|Home | About | Journals | Submit | Contact Us | Français|
Conception and design: HJM Kerkhof, TD Spector, AM Valdes, JBJ van Meurs, AG Uitterlinden, I Meulenbelt.
Analysis and interpretation of data: HJM Kerkhof, NE Lane, NK Arden, I Meulenbelt, N Yoshimura, A Tamm, MC Nevitt, UM Kujala, P Leino-Arjas, D Felson, FP Luyten, AM Valdes, JBJ van Meurs, M Kloppenburg, F Rivadeneira, S Lohmander, TD Spector.
Drafting of the article: HJM Kerkhof, AM Valdes, JBJ van Meurs
Critical revision of the article: all authors
Final approval of the article: all authors
Provision of study material or patients: I Meulenbelt, NK Arden, A Aromaa, A Carr, C Cooper, J Dai, M Doherty, SA Doherty, D Felson, A Gonzalez, DJ Hart, M Heliovaara, A Hofman, S Ikegawa, T Ingvarsson, Q Jiang, H Jonsson, I Jonsdottir, UM Kujala, NE Lane, P Leino-Arjas, S Lohmander, MC Nevitt, E Slagboom, TD Spector, K Stefansson, A Tamm, A Tsezou, JM Wilkinson, JBJ van Meurs, AG Uitterlinden, AM Valdes, N Yoshimura
Collection of data: T Akune, J Dai, A Gordon, A Harilainen, VB Hauksson, H Kawaguchi, M Kloppenburg, M Nakajima, D Shi, A Sudo, AE Tamm, A Uchida, I Meulenbelt, NK Arden, A Aromaa, A Carr, C Cooper, J Dai, M Doherty, SA Doherty, D Felson, A Gonzalez, DJ Hart, M Heliovaara, A Hofman, S Ikegawa, T Ingvarsson, Q Jiang, H Jonsson, I Jonsdottir, UM Kujala, NE Lane, P Leino-Arjas, S Lohmander, MC Nevitt, E Slagboom, TD Spector, K Stefansson, A Tamm, A Tsezou, JM Wilkinson, JBJ van Meurs, AG Uitterlinden, AM Valdes, N Yoshimura, HJM Kerkhof, JM Wilkinson.
To address the need for standardization of osteoarthritis (OA) phenotypes by examining the effect of heterogeneity among symptomatic (SOA) and radiographic osteoarthritis (ROA) phenotypes.
Descriptions of OA phenotypes of the 28 studies involved in the TREAT-OA consortium were collected. To investigate whether different OA definitions result in different association results, we created hip OA definitions used within the consortium in the Rotterdam Study-I and tested the association of hip OA with gender, age and BMI using one-way ANOVA. For radiographic OA, we standardized the hip, knee and hand ROA definitions and calculated prevalence's of ROA before and after standardization in 9 cohort studies. This procedure could only be performed in cohort studies and standardization of SOA definitions was not feasible at this moment.
In this consortium, all studies with symptomatic OA phenotypes (knee, hip and hand) used a different definition and/or assessment of OA status. For knee, hip and hand radiographic OA 5, 4 and 7 different definitions were used, respectively. Different hip OA definitions do lead to different association results. For example, we showed in the Rotterdam Study-I that hip OA defined as “at least definite JSN and one definite osteophyte” was not associated with gender (p=0.22), but defined as “at least one definite osteophyte” was significantly associated with gender (p=3×10−9). Therefore, a standardization process was undertaken for radiographic OA definitions. Before standardization a wide range of ROA prevalence's was observed in the 9 cohorts studied. After standardization the range in prevalence of knee and hip ROA was small. Standardization of SOA phenotypes was not possible due to the case-control design of the studies.
Phenotype definitions influence the prevalence of OA and association with clinical variables. ROA phenotypes within the TREAT-OA consortium were standardized to reduce heterogeneity and improve power in future genetics studies.
The Translational Research in Europe Applied Technologies for OsteoArthritis (TREAT-OA) consortium was established in January 2008 to address the generalisability and utility of genetic and biochemical risk factors (www.treatoa.eu). The two main goals of TREAT-OA are 1) to develop efficient diagnostics for risk and progression of osteoarthritis (OA) and 2) to identify new targets for therapeutic interventions. This will be done by identification of genes and biochemical markers consistently associated with risk and progression of OA, but also by defining the roles of these genes in molecular pathways involved in disease aetiology, for example by the development of in vivo transgenic animal OA model systems.
A major goal of the consortium is to identify new genes consistently associated with risk and progression of OA. To reach this goal, large-scale genome-wide association studies (GWASs) and meta-analyses are being performed. To date, research within the TREAT-OA consortium has resulted in the identification of a novel genetic locus on chromosome 7q22 that is associated with knee- and hand OA1, which was confirmed by a yet unpublished GWAS meta-analysis on knee OA. In addition, the ataxin 2 binding protein 1 gene2 and the prostaglandin-endoperoxide synthase 2 gene3 have been found associated with respectively hand and knee OA. One of the difficulties in these genetic analyses, and also in general in epidemiological research of OA is heterogeneity of the definition of the phenotype under study. Heterogeneity of the definition of the phenotype among different studies reduces power to find consistent associations in any disease4. Two working groups of HuGEnet and NCI-NHGRI have published recommendations for replication studies in genetic epidemiology studies5–7. One of their recommendations was to try to investigate the same or a very similar phenotype in replication studies. Specifically for OA, the American College of Rheumatology (ACR) criteria were developed to define clinical OA within a secondary care setting8 and the OARSI-OMERACT initiative proposed definitions for radiological progression of hip and knee OA9. The problem of heterogeneity in genetic association studies of OA has been highlighted10 and therefore standardized radiographic OA (ROA) phenotypes were used in our recent GWAS and subsequent meta-analysis1. However, symptomatic (SOA) and ROA phenotypes were both used within the same meta-analysis. For ROA, several grading systems exists, but the most widely and consistently used system is the Kellgren and Lawrence (K/L) grading system11. Among major cohort studies, K/L scores are interpreted differently, especially for the knee and hip, despite the fact that they all refer to the original description12–14.
In the current study, we will examine the effect of heterogeneity among symptomatic (SOA) and radiographic osteoarthritis (ROA) phenotypes and address the need for standardization of osteoarthritis phenotypes. We will provide recommendations for standardization of OA phenotypes.
We collected data for 28 studies currently involved in the TREAT-OA consortium on the following 9 items: 1) reference article, 2) study design, 3) ethnic origin, 4) country of origin, 5) joint site(s) studied 6) radiographic or symptomatic OA definition, 7) availability of age and/or BMI data, 8) percentage of women in the study and 9) availability of follow-up data. Table 1 describes the characteristics of all studies evaluated. A short description of each study is given in the supplementary data.
OA phenotypes can be categorized into symptomatic OA and radiographic OA, and this information was collected from all studies. Subsequently, we asked for the exact OA definition used in that particular study. For example, if a study used a K/L score and used the cut-off value defined by a summary grade of 2 or more to define OA cases, the exact description of a K/L of 2 was requested (e.g. definite osteophytes with possible JSN versus definite osteophyte(s) only) or a reference article was asked were the exact interpretation of the K/L score was given.
Within RSI radiographic features are scored separately for hip OA (such as osteophytes, sclerosis and joint space narrowing at the lateral, superior and axial site of the hip joint)15. In addition, total hip replacement and the presence of pain during the last month are recorded. To discover if differences in case definitions result in different association results, we created all hip OA case definitions used by studies of the consortium within RSI. Association analyses were performed to study the relationship between different OA definitions of the hip and age, gender and body mass index (BMI). One-way ANOVA was used to assess the relationship between hip OA and the clinical variables. The analyses were carried out using SPSS version 15.0.
Consensus on which ROA phenotype to use within the TREAT-OA consortium was based on the ROA definition as originally described by Kellgren and Lawrence and the feasibility of its use within each of the studies11. Total joint replacements (TJR) due to primary OA visible on radiographs are considered as OA. TJR due to fractures and other diseases were excluded as much as possible. After a consensus was reached between consortium members, the cohort studies either shared their data with our research group (Rotterdam Study) who standardized the definitions (data of TwinsUK, Chingford Study) or performed the standardization process themselves (other replication studies) if they were able and willing to standardize their ROA definition. The prevalence of OA was calculated by dividing the number of prevalent ROA cases over controls. Before standardization, controls were defined as the absence of OA, according to the definition used by each study, at the joint site studied. After standardization, controls were defined as the absence of OA, according to the standardized definition as described in the results section, at the joint site studied.
Since the start of the TREAT-OA consortium in 2008, the number of teams collaborating with the consortium has grown to include 28 teams participating as of April 2010. The studies originate from Europe, the United States of America and Asia. In 24 of the 28 studies (86%), the majority of subjects included are women (63% on average). With respect to genetic data, there are in total 11 studies with GWAS data, 2 studies in which part of the subjects have GWAS data and 15 studies without GWAS data. A short description of all studies involved in the consortium is given in the supplementary data.
In total, there were 11 studies using a symptomatic definition of OA and 15 studies with a radiographic definition. Two studies could not be classified as completely symptomatic or radiographic (SOA/ROA). For the GARP Study, subjects were recruited with clinical and radiographic confirmed OA at two or more joint sites among hand, spine, knee or hip16. If a subject was selected on the basis of hip and hand SOA for example, but the objective of a future study is knee OA, this subject might have ROA of the knee. We can neither call this pure knee ROA nor SOA since the subject was selected on the basis of having symptoms of generalized OA, but the assessment of knee OA is radiographic. In addition, the Finnish cases consist of 2 subsets with subjects affected with SOA and ROA of the hand and affected siblings with ROA of the hand17. In this study, we have chosen to classify the GARP Study and the Finnish cases as studies with SOA/ROA.
For knee OA, there are 14 studies using radiographic definitions of knee OA shown in Table 2a with a detailed description of the knee ROA definition. A total of 12 studies used the K/L score, of which 11 studies used a cut-off value of 2 to define knee ROA and 1 study used a more stringent cut-off of 3. Two studies, which are both high risk cohorts, used a definition of OA not according to a standard classification system. As is shown in Table 2a, four different interpretations are given for the K/L score of the knee considering a cut-off value of 2 although all studies used the original K/L atlas. In Table 2b–c, results are given for hand- and hip ROA respectively in a similar way as for knee ROA.
For hand ROA, most studies (7 out of 9) used the K/L score to define hand OA, with the exception of two studies16,17. The interpretation of this K/L score is the same for all these studies, but there are 4 different hand ROA definitions based on the number of joints included.
Hip ROA was defined by the (modified) Croft grade in 3 studies and by the K/L score in 4 studies. Also for hip ROA there is no consensus on the interpretation of the K/L score as 2 different interpretations are present among the studies. This includes both “definite JSN and a definite osteophyte” OR “one definite osteophyte”. The Croft grade cut off of 1 as a criterion for hip ROA, is defined as definite osteophytes and does not include JSN.
For knee OA, there are 10 studies using clinical definitions of knee OA, which are shown in Table 3a. In total, 4 of these 10 studies defined knee OA as ROA + symptoms, but the inclusion of patients was done in 4 different ways. For example, one study used a K/L score ≥ 2 (defined as one definite osteophyte) + medial joint space > 1 mm + pain to include patients, whilst another study used a K/L score ≥ 3 + symptomatic OA and treated on a regular basis. The other 6 studies included patients on the basis of total joint replacements due to primary OA or a combination of a TJR or ROA and clinical symptoms of OA.
In Table 3b–c, results are given for hand (n=2) and hip (n=8) SOA respectively in a similar way as for SOA of the knee. Also, these definitions differed for each study. In summary, hand SOA was defined by either ACR criteria or by patient records. Hip SOA was defined as a THR by 3 studies although the assessment was different for all 3 studies (i.e., based on hospital records versus based on the description of a rheumatologist). In addition, 2 studies defined SOA of the hip as symptoms of OA + ROA, but the definition of ROA is unclear and inclusion based on symptoms differs. Furthermore, there were 3 additional studies defining hip SOA again in another way (i.e., incident THR or either clinical records of SOA or a THR).
In Table 4, association results are given for the relationship between age, gender and BMI and different hip OA definitions. When hip OA was defined radiographically as “one definite osteophyte” subjects with hip OA were more frequently men compared to controls (mean difference of 10%, p=3×10−9), whilst subjects with a THR were more frequently women compared to controls (mean difference of 21%, p=0.001). When radiographic OA definitions were compared, we observed that hip ROA defined as “one definite osteophyte” were more frequently men compared to controls (p=3×10−9), whilst hip OA defined as “definite JSN and one definite osteophyte” was not associated with gender (p=0.22). When analyzing SOA, we did not observe clear differences in association results for the different definitions of SOA, but the number of cases for SOA is much lower than for ROA, therefore results should be taken with caution.
Consensus was reached for the knee and hip OA definition based on the ROA definition as originally described by Kellgren and Lawrence11 and at the feasibility within each of the studies. It was agreed that the knee ROA definition used within the TREAT-OA consortium is the original K/L score11 defined as “definite osteophytes and possible joint space narrowing” at the tibio-femoral (TF) joint. If studies did not score possible JSN as a separate feature, the definition used was: “at least 2 definite osteophytes OR one definite osteophyte plus definite JSN”. Hip ROA, which was the most poorly specified in the original scores, was defined as “at least definite joint space narrowing”. For hand ROA, consensus was not reached within the consortium, due to the fact that different studies graded different joints for hand OA, thus limiting the possibility to generate a single definition. As an alternative, thumb OA was put forward as an interesting phenotype to study, because of the high correlation with pain and disability18. Consensus was reached on a definition for thumb OA which is “at least one definite osteophyte (= original K/L grade ≥ 2) in either the left or right first carpometacarpal (CMC1) joint”.
In Table 5, the number of cases and controls for each study are given after standardization of phenotypes (both SOA and ROA). In total, there are 13,119 knee OA cases and 61,538 controls, 9,521 hip OA cases and 59,345 controls and 4,913 hand OA cases and 41,863 controls with DNA and phenotype data within the TREAT-OA consortium.
To evaluate the effect of standardization of the ROA phenotypes, we calculated the prevalence of knee and hip ROA in 8 Caucasian and 1 Japanese cohort study before and after standardization of the ROA definition. In Table 6, the mean age and BMI are shown for the 9 cohorts. The Framingham Osteoarthritis Study, The Hertfordshire Cohort Study, The Osteoporotic Fractures in Men Study, The Rotterdam study I, the ROAD Study and the Study of Osteoporotic Fractures are on average 14 years older than The Chingford Study, the Rotterdam Study III and TwinsUK. The result of the standardization of knee and hip OA phenotypes is shown in Figure 1. Results for the thumb OA phenotype are not shown since all studies use the same definition. The standardized hip OA definition is “at least definite JSN or a THR visible on the radiograph due to primary OA”. In the SOF and MrOS Study a minor adjustment was made and hip ROA was defined as: “at least medial JSN (grade≥3) or lateral JSN (grade≥2) or a THR visible on the radiograph due to primary OA”. The standardized knee ROA definition is “at least definite osteophytes and possible JSN or a TKR visible on the radiograph due to primary OA”.
Before standardization the prevalence of knee OA ranged between 10–55%, of hip OA between 2–33%. After standardization the prevalence of knee OA ranged from 8–25% and hip OA between 4–10%. When comparing cohorts with the same age range, the prevalence of knee ROA was 8–12% in the younger cohorts and 16–25% in the cohorts with subjects of an older age. To show that the differences in age are indeed the cause of the lower prevalence of knee OA in 3 cohort studies, we studied the prevalence of knee ROA in one relatively young and one old cohort with a wide age range, respectively TwinsUK and RS-I. The prevalence of knee ROA ranged from 10–15% in subjects aged 65 years and younger. In subjects aged 65 years and older, the prevalence ranged from 29–34% for the 2 studies.
A wide range of OA definitions were used in the 28 studies participating in the TREAT-OA consortium. Since heterogeneity in phenotype definitions will reduce power to find consistent associations, radiographic OA phenotypes were standardized within the consortium.
There are some research fields in which specific attention is given to phenotype definitions. This mainly concerns studies in the field of neuroscience (i.e., bipolar disorder or schizophrenia)19 and obesity20. In contrast, published research involving osteoarthritis, osteoporosis and heart disease does not usually discuss phenotype definitions. Our results showed that OA definitions should be standardized since association results differ when varying ROA and SOA definitions are used within the same study. In addition, it was recently shown that the ability to detect hip OA genetic associations is influenced by proper phenotyping21. We showed by standardizing of ROA phenotypes, that similar ROA prevalence's could be obtained.
For hip ROA, a distinction can be made between atrophic OA (presence of JSN without osteophytes), hypertrophic OA (presence of osteophytes without JSN) or a composite score (both JSN and osteophytes)22. It is known that these different forms of hip ROA have different risk factors23,24. In addition, atrophic OA shows to be a more progressive form of OA than hypertrofic OA25. Since some studies interpret a K/L score ≥ 2 as one definite osteophyte, whereas other studies interpret this as definite JSN and one definite osteophyte, a difference in association results would be expected. Although the standardized definition agreed upon by the consortium is based on JSN (hip ROA = at least definite JSN, with or without osteophytes), a majority of the subjects (78 and 80% in the Rotterdam Study-I and III, respectively) have both JSN and osteophytes. This definition can therefore also be seen as a composite score. Although less often used than the composite score of hip ROA, hypertrophic hip and atrophic hip ROA definitions should also be standardized. We suggest using “presence of at least one definite osteophyte at the femoral head without definite JSN” as preferred definition for hypertrofic OA and “definite JSN without the presence of any osteophytes at all locations” as atrophic OA which was also used in a previous study by Javaid et al.22.
It was difficult to reach consensus on the hand ROA definition, since different studies scored different joints. To overcome this problem, a subtype for clinically relevant OA was suggested within the consortium: thumb OA, associated with pain and disability18,26, will be used within the consortium. The definition of ROA of the thumb is “at least one definite osteophyte in either right or left CMC1 joint”.
We recommend for future studies on ROA to always specify the exact OA definition. A statement such as “we defined OA as a K/L ≥ 2” should be avoided or the interpretation of this K/L score should be given.
Since all studies involved in the consortium defined SOA differently, or at least assessed the OA status differently, it is likely that heterogeneity is a problem in studies on SOA. Standardization of SOA would in principle be possible if studies had pain, clinical assessment data for study subjects, as well as radiographic grade for the index joints, age, BMI, for both cases and controls. The design of some studies is such however that there is no radiographic characterization for cases and controls, which is necessary if SOA would be defined based on both symptoms and radiographs, and only a diagnosis of TJR for an indication of OA is present. These are extant studies and to collect homogenous SOA studies would require a huge investment of resources as well as time. However, there remains a lack of consensus and guidelines about how SOA should be assessed. For example, the American College of Rheumatology (ACR) defines signs of OA as stiffness <30 minutes, crepitus, bony tenderness, bony enlargement, no palpable warmth and pain in or around the joint. The presence of these traits in subjects over the age of 50 (preferably accompanied by radiographic evidence of OA) is commonly used in the design of randomized clinical trials (RCTs)27. But these criteria were developed in a clinic setting so the sensitivity and specificity of a diagnosis based on these criteria in a community or primary care settings, are as yet unknown.
Most of the SOA cases included in the TREAT-OA consortium are total joint replacement cases with a primary indication of OA. Although it is possible to define TJR as the main clinical outcome representative of severe symptomatic large joint OA in itself, as has been proposed for RCTs28, this might not be the best option. Recent studies on this topic have revealed considerable heterogeneity in the radiographic severity, functional disability and pain suffered by TJR candidates29. In addition, the pain and disability components among subjects undergoing TJR are significantly correlated with risk factors that also impact on ROA such as BMI, age, sex, whilst being poorly correlated with radiographic severity29,30. Further, not all patients with severe symptomatic OA can or are willing to get a TJR either because of lack of access to healthcare, or they may be afraid of surgery, or have co-morbidities that make them ineligible etcetera31. TJR patients are usually recruited in secondary care settings and might in some instances represent a non-random subset of severe symptomatic OA.
In summary, additional research is needed to reach consensus for in- and exclusion criteria and definitions of clinical/symptomatic OA studies. We suggest that more thought should be given to the establishment of clear guidelines for future research using symptomatic OA cohorts, as this would have implications not just for genetic studies, but also for the assessment of biomarkers, imaging and interventional studies.
Genome-wide association studies (GWAS) and meta-analyses have been1,32 and will continue to be performed within the TREAT-OA consortium in order to identify genes consistently associated with risk and progression of OA. Presently, there are few genes discovered for OA by means of GWAS, and this may be explained by heterogeneity of phenotypes and the limited sample size used in the discovery GWAS samples up to now. For example, in a previous GWAS, ROA and SOA definitions were used within one meta-analysis1. It has been shown before that ROA shows only modest correlation with clinical features of OA33,34. In addition, we showed in this study that the association between SOA and age, gender and BMI is different compared to ROA. Although the sample size would decrease using stratification methods, the statistical power might increase if there is a reduction in the heterogeneity in the phenotype definition. Therefore, we recommend that for future GWASs additional work is needed to standardize or stratify on ROA and SOA. Fortunately, in the TREAT-OA consortium studies on ROA have access to the source material and individual features of ROA are scored separately. This enables us to easily establish standardized phenotypes across cohorts.
Additionally, other phenotypes or possible predictors such as hypertrophic vs. atrophic forms of OA, joint shape, MRI based features, severe ROA (K/L≥3 versus K/L=0) or generalized OA may expand our definitions of the OA phenotypes and may increase the number of consistent associations in genetic studies. However, consensus among OA epidemiologist on OA phenotypes should be reached within the OA field, prior to the performance of these association studies.
In conclusion, standardization of radiographic OA phenotypes was carried out in the TREAT-OA consortium to reduce heterogeneity as much as possible. Standardization of symptomatic OA phenotypes, although desirable, was not possible due to the case-control study design of the studies. In the future, more precise OA phenotypes and stratification according to symptomatic and radiographic OA phenotypes are highly recommended.
This study is funded by the European Commission framework 7 programme TREAT-OA (grant 200800). The Rotterdam Study is supported by the Netherlands Organisation of Scientific Research NWO Investments (nr. 175.010.2005.011, 911-03-012), the Research Institute for Diseases in the Elderly (014-93-015; RIDE2), the Netherlands Genomics Initiative (NGI)/Netherlands Organisation for Scientific Research (NWO) (project nr. 050-060-810) and the Erasmus Medical Center and Erasmus University, Rotterdam. The deCODE study is funded by deCODE Genetics. The Academy of Finland, Finnish Ministry of Education and ORTON Research Institute, Invalid Foundation has supported the studies on Finnish OA cases and families. The Framingham Osteoarthritis Study is funded by the National Institutes of Health AR47785 and AG18393. The Oxford Study is supported by the Oxford NIHR Musculoskeletal Biomedical Research Unit. TwinsUK acknowledges financial support from the Wellcome Trust, the Department of Health via the National Institute for Health Research (NIHR) comprehensive Biomedical Research Centre award to Guy's & St Thomas' NHS Foundation Trust in partnership with King's College London and Arthritis Research Campaign. The Chingford Study is funded by the arc. The Hertfordshire Cohort Study is funded by the Medical Research Council (UK) and the Oxford NIHR Musculoskeletal Biomedical Research Unit. The Arthritis Research Council UK funded collection of some of the Nottingham OA study cases (grant 17661) and provided infrastructure support during the Nottingham OA study (grant 14581). the Estonian collection is funded by the Estonian Science Foundation grant No 5308, the Estonian Ministry of Social Affairs grants No 9.6-4/2035 and 12.1-5/597. The Swedish cohort studies were supported by the Swedish Research Council and Lund University. The GARP Study was supported by the Leiden University Medical Centre and the Dutch Arthritis Association. Pfizer Inc., Groton, CT, USA supported the inclusion of the GARP study.
The Osteoporotic Fractures in Men (MrOS) Study is supported by National Institutes of Health funding. The following institutes provide support: the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), the National Institute on Aging (NIA), the National Center for Research Resources (NCRR), and NIH Roadmap for Medical Research under the following grant numbers: U01 AR45580, U01 AR45614, U01 AR45632, U01 AR45647, U01 AR45654, U01 AR45583, U01 AG18197, U01-AG027810, R01 AR052000-01 A1, 2K24-AR04884, AR043052 and UL1 RR024140. DNA extraction for MrOS was supported by grant number R01-AR051124 from The National Institute of Arthritis and Musculoskeletal Diseases (NIAMS). The Study of Osteoporotic Fractures (SOF) is supported by National Institutes of Health funding. The National Institute on Aging (NIA) provides support under the following grant numbers: AG05407, AR35582, AG05394, AR35584, AR35583, R01 AG005407, R01 AG027576-22, 2 R01 AG005394-22A1, R01 AR052000-01 A1, 2K24-AR04884, AR043052 and 2 R01 AG027574-22A1. The authors are very grateful to all study participants, the staff from all studies and the participating physicians and pharmacists.
Grant supporter: European Commission framework 7 programme grant 200800 TREAT-OA and the Netherlands Genomics Initiative (NGI) grant 050-060-810 Netherlands Consortium of Healthy Ageing
Obtaining of funding: I Meulenbelt, E Slagboom, NK Arden, A Aromaa, A Carr, C Cooper, M Doherty, D Felson, A Gonzalez, A Hofman, S Ikegawa, Q Jiang, M Kloppenburg, UM Kujala, NE Lane, P Leino-Arjas, S Lohmander, MC Nevitt, HAP Pols, TD Spector, K Stefansson, A Tamm, A Tsezou, AG Uitterlinden, JM Wilkinson, N Yoshimura, AM Valdes, JBJ van Meurs.
Conflict of Interest statement: none declared.