Hand osteoarthritis (OA) mainly affects the distal interphalangeal (DIP), proximal interphalangeal (PIP) and thumb base joints,1 leading to joint pain, aching and/or stiffness. The prevalence estimates vary depending on the study population and the definition used.1 2 Currently, there is no approved disease-modifying treatment for OA,3–5 although several promising candidates are being tested.6

Using classification criteria is the standard method of assembling a homogenous group of people with the disease of interest for enrolment in clinical trials and observational studies. The 1990 American College of Rheumatology (ACR) criteria set is currently the only available classification criteria set for hand OA.7 The ACR criteria set uses clinical features only and advises against hand OA classification based on radiographs. It classifies hand OA as present almost exclusively based on joint involvement in the second–third fingers and the thumb base. The ACR criteria set may be less suited to classify hand OA in primary care settings or the general population as it was primarily developed to distinguish hand OA from rheumatoid arthritis (RA).7 Moreover, the ACR criteria set does not classify hand OA phenotypes such as interphalangeal or thumb base OA, which are two phenotypes that may require different treatment strategies. Based on these limitations, members of the European Alliance of Associations for Rheumatology (EULAR) taskforce for evidence-based recommendations on hand OA diagnosis ranked the development of new classification criteria as a top research priority.8

Hence, we aimed to develop new classification criteria sets incorporating radiographic features for overall hand OA, interphalangeal OA and thumb base OA in a population with hand pain, aching and/or stiffness.

Three phases for the development of the EULAR classification criteria sets were defined a priori (figure 1). The methods of phases 1–2 have been detailed in previous publications9 10 and are thus briefly summarised. The current report mainly focuses on phase 3. The EULAR/ACR methodology for the development of classification criteria was used.

Three phases in developing classification criteria for overall hand osteoarthritis (OA), interphalangeal OA and thumb base OA.

Using data from a multicentre observational study of persons with hand complaints due to hand OA (n=128) or other inflammatory and non-inflammatory conditions (n=70) as determined by the physician, logistic regression analyses were performed to identify self-reported, clinical, radiographic and laboratory features associated with hand OA (phase 1). We assessed the discrimination capacity of each feature in classifying hand OA.9

In phase 2, we used a consensus-driven decision-making approach to refine the criteria identified in phase 1 and determine their relative importance (ie, weight). Our multidisciplinary expert panel (n=21) included 13 rheumatologists, 2 primary care physicians, 2 surgeons, 2 occupational therapists, 1 physical therapist and 1 physician assistant. The experts were spread across Europe (n=17), North America (n=2), Asia (n=1) and Australia (n=1). Only 7 of 21 experts had also been actively involved in the data collection in phase 1. In addition to the 21 experts, 1 person developed all surveys and performed the analyses (IKH) and 8 experts were involved in the discussion about the results in phase 2 and/or a reliability exercise.10

Groups of case vignettes were created, presenting positive or negative findings from the diagnostic tests identified in phase 1. The experts ranked the cases according to their likelihood of having hand OA as cause of the complaints, both individually and in consensus. Sets of criteria and categories were drafted and tested in a series of surveys using the 1000minds software to determine the relative weights of the criteria and their categories.10

In phase 3, we refined the scoring system developed in phase 2 and determined the optimal cut-off for disease classification. The sensitivity and specificity of the proposed criteria were tested.

Rescaling was done to ensure feasibility and user-friendliness of the criteria. The impact of rescaling was evaluated by looking at the Spearman correlation between the total scores based on original weighting and the rescaled criteria. We also evaluated whether rescaling changed the ranking of patients. We used 3 sets of patient vignettes for overall hand OA, interphalangeal OA and thumb base OA, each with 30 patients with pain, aching and/or stiffness in at least 1 target joint on most days of the prior 6 weeks and without psoriasis (sets 2A–C).10

We used a consensus-based approach to identify the optimal cut-off. Using the proposed criteria score (range 0–100), 30 patient vignettes (set 2A) were ranked according to their likelihood of having hand OA. The 21 experts, who were unaware of the criteria score, were asked to examine the rankings of patients and to indicate the point at which the patients changed from ‘probable’ to ‘definite’ overall hand OA as the cause of the complaints. This exercise was repeated for interphalangeal OA and thumb base OA using other patient vignettes (sets 2B–C).

In addition, a data-driven approach based on analyses of phase 1 data was applied. We calculated the area under the curve (AUC) for the proposed criteria for overall hand OA (range: 0–100) using as a reference the physician’s evaluation of the primary cause of the hand complaints (ie, OA vs non-OA conditions). The optimal cut-off value was the point closest to the upper left corner of the receiver-operating curve. The data-driven approach was applied only to the overall hand OA criteria, as we lacked information regarding the physician’s specific opinion about interphalangeal OA and thumb base OA being the cause of the complaints. Two webinars were arranged to present and discuss the results of these surveys.

The proposed EULAR criteria were validated in two external cohorts: a Dutch hand OA cohort (Hand OA in Secondary Care Study, HOSTAS)11 and a Norwegian study of people with self-reported OA (Musculoskeletal pain in Ullensaker STudy, MUST).12 In HOSTAS, we restricted the analyses on sensitivity of the EULAR and ACR criteria to those with a high likelihood of having hand OA based on an evaluation by the physician (score ≥7 on a 0–10 scale). In the absence of a control group without hand OA in HOSTAS, we could not calculate specificity. In MUST, the sensitivity and specificity of the EULAR and ACR criteria were examined using persons with self-reported hand OA (ie, diagnosed with hand OA by a medical doctor and/or radiographs) as reference and persons without self-reported hand OA as controls.

The validation of the criteria was also complemented using data from phase 1.9 We restricted the analyses to cases with clearcut hand OA (score ≥7 on a 0–10 scale concerning the likelihood of having complaints due to hand OA based on the opinion by the physician) and cases where other conditions were clearly the cause of their hand complaints (score ≤3). We calculated the sensitivity and specificity of the EULAR and ACR criteria using the physician’s evaluation of hand OA or other conditions as the cause of complaints as reference.

Sensitivity analyses of radiographic ACR criteria were performed replacing bony enlargement with radiographic osteophytes.

Two European patient research partners were involved in phases 1–2. One patient research partner was replaced prior to phase 3, and two research partners (HL, EG) from Norway and the Netherlands were involved in preparations of this manuscript.

The proposed EULAR hand OA classification criteria are the final result of an international collaboration based on data-driven and consensus-based approaches, including expert opinion. The criteria are intended for use in clinical trials and observational studies.

The criteria include two mandatory criteria, which means that the criteria should be used in persons with hand symptoms, not better explained by another disease or acute injury. The requirement for hand symptoms is crucial in most clinical trial settings, where an effect on symptoms is often the primary outcome.

There are similarities and differences between the proposed EULAR criteria and the 1990 ACR criteria.7 While the original paper about the ACR criteria did not specify the duration of symptoms,7 it was mentioned that pain, aching or stiffness should be present on most days of the prior month in a later review paper.14 In the proposed EULAR criteria, we require pain, aching and/or stiffness to be present on most days of the previous 6 weeks. A cut-off of 6 weeks was also used in the ACR/EULAR classification criteria for RA.15 The benefit of the EULAR criteria is the ability to classify hand OA, independent of which DIP and PIP joints are affected, in contrast to the ACR criteria, which mainly concern the second–third DIP and PIP joints. Age and morning stiffness are included in the proposed EULAR criteria, but not in the ACR criteria. Including these features in the criteria set allows persons with higher age (and thus higher risk of hand OA) and short-lived morning stiffness to fulfil the criteria despite limited structural damage.

While the ACR criteria set includes clinical examination features, the proposed EULAR criteria set involves radiographic features. Radiographic features showed better discrimination of hand OA cases and controls than clinical features in phase 1.9 In line with these results, Cicuttini et al showed that radiographic osteophytes were better markers of OA in other joints than clinical nodes.16 Furthermore, due to poor agreement between clinical nodes and radiographic findings, these features were not interchangeable.16 According to clinical guidelines,8 17 radiographs are not needed for the diagnosis of hand OA. Importantly, these criteria are meant for use in research settings, and not clinical practice. We acknowledge that radiographs may not be feasible in all research settings, and the ACR criteria can still be an option to classify overall hand OA in these settings. Ultrasound may be helpful to detect OA pathology instead of radiographs. However, it is challenging to obtain reliable evaluation of cartilage in finger joints,18 and we therefore do not recommend to replace radiographs with ultrasound for the assessment of OA pathology.

Prolonged morning stiffness is often considered a sign of inflammatory joint diseases,19 20 and long morning stiffness was negatively associated with hand OA in phase 1.9 However, 17% of the participants in HOSTAS had morning stiffness of at least 1 hour, highlighting that long morning stiffness should not preclude the diagnosis of hand OA.13 The experts thus decided to remove long morning stiffness as an exclusion criterion. Information about morning stiffness is often not available in prior studies of hand OA. Hence, the experts agreed that it should be possible to apply the criteria without this information, using a slightly lower cut-off value for disease classification. The agreement between the original and modified criteria sets was excellent, suggesting a minor impact of this modification. Since the modified criteria sets are more heavily dependent on structural pathology, they may be less sensitive to classify hand OA in populations with less advanced disease. Hence, the experts recommend that morning stiffness should be included in the data collection of future studies and that the modified criteria sets should only be used in studies, in which information about morning stiffness is lacking.

Our experts considered concordance between symptoms and joint pathology important, while this criterion was not included in the ACR criteria. Hand symptoms are common with a wide variety of causes, and it may be challenging to distinguish symptoms from joints versus extraarticular tissues. Hence, our experts were reluctant to include persons with poor concordance between symptoms and radiographic findings in a clinical trial, since their symptoms may be caused by non-OA conditions. Since joints may have early OA pathology despite normal radiographs,21 the symptom-structure concordance criterion may push our criteria sets towards classification of more advanced disease.

The proposed EULAR criteria are meant for classification of established disease with a limited risk of false positive cases. Our goal was not to develop criteria for early symptomatic hand OA, which would have required analyses to differentiate not only hand OA versus mimickers, but also early versus established hand OA. Due to the lack of data on early hand OA symptoms and findings in a stage before the radiographic changes occur, it is challenging to develop criteria for early disease with high sensitivity and specificity. This would have required identification of typical early symptomatic hand OA features through patient surveys, expert surveys and/or additional data collections that were not part of this proposal. It is also important to bear in mind that hand OA differs from knee OA, as many patients have a combination of joints with early and established OA.

A major advantage of the proposed EULAR criteria is their ability to classify interphalangeal OA and thumb base OA separately. For persons with isolated thumb base OA, it is challenging, although not impossible, to fulfil the ACR and the EULAR criteria for overall hand OA. A separate criteria set for this phenotype is thus essential since studies may focus specifically on risk factors or treatments for thumb base OA. The interphalangeal OA criteria are identical to the overall hand OA criteria except that the thumb base joints are not included as target joints, that is, fewer target joints. Hence, the expert panel agreed that it was appropriate to have a slightly lower cut-off value for the classification of interphalangeal OA than for overall hand OA. This allows us to classify interphalangeal and thumb base OA in cases that do not fulfil the criteria for overall hand OA. This is reflected by the higher sensitivity of the criteria when applying the interphalangeal OA and/or thumb base OA criteria compared with the overall hand OA criteria (table 5).

We found better sensitivity and similar specificity of the proposed EULAR criteria compared with the ACR criteria in the phase 1 data set.9 These results should be treated with caution since we used phase 1 data to inform the development of the criteria. The promising results were supported by comparable sensitivity and specificity of the EULAR and ACR criteria in MUST.12 The lower specificity of both the EULAR and the ACR criteria in MUST is likely explained by the reference ‘self-reported hand OA’, as many people may have hand OA despite no diagnosis. In a study of patients with hand OA from secondary care (HOSTAS),11 the EULAR criteria for interphalangeal and/or thumb base OA demonstrated similar sensitivity as the ACR criteria, while the sensitivity of EULAR criteria for overall hand OA was slightly lower. These results are likely a reflection of high proportion of people with isolated thumb base OA. In sensitivity analyses, replacement of bony enlargement with radiographic osteophytes led to decreased sensitivity of the ACR criteria, suggesting that the evaluation of radiographic findings in this cohort was less sensitive than clinical examination. The sensitivity of the ACR criteria in HOSTAS is lower than previously reported due to different definitions of the mandatory criterion (online supplemental table 3).22 The participants had not indicated whether stiffness and aching were present on most days or not, and these symptoms were therefore not included.

In clinical trials, high specificity is more important than high sensitivity. The proposed EULAR criteria prespecify the exclusion of those with known other causes of hand pain, whereas ACR criteria do not. Other known causes of hand OA such as RA were either representing the control group (phase 1), excluded in advance (HOSTAS) or scarcely reported (MUST). Hence, the testing of validity was performed without classifying those with other conditions as non-hand OA cases. In a general population with hand pain, EULAR criteria would, by definition, have higher specificity. Classifying people with psoriasis as non-hand OA cases led to an increased specificity in our analyses, which exceeded the specificity of the ACR criteria (table 5).

The proposed EULAR criteria clearly have benefits in clinical trials and observational studies. They allow the classification of thumb base OA, which is poorly classified by the ACR criteria unless occurring in combination with interphalangeal OA. Furthermore, radiographic assessment of OA instead of clinical joint assessment may be done centrally by trained readers, and thus with higher reliability.23 People fulfilling the EULAR criteria only and not the ACR criteria were more likely to have radiographic findings in the majority of symptomatic joints, which is a benefit in clinical trials. The ACR criteria also allow persons to be classified as having OA if they have pain in metacarpophalangeal joints only and clinical features of OA in other joints, which may raise a clinical suspicion of RA.

In conclusion, we have presented the EULAR classification criteria for hand OA based on rigorous methodology by international experts. Separate criteria were developed for overall hand OA and two phenotypes (interphalangeal OA and thumb base OA), which may facilitate the entry of suitable candidates most likely to benefit from specific therapies in clinical trials and more homogenous populations in observational studies. Future studies with a control group without OA are needed for further validation of these criteria.