Clinical phenotypes in patients with knee osteoarthritis: a study in the Amsterdam osteoarthritis cohort

Osteoarthritis and Cartilage 23 (2015) 544e549

Clinical phenotypes in patients with knee osteoarthritis: a study in the Amsterdam osteoarthritis cohort M. van der Esch y *, J. Knoop y, M. van der Leeden y z x, L.D. Roorda y, W.F. Lems k ¶, D.L. Knol x #, J. Dekker z x yy y Amsterdam Rehabilitation Research Center, Reade, Amsterdam, The Netherlands z VU University Medical Center, Department of Rehabilitation Medicine, Amsterdam, The Netherlands x VU University Medical Center, EMGO Institute for Health and Care Research, Amsterdam, The Netherlands k VU University Medical Center, Department of Rheumatology, Amsterdam, The Netherlands ¶ Reade, Department of Rheumatology, Amsterdam, The Netherlands # VU University Medical Center, Department of Epidemiology and Biostatistics, Amsterdam, The Netherlands yy VU University Medical Center, Department of Psychiatry, Amsterdam, The Netherlands

a r t i c l e i n f o

s u m m a r y

Article history: Received 15 August 2014 Accepted 8 January 2015

Objective: To identify and validate previously established phenotypes of knee osteoarthritis (OA) based on similarities in clinical patient characteristics. Methods: Knee OA patients (N ¼ 551) from the Amsterdam OA (AMS-OA) cohort provided data. Four clinical patient characteristics were assessed: upper leg muscle strength, body mass index (BMI), radiographic severity (Kellgren/Lawrence [KL] grade), and depressive mood (the Hospital Anxiety and Depression Scale [HADS] questionnaire). Cluster analysis was performed to identify the optimal number of phenotypes. Differences in clinical characteristics between the phenotypes were analyzed with ANOVA. Results: Cluster analysis identified five phenotypes of knee OA patients: “minimal joint disease phenotype”, “strong muscle strength phenotype”, “severe radiographic OA phenotype”, “obese phenotype”, and “depressive mood phenotype”. Conclusions: Among patients with knee OA, five phenotypes were identified based on four clinical characteristics. To a high degree, the results are a replication of earlier findings in the OA Initiative, indicating that these five phenotypes seem a stable, valid, and clinically relevant finding. © 2015 Osteoarthritis Research Society International. Published by Elsevier Ltd. All rights reserved.

Keywords: Osteoarthritis Knee [Mesh] Muscle strength [Mesh] Phenotype [Mesh] Obesity [Mesh] Depressive disorder [Mesh]

Introduction Knee osteoarthritis (OA) is a heterogeneous disease with distinct characteristics during the various stages of disease progression1,2. It has been hypothesized that knee OA consists of different subgroups or phenotypes3. To identify subgroups or phenotypes, several approaches are available, focusing on etiological and risk factors (metabolic, traumatic, inflammatory, and subchondral bone turnover)4, focusing on genetic factors5, or focusing on clinical characteristics6. The latter approach seems particularly relevant, as the ultimate goal of identifying clinical phenotypes is that clinicians

* Address correspondence and reprint requests to: M. van der Esch, Amsterdam Rehabilitation Research Center, Reade, P.O. Box 58271, 1040 HG Amsterdam, The Netherlands. Tel: 31-20-589-62-91; Fax: 31-20-589-63-16. E-mail address: [email protected] (M. van der Esch).

can identify and manage subgroups of knee OA patients in their practices6. There is, however, a paucity of studies identifying clinical, homogenous knee OA subgroups. Most studies focused on a single clinical characteristics such as pain7, knee malalignment8, and race and gender9. Only one recent study from our own group based on data of the OA Initiative (OAI) included several clinical characteristics at the same time10. In the last-mentioned study, five clinically relevant phenotypes were identified with data from the open-population-based “progression subcohort” of the OAI cohort10. These five phenotypes were “minimal joint disease phenotype”, “strong muscle phenotype”, “non-obese and weak muscle phenotype”, “obese and weak muscle phenotype”, and “depressive phenotype”. These phenotypes were based on four patient characteristics i.e., upper leg muscle strength, body mass index (BMI), radiographic OA, and depressive mood. The four clinical characteristics are regularly

http://dx.doi.org/10.1016/j.joca.2015.01.006 1063-4584/© 2015 Osteoarthritis Research Society International. Published by Elsevier Ltd. All rights reserved.

M. van der Esch et al. / Osteoarthritis and Cartilage 23 (2015) 544e549

assessed and relevant to clinical practice6, and have a major impact on disease progression and clinical outcome6. In clinical practice, patients differ highly on these characteristics, necessary for cluster analysis11. For reasons of robustness, there is a need to determine whether the five phenotypes can be replicated in another sample. The present study aimed to identify and validate previously established phenotypes of knee OA based on similarities in clinical patient characteristics, i.e., upper leg muscle strength, BMI, radiographic OA, and depressive mood in patients with knee OA from a clinical population. Methods Patients and methods The Amsterdam OA (AMS-OA) cohort contains patients with OA of the knee and/or hip who have been referred to a secondary care outpatient rehabilitation center (Reade, Center for Rehabilitation and Rheumatology, Amsterdam, the Netherlands)12. The examination protocol involves assessments by rheumatologists, radiologists, and rehabilitation physicians. Demographic, clinical, radiographic, biomechanical, and psychosocial factors related to OA were assessed. Total knee replacement, rheumatoid arthritis, or any other form of inflammatory arthritis (i.e., crystal arthropathy or septic arthritis) were exclusion criteria. In the present study, 551 patients from the AMS-OA cohort with a unilateral or bilateral diagnosis of knee OA according to the American College of Rheumatology (ACR) were included13. The group of patients was defined as established knee OA due to the combination of confirmed diagnosis of knee OA according to the ACR criteria and the presence of knee OA related symptoms making a visit for secondary care necessary. All patients provided written, informed consent according to the Declaration of Helsinki. The study was approved by the Slotervaart Hospital/Reade Institutional Review Board. Clustering variables Muscle strength was used in analyses as the mean score of left and right lower limb isokinetic muscle strength (quadriceps and hamstring strength in Newton meters), because the correlation between left and right muscle strength for extension and flexion muscle strength was r ¼ .64 and r ¼ .72 (P < .001), respectively14. BMI was calculated by dividing mass (in kilograms) by squared height (in meters). Radiographic severity of knee OA (ROA) was scored by an overall severity grade (Kellgren and Lawrence grade (KL))15 ranging from 0 to 4, based on weight-bearing fixed-flexion knee radiograph16. The inter-rater agreement of scoring the radiographic features was good17. In analysis the most severely damaged knee was used. Depressive mood was based on the Hospital Anxiety and Depression Scale (HADS) questionnaire, which is one of the most frequently used questionnaires for depressive mood experiences with very strong psychometric properties18. The 7-item subscale measures depressive mood on a 4-point response scale (from 0, no symptoms, to 3, maximum symptoms), with possible scores for the subscale ranging from 0 to 21. The HADS is a valid and reliable questionnaire for detecting mood disorder in people with OA19. Statistical analysis Identification of subgroups K-means (or its equivalent “non-hierarchical”) cluster analysis was used to identify homogeneous subgroups of knee OA patients (i.e., phenotypes). Cluster analysis is a data analysis technique aimed at grouping entities on the basis of their similarity with

545

respect to selected variables. Members of the resulting groups are as similar as possible to others within their group (high withingroup homogeneity) and as different as possible to those in other groups (low between-group homogeneity). This analysis implicates that members are permitted to be member of a single group (i.e., phenotype)20. The following clinical characteristics were included in cluster analysis: mean of left and right lower limb muscle strength (quadriceps and hamstring strength), BMI, KL grade of the most severely damaged knee, and score on the HADS questionnaire (depressive mood) (see Table I for more information). Since cluster analysis techniques are sensitive to outliers, outlier cases should be identified and removed from the dataset before cluster analysis21. The FASTCLUS procedure of the SAS package22 was used. An estimation of the optimal number of clusters (between three and eight clusters) was based on the Calinski and Harabasz pseudo F-statistic23, which is the best performing clustering criterion according to Milligan and Cooper24. Additionally, a second clustering criterion was used, as recommended by Milligan and Cooper24. This second criterion was the Beale's F-ratio, which tests the hypothesis of the existence of an additional cluster vs the already detected clusters25. The addition of one cluster in the cluster solution should continue until the hypothesis e tested by Beale's F ratio e was rejected25. This analysis was continued till the highest pseudo F value was found, considering that this was the most adequate number of clusters. Spearman's correlation coefficients were computed to examine the bivariate associations between the four clinical characteristics: muscle strength, KL grade, BMI and HADS scores. Differences of muscle strength, KL grade, BMI, and HADS scores between the clusters (phenotypes) were evaluated with analysis of variance (ANOVA) and a Bonferroni post-hoc analysis. Results No outliers were found and therefore a total of 551 patients were included in the study. The characteristics of the study group are displayed in Table I. The mean age was 61.7 (SD ¼ 8.8) years and 68 % of the patients were women. Knee symptoms were present for more than 5 years in 72% of the patients. Radiographic evidence of knee OA (i.e., KL grade  2) was found in 63% of the patients. Mean muscle strength (Nm) and HADS-depressive mood were 70.26 (SD ¼ 35.81) and 4.81 (SD ¼ 3.47), respectively. The mean values of Table I Characteristics of study group (mean and standard deviation, unless otherwise stated) Characteristic

Mean (SD)

N Age, years Gender (% female) Duration of knee symptoms: Less than 5 years (%) More than 5 years (%) Bilateral knee pain (% yes) Co-occurrence of hip pain (% yes) Number of comorbidities (median and IQR) Muscle strength, Nm Body mass index, kg/m2 Obesity (BMI  30.0) (%) Radiographic knee OA: K/L score 0 (%) K/L score 1 (%) K/L score 2 (%) K/L score 3 (%) K/L score 4 (%) HADS depressive mood score

551 61.7 (8.8) 68

IQR ¼ interquartile range.

28 72 67 15 2 (1e3) 70.3 (35.8) 30.4 (6.2) 46 5 32 27 22 14 4.8 (3.5)

546

M. van der Esch et al. / Osteoarthritis and Cartilage 23 (2015) 544e549

Table II Clustering criteria for solution of three or more clusters (five-cluster solution in bold) Number of clusters

Calinski and Harabasz F 23

Cluster comparison

Beale's F (df*, dfy)25

Beale's Fp-value26

3 4 5 6

189.23 231.00 232.42 207.13

e 3 vs 4 4 vs 5 5 vs 6

e 1.46 (43, 274) 1.66 (29, 244) 0.75 (22, 223)

e 0.040 0.030 0.782

Bold formatted numbers showed a significant improvement of the model's fit of a five-cluster solution compared to a four-cluster and a six-cluster solution. * numerator degrees of freedom. y denominator degrees of freedom.

muscle strength were not significantly different between uni- and bilateral radiographic OA: mean (SD) 74.48 (38.68) Nm vs 69.16 (35.07), respectively; P ¼ .193. A BMI  30 was found in 46% of the study group. Spearman's correlation coefficients of the four clinical characteristics showed weak associations between muscle strength and depressive mood (r ¼ .17, P < .001) and between KL grade and BMI (r ¼ .12, P ¼ .006). No significant associations were found between KL grade and muscle strength (r ¼ .06; P ¼ .172) and KL grade and depressive mood (r ¼ .07; P ¼ .103). No significant associations were found between muscle strength and BMI (r ¼ .06; P ¼ .147), and between BMI and depressive mood (r ¼ .03; P ¼ .441). The Calinski and Harabasz pseudo F-statistics and Beale's F-ratio statistics are shown in Table II. Values of both clustering criteria were the highest in a five-cluster solution. Both F-statistics revealed that a five-cluster solution showed a significant improvement of the model's fit compared to a four-cluster solution, while a sixcluster solution did not show a significant improvement of the model's fit compared to a five-cluster solution. Therefore, the most adequate number of phenotypes was five. The five identified phenotypes were named 1. ”minimal joint disease phenotype”, 2. “strong muscle strength phenotype”, 3.”severe radiographic OA phenotype”, 4. “obese phenotype”, and 5.”depressive mood phenotype”. The four clinical characteristics (i.e., upper leg muscle strength, BMI, radiographic severity grade, and depressive mood) of the five phenotypes and total study group are shown in Table III. The differences between the five phenotypes on each clinical characteristic are presented in Fig. 1 and displayed in Table III. The between-subgroup differences were significant for muscle

strength, F (4, 539) ¼ 192.87, P < .001. Post-hoc analyses (Bonferroni) showed that muscle strength in phenotype 2 was significantly greater than in phenotypes 1, 3, 4, and 5. The between-subgroup differences were not significant for phenotypes 1, 3, 4, and 5. The between-subgroup differences were significant for BMI, F (4, 545) ¼ 159.75, P < .001. Post-hoc analysis showed that BMI in phenotype 4 was significantly higher than in phenotypes 1, 2, 3, and 5, while BMI in phenotypes 1, 2, and 5 were not significantly different. BMI of phenotype 3 was significantly higher than phenotype 1, but not higher than BMI of phenotypes 2 and 5. The between-subgroup differences were significant for KL grade, F (4, 537) ¼ 155.01, P < .001. Post-hoc analyses showed that the KL grade in phenotype 3 was significantly higher than in phenotypes 1, 2, 4, and 5. The between-subgroup differences were not significant for phenotypes 1 and 5 and for phenotypes 2 and 4. The KL grade of phenotypes 2 and 3 was significantly different from the KL grade of phenotypes 1 and 5. The between-subgroup differences were significant for the HADS score, F (4,514) ¼ 114.07, P < .001. Post-hoc analysis showed that the HADS score in phenotype 5 was significantly higher than in phenotypes 1, 2, 3, and 4. The HADS score in phenotype 4 was also significantly higher than in phenotypes 1, 2, and 3, while the HADS scores in phenotypes 1, 2, and 3 were not significantly different.

Discussion The aim of the study was to identify phenotypes based on similarities in four clinical characteristics in a group of patients with established knee OA. Cluster analysis showed five phenotypes: “minimal joint disease phenotype”, “strong muscle strength phenotype”, “severe radiographic OA phenotype”, “obese phenotype”, and “depressive mood phenotype”. We used the same four clinical characteristics (i.e., upper leg muscle strength, BMI, radiographic severity grade, and depressive mood) as in our previous study on the OAI data10. These four characteristics were not associated, except for weak associations between muscle strength and depressive mood and between KL grade and BMI. Therefore, the association between clinical characteristics did not add statistical noise to the analysis and did not corrupt the cluster structure25. The present study revealed a fivecluster solution, indicating that the optimal number of clusters in the AMS-OA data as well as in the OAI data was five10. It can be claimed with more certainty that the finding of five clinical

Table III Clinical characteristics of five identified phenotypes and total study group in the AMS-OA cohort

n Isokinetic muscle strength, Nm BMI, kg/m2 K/L 0e1 K/L 2 K/L 3 K/L 4 Depressive mood (HADS)

1: Minimal joint disease phenotype

2: Strong muscle strength phenotype

3: Severe radiographic OA phenotype

4: Obese phenotype

5: Depressive phenotype

Total study group

154 57.0 ± 20.8

114 124.3 ± 23.1*

116 53.0 ± 22.8

81 61.2 ± 27.3

86 55.3 ± 22.8

551 70.3 ± 35.8

27.5 ± 4.0 69% 31% 0% 0% 3.3 ± 2.0

28.8 ± 3.8 22% 35% 31% 12% 3.4 ± 2.5

29.6 ± 4.3 0% 0% 55%z 45%z 4.0 ± 2.7

41.3 ± 5.0y 31% 35% 23% 11% 5.2 ± 3.4

28.5 ± 4.4 56% 37% 7% 0% 10.2 ± 2.3x

30.4 ± 6.2 37% 27% 22% 14% 4.8 ± 3.5

K/L ¼ Kellgren/Lawrence grade for radiographic severity; BMI ¼ body mass index; HADS ¼ Hospital Anxiety and Depression Scale. Bold formatted numbers are significantly different from other phenotypes and characterize the phenotype. * Muscle strength was significantly different from phenotypes 1, 3, 4, 5. No significant difference between phenotypes 1, 3, 4, 5. y BMI was significantly different from phenotypes 1, 2, 3 and 5. Phenotype 1, 2 and 5 were not significantly different, phenotype 3 was significantly different from phenotype 1. z KL was significantly different from phenotypes 1, 2, 4 and 5. No significant difference between phenotypes 1 and 5 and between phenotypes 2 and 4. x Depressive mood was significantly different form phenotypes 1e4. Phenotype 4 was significantly different from phenotypes 1e3. No significant difference between phenotypes 1e3.

M. van der Esch et al. / Osteoarthritis and Cartilage 23 (2015) 544e549

547

Fig. 1. Comparison of the five phenotypes on each clinical characteristic. A. Isokinetic muscle strength (Nm). Phenotype 2 was significantly different (P < .001) from phenotypes 1, 3, 4, and 5. B. BMI. Phenotype 4 was significantly different (P < .001) from phenotypes 1, 2, 3, and 5. Phenotype 3 was significantly different from 1, 2 and 5 (P < .05). C. Radiographic OA: Kellgren/Lawrence (0e4). Phenotype 3 was significantly different (P < .001) from phenotypes 1, 2, 4, and 5. Phenotypes 2 and 4 were significantly different from 1 to 5 (P < .001). D. HADS questionnaire: depressive mood). Phenotype 5 was significantly different (P < .001) from phenotypes 1, 2, 3, and 4. Phenotype 4 was significantly different from phenotypes 1, 2, and 3 (P < .05). Phenotypes: 1 ¼ “minimal joint disease”, 2 ¼ “strong muscle strength”, 3 ¼ “severe radiographic OA”, 4 ¼ “obese”, 5 ¼ “depressive mood”. Error bars: 95% CI.

phenotypes, based on the four characteristics, is a valid result: not a finding by chance, but a definitive, repeatable result. The clinically heterogeneous population of knee OA patients seems to exist of five homogeneous subgroups of patients. The clinical characteristics can be used to interpret these five phenotypes. The “minimal joint disease” phenotype is a subgroup without prominent clinical or radiological characteristics, despite the symptoms necessitating a visit to a second care rehabilitation center and despite the clinically diagnosed OA. The “strong muscle strength” phenotype is a subgroup with high muscle strength as most prominent characteristic, and relatively more radiographic severity. The “severe radiographic OA” phenotype is a subgroup with degeneration of cartilage and bone as most prominent characteristic. The “obese” phenotype is a subgroup with high BMI as most prominent characteristic, and relatively severe radiographic OA and relatively high depressive mood (HADS). Finally, the “depressive mood phenotype” is a subgroup with depressive mood as the most prominent characteristic, irrespective of the other three characteristics.

The findings are to a large extent similar to what was found by our research group in the OAI cohort. Both studies had similar “minimal joint disease”, “strong muscle strength”, and “depressive mood” phenotypes. In the OAI study, the “severe radiographic OA” phenotype was named the “non-obese and weak muscle strength” phenotype, while the “obese” phenotype was named the “obese and weak muscle strength” phenotype. The renaming of these two phenotypes (“non-obese and weak muscle strength” into “severe radiographic OA”, and “obese and weak muscle strength” into “obese”) was respectively based on the more pronounced appearance of radiographic OA and BMI characteristics in the current study. The difference in phenotypes might be a real difference, but could also be explained by a change in interpretation of the data, primarily based on advancing insight. Taking into account the results of the current study, the data of the OAI study were reanalyzed. We noticed that the “non-obese and weak muscle” phenotype of the OAI study also showed more extent of OA radiographic severity10, and that in the “obese and weak muscle” phenotype, muscle strength was not significantly different from the

548

M. van der Esch et al. / Osteoarthritis and Cartilage 23 (2015) 544e549

other phenotypes10. Therefore, these two phenotypes could have been labeled as “severe radiographic OA” and “obese” phenotype as well. In general, knee OA patients will benefit from muscle strengthening exercises26,27. Based on the results, however, we believe that segregating patients on four clinical characteristics into five phenotypes may help in finding the best targeted personalized care in knee OA. Theoretically, patients within the “minimal joint disease” phenotype may need a treatment approach including education, pain coping skill training, and self-management, as has been used in chronic pain conditions28,29. For patients within the “strong muscle strength” phenotype, appropriate education and pain medication might be a targeted solution30e32, and muscle strengthening exercises would not be the treatment of first choice. Although the influence of structural degradation in the indication for surgery in OA is controversial, the “severe radiographic OA” phenotype might help physicians and surgeons in making decisions for surgery33. Patients of the “obese” phenotype may need dietary changes, alongside to exercise therapy34. Additionally, attention for a slightly increased depressive mood might also be included in the treatment approach for the “obese” phenotype. Finally, patients of the “depressive mood” phenotype may benefit from treatment for their depressive mood, alongside their OA specific management35,36. These speculations on targeted personalized care in knee OA need to be tested in randomized controlled trials in patients with predefined phenotypes before being applied in clinical care. The choice of clinical characteristics included in the cluster analysis may affect the results. We used upper leg muscle strength, BMI, severe radiographic OA, and depressive mood to identify clinical phenotypes. Other phenotypes could have been found had other clinical characteristics been included in the analysis. Biomechanical characteristics (e.g., knee alignment, joint stability), bone involvement, clinical synovitis, or psychosocial characteristics (anxious mood, self-efficacy) could have been included in cluster analyses, potentially yielding other phenotypes4. The four included clinical characteristics, however, are easy to diagnose, with simple assessment measurements6. In identifying meaningful phenotypes the compartmental difference between tibiofemoral OA and patellofemoral OA might also be important37. The KL grading depends on scoring the radiographic features of the tibiofemoral compartments and not on scoring of the features of the patellofemoral compartment. Future phenotype studies should account for the difference in compartmental grading. A potential limitation of the study is the cross-sectional design: longitudinal follow-up of phenotypes is interesting for examining the stability of the clusters found over time. Longitudinally, patients might change from phenotype, showing changes in disease trajectories. Knowledge on the change in membership of a phenotype over time is important for causal reasons and has implications for personalized care. Therefore, longitudinal studies are needed to test the stability of the clusters over time and the membership of patients to these clusters over time. Furthermore, whereas cluster analysis is an objective means for the subgrouping of patients with knee OA, its potential shortcomings must be kept in mind. One of the shortcomings is that no accepted criterion to determine the number of clusters and the sample size of patients exists. In our study, however, both the pseudo F-statistic and Beale's F-ratio statistic indicated that the five-cluster solution was the best solution. Regarding minimum sample size of patients, there is no generally accepted rule of thumb. The objective of cluster analysis is to construct internally homogeneous and clearly distinct clusters. We used a limited set of four clinical characteristics in a population of 551 patients: this resulted in internally homogeneous and clearly distinctable clusters. Finally, a limitation is the setting of patient recruitment of this cohort in one secondary care rehabilitation

center. Large cohorts, containing detailed information on patients recruited in multiple settings are costly to establish. An option could be to merge multiple cohorts obtained in different settings on comparable clinical characteristics, to ensure representation of different phenotypes of patients. In conclusion, among patients with knee OA, five phenotypes were identified based on four clinical characteristics. To a high degree, the results are a replication of earlier findings in the OAI. The five phenotypes seem a stable, valid, and clinically relevant finding. Contributors All authors contributed substantially to the conception and design of the study and interpreted data. LDR collected data. MvdE, JK, and DLK analyzed data. MvdE, JK, MvdL, DLK, LDR, WFL, and JD prepared the first draft of the report. All authors contributed to the report and approved the final version. Funding No funding. Competing interests The authors have no conflicts of interest to disclose. Statement No commercial party having a direct financial interest in the results of the research supporting this article has or will confer on the authors or on any organization with which the authors are associated. References 1. Karsdal MA, Christiansen C, Ladel C, Henriksen K, Kraus VB, Bay-Jensen AC. Osteoarthritis e a case for personalized health care? Osteoarthritis Cartilage 2014;22(1):7e16. 2. Driban JB, Sitler MR, Barbe MF, Balasubramanian E. Is osteoarthritis a heterogeneous disease that can be stratified into subsets? Clin Rheumatol 2010;29(2):123e31. 3. Bijlsma JW, Berenbaum F, Lafeber FP. Osteoarthritis: an update with relevance for clinical practice. Lancet 2011;377(9783): 2115e26. 18. 4. Bierma-Zeinstra SM, Verhagen AP. Osteoarthritis subpopulations and implications for clinical trial design. Arthritis Res Ther 2011;13(2):213. 5. van Meurs JB, Uitterlinden AG. Osteoarthritis year 2012 in review: genetics and genomics. Osteoarthritis Cartilage 2012;20(12):1470e6. 6. Felson DT. Identifying different osteoarthritis phenotypes through epidemiology. Osteoarthritis Cartilage 2010;18(5): 601e4. 7. Hern andez-Molina G, Neogi T, Hunter DJ, Niu J, Guermazi A, Reichenbach S, et al. The association of bone attrition with knee pain and other MRI features of osteoarthritis. Ann Rheum Dis 2008;67(1):43e7. 8. Sharma L, Song J, Dunlop D, Felson D, Lewis CE, Segal N, et al. Varus and valgus alignment and incident and progressive knee osteoarthritis. Ann Rheum Dis 2010;69(11):1940e5. 9. Nelson AE, Golightly YM, Renner JB, Schwartz TA, Kraus VB, Helmick CG, et al. Brief report: differences in multi joint symptomatic osteoarthritis phenotypes by race and sex: the Johnston County Osteoarthritis Project. Arthritis Rheum 2013;65(2):373e7.

M. van der Esch et al. / Osteoarthritis and Cartilage 23 (2015) 544e549

10. Knoop J, van der Leeden M, Thorstensson CA, Roorda LD, Lems WF, Knol DL, et al. Identification of phenotypes with different clinical outcomes in knee osteoarthritis: data from the osteoarthritis initiative. Arthritis Care Res (Hoboken) 2011;63(11):1535e42. 11. Everitt BS, Landau S, Leese M. Cluster Analysis. 4th edn. London: Arnold; 2001. 12. van der Esch M, Knoop J, van der Leeden M, Voorneman R, Gerritsen M, Reiding D, et al. Self-reported knee instability and activity limitations in patients with knee osteoarthritis: results of the Amsterdam osteoarthritis cohort. Clin Rheumatol 2012;31(10):1505e10. 13. Altman R, Asch E, Bloch D, Bole G, Borenstein D, Brandt K, et al. Development of criteria for the classification and reporting of osteoarthritis. Classification of osteoarthritis of the knee. Diagnostic and Therapeutic Criteria Committee of the American Rheumatism Association. Arthritis Rheum 1986;29:1039e49. 14. van der Esch M, Steultjens MPM, Harlaar J, Knol DL, Lems W, Dekker J. Joint proprioception, muscle strength and functional ability in patients with osteoarthritis of the knee. Arthritis Rheum 2007;57:787e93. 15. Altman RD, Hochberg M, Murphy Jr WA, Wolfe F, Lequesne M. Atlas of individual radiographic features in osteoarthritis. Osteoarthritis Cartilage 1995;3(Suppl A):3e70. 16. Buckland-Wright C. Protocols for precise radio-anatomical positioning of the tibiofemoral and patellofemoral compartments of the knee. Osteoarthritis Cartilage 1995;3(Suppl A): 71e80. 17. van der Esch M, Knol DL, Schaffers IC, Reiding DJ, van Schaardenburg D, Knoop J, et al. Osteoarthritis of the knee: multicompartmental or compartmental disease? Rheumatology (Oxford) 2014;53(3):540e6. 18. Axford J, Butt A, Heron C, Hammond J, Morgan J, Alavi A, et al. Prevalence of anxiety and depression in osteoarthritis: use of the hospital anxiety and depression scale as a screening tool. Clin Rheumatol 2010;29(11):1277e83. 19. Bjelland I, Dahl AA, Haug TT, Neckelmann D. The validity of the hospital anxiety and depression scale. An updated literature review. J Psychosom Res 2002;52(2):69e77. 20. Clatworthy J, Buick D, Hankins M, Weinman J, Horne R. The use and reporting of cluster analysis in health psychology: a review. Br J Health Psychol 2005;10(Pt 3):329e58. 21. Jain AK, Dubes RC. Algorithms for Clustering Data. 1st edn. Englewood Cliffs (NJ): Prentice-Hall, Inc; 1988. 22. SAS Institute Inc. SAS/STAT® 9.2 User's Guide. 2nd edn. Cary, NC: SAS Institute Inc; 2009. 23. Calinski RB, Harabasz J. A dendrite method for cluster analysis. Commun Stat 1974;3:1e27. 24. Milligan GW, Cooper MC. An examination of procedures for determining the number of clusters in a data set. Psychometrika 1985;50(2):159e79. 25. Beale EM. Euclidean cluster analysis. Bull Int Stat Inst 1969;43: 92e4.

549

26. Fransen M, McConnell S. Exercise for osteoarthritis of the knee. Cochrane Database Syst Rev 2008;8(4):CD004376. 27. Knoop J, Dekker J, van der Leeden M, van der Esch M, Thorstensson CA, Gerritsen M, et al. Knee joint stabilization therapy in patients with osteoarthritis of the knee: a randomized, controlled trial. Osteoarthritis Cartilage 2013;21(8): 1025e34. 28. Kroon FP, van der Burg LR, Buchbinder R, Osborne RH, Johnston RV, Pitt V. Self-management education programmes for osteoarthritis. Cochrane Database Syst Rev 2014;15(1): CD008963. 29. Bennell KL, Ahamed Y, Bryant C, Jull G, Hunt MA, Kenardy J, et al. A physiotherapist-delivered integrated exercise and pain coping skills training intervention for individuals with knee osteoarthritis: a randomised controlled trial protocol. BMC Musculoskelet Disord 2012;24(13):129. 30. Coleman S, Briffa NK, Carroll G, Inderjeeth C, Cook N, McQuade J. A randomised controlled trial of a selfmanagement education program for osteoarthritis of the knee delivered by health care professionals. Arthritis Res Ther 2012;14(1):R21. 31. Driban JB, Boehret SA, Balasubramanian E, Cattano NM, Glutting J, Sitler MR. Medication and supplement use for managing joint symptoms among patients with knee and hip osteoarthritis: a cross-sectional study. BMC Musculoskelet Disord 2012;13:47. 32. Snijders GF, van den Ende CH, van den Bemt BJ, van Riel PL, van den Hoogen FH, den Broeder AA, NOAC study group. Treatment outcomes of a Numeric Rating Scale (NRS)-guided pharmacological pain management strategy in symptomatic knee and hip osteoarthritis in daily clinical practice. Clin Exp Rheumatol 2012;30(2):164e70. 33. Maillefert JF, Roy C, Cadet C, Nizard R, Berdah L, Ravaud P. Factors influencing surgeons' decisions in the indication for total joint replacement in hip osteoarthritis in real life. Arthritis Rheum 2008;59(2):255e62. 34. Jenkinson CM, Doherty M, Avery AJ, Read A, Taylor MA, Sach TH, et al. Effects of dietary intervention and quadriceps strengthening exercises on pain and function in overweight people with knee pain: randomized controlled trial. BMJ 2009;18(339):B3170. 35. Gleicher Y, Croxford R, Hochman J, Hawker G. A prospective study of mental health care for comorbid depressed mood in older adults with painful osteoarthritis. BMC Psychiatry 2011;11:147. 36. Iyengar RL, Gandhi S, Aneja A, Thorpe K, Razzouk L, Greenberg J, et al. NSAIDs are associated with lower depression scores in patients with osteoarthritis. Am J Med 2013;126(11). 1017.e11-8. 37. Hinman RS, Crossley KM. Patellofemoral joint osteoarthritis: an important subgroup of knee osteoarthritis. Rheumatology (Oxford) 2007;46(7):1057e62.