Objective To develop evidence-based recommendations on the use of imaging of the joints in the clinical management of rheumatoid arthritis (RA).
Methods The task force comprised an expert group of rheumatologists, radiologists, methodologists and experienced rheumatology practitioners from 13 countries. Thirteen key questions on the role of imaging in RA were generated using a process of discussion and consensus. Imaging modalities included were conventional radiography, ultrasound, MRI, CT, dual-emission x-ray absorptiometry, digital x-ray radiogrammetry, scintigraphy and positron emission tomography. Research evidence was searched systematically for each question using MEDLINE, EMBASE and Cochrane CENTRAL. The experts used the evidence obtained from the relevant studies to develop a set of 10 recommendations. The strength of recommendation was assessed using a visual analogue scale.
Results A total of 6888 references was identified from the search process, from which 199 studies were included in the systematic review. Ten recommendations were produced encompassing the role of imaging in making a diagnosis of RA, detecting inflammation and damage, predicting outcome and response to treatment, monitoring disease activity, progression and remission. The strength of recommendation for each proposition varied according to both the research evidence and expert opinion.
Conclusions Ten key recommendations for the role of imaging in the management of RA were developed using research-based evidence and expert opinion.
Statistics from Altmetric.com
Structural damage in rheumatoid arthritis (RA) can occur early in the disease. Prompt treatment has been shown to reduce inflammation thereby limiting structural damage.1 ,2 Although conventional radiography (CR) has been considered the gold standard for imaging in RA, its sensitivity for structural damage in RA diagnosis is low, and disease activity cannot be assessed.3 Significant advances have been made within the field of imaging in rheumatic diseases over the past decade.4
A European League Against Rheumatism (EULAR) task force was therefore convened to develop evidence-based recommendations on the use of imaging of the joints in the clinical management of RA.
An expert group of rheumatologists, radiologists, methodologists and experienced rheumatology practitioners (19 people, representing 13 countries) participated in the study. The objectives were to formulate key clinical questions relating to the role of imaging in RA, to identify and critically appraise the available evidence, and to generate recommendations based on both evidence and expert opinion.
At the initial task force meeting, members contributed clinically relevant questions related to key aspects of the use of imaging in RA. The research questions were agreed by consensus and 13 final research questions were selected, which encompassed the role of imaging in making a diagnosis of RA, detecting inflammation and damage, predicting outcome and response to treatment, monitoring disease progression, and remission (see supplementary material, S1. Research questions, available online only).
A systematic search of articles was performed and the bibliographies of included papers were hand searched for evidence of other studies for inclusion. Specific medical subject headings and additional keywords were used to identify all relevant studies (see supplementary material, S2. Search strategy, available online only).
Titles and abstracts of all citations identified were screened, and potentially relevant articles were reviewed in full text using predetermined inclusion and exclusion criteria. Studies, published in English, on the use of imaging in adults (≥18 years of age) with a clinical diagnosis of RA were included. Imaging modalities included were CR, ultrasound, MRI, CT, dual-emission x-ray absorptiometry (DXA), digital x-ray radiogrammetry (DXR), scintigraphy and positron emission tomography (PET). Study types included randomised controlled trials, systematic reviews, controlled clinical trials, cohort, case–control and diagnostic studies. Studies were considered for inclusion when they provided information on the role of imaging in making a diagnosis of RA, detecting inflammation and damage, predicting outcome and response to treatment, monitoring disease progression and remission.
Following presentation of the data from the literature review, the experts produced 10 recommendations based on the 13 clinical questions with final agreement by a process of discussion and consensus. The experts scored the perceived strength of recommendation (SOR) for each proposition using a 0–10 visual analogue scale (VAS; 0=not recommended at all, 10=fully recommended). Scores reflected both research evidence and clinical expertise.5
Evidence was categorised according to study design using a hierarchy of evidence in descending order according to quality.6 Greater emphasis was given to the best available evidence when answering questions, although all data were collected and reviewed.
Recommendations for future research were agreed by consensus following presentation of the literature review.
The search of databases (performed in June 2011) resulted in 6888 records, of which 2567 were duplicates. Of the remaining 4321 articles, 3975 were excluded based on title or abstract, leaving 346 articles for detailed review. All full text articles written in English were retrieved for review; 175 articles were excluded after reviewing the full text leaving 171 articles for inclusion (see supplementary figure S3, available online only). The hand search identified 28 additional articles for inclusion, resulting in a total of 199 articles for inclusion. Articles that were relevant to more than one research question were included in the review more than once. The number of articles included in each question is shown in supplementary table S4 (available online only).
Ten recommendations were produced, and the final wording of the propositions was adjusted using e-mail exchange and at the closing meeting of the group. The recommendations, SOR (mean VAS and 95% CI) and level of evidence are presented in table 1.5 A full reference list for articles included in each recommendation is given in the supplementary material, S5 (available online only).
Making a diagnosis of RA (in patients with at least one joint with definite clinical synovitis):
Recommendation 1: When there is diagnostic doubt, CR, ultrasound or MRI can be used to improve the certainty of a diagnosis of RA above clinical criteria alone.
Strength of recommendation: 9.1 (95% CI 8.6 to 9.6)
Five observational studies described the impact of imaging on confirming a diagnosis of RA when the diagnosis could not be confirmed using conventional methods, two with ultrasound and three with MRI. Three of these studies examined the hand joints (wrist, metacarpophalangeal and proximal interphalangeal joints), but none compared sites.7–11 One study showed that ultrasound synovitis improved the certainty of RA diagnosis from 42.0% to 53.2% (p 0.17),7 and another described how synovitis seen with ultrasound helped confirm (65.2%) or change the diagnosis (11.1%); ultrasound was superior to clinical examination in 75% of patients.8 Compared to clinical classification criteria, the demonstration of MRI synovitis increased the diagnosis of RA,9 ,10 and was more valuable than anti-cyclic citrullinated peptide antibody (ACPA) determination in the absence of rheumatoid factor (RF).11
Recommendation 2: The presence of inflammation seen with ultrasound or MRI can be used to predict the progression to clinical RA from undifferentiated inflammatory arthritis.
Strength of recommendation: 7.9 (95% CI 6.7 to 9.0)
Several studies assessed the prognostic value of imaging in patients with undifferentiated inflammatory arthritis (UIA), mainly using ultrasound or MRI. A recent systematic review identified 11 studies relating to MRI.12 The presence of bone oedema or both synovitis and erosion on MRI increased the likelihood of developing RA (positive likelihood ratio 4.5 and 4.8, respectively), whereas the absence of MRI synovitis decreased the probability of progression to RA (negative likelihood ratio 0.2). A prediction model including clinical hand arthritis, morning stiffness, positivity for RF and bone oedema on MRI correctly predicted progression to RA in 82% of UIA patients.13 MRI flexor tenosynovitis has also been described as a predictor of early RA (sensitivity 0.60, specificity 0.73).14 Of the three strongest predictors of RA (MRI flexor tenosynovitis, RF and ACPA), ACPA was found to be the strongest predictor (OR 13.8) and flexor tenosynovitis the weakest (OR 5.0), but its additional value in diagnosing RA was significant.
In a longitudinal study ultrasound significantly increased the detection of joint involvement in all joint regions. When combined with the Leiden prediction rule,15 power Doppler counts significantly improved area under the curve (AUC) values for the prediction of progression to RA (0.905 to 0.962).16 Salaffi et al17 described the likelihood of progression of UIA to RA using the presence of power Doppler on ultrasound (scores higher than grade 1), with OR 9.9 if one joint was involved, and 48.7 if more than three were involved; OR with high titre ACPA or RF was 10.9.
Detecting inflammation and damage:
Recommendation 3: Ultrasound and MRI are superior to clinical examination in the detection of joint inflammation; these techniques should be considered for a more accurate assessment of inflammation.
Strength of recommendation: 8.7 (95% CI 7.8 to 9.7)
This recommendation examines the added benefit of assessing joint inflammation by imaging over clinical examination. Sensitivity and specificity were initially extracted from the data; however, as clinical examination was used as the reference these results are difficult to use clinically. To overcome this we recorded detection rates; for example, how many times more (>onefold) or less (<onefold) does imaging detect inflammation over clinical examination. Our chosen approach may increase the number of false positive results.
We identified 51 studies comparing imaging and clinical examination in the detection of inflammation in various joints; 29 with ultrasound,18–36 16 with MRI,21 ,26 ,29 ,30 ,37–44 14 with scintigraphy,41 ,45–4,7 and two with PET (table 2). In general, ultrasound and MRI detected joint inflammation more frequently than clinical examination; the mean detection rate for synovitis at the hand and wrist was 2.18-fold for ultrasound and 2.20-fold for MRI.30 Using scintigraphy and PET were found to provide little benefit over clinical examination.
Recommendation 4: CR of the hands and feet should be used as the initial imaging technique to detect damage. However, ultrasound and/or MRI should be considered if CR do not show damage and may be used to detect damage at an earlier time point (especially in early RA).
Strength of recommendation: 9.0 (95% CI 8.4 to 9.6)
Three studies compared tissue damage (erosions or loss of joint space) detected by imaging with abnormal clinical examination. Caution is needed when interpreting these studies as bony involvement shown on imaging was compared with clinical signs of inflammation as reference.
Prognosis in RA: predicting outcome:
Recommendation 5: MRI bone oedema is a strong independent predictor of subsequent radiographic progression in early RA and should be considered for use as a prognostic indicator. Joint inflammation (synovitis) detected by MRI or ultrasound as well as joint damage detected by CR, MRI or ultrasound can also be considered for the prediction of further joint damage.
Strength of recommendation: 8.4 (95% CI 7.7 to 9.2)
Forty-eight longitudinal studies described how baseline changes in imaging predicted outcome, in particular erosive progression; 26 with MRI, 11 with ultrasound, 19 with CR, seven with DXA or DXR and three with scintigraphy. Of these, 46 studies examined the hands and 14 also included the feet; none compared the benefit of imaging different joints.
Bone marrow oedema
Of baseline MRI features, bone marrow (BM) oedema was a strong, independent predictor of erosive progression. Hetland et al48 ,49 have provided compelling data supporting this association; baseline MRI BM oedema was the only independent predictor of radiographic change in their 2 and 5-year follow-up studies (coefficient 0.75, p<0.001; and coefficient 0.82, p<0.001, respectively). Haavardsholm et al50 also identified baseline MRI BM oedema (score >2 RAMRIS units) as an independent predictor of radiographic (OR 2.77, 95% CI 1.06 to 7.21) as well as MRI erosive progression (unstandardised β, B 0.21, 95% CI 0.08 to 0.34). This is supported by McQueen et al51 who described BM oedema to be predictive of MRI erosive progression, OR 6.47, p<0.001. This study also demonstrated that the development of radiological erosions at 1 year was highly unlikely in the absence of baseline MRI inflammatory changes (negative predictive value 0.92). Patients with erosive progression on CT also have higher baseline MRI BM oedema scores (relative risk (RR) of CT progression 3.8, 95% CI 1.5 to 9.3).52
Baseline synovitis, detected by MRI or ultrasound, is a predictor of erosive progression. Dohn et al52 reported the RR of CT erosive progression with baseline ultrasound grey-scale synovitis as 11.2, 95% CI 0.65 to 195.7, p 0.1, baseline ultrasound power Doppler activity RR 7.6, 95% CI 0.91 to 63.2, p 0.061, and baseline MRI synovitis RR 0.68, 95% CI 0.04 to 11.5, p 0.79. 52 The predictive value of baseline ultrasound grey-scale synovitis for MRI erosive progression performed better than MRI synovitis with positive likelihood ratios of 1.75 and 1.47, respectively, and accuracy of 70% and 62%, respectively.53 Conaghan et al54 described a close correlation between the degree of MRI synovitis and the number of new erosions, with the AUC for MRI synovitis the only significant predictor of erosive progression (AUC for MRI synovitis r 0.420, p<0.007).
Baseline tenosynovitis on ultrasound appears to be predictive of erosive progression at 1 year (OR 7.18) and 3 years (OR 3.4).55 This effect has not been seen with MRI tenosynovitis,56 but baseline MRI tendinopathy has been shown to be predictive of tendon rupture at 1 year (OR 1.57, p 0.02) and 6 years (OR 1.52, p 0.03).57
Baseline erosions detected by various imaging techniques appear to be predictive of further erosions at 6 months; MRI erosions (β 0.63, p<0.001), radiographic erosions (β 0.68, p 0.04), with ultrasound erosions less significant (β 0.57, p 0.07).58 Several studies have reported that baseline MRI erosions are predictive of erosive progression;59 ,60–62 and the absence of baseline MRI erosions predicts that radiographic or MRI erosions are unlikely (negative predictive value 1.0).61 Baseline radiographic erosions independently predict further radiographic progression (at 3 years, OR 8.47; at 10 years, OR 5.64–18.1).63–65 In addition, the baseline Larsen score is shown to predict an annual radiological progression rate greater than the median (OR 2.6, 95% CI 1.3 to 5.3).65
Digital x-ray radiogrammetry/dual-emission x-ray absorptiometry
Early hand bone loss measured by change in estimated bone mineral density in the first year of disease by DXR appears to be an independent predictor of erosive progression, even up to 20 years.53 ,66 ,67 Baseline femoral neck osteopenia or osteoporosis are also predictive of radiographic erosive progression.68
Baseline inflammatory disease measured by scintigraphy appears to be associated with radiographic progression.69 In addition, multiple regression analysis has demonstrated that progression of radiographic joint destruction was primarily predicted by 99mTc-IgG scintigraphy; joint swelling, ESR and IgM RF were not predictive. This suggests that scintigraphy may be superior to conventional clinical and laboratory measurements in the prediction of joint destruction.70 However, when comparing scintigraphy to other baseline imaging predictors of progression, baseline MRI BM oedema score (Spearman's correlation, r 0.67), MRI synovitis score (r 0.57), and 99mTc-NC scintigraphy uptake (r 0.45) were predictive of change in MRI erosion score from baseline to 2 years. In the multivariate analysis, the BM oedema score was the only baseline variable that predicted erosive progression (OR 4.2, 95% CI 1.3 to 13.8).71
Prognosis in RA: Predicting response to treatment:
Recommendation 6: Inflammation seen on imaging may be more predictive of a therapeutic response than clinical features of disease activity; imaging may be used to predict response to treatment.
Strength of recommendation: 7.8 (95% CI 6.7 to 8.8)
Two prospective cohort studies have assessed the use of clinical measures and imaging to predict response to anti-tumour necrosis factor (TNF) therapy. Ellegaard et al72 measured ultrasound Doppler activity and clinical parameters at baseline to predict which patients would benefit from treatment, assessed by treatment persistence at 1 year. They identified ultrasound Doppler activity to be the only baseline parameter to predict treatment persistence (p 0.024); baseline clinical measures including tender and swollen joint counts, C-reactive protein, 28-joint disease activity score (DAS28) and health assessment questionnaire showed no significant association. Elzinga et a73l used changes in PET uptake 2 weeks after treatment to predict future treatment response, according to DAS28. A significant correlation was seen between the changes in PET activity at 2 weeks and DAS28 at 14 and 22 weeks after treatment (r 0.62, p<0.05; r 0.65, p<0.01 respectively).
Monitoring disease progression:
Recommendation 7: Given the improved detection of inflammation by ultrasound and MRI than by clinical examination, they may be useful in monitoring disease activity.
Strength of recommendation: 8.3 (95% CI 7.4 to 9.1)
No published data were identified that specifically addressed how imaging should be used to monitor RA disease activity. In the absence of this information, data were extracted on each factor separately.
Comparison of the ability of imaging to detect inflammation
Several studies compared ultrasound and MRI in the detection of joint inflammation, with MRI considered the reference technique. There seems to be significant association between these modalities,23 ,24 but aside from access to imaging, there may be advantages to using each technique in certain situations. For example, ultrasound has been shown to detect more joint and tendon sheath effusions than MRI,58 whereas MRI appears to be more sensitive in identifying tenosynovitis.74 Comparisons of conventional high-field MRI with dedicated, low-field extremity MRI have shown high agreement for synovitis, with lower agreement for BM oedema and tenosynovitis detected by low-field MRI, with high-field MRI as reference.75 ,76 Low-field MRI without contrast also demonstrates poor sensitivity in the detection of synovitis, compared with power Doppler ultrasound.77 Only one study compared scintigraphy with more modern imaging techniques, and showed strong correlation between uptake on scintigraphy and inflammatory changes seen on MRI.78
Responsiveness to change in inflammation
Ultrasound and MRI appear to show good responsiveness to change. A study of responsiveness of MRI and ultrasound to change in inflammation with treatment has shown that MRI synovitis (standardised response mean (SRM) −0.79 to −0.92), MRI tenosynovitis (SRM −0.70 to −1.02) and BM oedema (SRM −1.05–−1.24) were responsive to change, but ultrasound inflammation (synovitis, tenosynovitis and effusion) was less responsive (SRM −0.37–−0.54).26 A study by Haavardsholm et al79 reported MRI to have a higher potential to detect change in wrist BM oedema than in synovitis over 1 year. The smallest detectable difference for a range of ultrasound measures including power Doppler was low in a large 1-year observational multiple-reader study of RA patients treated with anti-TNF agents, demonstrating both the reliability of this measure and the ability to detect individual-level important change. At the group level, there were significant changes in all ultrasound synovial assessments in parallel with DAS28 changes.80 When comparing the changes in power Doppler and grey-scale ultrasound activity with response to treatment, grey-scale ultrasound appears to perform better,81 as does the addition of contrast enhancement.82
Which joints to assess
Only one study directly compared the assessment of inflammation by imaging different areas; Calisir et al40 described MRI synovitis and BM oedema in the hands and feet of patients with early RA and found no significant difference in MRI inflammation in these regions.
Recommendation 8: The periodic evaluation of joint damage, usually by radiographs of the hands and feet, should be considered. MRI (and possibly ultrasound) is more responsive to change in joint damage and can be used to monitor disease progression.
Strength of recommendation: 7.8 (95% CI 6.8 to 8.9)
As for the previous recommendation, there were no specific data on the recommended frequency of imaging in the assessment of progressive joint damage.
Comparison of the ability of imaging to detect damage
Dohn et al52 performed comparison studies of the ability of CR, CT, ultrasound and MRI to detect erosive damage.53 ,83 With CT as the reference technique, CR was shown to have an accuracy of 81%, MRI of 89% and ultrasound of 80%, with high specificities and lowest sensitivity for CR.52 ,83 A previous systematic review has described ultrasound to be more effective for erosion detection than CR, with comparable efficacy to MRI.84 A summary of data comparing the different imaging modalities in the detection of erosions is given in table 3.24 ,26 ,29 ,39 ,43 ,52 ,58 ,75 ,76 ,83 ,85–102
Studies assessing tendon damage have shown ultrasound to be more sensitive than MRI in the detection of finger extensor tendon tears later confirmed at surgery;103 and moderate agreement between ultrasound and MRI (used as the reference technique) in the assessment of shoulder tendon involvement.104
Responsiveness to change in damage
CR is the standard imaging technique used to detect and monitor joint damage. There are some data suggesting that CR is responsive to change in erosions on an individual level, particularly after the first 12 months of disease.26 Radiographic progression appears to be most rapid in the first 2 years of disease, with 75% of all damage seen in the first 5 years of a 10-year study.105 MRI seems to be more responsive to change at earlier time points, but measures of annual progression rates are similar with MRI and CR.26 This is supported by Østergaard et al,106 who found that 78% of new radiographic bone erosions were seen at least 1 year earlier by MRI, in fact MRI detection of new erosions preceded CR by a median of 2 years.
Which joints to assess
Early erosive changes on CR appear to be more common in the feet than in the hands, but from year 3 onwards these areas are more equally affected.105 ,107
Recommendation 9: Monitoring of functional instability of the cervical spine by lateral radiograph obtained in flexion and neutral should be performed in patients with clinical suspicion of cervical involvement. When the radiograph is positive or specific neurological symptoms and signs are present, MRI should be performed.
Strength of recommendation: 9.4 (95% CI 8.9 to 9.8)
Thirteen studies described the assessment of cervical spine involvement in RA, summarised in table 4.108–120 No studies explored the appropriate frequency for monitoring change in the cervical spine; Yurube et al118 investigated baseline features on CR predictive of future cervical instability and found that patients with baseline deforming hand changes, cervical vertical subluxation (VS), and subaxial subluxation showed more progression in VS and subaxial subluxation at 5 years, and Reijnierse et al120 identified that baseline MRI atlas erosions and reduced subarachnoid space were associated with clinical neurological dysfunction at 1 year.
Comparison studies of different imaging modalities of the cervical spine have shown variation in the detection of the different pathologies, according to the imaging technique used. Fezoulidis et al108 found CR and CT to be comparable and better than MRI in detecting atlanto-axial and atlanto-occipital lesions, but MRI to be superior in identifying odontoid lesions. MRI also seems to be better at showing erosions of the dens.117
Independent of the imaging modality used, dynamic lateral views of the cervical spine are more useful than static, neutral views in detecting atlanto-axial subluxation (AAS), in particular anterior AAS.111 Flexion and neutral views are used commonly, with evidence to suggest greater change in the atlanto-dental interval with these views.110 The open mouth view is used for imaging the odontoid peg and to assess for lateral and rotatory AAS; whereas posterior AAS can be measured with neutral and extension views, and VS with a lateral neutral view, although these types of AAS are much less common than anterior AAS.117 When using CR to assess odontoid erosions, lateral cervical spine views appear to be more sensitive than open mouth views.114
Imaging in clinical remission:
Recommendation 10: Ultrasound and MRI can detect inflammation that predicts subsequent joint damage, even when clinical remission is present and can be used to assess persistent inflammation.
Strength of recommendation: 8.8 (95% CI 8.0 to 9.6)
The role of imaging in the detection of inflammation and subsequent prediction of outcome has been discussed previously (recommendation 5). There is good evidence to support the disparity between clinical remission and evidence of ongoing inflammation seen with various imaging modalities. Power Doppler activity has been found in 15–62% of patients in clinical remission according to DAS28, American College of Rheumatology or simplified disease activity index remission criteria,121–124 MRI synovitis in 96% and BM oedema in 52%.124 ,125 In one study, 60% of patients in disease activity score remission had increased uptake on scintigraphy.126
The significance of persistent inflammation, shown in a number of studies, is summarised in table 5.127–133 The presence of ultrasound synovial hypertrophy, power Doppler activity and MRI synovitis at baseline in clinical remission has been shown to be significantly associated with structural progression at 1 year, even in asymptomatic joints.127 Baseline ultrasound inflammatory activity in clinical remission also seems predictive of future disease flare, with 20% of patients experiencing a flare within 12 months in the absence of baseline ultrasound power Doppler activity, compared with 47% in patients with baseline power Doppler activity (p 0.009).128 Although radiographic progression can still be seen in clinical remission, individuals with sustained clinical remission show fewer signs of structural progression compared with patients with clinically relapsing disease.131–133
Future research agenda
The most important topics for future research according to currently available evidence and clinical practice were formulated by the group, shown in table 6.
These are the first recommendations produced by a EULAR task force on imaging in RA clinical practice. The recommendations were developed by an international group of experts with detailed literature review, and aimed to address clinical questions relevant to current practice. We acknowledge there is still a large amount of research required to optimise the use of imaging tools in routine clinical practice, in particular which joints should be used for disease assessment and monitoring and consideration of the feasibility, costs and appropriate training required to use ultrasound and MRI in clinical practice. In view of a lack of literature at the time of the review, these recommendations have not focused on detecting joint space narrowing, which is important to consider in view of the impact on functional status.134 We have made specific reference to this in our proposed future research agenda.
In summary, we have developed 10 recommendations on various aspects of imaging in RA. These are based on the best available evidence and clinical expertise supported by an international panel of experts. We aimed to produce recommendations that are practical and valuable to clinical practice.
The authors would like to thank Louise Falzon for her work in the development of the literature search strategy.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online appendix
Handling editor Tore K Kvien
Contributors ANC and CJE performed the literature review and produced drafts of the manuscript with advice from PGC and MO. All authors were involved in the production of the recommendations, and have reviewed the final manuscript.
Funding The authors would like to thank EULAR for financial support for this work.
Competing interests ANC, MØ, DvdH, PVB, KF, WG, EAH, GH, AGJ, RBML, EN, PJO'C, BO, KP, WAS, JSS, SS, IW: none declared. CJE: Speakers bureau for Roche, BMS, Pfizer, Abbott, UCB, Samsung, MSD, GSK; MADA: Consulting fees and PI of international multicentre study on ultrasound for Bristol-Myers Squibb; speakers bureau for Roche, BMS, Pfizer, Abbott and UCB; research grant on ultrasound from PHRC; book royalties from Elsevier; PGC: speakers bureau or advisory boards for BMS, Pfizer and Roche.
Provenance and peer ereview Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.