• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Ann Rheum Dis. Apr 1, 2012; 71(4): 484–492.
Published online Apr 1, 2012. doi:  10.1136/annrheumdis-2011-200329
PMCID: PMC3298664

2012 provisional classification criteria for polymyalgia rheumatica: a European League Against Rheumatism/American College of Rheumatology collaborative initiative

Abstract

The objective of this study was to develop EULAR/ACR classification criteria for polymyalgia rheumatica (PMR). Candidate criteria were evaluated in a 6-month prospective cohort study of 125 patients with new onset PMR and 169 non-PMR comparison subjects with conditions mimicking PMR. A scoring algorithm was developed based on morning stiffness >45 minutes (2 points), hip pain/limited range of motion (1 point), absence of RF and/or ACPA (2 points), and absence of peripheral joint pain (1 point). A score ≥4 had 68% sensitivity and 78% specificity for discriminating all comparison subjects from PMR. The specificity was higher (88%) for discriminating shoulder conditions from PMR and lower (65%) for discriminating RA from PMR. Adding ultrasound, a score ≥5 had increased sensitivity to 66% and specificity to 81%. According to these provisional classification criteria, patients ≥50 years old presenting with bilateral shoulder pain, not better explained by an alternative pathology, can be classified as having PMR in the presence of morning stiffness>45 minutes, elevated CRP and/or ESR and new hip pain. These criteria are not meant for diagnostic purposes.

Polymyalgia rheumatica (PMR) is a common inflammatory rheumatic disease of older individuals and a common indication for long-term corticosteroid therapy.13 PMR is also subject to wide variations of clinical practice, due to the considerable uncertainty related to diagnosis, course, and management in primary and secondary care.47 There is no diagnostic laboratory test, inflammatory markers are not specific, and clinicians often turn to the corticosteroid response as a ‘test of treatment’ to establish the diagnosis.1 2 8

Difficulties in diagnosing and classifying patients with PMR are inherent in its definitions.9 10 The proximal pain and stiffness syndrome can occur at presentation in many other rheumatological inflammatory illnesses in older people.1 2 7 9 Approximately half of patients diagnosed with PMR may have distal manifestations such as peripheral arthritis, hand swelling with pitting oedema and carpal tunnel syndrome.1 2 7 11 Polymyalgic presentation is common in late-onset rheumatoid arthritis (RA) and spondyloarthritis and is also associated with giant cell arteritis in 10–30% of cases.1 2 Heterogeneity in the disease course, uncertainty regarding disease assessment parameters, and evolution of alternative diagnoses on follow-up complicate the management of PMR.9 1215 For the above reasons a safe and specific approach preferring a relative underdiagnosis to an overdiagnosis is needed in PMR.9 11

Uniform responsiveness to low doses of corticosteroids has been assumed to be a cardinal feature of PMR. However, there is little hard evidence to substantiate this assertion. A previous report showed that 3 weeks after starting prednisolone 15 mg a day only 55% showed a complete response to therapy.10 This also emphasises that clinical trials of novel effective agents are needed in PMR.

The lack of standardised classification criteria has been a major factor hampering the development of rational therapeutic approaches12 16 17 and causing difficulties in evaluating patients in clinical studies. In response to a European League Against Rheumatism (EULAR)/American College of Rheumatology (ACR) initiative, a criteria development work group convened in 2005.

A systematic literature review, a three-phase hybrid consensus process and a wider survey were undertaken to identify candidate criteria items.18 The first stage consisted of a meeting of 27 international experts who anonymously rated 68 potential criteria identified through literature review. In the second round the experts were provided with the results of the first round and re-rated the criteria items. In the third round, the wider acceptance of the chosen criteria (>50% support) was evaluated using a survey of 111 rheumatologists and 53 non-rheumatologists in North America and western Europe.

In round three, over 70% of respondents agreed on the importance of seven core criteria (all achieving 100% support in round two). These were aged 50 years or older, symptom duration 2 weeks or longer, bilateral shoulder and/or pelvic girdle aching, duration of morning stiffness more than 45 min, elevated erythrocyte sedimentation rate (ESR), elevated C-reactive protein (CRP) and rapid corticosteroid response. More than 70% agreed on assessing pain and limitation of shoulder (84%) and/or hip (76%) on motion, but agreement was low for peripheral signs (eg, carpal tunnel syndrome, tenosynovitis, peripheral arthritis).

The group reached consensus on the need for a prospective cohort study to evaluate the disease course from presentation in patients included on the basis of proximal pain and stiffness with evaluations over a 6-month period while receiving a standardised corticosteroid treatment regimen.7 1825 The group also agreed to assess musculoskeletal ultrasound as part of the PMR classification criteria. In this paper, we present the results from this prospective study and propose new classification criteria for PMR.

Study design and methods

Consensus decisions about study design

A priori, the work group decided that a specific approach should be adopted for classifying newly presenting patients with bilateral shoulder pain as PMR.18 The group agreed on the following:

  1. Patients presenting with polymyalgic syndrome should have stepwise evaluation on the basis of inclusion and exclusion criteria, response to a standardised corticosteroid challenge and follow-up confirmation. The criteria items would be agreed upon clinical features and laboratory investigations.
  2. The need to standardise response to corticosteroid therapy in PMR. Because the goal of classification criteria is to identify patients for enrollment into clinical trials before any treatment, the response to corticosteroid therapy should be used in verifying the classification of a patient as having PMR, although it should not be used as a classification criterion. Because scientific evidence is poor as to what constitutes such a response it was agreed (>75% agreement) this response would be defined as greater than 75% global response in clinical and laboratory parameters within 7 days to corticosteroid challenge with oral 15 mg prednisone or prednisolone and subsequent resolution of inflammatory indices.18
  3. That a prospective study would be needed to evaluate the disease course from presentation in patients included on the basis of the mandatory ‘core’ criteria of proximal pain and stiffness. New-onset bilateral shoulder pain was selected as the main eligibility criterion as the percentage of PMR presenting with hip pain without shoulder pain was very small (<5%), and that as hip girdle pain is due to a wide range of conditions it would require the enrollment of an impracticably large number of comparator patient groups.18 The study would evaluate at prespecified intervals symptoms, examination, investigations and their evolution with standardised corticosteroid treatment in a prospective cohort of patients with new-onset bilateral shoulder pain (comparing the PMR case cohort with the comparator cohort of mimicking conditions) over a 6-month period.
  4. Musculoskeletal ultrasound should be evaluated in a substudy as a feasible mode of investigation of possible PMR. A secondary objective of this substudy would be the evaluation of clinical and patient-based outcomes in PMR over a 6-month period.26 27

Study population

The study was a prospective cohort study that included a cohort of patients with new-onset PMR and a comparison cohort of non-PMR patients with various conditions mimicking PMR. Study subjects were recruited from 21 community-based and academic rheumatology clinics in 10 European countries and the USA. Inclusion criteria for PMR patients were age 50 years or older, new-onset bilateral shoulder pain and no corticosteroid treatment (for any condition) within 12 weeks before study entry, fulfilling all the inclusion and exclusion criteria defined by our previous report and in accordance with the judgement of the participating investigator that the patient had PMR.18 Every effort was made to choose patients across the spectrum of disease severity. Corticosteroid treatment for PMR patients was initiated according to a predefined treatment protocol starting with 15 mg a day oral prednisone for weeks 1 and 2, 12.5 mg a day for weeks 3 and 4, 10 mg a day for weeks 6–11, 10 mg/7.5 mg every other day for weeks 12–15, 7.5 mg a day for weeks 16–25 and tapering according to treatment response from week 26 onwards. The gold standard for the pre-steroid diagnosis of PMR was established as above at presentation and when the diagnosis was maintained without an alternative diagnosis at week 26 of follow-up.

The non-PMR comparison cohort included conditions representative of the types that need to be distinguished from PMR, in both primary and secondary care. Inclusion criteria for the non-PMR comparison cohort were age 50 years or older, new-onset bilateral shoulder pain and a diagnosis of either inflammatory or non-inflammatory conditions, including new-onset RA, connective tissue diseases, various shoulder conditions (eg, bilateral rotator cuff syndrome and/or adhesive capsulitis, rotator cuff tear, glenohumeral osteoarthritis), fibromyalgia, generalised osteoarthritis and others. Patients known to have the condition for more than 12 weeks before the baseline evaluation (except fibromyalgia and chronic pain) were not eligible for inclusion. PMR patients with clinical suspicion of giant cell arteritis were included as part of the comparison cohort because these patients required different corticosteroid doses. Patients in the comparison cohort were included on the basis of clinician diagnosis and not on formal criteria. No guidelines were provided for treatment of the conditions in the comparison cohort.

Ethics board approval was obtained at all participating institutions before initiation of the study, and all participants gave written informed consent before enrollment.

Follow-up and data collection

PMR patients were evaluated at baseline, and at 1, 4, 12 and 26 weeks. At each follow-up visit, clinical evaluation included response to corticosteroid therapy and opinion on the emergence of alternative diagnoses. Patients not considered as having PMR at any time were evaluated and treated according to accepted clinical practice. They were excluded from the PMR cohort and included in the non-PMR comparison cohort. Patients in the comparison cohort were evaluated at baseline and at 26 weeks.

Data were collected using standardised data collection forms and questionnaires translated into national languages. Data collection included the candidate inclusion/exclusion criteria items for classification of PMR, physical examination and assessment of corticosteroid response. Criteria items were age 50 years or older, symptom duration 2 weeks or more, bilateral shoulder and/or pelvic girdle aching, recent weight loss greater than 2 kg, duration of morning stiffness more than 45 min, elevated ESR, elevated CRP and rapid response of symptoms to corticosteroids (>75% global response within 1 week to prednisolone/prednisone 15–20 mg a day). Pain was assessed using a horizontal 100 mm visual analogue scale (VAS) in four separate locations (shoulder, pelvic, neck and overall) with zero indicating no pain and 100 indicating worst pain. Morning stiffness (in the past 24 h) was assessed in minutes; functional status and quality of life were assessed by the modified health assessment questionnaire (MHAQ) and short form 36. A 100-mm VAS was also used for recording global wellbeing measures (patient and physician global) and fatigue. Physical examination included the presence or absence of tenderness, pain on movement and limitation of the shoulders and hips. Aspects of corticosteroid therapy including dose, therapeutic response and change in dose and therapy discontinuation were documented.

Data regarding laboratory measures (including ESR, CRP, rheumatoid factor (RF) and anticitrullinated protein antibody (ACPA)) were obtained from clinically ordered tests performed at each study centre. As the laboratory assays used at each centre varied, test results were classified as normal/abnormal using the reference ranges from each centre (see supplementary appendix table S1, available online only). Both PMR and non-PMR subjects underwent ultrasound evaluation of shoulders and hips at baseline and at 26 weeks. Evaluations were made according to EULAR guidelines28 to assess for features previously reported to be associated with PMR, including bicipital tenosynovitis, subacromial and subdeltoid bursitis, trochanteric bursitis and glenohumeral and hip effusion. A rheumatologist or radiologist experienced in musculoskeletal ultrasound performed the ultrasound examination using linear probes with the frequency range 6–10 MHz for shoulders and linear or curved array probes with the frequency range 5–8 MHz for hips.

Statistical analysis

Descriptive statistics (means, percentages, etc.) were used to summarise the candidate criteria data. Demographics (age and gender) and candidate criteria were compared between PMR and comparison subjects using χ2 and rank sum tests. Several statistical approaches were considered in order to develop a scoring algorithm for PMR and to assess the proposed classification criteria in patients judged by expert clinician investigators to have PMR 26 weeks after enrolment.29

First, logistic regression models to distinguish PMR patients from all comparison subjects and each subset of comparison subjects were examined. The c-statistic, a measure of concordance analogous to the area under the receiver operating characteristic curve, was used to assess the ability of each individual criterion to distinguish between PMR and comparison subjects. The c-statistic ranges from 0.5 to 1, with 0.5 indicating a criterion that provides no information. The sensitivity, specificity and the positive and negative likelihood ratios were also examined.

Second, exploratory factor analysis was used to examine the interdependencies between the candidate criteria.30 Maximum likelihood factor analysis with varimax rotation was used. Maximum likelihood tests were used to examine goodness of fit (eg, to determine the number of factors). This method is thought to be superior to the eigenvalue greater than 1 or Catell's scree plot method for selecting the number of factors.30 For each factor, variables with factor loadings greater than 0.5 were examined and found to represent a similar domain. This technique allowed a reduction of the number of variables, as one variable from each factor was examined in multivariable logistic regression models. In addition, to avoid discarding a relevant domain as identified by expert consensus, a few of the variables with factor loading between 0.4 and 0.5 were also considered in the multivariable models.

Classification trees including the variables determined by the factor analysis were also considered, but were not found to be optimal for distinguishing PMR from comparison subjects.31 32 An integer scoring algorithm was defined based on the odds ratios in the final multivariable logistic regression model. Performance characteristics (sensitivity, specificity, etc.) of this scoring algorithm were assessed. In addition, the utility of ultrasound assessments for classifying PMR patients was examined using factor analysis and adding potential ultrasound criteria to the scoring algorithm. Odds ratios for clinical criteria varied somewhat in the scoring algorithm that included ultrasound criteria. Scoring weights based on both models were considered and found to perform similarly, so a common set of scoring weights was used for the clinical items in both algorithms in order to ease comparison and application of the criteria.

Finally, gradient boosting regression tree models, which are a machine learning technique, were examined to determine whether a better prediction could be achieved using a more complex algorithm.33

Results

At baseline, 128 patients were recruited into the PMR cohort and 184 patients were recruited into the non-PMR comparison cohort. During follow-up, 10 PMR patients were reclassified as not having PMR and moved into the non-PMR cohort. Similarly, eight non-PMR comparison cohort patients were reclassified as having PMR and moved into the PMR cohort. In addition, seven patients (one PMR and six non-PMR) were excluded due to missing information, and two non-PMR subjects with age less than 50 years and nine non-PMR subjects with no shoulder pain were also excluded. Therefore, the final analysis was based on 125 PMR and 169 non-PMR subjects.

The diagnoses of the 169 non-PMR subjects (see supplementary appendix table S2, available online only) were new-onset RA (49, 29%), new-onset other seronegative arthritis (20, 12%), new-onset connective tissue diseases or vasculitis (9, 5%), shoulder conditions (52, 31%), chronic pain (26, 15%), infection (5, 3%), previously undiagnosed malignancy (4, 2%), and two each of endocrinopathy and neurological disorders.

The distribution of the candidate classification criteria for the 125 PMR patients and 169 non-PMR comparison subjects (all, RA only and shoulder condition only) is displayed in table 1. Criteria items present in over 80% of the PMR subjects were 2 weeks or greater duration of symptoms, bilateral shoulder pain and elevated CRP and/or ESR. Relevant clinical features that best discriminated RA from PMR were peripheral synovitis, the presence of RF and/or ACPA and hip pain/limited range of motion. Features best discriminating shoulder conditions from PMR were hip pain/limited range of motion, morning stiffness and elevated CRP and/or ESR.

Table 1
Distribution of candidate criteria for 125 PMR patients and 169 comparison subjects*

Development of a scoring algorithm

Table 2 shows the results of univariate logistic regression models to distinguish PMR from all comparison subjects, RA only and shoulder conditions only. Criteria items related to hip involvement (pain, tenderness, limited movement) had significant ability to discriminate PMR from all comparison subjects, RA and shoulder conditions based on the c-statistic. Early morning stiffness, MHAQ, weight loss and raised laboratory markers of inflammation distinguished PMR from comparison subjects, particularly those with shoulder conditions. The presence of ACPA and/or RF, peripheral synovitis and joint pains had significant ability (with high c-statistic) to distinguish PMR from RA. In addition, the odds ratios for abnormal CRP and/or ESR were particularly high. This resulted because only five PMR patients did not have abnormal CRP and/or ESR, perhaps reflecting that the diagnosis of PMR is less certain in the presence of normal CRP and ESR. Therefore, abnormal CRP or ESR would be included as a required criterion in the scoring algorithm for PMR. Similarly, as all subjects in the study were required to have shoulder pain, this was also included as a required criterion for PMR.

Table 2
Univariate logistic regression models to distinguish subjects with PMR from comparison subjects

Factor analysis revealed that four factors were sufficient to represent all the criteria (see supplementary appendix table S3, available online only). These four factors were: hip pain/tenderness, peripheral synovitis or other joint pain, morning stiffness and shoulder tenderness. Note that duration of symptoms, neck aching, carpal tunnel syndrome, weight loss, presence of RF/ACPA and MHAQ, did not play a prominent role in any of the factors.

Next, multivariable logistic regression models were used to determine the importance of each criterion when assessed simultaneously (table 3). Three models were considered: (1) including the four factors identified in factor analysis; (2) removing shoulder tenderness and adding presence of RF/ACPA; and (3) subsequently adding MHAQ. While the subsequent additions appeared to be significant, they also negatively impacted the hip pain criteria. The utility of the inclusion of MHAQ was questionable, as hip pain was easier to assess than MHAQ in a clinical setting. Therefore, the second model, which included hip pain, other joint pain, morning stiffness and abnormal RF/ACPA was deemed the best multivariable model.

Table 3
Multivariable logistic regression models

Additional analyses were performed using classification trees and assessing combinations of criteria. Classification trees were examined using three sets of potential variables: (1) the four identified factors; (2) adding presence of RF/ACPA; and (3) subsequently adding MHAQ. The resulting trees were deemed inadequate. For example, the second tree, which was fit using the same five variables that were included in our scoring algorithm, had a sensitivity of 66% and specificity of 66%. This specificity was lower than the scoring algorithm developed from the logistic models. In addition, this tree only included morning stiffness and absence of RF and/or ACPA, so it was inadequate because it excluded other domains deemed necessary for content validity. Content validity requires that the set of criteria identified is comprehensive. This was the case in all three trees. Therefore, classification trees were deemed to have poorer performance and content validity than the logistic regression models.

A scoring algorithm was developed (table 4) based on the multivariable logistic regression model presented in table 3 and included morning stiffness for more than 45 min (two points), hip pain/limited range of motion (one point), absence of RF and/or ACPA (two points) and the absence of peripheral joint pain (one point). The score was evaluated using all PMR subjects (including the five with normal CRP/ESR) and all comparison subjects. This was done to account properly for the influence of CRP/ESR on the performance of the scoring algorithm. A score of 4 or greater had 68% sensitivity and 78% specificity for discriminating all comparison subjects from PMR. The specificity was higher (88%) for discriminating shoulder conditions from PMR and lower (65%) for discriminating RA from PMR. The c-statistic for the scoring algorithm was 81%. A total of 40 (32%) PMR patients and 38 (22%) comparison subjects was incorrectly classified. The positive predictive value was 69% and the negative predictive value was 77%. Supplementary appendix table S4 (available online only) shows the sensitivity and specificity for all possible cut points of the scoring algorithm.

Table 4
Scoring algorithm with and without optional ultrasound criterion—required criteria: age 50 years or greater, bilateral shoulder aching and abnormal CRP and/or ESR

Finally, gradient boosting regression tree models, which are a model averaging technique, were examined to determine whether a better prediction could be achieved using a more complex algorithm. The resulting c-statistic from the gradient boosting model was 80%, which was quite comparable to the results for our scoring algorithm. This indicated that we will not be able to make better predictions from these data with another modelling approach.

Ultrasound findings

Ultrasound was performed in 120 PMR subjects, 154 of the comparison subjects (including 46 with RA and 47 with shoulder conditions) and 21 additional controls (not included in our study cohorts) who did not have shoulder conditions. Patients with PMR were more likely to have abnormal ultrasound findings in the shoulder (particularly subdeltoid bursitis and biceps tenosynovitis), and somewhat more likely to have abnormal findings in the hips than comparison subjects as a group (table 5). PMR could not be distinguished from RA on the basis of ultrasound, but could be distinguished from non-RA shoulder conditions and subjects without shoulder conditions.

Table 5
Ultrasound findings in 120 patients with PMR, 154 comparison subjects (including 46 with RA and 47 with shoulder conditions) and 21 subjects without shoulder conditions

Assessing the utility of ultrasound in classifying PMR

Factor analysis (see supplementary appendix table S5, available online only) revealed several strong factors potentially useful for classifying patients with PMR from the ultrasound data. The inclusion of ultrasound findings in the scoring algorithm resulted in improving the c-statistic to 82%. A score of 5 or greater had 66% sensitivity and 81% specificity for discriminating all comparison subjects from PMR. The specificity was higher (89%) for discriminating shoulder conditions from PMR and lower (70%) for discriminating RA from PMR. A total of 41 (34%) PMR patients and 30 (19%) comparison subjects was incorrectly classified. The positive predictive value was 72% and the negative predictive value was 75%. Therefore, ultrasound findings were useful in discriminating PMR from shoulder conditions, but less so in discriminating PMR from RA. Table 4 also shows the scoring algorithm including the ultrasound criteria. Supplementary appendix table S4 (available online only) shows the sensitivity and specificity for all possible cut points of the scoring algorithm.

Response to corticosteroids in PMR

Complete corticosteroid response at 4 weeks was seen in 71% of patients and the response was sustained in 78% of responders at 26 weeks. As expected by the plan for tapering of the corticosteroids, the median prednisone dose decreased from 15 mg at baseline to 5 mg at 26 weeks. Response to treatment (percentage improvement in global pain VAS at weeks 4 and 26) was highly correlated with percentage improvement in other VAS measures (correlation >0.5 and p<0.001 at weeks 4 and 26), but was not correlated with percentage change in corticosteroid dose (p=0.20 at week 4 and p=0.47 at week 26). There was no association between the points obtained on either scoring algorithm and the response to corticosteroids at 4 weeks (Spearman correlation coefficient 0.09; p=0.38 for the scoring algorithm including ultrasound) and 26 weeks (data not shown), indicating that corticosteroid response can not be used as part of PMR classification.

We also re-evaluated our risk score model for the scoring algorithm using only the PMR subjects who responded to corticosteroids. When the final risk score model was re-computed using only the subset of subjects with PMR who responded to corticosteroids (and all of the comparison subjects), the odds ratios for each of the criteria remained essentially unchanged. In addition, the specificity was identical and the sensitivity increased by an insignificant amount (0.5%).

Blinded re-evaluation of selected PMR patients and non-PMR controls

The re-evaluation exercise showed that most candidate criteria items performed well in discriminating PMR patients from controls. However, a third of the sample of PMR patients/comparison subjects was difficult to classify. The high c-statistic levels associated with the corticosteroid response and post-treatment CRP and ESR suggested that the uncertainty originated from the pivotal role of corticosteroids in the investigator assessment, in deciding whether a patient does or does not have PMR. It raises questions such as whether PMR always responds adequately to corticosteroids and whether polymyalgic RF-positive disease without peripheral synovitis can occur (see supplementary appendix 2, available online only, for details).

Discussion

This is the first international multicentre prospective study examining consensus-based candidate classification criteria for PMR proposed by an international work group. Findings indicate that patients 50 years of age or older presenting with new bilateral shoulder pain (not better explained by an alternative diagnosis) and elevated CRP and/or ESR can be classified as having PMR in the presence of morning stiffness for more than 45 min, and new hip involvement (pain, tenderness, limited movement). The absence of peripheral synovitis or of positive RA serology increases the likelihood of PMR. While recognising that RF particularly may be present in patients with PMR, the absence of RF serology is useful in distinguishing PMR from RA in older patients for classification purposes.18 34 Ultrasound findings of bilateral shoulder abnormalities (subacromial bursitis/bicipital tenosynovitis/glenohumeral effusion) or abnormalities in one shoulder and hip (hip effusion, trochanteric bursitis) may significantly improve the specificity of the clinical criteria. These criteria are not meant for diagnostic purposes and have not been tested as diagnostic criteria.

Newer concepts of PMR are revealed by this and other recent studies—heterogeneity at presentation and course, lack of uniform responsiveness to low-dose steroids and overlap with inflammatory arthritis. However, we feel that at present these classification criteria provide a basic framework for developing clinical trials of novel therapies in PMR.

How should the PMR classification criteria be applied?

The target population will be patients aged 50 years or older presenting with new-onset (<12 weeks) bilateral shoulder pain and abnormal acute phase response. The criteria may only be applied to those patients in whom the symptoms are not better explained by an alternative diagnosis. Mimicking conditions include the inflammatory and non-inflammatory conditions studied as comparators in this report.

Four clinical and laboratory criteria along with optional ultrasound criteria (table 4 and table 6) can be applied to eligible patients to identify patients with PMR suitable for low-dose corticosteroid therapy. The scoring scale is 0–6 (without ultrasound) and 0–8 (with ultrasound). In the absence of competing diagnoses, a score of 4 or greater (without ultrasound), or 5 or greater (with ultrasound) is indicative of PMR. Patients with a score of less than 4 (based on clinical plus laboratory criteria) cannot be considered to have PMR. Ultrasound improves the specificity of PMR diagnosis, and shows particularly good performance in differentiating PMR from non-inflammatory conditions and thus is a recommended investigation for PMR.

Table 6
PMR classification criteria scoring algorithm—required criteria: age 50 years or older, bilateral shoulder aching and abnormal CRP and/or ESR*

Classification criteria for PMR should be useful for identifying patients appropriate for enrollment into clinical trials of novel medications for the treatment of PMR, and studying long-term outcomes in more homogeneous patient cohorts. Our analyses indicate that even typical PMR patients at presentation may vary in their response to low-dose corticosteroid therapy, indicating that corticosteroid response is not reliable as a classification feature for PMR. This is similar to other rheumatic diseases such as RA, in which phenotypically similar patients may exhibit different responses to disease-modifying antirheumatic drugs.

Strengths and limitations

The strengths of the study relate to harnessing international effort to address a disease area subject to wide variation of practice; to develop agreement on the definition of what may or may not be treated as PMR and what needs further evaluation.

Our study methodology satisfies the ACR guidelines for the development of classification criteria for rheumatic diseases.35 The consensus-based candidate criteria were generated by a multispecialty international group whose views were supported by the results of a wide trans-Atlantic survey. The work group suggested a prospective study designed to separate PMR patients from comparison subjects in patients included on a single eligibility criterion (new bilateral shoulder pain in subjects ≥50 years).

This ensured an inception cohort longitudinal observational design wherein the PMR cohort could be compared with the comparison cohort at similar chronological time points of disease. All PMR patients were evaluated before treatment with corticosteroids, were treated with a standard corticosteroid schedule and assessed at predetermined time points. Our study is in keeping with the EULAR/ACR goal of developing rheumatic disease classification criteria as opposed to diagnostic criteria. We focused on subjects with new-onset/incident disease, and a 6-month longitudinal follow-up allowed an accurate evaluation of the disease course and diagnoses. Previous criteria for PMR were developed using cross-sectional comparisons. Only two of the previous criteria were developed through an evaluative process.20 36 Neither had a definition of the ‘gold standard’ diagnosis other than the physician considered the patient to have ‘unequivocal’ PMR.

Another strength of our proposed classification criteria is the imaging component. Musculoskeletal ultrasound has promise due to widespread availability, feasibility and good research evidence.18 The PMR work group standardised the examination of shoulders and hips by ultrasound for the purposes of the current study.37 Our findings indicate that ultrasound evaluation of hips and shoulders adds significantly to the evaluation of the polymyalgic syndrome. The lack of a ‘gold standard’ and the challenge of circularity was addressed through a blinded multi-rater re-evaluation exercise in selected cases and comparison subjects (in supplementary appendix 2, available online only). Most of the candidate criteria items performed well. The misclassification of several subjects reflects the difficulty in discriminating PMR from other inflammatory conditions such as RA. This uncertainty originates from the pivotal role of (and circular reasoning related to using) the corticosteroid response in deciding whether a patient does or does not have PMR. Although the proposed scoring algorithm has high specificity for identifying PMR patients,, it is nevertheless unable to predict the subsequent corticosteroid response, suggesting heterogeneity of disease course and treatment response. This has also been reported previously.10 38

While we were able to scrutinise the basis for PMR diagnosis, no formal criteria were required for diagnosis of the comparison conditions. This is a limitation of the study. However, the study was led in all centres by experienced rheumatologists with major clinical and research interest in PMR and related conditions. We did not include hip pain without shoulder pain as an eligibility criterion for reasons discussed in the ‘Methods’ section. Funding constraints limited the follow-up duration to only 6 months. We were also limited by lack of funding for the central measurement of laboratory data. Values of ESR, CRP, RF and ACPA were based on local laboratory assays. Our study approach reflects a pragmatic view, which perhaps lends wider applicability to the results of the study.

Our classification algorithm had a c-statistic of 81%, which exceeds the threshold of 80% that is conventionally considered to be useful in clinical decision-making. However, we suggest that the criteria are regarded as provisional at this point awaiting validation in a separate cohort.

In conclusion, patients aged 50 years or older presenting with bilateral shoulder pain and elevated CRP and/or ESR can be classified as having PMR in the presence of morning stiffness for more than 45 min, and new hip pain in the absence of peripheral synovitis or positive RA serology. Using ultrasound, a score of 5 or greater had 66% sensitivity and 81% specificity for discriminating all comparison subjects from PMR. In our view, this approach can now be used to test eligibility for trials with newer therapies in PMR. A number of future research questions are highlighted, including: (1) Should PMR be considered as a part of the spectrum of late-onset inflammatory arthritis? (2) Can polymyalgic disease without peripheral synovitis occur in RF-positive disease? (3) Can we diagnose PMR in patients with normal acute phase response? (4) What is the role of the early introduction of disease-modifying antirheumatic drugs in PMR?

We have collected and stored blood samples from both the case and the comparator groups in the study. We hope to develop research proposals using these biospecimens to test several candidate biomarkers in PMR. These proposals will also examine the acute phase response; whether stratification of the response and differential levels of acute phase reactants and cytokines may function as additional classification criteria items.

Supplementary Material

Supplemental Figures:

Acknowledgments

The investigators would like to acknowledge EULAR, ACR, the Mayo Foundation and the Biobanque de Picardie in Amiens, France, for material and intellectual support of this project. The investigators also wish to recognise the individual and uncompensated efforts of the staff at each participating institution, without which this study could not have been completed.

Footnotes

Contributors All authors contributed and were involved in the conduct of this study, in study design and/or subject recruitement, analysis and manuscript preparation.

Funding Funding for this study was provided by EULAR, ACR, the Mayo Foundation, and was supported further by the individual and uncompensated efforts of participating investigators and their staffs and institutions, and the Biobanque de Picardie in Amiens, France. GE-F and MCC were supported by Ministerio de Ciencia y Tecnologia (SAF 08/04328).

Competing interests None.

Ethics approval The study was approved by the ethics boards of the participating investigor institutions, which are listed.

Patient consent Obtained.

Provenance and peer review Not commissioned; externally peer reviewed.

References

1. Salvarani C, Cantini F, Boiardi L, et al. Polymyalgia rheumatica and giant-cell arteritis. N Engl J Med 2002;347:261–71 [PubMed]
2. Weyand CM, Goronzy JJ. Giant-cell arteritis and polymyalgia rheumatica. Ann Intern Med 2003;139:505–15 [PubMed]
3. Doran MF, Crowson CS, O'Fallon WM, et al. Trends in the incidence of polymyalgia rheumatica over a 30 year period in Olmsted County, Minnesota, USA. J Rheumatol 2002;29:1694–7 [PubMed]
4. Chakravarty K, Elgabani SH, Scott DG, et al. A district audit on the management of polymyalgia rheumatica and giant cell arteritis. Br J Rheumatol 1994;33:152–6 [PubMed]
5. Maradit Kremers H, Reinalda MS, Crowson CS, et al. Use of physician services in a population-based cohort of patients with polymyalgia rheumatica over the course of their disease. Arthritis Care Res 2005;53:395–403 [PubMed]
6. Gamez-Nava JI, Gonzalez-Lopez L, Davis P, et al. Referral and diagnosis of common rheumatic diseases by primary care physicians. Br J Rheumatol 1998;37:1215–19 [PubMed]
7. González-Gay MA, García-Porrúa C, Vázquez-Caruncho M, et al. The spectrum of polymyalgia rheumatica in northwestern Spain: incidence and analysis of variables associated with relapse in a 10 year study. J Rheumatol 1999;26:1326–32 [PubMed]
8. Cantini F, Salvarani C, Olivieri I. Erythrocyte sedimentation rate and C-reactive protein in the diagnosis of polymyalgia rheumatica. Ann Intern Med 1998;128:873–4 [PubMed]
9. Dasgupta B, Borg FA, Hassan N, et al. BSR and BHPR guidelines for the management of polymyalgia rheumatica. Rheumatology (Oxford) 2010;49:186–90 [PubMed]
10. Hutchings A, Hollywood J, Lamping DL, et al. Clinical outcomes, quality of life, and diagnostic uncertainty in the first year of polymyalgia rheumatica. Arthritis Rheum 2007;57:803–9 [PubMed]
11. Dasgupta B, Matteson EL, Maradit-Kremers H. Management guidelines and outcome measures in polymyalgia rheumatica (PMR). Clin Exp Rheumatol 2007; 25(6 Suppl 47):130–6 [PubMed]
12. Matteson EL. Clinical guidelines: unraveling the tautology of polymyalgia rheumatica. Nat Rev Rheumatol 2010;6:249–50 [PubMed]
13. Caporali R, Montecucco C, Epis O, et al. Presenting features of polymyalgia rheumatica (PMR) and rheumatoid arthritis with PMR-like onset: a prospective study. Ann Rheum Dis 2001;60:1021–4 [PMC free article] [PubMed]
14. Kremers HM, Reinalda MS, Crowson CS, et al. Relapse in a population based cohort of patients with polymyalgia rheumatica. J Rheumatol 2005;32:65–73 [PubMed]
15. Delecoeuillerie G, Joly P, Cohen de Lara A, et al. Polymyalgia rheumatica and temporal arteritis: a retrospective analysis of prognostic features and different corticosteroid regimens (11 year survey of 210 patients). Ann Rheum Dis 1988;47:733–9 [PMC free article] [PubMed]
16. Jones JG, Hazleman BL. Prognosis and management of polymyalgia rheumatica. Ann Rheum Dis 1981;40:1–5 [PMC free article] [PubMed]
17. Dasgupta B, Hutchings A, Matteson EL. Polymyalgia rheumatica: the mess we are now in and what we need to do about it. Arthritis Rheum 2006;55:518–20 [PubMed]
18. Dasgupta B, Salvarani C, Schirmer M, et al. Developing classification criteria for polymyalgia rheumatica: comparison of views from an expert panel and wider survey. J Rheumatol 2008;35:270–7 [PubMed]
19. Ferraccioli GF, Di Poi E, Damato R. Steroid sparing therapeutic approaches to polymyalgia rheumatica-giant cell arteritis. State of the art and perspectives. Clin Exp Rheumatol 2000;18(4 Suppl 20):S58–60 [PubMed]
20. Bird HA, Esselinckx W, Dixon AS, et al. An evaluation of criteria for polymyalgia rheumatica. Ann Rheum Dis 1979;38:434–9 [PMC free article] [PubMed]
21. Leeb BF, Bird HA, Nesher G, et al. EULAR response criteria for polymyalgia rheumatica: results of an initiative of the European Collaborating Polymyalgia Rheumatica Group (subcommittee of ESCISIT). Ann Rheum Dis 2003;62:1189–94 [PMC free article] [PubMed]
22. Kyle V, Hazleman BL. The clinical and laboratory course of polymyalgia rheumatica/giant cell arteritis after the first two months of treatment. Ann Rheum Dis 1993;52:847–50 [PMC free article] [PubMed]
23. Dejaco C, Duftner C, Cimmino MA, et al. Definition of remission and relapse in polymyalgia rheumatica: data from a literature search compared with a Delphi-based expert consensus. Ann Rheum Dis 2011;70:447–53 [PMC free article] [PubMed]
24. Dasgupta B, Hutchings A, Hollywood J, et al. Clinical outcomes, quality of life and diagnostic uncertainty in the first year in polymyalgia rheumatica: a prospective cohort study. Rheumatology 2006;45(Suppl 1):i170
25. Weyand CM, Fulbright JW, Evans JM, et al. Corticosteroid requirements in polymyalgia rheumatica. Arch Intern Med 1999;159:577–84 [PubMed]
26. Cantini F, Salvarani C, Olivieri I, et al. Inflamed shoulder structures in polymyalgia rheumatica with normal erythrocyte sedimentation rate. Arthritis Rheum 2001;44:1155–9 [PubMed]
27. Cantini F, Niccoli L, Nannini C, et al. Inflammatory changes of hip synovial structures in polymyalgia rheumatica. Clin Exp Rheumatol 2005;23:462–8 [PubMed]
28. Backhaus M, Burmester GR, Gerber T, et al. Guidelines for musculoskeletal ultrasound in rheumatology. Ann Rheum Dis 2001;60:641–9 [PMC free article] [PubMed]
29. Steyerberg EW. Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating. Statistics for Biology and Health. New York: Springer, 2010:57–71
30. Fabrigar LR, Wegener DT, MacCallum RC, et al. Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods 1999;4:272–99
31. Breiman L. Random forests. Machine Learning 2001;45:5–32
32. Therneau TM, Atkinson EJ. An Introduction to Recursive Partitioning using the RPART Routines. Department of Health Sciences Research Technical Report Series: no 66. Rochester, Minnesota: Mayo Clinic, 1997:1–52
33. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd edn New York: Springer Science + Business Media, 2009:337–88
34. Lopez-Hoyos M, Ruiz de Alegria C, Blanco R, et al. Clinical utility of anti-CCP antibodies in the differential diagnosis of elderly-onset rheumatoid arthritis and polymyalgia rheumatica. Rheumatology (Oxford) 2004;43:655–7 [PubMed]
35. Classification and Response Criteria Subcommittee of the American College of Rheumatology Committee on Quality Measures Development of classification and response criteria for rheumatic diseases. Arthritis Rheum 2006;55:348–52 [PubMed]
36. Nobunaga M, Yoshioka K, Yasuda M, et al. Clinical studies of polymyalgia rheumatica. A proposal of diagnostic criteria. Jpn J Med 1989;28:452–6 [PubMed]
37. Scheel AK, Matteson EL, Dasgupta B, et al. Reliability exercise for the polymyalgia rheumatica classification criteria study: the Oranjewoud Ultrasound Substudy. Int J Rheumatol 2009;2009:738931. [PMC free article] [PubMed]
38. Hernández-Rodríguez J, Cid MC, López-Soto A, et al. Treatment of polymyalgia rheumatica: a systematic review. Arch Intern Med 2009;169:1839–50 [PubMed]

Articles from BMJ Open Access are provided here courtesy of BMJ Group
PubReader format: click here to try

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

  • MedGen
    MedGen
    Related information in MedGen
  • PubMed
    PubMed
    PubMed citations for these articles
  • Substance
    Substance
    PubChem Substance links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...