Boutron I, Tubach F, Giraudeau B, Ravaud P. Methodological Differences in Clinical Trials Evaluating Nonpharmacological and Pharmacological Treatments of Hip and Knee Osteoarthritis. JAMA. 2003;290(8):1062–1070. doi:10.1001/jama.290.8.1062
Author Affiliations: Département d'Epidémiologie, Biostatistique et Recherche Clinique, Groupe Hospitalier Bichat–Claude Bernard (Assistance Publique des Hôpitaux de Paris), Faculté Xavier Bichat, Université Paris VII (Drs Boutron, Tubach, and Ravaud), and INSERM EMI 0357 (Drs Boutron, Tubach, and Ravaud), Paris, France; and INSERM CIC 202, Tours, France (Dr Giraudeau).
Context Randomized controlled trials have been developed essentially in the
context of pharmacological treatments (ie, oral drugs; intra-articular injection;
and topical, intramuscular, and intravenous treatments), but assessment of
the effectiveness of nonpharmacological treatments (ie, surgery, arthroscopy,
joint lavage, rehabilitation, acupuncture, and education) presents specific
Objectives To compare the quality of articles of nonpharmacological and pharmacological
treatments of hip and knee osteoarthritis and to identify specific methodological
issues related to assessment of nonpharmacological treatments.
Design and Setting We searched MEDLINE and the Cochrane Central Register of Controlled
Trials for articles of randomized controlled trials published between January
1, 1992, and February 28, 2002, in 28 general medical and specialty journals
with high impact factors and assessing nonpharmacological and pharmacological
treatments in patients with hip or knee osteoarthritis.
Main Outcome Measures The quality of the methods reported in the selected articles was assessed
by 2 independent reviewers using the Jadad scale, the Delphi list, and guidelines
found in the Users' Guides to the Medical Literature.
Investigators also used a checklist of items developed by the authors to analyze
Results A total of 110 articles were included in the analysis; 50 (45.5%) assessed
nonpharmacological treatments and 60 (54.5%) assessed pharmacological treatments.
Reports of nonpharmacological treatments had a lower global quality score
than did reports of pharmacological treatments as measured by the Jadad scale
(mean [SD] score, 1.4 [1.3] vs 3.0 [1.3]) and the Delphi list (mean [SD] score,
5.2 [1.5] vs 7.5 [1.1]). Lack of reporting adequate random sequence generation
and intention-to-treat analyses were found in both nonpharmacological and
pharmacological articles. Nonpharmacological treatments were less often compared
with a placebo than were pharmacological treatments (28.0% of articles vs
71.7%). Compared with pharmacological articles, nonpharmacological articles
less often described blinding of patients (26.0% vs 96.7%), care providers
(6.0% vs 81.7%), and outcome assessors (68.0% vs 98.3%). Care providers' skill
levels could influence treatment effect in 84.0% of nonpharmacological articles
vs 23.3% of pharmacological articles.
Conclusions In this analysis of reports of hip and knee osteoarthritis therapy,
nonpharmacological articles scored lower than pharmacological articles in
terms of quality. Assessments of nonpharmacological treatments must take into
consideration additional methodological issues.
Randomized controlled trials (RCTs) are widely accepted as the most
reliable method of determining the effectiveness of specific therapies.1 The design, conduct, and analysis of clinical trials
aim at providing valid results, which implies that the treatment effect reported
represents its true direction and magnitude2 and
that the trial minimizes bias, which can distort results.3,4
Hip and knee osteoarthritis are together a major cause of disability5 and can be treated either by nonpharmacological treatment,
such as surgery, rehabilitation, joint lavage, acupuncture, behavioral interventions,
or spa therapy, or pharmacological treatment, such as oral drug, intra-articular
injection, or topical treatments.
Randomized controlled trials have been developed essentially in the
context of pharmacological treatment, probably because of the pressure from
regulatory agencies on pharmaceutical companies to perform trials before releasing
a new drug.6 Assessing the effectiveness of
nonpharmacological treatment presents specific issues.6,7 Randomized
controlled trials are usually more easily conducted to assess pharmacological
treatment because investigators can standardize the dosage, measure compliance,
produce identical placebos, and blind patients, care providers, and outcome
assessors. In nonpharmacological treatment, it is often technically or ethically
difficult to perform a sham intervention, and the blinding of patients and
care providers is frequently impossible, whereas the placebo effect of nonpharmacological
treatment is probably important. For example, reports of RCTs assessing joint
lavage8,9 describe the effectiveness
of this treatment in knee osteoarthritis. However, these studies were performed
without a sham intervention in the control group, and patients were not blinded.
These results are inconsistent with those of another RCT evaluating joint
lavage that used a sham intervention in the control group and blinded patients
and outcome assessors.10 These conflicting
results could be linked to the choice of the control intervention or to the
results' variability. Moreover, contrary to pharmacological treatment, in
nonpharmacological treatment, care providers are an integral part of the treatment;
the success of the treatment depends on the care providers' skills, experience,
and enthusiasm. For example, hip and knee arthroplasty outcomes are well known
to depend on surgeons' experience and hospital surgical volume.11,12 Finally,
nonpharmacological treatment is usually complex and difficult to standardize,
and technical modification may occur as the procedure evolves. These methodological
issues are usually not taken into account in assessment of the quality of
articles evaluating nonpharmacological treatment.
The goals of this study were to compare the quality of reports of nonpharmacological
treatment and pharmacological treatment of hip and knee osteoarthritis and
to identify specific methodological issues in assessment of nonpharmacological
We searched MEDLINE and the Cochrane Central Register of Controlled
Trials using the search terms osteoarthritis OR osteoarthritic and hip OR knee, with a limitation to clinical trials. We identified and selected all reports of RCTs assessing nonpharmacological
treatment and pharmacological treatment in patients with hip or knee osteoarthritis
published between January 1, 1992, and February 28, 2002, in the following
journals based on impact factors reported in 2001: (1) the 10 highest-impact-factor
general and internal medicine journals (New England Journal
of Medicine, JAMA, The Lancet, Annals of Internal Medicine, Annual Review of Medicine, Archives of Internal Medicine, BMJ, American Journal of Medicine, Medicine, and Proceedings of the Association of American
Physicians); (2) the 6 highest-impact-factor rheumatologic journals
(Arthritis and Rheumatism, Seminars
in Arthritis and Rheumatism, Annals of the Rheumatic
Diseases, Rheumatology [Oxford, England], Journal
of Rheumatology, and Rheumatic Diseases Clinics of
North America); (3) the 6 highest-impact-factor orthopedic journals
(Osteoarthritis and Cartilage/OARS, Osteoarthritis Research
Society; Journal of Orthopaedic Research: Official Publication of the Orthopaedic
Research Society; Journal of Bone & Joint Surgery, American Volume; Spine; Gait & Posture; and Journal of Bone & Joint Surgery, British Volume); and (4) the 6
highest-impact-factor rehabilitation journals (Archives
of Physical Medicine and Rehabilitation, Supportive
Care in Cancer: Official Journal of the Multinational Association of Supportive
Care in Cancer, Journal of Electromyography and Kinesiology:
Official Journal of the International Society of Electrophysiological Kinesiology, Physical Therapy, Journal
of Rehabilitation Research and Development, and Scandinavian Journal of Rehabilitation Medicine).
We chose these journals because a high impact factor is a good predictor
of high methodological quality of journal articles13 and
because our goal was not to be exhaustive but, rather, to raise awareness
of methodological issues when assessing nonpharmacological treatment.
Retrieved articles were assessed by 1 of us (I.B.), who screened the
titles and abstracts to identify the relevant studies. Articles were included
only if the study was identified as an RCT, published as a full-text article,
and assessed nonpharmacological treatment or pharmacological treatment of
hip or knee osteoarthritis. Case series, uncontrolled studies, and articles
published as abstracts only, editorials, news, or correspondence sections
were excluded. Articles were screened for duplicate publication (ie, the same
trial with results from different lengths of follow-up published twice), and,
in these cases, only the more recent article was selected for inclusion.
Two independent reviewers (I.B. and F.T.) assessed the quality of the
methods in the selected articles using the Jadad scale14 and
the Delphi list15 because of their validity
and widespread use. They also used the Users' Guides to
the Medical Literature.16 However, to
our knowledge, no quality assessment tool specific to nonpharmacological treatment
is available. Therefore, the reviewers also assessed articles using a checklist
of items developed by the authors to target methodological issues when assessing
nonpharmacological treatment. (This checklist is available at http://www.bichat.inserm.fr/fichierpdf/emi0357/checklistofitems1.pdf.) With the checklist, data were obtained for year of publication, funding
sources (public or private), number of centers involved, type of treatment
assessed (oral drug administration, topical treatment, intra-articular injection,
surgery, arthroscopy, joint lavage, acupuncture, rehabilitation, or behavioral
intervention), and whether data in the CONSORT (Consolidated Standards of
Reporting Trials) diagram17 were reported in
a flowchart or in the text. Information about study design, randomization
mode, and appropriateness of allocation concealment was collected. Randomization
sequence generation was considered adequate if selection bias was prevented
by use of, for example, a table of random numbers, random numbers generated
by computer, coin tossing, or shuffling cards. Allocation was considered adequately
concealed if patients and the investigators who enrolled the patients could
not foresee assignments because of use of, for example, centralized randomization,
pharmacy control, opaque sealed envelopes, or numbered or coded bottles or
Reviewers examined whether treatment was individualized, which supposed
that the treatment's dosage or mode was modified according to patient tolerance
or comorbidity and to clinical efficacy. Reviewers evaluated whether, in their
opinion, the intervention was described with enough detail to be reproducible.
In situations in which care providers could influence the success of the treatment,
reviewers analyzed whether the learning curve and the care providers' experience
were taken into account and whether care providers were trained.
Compliance with repeated interventions (ie, treatment necessitating
iterative interventions, such as drug treatment or physiotherapy rather than
surgery) was evaluated. Quantitative and qualitative compliance were distinguished.
Quantitative compliance assessed, for example, whether a patient attended
all the physiotherapy sessions or took all of the oral drugs prescribed, and
qualitative compliance assessed whether a patient correctly performed the
intervention (eg, if the home-based exercises performed by a patient were
in accordance with the program prescribed). The method used to measure compliance
(eg, drug dosage, pill counts, patient questioning, or diaries) was recorded.
Reviewers examined the control intervention used and whether the potential
placebo effect of each intervention was similar, in their opinion. For example,
reviewers considered that the placebo effect of use of a nonsteroidal anti-inflammatory
drug and that of an indistinguishable placebo were similar. However, the potential
placebo effect of joint lavage, performed under aseptic conditions in an operating
theater with use of local anesthesia and infusion of 1 L of saline solution,
compared with a single intra-articular injection is probably different. Information
concerning the similarity of concomitant treatment in each group (unintended
additional care provided to either comparison group) and the occurrence or
likelihood of contamination (the intervention provided to the control group)
between the different treatment groups was also considered.
Reporting of blinding of patients, care providers, and outcome assessors
and whether blinding was tested were also studied. Blinding refers to keeping
participants, care providers, and outcome assessors unaware of the assigned
intervention so that they are not influenced by that knowledge.18
Reviewers examined whether a sample size justification or an evaluation
of the power of the study a posteriori was reported and whether an intention-to-treat
analysis19 was reported and/or performed. Finally,
we aimed at following the approach of the Users' Guides
to the Medical Literature16 to subjectively
appraise the validity of the study. According to these guides, the final assessment
of validity is never a "yes" or "no" decision but a continuum, ranging from
strong studies that avoid bias to weak studies that likely yield a biased
estimate of effect. For this purpose, reviewers gave a subjective evaluation
of the study's quality on a numerical rating scale ranging from 1 to 10 by
answering the question, "To what extent were systematic errors or bias avoided
in this report?" Since this global evaluation involves subjectivity, we cannot
exclude a bias against nonpharmacological treatment studies.
Before undertaking the study, the 2 reviewers practiced evaluation of
a distinct set of 10 articles. Then, during a meeting, they discussed the
interpretation of the different scales to resolve any differences in scoring.
During the study, each reviewer independently examined the selected articles
in a different computer-generated random sequence. Reviewers assessed the
title and the "Methods" and "Results" sections. Reviewers were not blinded
to the journal name and authors, as evidence concerning the effect of masking
on assessments of trial quality is inconsistent.20,21 Discrepancies
in the assessment of the selected articles between the 2 reviewers were resolved
by consensus. For each inconsistent item, reviewers read the article again
and came to an agreement. The data presented herein resulted from this consensus.
The quality of the selected articles was determined by use of the mean of
the 2 assessors' global appreciation of the articles' quality on a numerical
rating scale ranging from 1 to 10, the Jadad scale (score, 0-5), and the Delphi
list's overall quality score, consisting of the number of items satisfied
and ranging from 0 to 9.22 On all scales, a
high score indicates high quality.
Descriptive statistics (means, SDs, and minimum and maximum values)
were used for continuous variables. Categorical variables were described with
frequencies and percentages. The degree of agreement between the 2 reviewers
was determined with use of the κ coefficient for categorical variables.
Interrater reliability was assessed by use of the intraclass correlation coefficient
(ICC) for continuous variables. All data analyses were performed using SAS
version 8.2 (SAS Institute Inc, Cary, NC).
Of 198 articles identified, 119 were selected for assessment (Figure 1). The 79 excluded articles were
abstracts only (n = 5), were duplicate publications (n = 3), were not RCTs
(n = 12), did not assess a therapeutic intervention (n = 45), did not assess
treatment of hip or knee osteoarthritis (n = 13), or were phase 2 trials (n
= 1). Nine articles were secondarily excluded after obtaining the full text
because they were not RCTs (n = 3) or because they were subgroup analyses
(n = 1) or extended follow-ups of RCTs described in other articles (n = 5).
One hundred ten articles were included in the analysis. Fifty articles (45.5%)
assessed nonpharmacological treatment: surgery (n = 17), arthroscopic lavage
(n = 3), joint lavage (n = 3), rehabilitation (n = 23), education (n = 2),
spa therapy (n = 1), and acupuncture (n = 1). Rehabilitation interventions
were related to physiotherapy (n = 14), technical devices (n = 5), and transcutaneous
electrical nerve stimulation or laser therapy (n = 4). A total of 60 articles
(54.5%) assessed pharmacological treatment: oral drug administration (n =
41); topical (n = 3), intramuscular (n = 1), and intravenous (n = 1) treatments;
and intra-articular injection (n = 14).
Interrater reliability was good for random sequence generation (agreement,
92.7%; κ = 0.86; 95% confidence interval [CI], 0.77-0.95), allocation
concealment (agreement, 89.1%; κ = 0.79; 95% CI, 0.67-0.90), patient
blinding (agreement, 96.4%; κ = 0.92; 95% CI, 0.85-1.00), care provider
blinding (agreement, 96.4%; κ = 0.93; 95% CI, 0.86-1.00), and outcome
assessor blinding (agreement, 97.3%; κ = 0.90; 95% CI, 0.78-1.00). Determining
whether the study was performed according to an intention-to-treat analysis
resulted in a lower κ value (agreement, 70.9%; κ = 0.59; 95% CI,
0.48-0.71). Interrater reliability assessed by use of the ICC was good for
the Jadad scale (ICC, 0.84; 95% CI, 0.78-0.89), the Delphi list (ICC, 0.88;
95% CI, 0.83-0.92), and the numerical rating scale (ICC, 0.62; 95% CI, 0.49-0.72).
Only 10 articles (9.1%) were published in a general medical journal
(JAMA, The Lancet, BMJ, Annals of Internal Medicine, and Archives
of Internal Medicine). The other articles were published mainly in
the Journal of Rheumatology (20.9%), Osteoarthritis and Cartilage (16.4%), Arthritis & Rheumatism (11.8%), Journal of Bone &
Joint Surgery, British Volume (11.8%), Rheumatology (10.0%), and Annals of Rheumatic Disease (9.1%).
Financial support was totally or partially private in 57 articles (51.8%),
public in 26 articles (23.6%), and not reported in 27 articles (24.6%). Pharmacological
treatment funds were mainly private (75%) or not reported (23.3%), whereas
in nonpharmacological treatment, funds were provided by public support in
25 articles (50.0%) and private support in 12 articles (24.0%) and were not
reported in 13 articles (26.0%). Half of the articles concerned multicenter
trials. Multicenter trials were reported more often in pharmacological treatment
than in nonpharmacological treatment articles (68.3% vs 30.0%).
Items from the CONSORT diagram (flow of participants through each stage
of the trial) were reported in the text or in a flowchart in 48 of the 75
articles published since the CONSORT statement was published in 1996.23 These data were reported in 71.4% (30/42) of the
pharmacological treatment articles and 54.5% (18/33) of the nonpharmacological
treatment articles and in only 9.1% (1/11) of the surgical articles.
Whatever tool was used for assessment, the quality scores were better
for articles on pharmacological treatment than those on nonpharmacological
treatment (Figure 2). On the Jadad
scale, pharmacological treatment articles had a mean (SD) score of 3.0 (1.3)
vs 1.4 (1.3) for nonpharmacological treatment articles. Lack of blinding in
nonpharmacological treatment articles explained most of this difference. On
the Delphi list, pharmacological treatment articles had a mean (SD) score
of 7.5 (1.1) vs 5.2 (1.5) for nonpharmacological treatment articles. On the
numerical rating scale, pharmacological treatment articles had a mean (SD)
score of 7.0 (1.7) vs 4.9 (2.0) for nonpharmacological treatment articles.
Moreover, as shown in Figure 2,
reports of surgery/arthroscopy/lavage had the lowest quality scores, those
of rehabilitation and intra-articular injection had similarly low quality
scores, and other pharmacological treatment articles had the highest scores.
More information is available at http://www.bichat.inserm.fr/fichierpdf/emi0357/dataonarticlesassessed.pdf.
Study Design. The studies were all of parallel-group
(n = 104) or crossover (n = 6) design. All of the selected articles involved
randomization of patients; however, generation of randomization sequence was
adequate in only 49.1% of the articles and was concealed from the investigators
who enrolled the patients in only 20.9% (Table 1). There were no differences between nonpharmacological treatment
and pharmacological treatment articles in the adequacy of random sequence
generation. Adequate allocation concealment was similarly reported in nonpharmacological
treatment and pharmacological treatment articles. However, allocation concealment
was more often inadequate in nonpharmacological treatment articles than in
pharmacological treatment articles, whereas allocation concealment was more
often not reported in pharmacological treatment articles than in nonpharmacological
treatment articles (Table 1).
Description of Interventions.Reproducibility and Individualization. The intervention was more often
described with enough detail to be reproducible in pharmacological treatment
than in nonpharmacological treatment articles (Table 1). Among nonpharmacological treatment articles, surgical
treatments were less often considered reproducible (Table 2). The intervention was almost never individualized in pharmacological
treatment articles, whereas one third of the nonpharmacological treatment
articles described individualized interventions (Table 1). Finally, the technical quality of the nonpharmacological
treatment was never evaluated in nonpharmacological treatment articles.
Care Providers. Care provider skill level and
experience could influence the treatment effect, including all surgical interventions,
in most of the nonpharmacological treatment articles but only in pharmacological
treatment articles assessing intra-articular injection (Table 1 and Table 2).
In contrast, care provider experience was reported in only 6 articles, hospital
volume was reported in only 1 study, and the learning curve of care providers
was never taken into account. Finally, care provider training before the beginning
of the trial was mentioned in only 2 articles, one assessing intra-articular
injection and the other assessing rehabilitation.
Compliance. We evaluated compliance with only
repeated interventions (ie, treatment necessitating iterative interventions,
such as drug treatment or physiotherapy), which comprised 57 (95.0%) of the
pharmacological treatment articles and 26 (52.0%) of the nonpharmacological
treatment articles. Among these articles, reporting of compliance was similar
between pharmacological treatment articles (32 [56.1%] of 57) and nonpharmacological
treatment articles (16 [61.5%] of 26). However, qualitative compliance was
never evaluated in these articles. Compliance was assessed with at least 1
objective criterion in all pharmacological treatment articles (pill counts
or drug dosage) vs in 7 (43.7%) of 16 nonpharmacological treatment articles
(number of sessions attended for physiotherapy or a timer recording hours
of use for transcutaneous electrical nerve stimulation).
Control Intervention. The control intervention
was more often reported to be a placebo in at least 1 group in pharmacological
treatment articles, whereas in nonpharmacological treatment articles, experimental
treatments were more often compared with active control treatments or with
usual care or waiting lists (Table 1).
Surgical interventions were always compared with an active control intervention
and never with a placebo (Table 2).
The potential placebo effect of the different treatments being compared was
considered to be similar more often in pharmacological treatment than in nonpharmacological
treatment articles (Table 1).
Concomitant Treatments. The description of
concomitant treatments (ie, additional care outside of the intervention provided
to either comparison group) was reported more often in pharmacological treatment
articles (58.3%) than in nonpharmacological treatment articles (24.0%). Contamination
between groups was reported in only 3 articles.
Blinding. Patients were almost always reported
to be blinded in pharmacological treatment articles, but only about one quarter
of the nonpharmacological treatment articles described blinding (Table 1). Care providers were reported
to be blinded in 81.7% of the pharmacological treatment articles but were
rarely blinded in nonpharmacological treatment articles (Table 1). When patients and care providers were reported to be blinded,
care provider blinding was never tested, and patient blinding was tested in
only 1 pharmacological treatment study and 1 nonpharmacological treatment
study. When patients were not blinded, in only 2 nonpharmacological treatment
articles were they instructed not to inform the outcome assessor about the
treatment they received. Finally, outcome assessors were less often blinded
in nonpharmacological treatment articles than in pharmacological treatment
articles (Table 1).
Outcome Assessment. In only 3 nonpharmacological
treatment and 4 pharmacological treatment articles were outcome assessors
trained, and outcomes were never reported to be assessed by an end-point review
committee (an independent committee that ultimately decides whether a participant
meets the criteria for a study's outcome). Occurrence of adverse effects was
reported more often in pharmacological treatment articles than in nonpharmacological
treatment articles (83.3% vs 46.0%). Adverse effects were assessed by an independent
committee in only 2 articles.
Intention-to-Treat Analysis and Sample Size Justification. Statistical analysis was performed according to an intention-to-treat
principle24 (ie, all randomized participants
were included in the analysis and kept in their original group) in only 30%
of all articles (Table 1). A sample
size justification or an estimation of the power of the study a posteriori
was reported more often in pharmacological treatment articles than in nonpharmacological
treatment articles and was especially rare in surgical articles.
We analyzed the "journal effect" and found that articles published in
general and internal medicine journals (n = 10) and rheumatologic journals
(n = 57) had a higher quality score than articles published in orthopedic
journals (n = 36) and rehabilitation journals (n = 7) according to the Jadad
scale (mean [SD] score of 2.8 [1.3] vs 1.5 [1.4]), the Delphi list (7.0 [1.2]
vs 5.5 [1.9]), and the numerical rating scale (6.7 [1.8] vs 5.1 [2.1]).
This study assessed the methodological quality of all RCTs published
on the topic of hip and knee osteoarthritis during a 10-year period in high-impact
general medical and specialty journals.
Several studies have assessed the methodological quality of a broad
range of reports of randomized trials in several areas of health care. Whatever
the domain assessed, the overall quality of published RCTs is poor. Methodological
problems concerning randomization and intention-to-treat analysis are common.1,25,26 Moreover, results
of these studies showed that such methodological deficiencies could influence
the effect size. Inadequate random sequence generation and lack of allocation
concealment and double-blinding yielded larger treatment effects.3,24,27 However, to our knowledge,
no study has compared the methodological quality of nonpharmacological treatment
and pharmacological treatment articles. We focused on the methodological quality
of articles of RCTs assessing hip and knee osteoarthritis treatments because
these treatments cover a wide range of nonpharmacological treatment (eg, surgery,
arthroscopy, joint lavage, exercise therapy, physiotherapy, orthosis, spa
therapy, acupuncture, and education) and pharmacological treatment (eg, oral
drug administration, intra-articular injection).
Our analysis showed that the methodological quality of reports of nonpharmacological
treatment trials was lower than that of reports of pharmacological treatment
trials. Methodological deficiencies such as lack of reporting adequate random
sequence generation, adequate allocation concealment, and intention-to-treat
analysis were common in both nonpharmacological treatment and pharmacological
treatment articles.19 These deficiencies could
be reduced easily in the trials themselves (through changes in conduct when
necessary) and in the reporting of the trials. However, some specific issues,
such as choice of control intervention, lack of double-blinding, care provider
effect, and standardization problems, could be difficult to resolve when assessing
When assessing nonpharmacological treatment, a placebo or sham intervention
can be difficult or impossible to perform for ethical or technical reasons.
Ethical concerns are substantial in trials assessing surgical procedures when
sham interventions are prohibitive.6,28 In
the trials investigated herein, surgery was always compared with another surgical
procedure and never with a placebo. Recently, Moseley et al29 pointed
out the difficulties of performing placebo-controlled trials of surgery when
evaluating arthroscopic intervention for knee osteoarthritis. In that trial,
the control group underwent simulated arthroscopic surgery, where small incisions
were made but no instruments were inserted. Practical issues are also important,
and implementing a sham intervention often requires creative solutions. For
example, to perform double-blind placebo-controlled trials assessing acupuncture,
Streitberger and Kleinhenz30 developed a placebo
acupuncture needle that gave patients the feeling of penetration but did not
penetrate the skin. Thus, it appears important to develop research in this
Blinding patients and care providers is usually possible in drug trials
with a matching placebo but is often impossible to perform in nonpharmacological
treatment trials. Although some investigators have proposed solutions such
as use of standardized wound dressings31 when
assessing laparoscopic cholecystectomy, surgeons usually know which intervention
has been done and patients usually know which rehabilitation program they
To avoid these biases, efforts could be made to blind outcome assessors.
Use of the Prospective Randomized Open Blinded End-point (PROBE) study design,
which is based on blinded end points, could be an alternative.32 However,
treatments of hip or knee osteoarthritis are usually assessed from patient-reported
symptoms, and if patients are not blinded, outcome assessors cannot be blinded.
Results of several studies that did not involve double-blinding yielded exaggerated
estimates of treatment effects,3,33 while
results of other studies showed no effect.27 Heterogeneity
in who was blinded and in the outcomes assessed may be responsible for these
discrepancies. Blinding is particularly important when outcome measures involve
patient-reported symptoms such as pain but is less important for objective
criteria such as death because of little detection bias.
Nonpharmacological treatments are usually complex; that is, they include
several components and/or several health care professionals and are often
individualized. The active component of such interventions is therefore difficult
to identify, and the treatments are difficult to replicate. A detailed standardization
of the intervention is necessary, and the technical quality of the intervention
should be evaluated.34 Contrary to pharmacological
treatment, in which the effect of health care professionals can generally
be regarded as secondary, in nonpharmacological treatment the health care
professional is an integral part of the intervention. The success of the intervention
depends on care provider skill, experience, and training. Variation in care
provider skills in each arm of a trial can be confounded with the treatment
effect.35 Most nonpharmacological treatment,
especially surgery, involves complex procedures, and quality in performance
requires frequent repetitions.36 No trial evaluated
in our study assessed the learning curve, and only 6 articles described care
Among nonpharmacological treatment articles, surgical articles had the
lowest-quality scores. Some methodological deficiencies, such as low reporting
of data in the CONSORT diagram, adequate allocation concealment, and detailed
description of the intervention, should not be more difficult to resolve than
in other nonpharmacological treatment trials. However, surgical trials also
present specific issues. In these trials, care providers can always influence
the treatment effect and are never blinded. In the articles analyzed, surgical
interventions were never compared with a placebo, probably for ethical reasons.
For example, patients included in trials assessing surgical interventions
that are irreversible will not have an opportunity to benefit from results
of the trial.
Our study was limited because we assessed only the reports of RCTs,
not the trials themselves. However, failure to report the methods of a trial
does not necessarily mean that investigators did not carry out these methods.
Some methodological deficiencies may lie in the reporting of trials rather
than in their performance.37
This study allowed us to point out the difficulties in assessing nonpharmacological
treatment. Some improvement could occur with use of adequate random allocation
generation and concealment and intention-to-treat analyses. However, methodological
issues concerning choice of a control intervention; blinding of patients,
care providers, and outcome assessors; standardization of interventions; and
taking into account care provider skill are more difficult to resolve. Finally,
tools used to measure quality, such as the Jadad scale and the Delphi list,
are probably not appropriate for assessing the quality of nonpharmacological
treatment articles because they focus on randomization, blinding, and intention-to-treat
analysis and do not take into account additional methodological issues, such
as learning curves, reproducibility, and quality assessment of the intervention.
Because nonpharmacological treatment represents a wide range of treatments
available to patients, it is important to enhance research in this domain
and to develop specific tools appropriate but no less rigorous for assessing
nonpharmacological treatment trials. Moreover, despite the challenges posed
by nonpharmacological treatment studies, the same expectations of quality
should be applied to nonpharmacological treatment trials as are applied to
pharmacological treatment trials.