185
–
–
–
1293
2
2
2
–
–
1
–
7
181
103
180
–
–
222
–
686
9
5
–
–
–
1
–
15
469
163
–
–
–
100
–
732
OBS, observational study; N, number of treatment arms; Pat., number of patients in treatment arms.
F. 1. Survival curves representing the percentage of patients
withdrawing from each agent because of inefficacy, toxicity or
other reasons. Six-month interval data were generated from
studies providing withdrawal information for intervals larger
than 12 months. The data are from observational studies and
randomized controlled trials.
because of lack of efficacy, treatment-associated adverse
events or other factors. When withdrawals due to lack
of efficacy alone were considered, there were more
patients withdrawing from MTX than GST, and it was
noted that a large proportion of patients who withdrew
from MTX did so for reasons other than inefficacy
or toxicity.
This is not the first investigation to make a formal
attempt to review withdrawals from treatment with
selected DMARDs [2, 9, 10], but it is the first to
combine data from observational studies and RCTs,
and to specifically compare rates of withdrawal due to
lack of efficacy, toxicity and other reasons. Wolfe [11]
noted in a review that estimates of 5-yr treatment
withdrawal vary considerably, those for MTX ranging
from 30 to 75%. Wolfe combined the 5-yr estimates of
MTX therapy, weighting them according to the sample
size of patients at risk at baseline; however, this may
not be appropriate as the precision of the survival
F. 2. Survival curves representing the percentage of patients
withdrawing from each agent because of inefficacy. Six-month
interval data were generated from studies providing withdrawal
information for intervals longer than 12 months. The data are
from observational studies and randomized controlled trials.
estimates is determined by the number of patients at
risk not only at baseline but throughout the study
period. Felson et al. [10] compared withdrawal rates
between DMARDs but restricted their review to RCTs.
Furthermore, they compared overall withdrawal rates
and withdrawals due to toxicity, but did not calculate
combined survival curves.
Some investigations that were not included in the
present meta-analysis—because they did not disclose
the actual number of patients withdrawing—documented higher rates of MTX maintenance at the 5-yr
interval. For example, Wolfe et al. [12] estimated that
~45% of subjects remained on MTX at 5 years, and
Pincus et al. [13] put the figure at ~65% beyond 5 yr.
Inclusion of these data would have slightly increased
the combined percentage of patients continuing MTX
for up to 5 yr. In fact, the year of publication independently contributed to survival differences for MTX and
SSZ. Withdrawal rates on MTX and SSZ have therefore
declined in recent years. Year of publication was
Meta-analysis of RA treatment termination rates
F. 3. Survival curves representing the percentage of patients
withdrawing from each agent because of toxicity. Six-month
interval data were generated from studies providing withdrawal
information for intervals longer than 12 months. The data are
from observational studies and randomized controlled trials.
F. 4. Percentage of patients withdrawing due to inefficacy,
toxicity or other reasons in observational studies and randomized controlled trials. The comparison was limited to 24
months’ duration and showed no statistically significant
difference in withdrawal rates between observational studies
and randomized controlled trials for all three agents.
insignificant for HCQ and GST, documenting no change
in survival on therapy for these drugs over the years.
This is confirmed by the similarity of the present 5-yr
treatment survival estimate of 23% to those by Wolfe
979
et al. [12] and Pincus et al. [13], who noted, respectively,
that 20 and 28% of patients continued GST therapy for
up to 5 yr.
Even though the combined rates of survival on
therapy presented here were obtained from several studies, only two treatment arms each for GST [14, 15] and
SSZ [15, 16 ], and three for MTX [17–19], contributed
information for up to 60 months. However, survival on
therapy curves for the combined studies reveal a consistent pattern within the first 2 yr and most are parallel,
suggesting that additional studies of the same agents
would be unlikely to make a considerable difference.
More detailed information concerning the disease status
and demographic characteristics of patient populations
would probably facilitate analysis of subgroups exhibiting different rates of survival on therapy, but that level
of detail could not, unfortunately, be achieved. Thus,
this meta-analysis can offer only a rough guide to
physicians who want to know how long a particular
patient is likely to continue treatment with an individual
DMARD.
A further limitation of the present study is the lack
of adjustment for meaningful covariables in the statistical comparisons. Although information was sought on
common variables such as joint count, rheumatoid factor
status, disease duration and the number of prior
DMARDs, it had too seldom been collected by investigators to allow more detailed analysis. Earlier studies
might have included patients with more severe disease,
leading to lower treatment survival rates, particular
with GST.
Examination of the underlying reasons for treatment
withdrawal in the present meta-analysis indicated that,
after adjustment for year of publication, patients on
GST were significantly less likely to withdraw from
therapy because of lack of efficacy than MTX patients.
These data do not reveal how many patients achieved
remission or significant clinical improvement during
treatment with these two agents, but they do encourage
us to take a closer look. GST is one of the least favoured
DMARDs [20], despite efficacy comparable to MTX
T 2. Statistical comparisons of treatment withdrawal rates between drugs by means of the log-rank, Wilcoxon and Cox proportional
hazards regression tests
GST vs MTX
P
All reasons
Toxicity/inefficacy
Inefficacy
Toxicity
Log-rank
Wilcoxon
Cox’s proportional
Log-rank
Wilcoxon
Cox’s proportional
Log-rank
Wilcoxon
Cox’s proportional
Log-rank
Wilcoxon
Cox’s proportional
RR, risk ratio (see text for explanation).
hazards
hazards
hazards
hazards
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.36
0.02
0.0077
< 0.0001
< 0.0001
< 0.0001
RR
1.40
1.68
0.73
2.28
SSZ vs MTX
P
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
RR
1.60
1.84
1.89
1.68
SSZ vs GST
P
0.10
0.0006
0.008
0.45
0.15
0.10
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.0031
< 0.0001
RR
1.14
1.09
2.57
0.74
980
A. Maetzel et al.
having been demonstrated by RCTs [21, 22], on large
databases [23, 24] and in the meta-analysis by Felson
et al. [10]. The present results suggest that toxicity is
the main reason for GST being out of favour. It is likely
that most of the problems relate to the cutaneous rash
that, it has been speculated, occurs more often in patients
more likely to get into remission [9]. However, GSTassociated adverse events can be managed in such a way
as to avoid discontinuation of therapy [25] or promote
initiation of a second course [25–27]. Furthermore,
withdrawal rates reported in the literature may obscure
the true efficacy of GST, as it has been observed that
patients who withdraw from GST are more likely to
experience a sustained improvement than patients withdrawing from MTX [28]. This finding may be a direct
consequence of using implicit judgement to classify
patients as toxicity or inefficacy withdrawals. In this
meta-analysis we had to rely on the original author’s
classification of withdrawals, which was often performed
without clear guidance by specific criteria, which in turn
leaves the possibility for classification biases that may
artificially favour one drug over others.
It is encouraging that, contrary to current opinion,
there was no apparent difference in withdrawal rates
between observational studies and RCTs. Theoretically,
withdrawal rates in RCTs could be either higher
(because of protocol-mandated withdrawals or patients
fearing that they are taking placebo) or lower (because
subjects participating in RCTs receive more care and
attention than is usual in the real world ). However,
although the present results show that equal trust can
be placed in reported withdrawal rates, whether they
originate from RCTs or observational studies, the relevant comparison was possible for only up to 24 months
of follow-up.
The product limit calculation of treatment survival
probabilities is a routine method employed in the
reporting of treatment success of various DMARDs.
However, it is possible that the method is often applied
to data which violate some basic assumptions for its use
[29]. For example, survival estimates in many older
studies may be biased downwards, as these patients
generally received their first DMARD late in the disease
course, but also because survival on a specific treatment
may improve as rheumatologists gain experience with it
[29]. This better survival for patients will not be given
its due respect when all patients are grouped together
for the purpose of the survival analysis. The results
presented here will only repeat the same potential biases
that may afflict the underlying studies, unless these
biases cancel each other out.
The decision to withdraw therapy is often made
implicitly in observational studies. Those that adopt
explicit definitions of clinical response and follow
common methodological reporting standards will
enhance the short-term results provided by RCTs and
supply valuable information on the long-term performance of antirheumatic therapies to physicians, patients
and researchers.
Acknowledgements
The authors wish to acknowledge the assistance of Dr
Maria Suarez-Almazor and the RA Subgroup of the
Cochrane Musculoskeletal Review Group in providing
a second set of references to be assessed as part of this
review, and the assistance of William Francis in editing
the manuscript.
Disclosure
This study was supported by an unrestricted grant to
the Arthritis and Autoimmunity Research Centre from
Aventis Canada Inc. The terms of the contract stipulated
that the authors should retain the right to absolute
control of the methods, conclusions and means of publication of the study. Drs Bombardier, Tugwell, Strand
and Wells are consultants to Aventis. Dr A Maetzel is
supported by a Ph.D. fellowship (health research) from
the Medical Research Council of Canada.
References
1. Felson DT, Anderson JJ, Boers M et al. The American
College of Rheumatology preliminary core set of disease
activity measures for rheumatoid arthritis clinical trials.
The Committee on Outcome Measures in Rheumatoid
Arthritis Clinical Trials. Arthritis Rheum 1993;36:729–40.
2. Hawley DJ, Wolfe F. Are the results of controlled clinical
trials and observational studies of second line therapy in
rheumatoid arthritis valid and generalizable as measures
of rheumatoid arthritis outcome? Analysis of 122 studies
[see comments]. J Rheumatol 1991;18:1008–14.
3. Pincus T, Stein CM. What is the best source of useful
data on the treatment of rheumatoid arthritis: clinical
trials, clinical observations, or clinical protocols? [editorial;
comment]. J Rheumatol 1995;22:1611–7.
4. Jenkins DP, Cooke MW, Glucksman EE. Audit of upper
limb fracture management in an accident and emergency
department. J Accident Emerg Med 1994;11:105–8.
5. Kirklin JW, Blackstone EH, Naftel DC, Turner ME.
Influence of study goals on study design and execution.
Controlled Clin Trials 1997;18:488–93.
6. Herman J. Advancing the standards of clinical research:
the urgent need for new methods and better data. J Eval
Clin Pract 1997;3:223–7.
7. Black N. Why we need observational studies to evaluate
the effectiveness of health care. Br Med J 1996;312:1215–8.
8. Streiner DL. Stayin’ alive: an introduction to survival
analysis. C J Psychiatry Rev Can Psychiatrie
1995;40:439–44.
9. Caspi D, Tishler M, Yaron M. Association between gold
induced skin rash and remission in patients with rheumatoid arthritis [see comments]. Ann Rheum Dis
1989;48:730–2.
10. Felson DT, Anderson JJ, Meenan RF. The comparative
efficacy and toxicity of second-line drugs in rheumatoid
arthritis. Results of two metaanalyses [see comments].
Arthritis Rheum 1990;33:1449–61.
11. Wolfe F. The epidemiology of drug treatment failure in
rheumatoid arthritis. [Review]. Baillieres Clin Rheumatol
1995;9:619–32.
12. Wolfe F, Hawley DJ, Cathey MA. Termination of slow
acting antirheumatic therapy in rheumatoid arthritis: a
Meta-analysis of RA treatment termination rates
13.
14.
15.
16.
17.
18.
19.
20.
21.
14-year prospective evaluation of 1017 consecutive starts.
J Rheumatol 1990;17:994–1002.
Pincus T, Marcum SB, Callahan LF. Longterm drug
therapy for rheumatoid arthritis in seven rheumatology
private practices: II. Second line drugs and prednisone.
J Rheumatol 1992;19:1885–94.
Ferraccioli GF, Salaffi F, Nervetti A, Manganelli P. Longterm outcome with gold thiosulphate and tiopronin in 200
rheumatoid patients. Clin Exp Rheumatol 1989;7:577–81.
Porter DR, McInnes I, Hunter J, Capell HA. Outcome of
second line therapy in rheumatoid arthritis. Ann Rheum
Dis 1994;53:812–5.
McEntegart A, Porter D, Capell HA, Thomson EA.
Sulfasalazine has a better efficacy/toxicity profile than
auranofin—evidence from a 5 year prospective, randomized trial. J Rheumatol 1996;23:1887–90.
Hoffmeister RT. Methotrexate therapy in rheumatoid
arthritis: 15 years experience. Am J Med 1983;75:69–73.
Scully CJ, Anderson CJ, Cannon GW. Long-term methotrexate therapy for rheumatoid arthritis. [Review]. Semin
Arthritis Rheum 1991;20:317–31.
Weinblatt ME, Kaplan H, Germain BF et al. Methotrexate
in rheumatoid arthritis. A five-year prospective multicenter
study. Arthritis Rheum 1994;37:1492–8.
Maetzel A, Bombardier C, Strand V, Tugwell P, Wells G.
How US and Canadian rheumatologists treat moderate
or aggressive rheumatoid arthritis: a survey. J Rheumatol
1998;25:2331–8.
Rau R, Herborn G, Karger T, Menninger H, Elhardt D,
Schmitt J. A double blind randomized parallel trial of
22.
23.
24.
25.
26.
27.
28.
29.
981
intramuscular methotrexate and gold sodium thiomalate
in early erosive rheumatoid arthritis. J Rheumatol
1991;18:328–33.
Rau R, Herborn G, Menninger H, Blechschmidt J.
Comparison of intramuscular methotrexate and gold
sodium thiomalate in the treatment of early erosive
rheumatoid arthritis: 12 month data of a double-blind
parallel study of 174 patients. Br J Rheumatol
1997;36:345–52.
Fries JF, Spitz PW, Mitchell DM, Roth SH, Wolfe F,
Bloch DA. Impact of specific therapy upon rheumatoid
arthritis. Arthritis Rheum 1986;29:620–7.
Wolfe F, Hawley DJ, Cathey MA. Measurement of gold
treatment effect in clinical practice: evidence for effectiveness of intramuscular gold therapy. J Rheumatol
1993;20:797–802.
Klinkhoff AV, Teufel A. How low can you go? Use of
very low dosage of gold in patients with mucocutaneous
reactions. J Rheumatol 1995;22:1657–9.
Klinkhoff AV, Teufel A. The second course of gold.
J Rheumatol 1995;22:1655–6.
Svensson A, Theander J. Skin rashes and stomatitis due
to parenteral treatment of rheumatoid arthritis with
sodium aurothiomalate. Ann Rheum Dis 1992;51:326–9.
Sander O, Herborn G, Bock E, Rau R. Prospective six
year follow up of patients withdrawn from a randomised
study comparing parenteral gold salt and methotrexate.
Ann Rheum Dis 1999;58:281–7.
Utley M, Gallivan S, Young A et al. Potential bias in
Kaplan–Meier survival analysis applied to rheumatology
drug studies. Rheumatology 2000;39:1–2.