A systematic review of the relationship between subchondral bone features, pain and structural pathology in peripheral joint osteoarthritis

Introduction Bone is an integral part of the osteoarthritis (OA) process. We conducted a systematic literature review in order to understand the relationship between non-conventional radiographic imaging of subchondral bone, pain, structural pathology and joint replacement in peripheral joint OA. Methods A search of the Medline, EMBASE and Cochrane library databases was performed for original articles reporting association between non-conventional radiographic imaging-assessed subchondral bone pathologies and joint replacement, pain or structural progression in knee, hip, hand, ankle and foot OA. Each association was qualitatively characterised by a synthesis of the data from each analysis based upon study design, adequacy of covariate adjustment and quality scoring. Results In total 2456 abstracts were screened and 139 papers were included (70 cross-sectional, 71 longitudinal analyses; 116 knee, 15 hip, six hand, two ankle and involved 113 MRI, eight DXA, four CT, eight scintigraphic and eight 2D shape analyses). BMLs, osteophytes and bone shape were independently associated with structural progression or joint replacement. BMLs and bone shape were independently associated with longitudinal change in pain and incident frequent knee pain respectively. Conclusion Subchondral bone features have independent associations with structural progression, pain and joint replacement in peripheral OA in the hip and hand but especially in the knee. For peripheral OA sites other than the knee, there are fewer associations and independent associations of bone pathologies with these important OA outcomes which may reflect fewer studies; for example the foot and ankle were poorly studied. Subchondral OA bone appears to be a relevant therapeutic target. Systematic review PROSPERO registration number: CRD 42013005009 Electronic supplementary material The online version of this article (doi:10.1186/s13075-015-0735-x) contains supplementary material, which is available to authorized users.


Introduction
Osteoarthritis (OA), the most common form of arthritis, is a major cause of chronic pain and disability. OA confers a huge burden on both individuals and health economies [1,2]. There are currently no licensed disease-modifying osteoarthritis drugs (DMOADs) but ideally these should both inhibit structural progression and improve symptoms and/or function [3,4]. While hyaline cartilage loss is the hallmark pathology, clinical OA usually involves multiple tissues. Describing the relationships of these tissues with structural progression and symptoms may identify potential tissue targets.
The subchondral bone in particular is intimately associated with hyaline cartilage and therefore a tissue of great potential interest. Conventional radiographs are known to be relatively insensitive to the structural features of OA [5], in part because they do not assess three-dimensional (3D) bone structure [6]. A number of non-conventional radiographic imaging modalities accurately demonstrate in vivo subchondral bone pathological changes, including magnetic resonance imaging (MRI), computed tomography (CT), dual-energy x-ray absorptiometry (DXA), scintigraphy and positron emission tomography (PET) [5,[7][8][9][10][11][12][13]. Hunter and colleagues found a moderate association between bone marrow lesions (BMLs), structural progression and longitudinal change in pain in a systematic review focused on MRI biomarkers and knee OA [7]. In another systematic review Kloppenburg and colleagues examined associations between MRI features and knee pain, but not structural pathology [14].
We therefore wished to comprehensively review the literature on subchondral bone structure assessed with all non-conventional radiographic imaging modalities, examining the common sites of peripheral OA and describing the relationships between imaging-detected subchondral bone features and joint replacement, structural progression and pain.

Systematic literature search
A systematic literature search of Medline (from 1950), EMBASE (from 1980) and the Cochrane library databases until September 2014 was performed. A full description of the search terms used is recorded in Additional file 1: Table S1. An abbreviation of the full search terms used was 'knee, hip, hand, foot and ankle' and 'osteoarthritis' and 'subchondral bone' manifestations of OA ('bone marrow lesion' , 'osteophyte' , 'bone cyst' , 'bone area' , 'bone shape' , 'bone attrition' , bone morphometry and mineral density) and 'MRI' or 'CT' or 'DXA' or 'scintigraphy' or 'PET'. The search term 'bone shape' was not restricted to non-conventional radiographic imaging. The final search was restricted to humans. There was no language restriction and abstracts were not excluded. Exclusion criteria are listed in Fig. 1. Any analysis of fewer than 20 patients with confirmed OA was excluded to remove papers at risk of study imprecision. The inclusion criteria were in vivo observational studies of a human population with clinical and/or radiographic OA, which included an imaging description of the adjacent subchondral bone pathology to the osteoarthritic joint and the relationship of this with pain, structural progression or joint replacement. Analyses describing the relationship between OA bone manifestations and structural severity (cross-sectional) or progression (prospective cohorts) in populations without clinical and radiographic OA were included to incorporate early structural features of joint degeneration. The outcome measures of structural severity or progression included cartilage defects, cartilage thickness, cartilage volume, denuded subchondral bone, Kellgren-Lawrence (KL)grade, joint space width and joint space narrowing. Other outcome measures included joint replacement and any pain measures.
The articles identified by the preliminary search were screened by two reviewers (DH, AB) for relevance and for references not identified by the preliminary search, although no additional citations were found. Discordance in opinion was resolved by a third reviewer (SK). We applied the methods for reporting meta-analyses of observational studies in epidemiology that are recommended by the Cochrane collaboration [15,16].

Data extraction
Data extraction was performed by two reviewers (DH, AB) as described in the Supplementary methods 'data extraction' (see Additional file 1).

Quality assessment
The quality of each observational study was independently assessed by two reviewers (TC, AB), as described in Supplementary methods 'Quality assessment' (see Additional file 1).

Best evidence synthesis
Statistical pooling of the data was considered inappropriate in light of the heterogeneous study populations, methodological quality and bone feature or outcome measurements for OA. Therefore a qualitative summary of the evidence for each bone feature (e.g., BML) and its association with pain or structural progression and joint replacement was provided based on the study design, adequacy of adjustment for confounders (age, body mass index and gender) and quality score as described in the Supplementary methods 'Best evidence synthesis' (see Additional file 1).
Studies that investigated the association between multiple bone features and OA pain or structural progression outcomes were considered as a single study for each bone feature. Included studies that established significant correlation between bone and pain, structural progression or joint replacement were described as positive (+) or negative (−) accordingly. If no association or inconclusive findings were described this was reported as no association (NA) or no conclusion (NC) respectively.

Systematic literature search and selection
The Preferred reporting items for systematic reviews and meta-analyses (PRISMA) diagram in Fig. 1 describes the literature flow. Following exclusion of duplicates and triplicates, 2,456 articles met the search criteria. After applying inclusion/exclusion criteria, 139 articles were included for data extraction and quality scoring. In total, 71 papers provided longitudinal data (55 cohorts, 16 case−controls), 70 provided cross-sectional data, and two papers provided both.

Quality assessment of studies
Concordance of opinion in quality scoring was observed in 2,040 (89 %) of the 2,242 scoring items assessed, which are recorded in Additional file 1: Tables S3-S5. The majority of discordant scoring was for study design (criteria 17) and data presentation (criteria 18). Quality scores were converted to percentages of the maximum scores for each class of paper. The mean (range) quality score was 59 % (29-79), 54 % (22-83) and 59 % (47-76) for crosssectional, cohort and case-control studies, respectively.

Relationship between knee bone feature and structural progression
The association of bone features with structural progression and joint replacement are described in Tables 1 and 5.         Positive correlation reported between bone feature and outcome measure (+); negative correlation reported between bone feature and outcome measure (−). BMD bone mineral density, BMI body mass index, BML bone marrow lesion, BOKS Boston osteoarthritis of the knee study, BLOKS Boston-Leeds osteoarthritis knee score, BVF bone volume fraction, C a feature or outcome described in cross-section, CT computed tomography, DXA dual-energy x-ray absorptiometry, GARP Genetics, osteoarthritis and progression study, JSN joint space narrowing, JSW joint space width, KL Kellgren-Lawrence, KOSS knee osteoarthritis scoring system, L a feature or outcome described longitudinally, MAK-2 mechanical factors in arthritis of the knee 2. NC no conclusion could be found for an association between bone feature and outcome measure, SWAN in prospective cohorts with high-quality, well-adjusted analyses the presence and increasing size of baseline BMLs and incidence of BMLs conferred greater odds of structural progression [67,[70][71][72][73]. Similarly increasing baseline BML size increased the risk of total knee replacement (TKR) and expedited the outcome of TKR [44,[74][75][76]. The association between BMLs and structural progression of OA was maintained in cohorts without clinical features of knee OA [67] and in analyses with poorer quality or statistical adjustment [25,28,[76][77][78][79][80][81][82][83][84][85][86][87][88][89][90].
Osteophytes MRI (three cohorts, eight cross-sectional studies): in one prospective cohort with high quality and well-adjusted analysis, the increasing size of osteophytes conferred greater odds of structural progression of OA [106]. In lower quality, inadequately adjusted, prospective cohorts, increasing osteophyte size increased the risk of incident TKR and structural progression of OA [28,45]. The increasing size and presence of osteophytes was associated with greater structural progression or severity in all included analyses [22,26,28,45,64,65,97,101,103,[106][107][108].
In summary, osteophytes are independently associated with knee structural progression and are associated with TKR incidence.

Bone attrition
MRI (one cohort, two cross-sectional, one case-control study): one prospective, well-adjusted, but below-averagequality cohort analysis found an association with baseline attrition severity and structural progression that became insignificant after covariate adjustment [82]. The unadjusted cross-sectional analyses and case-control analysis found similar associations with structural severity [103,109,110]. In summary, bone attrition is associated, but not independently so, with structural progression.

Bone shape/dimension
MRI (eight cohort, seven cross-sectional, four case-control studies): in prospective cohorts with high quality welladjusted analyses, greater baseline tibial plateau bone area conferred greater odds of structural progression of OA and incidence of TKR [18,20,111,112]. The same association was observed in a lower quality, prospective-cohort, well-adjusted analysis [113] and in a study of the knee in patients who predominantly had no radiographic evidence of knee OA [18]. The mismatch ratio of the femoral and tibial articulating areas was not associated with structural progression after adjustment [114], but the trochlear sulcus angle and shape was associated with cross-sectional patellofemoral structural severity demonstrated on MRI [115,116]. All cross-sectional [23,66,107,117] and case-control [118][119][120][121] analyses of tibial bone area or 3D knee bone shape found association with structural severity [23,66,107,[117][118][119][120][121]. In summary, tibial bone area is independently associated with structural progression of OA of the knee and incidence of TKR.

Bone cyst
MRI and CT (five cohort, five cross-sectional) studies: two prospective cohorts with well-adjusted but belowaverage-quality analyses of cysts reported no association with structural progression of OA before or after adjustment [82,93]. Two prospective cohorts with low quality unadjusted analyses of cysts found an association with structural progression of OA [77,88]. Cross-sectional well-adjusted [65] and unadjusted [22,40,101,122] cyst analyses found an association with structural severity. In summary, after covariate adjustment there is no independent association between cysts and structural progression of OA.

Trabecular bone morphometry
MRI (one cohort, five cross-sectional studies): one prospective cohort, unadjusted, below-average-quality analysis reported increasing bone volume fraction, trabecular number and thickness and decreasing trabecular spacing were associated with structural progression [53]. The same bone changes were associated with structural severity in cross-sectional unadjusted analyses [34,49,50,54,123]. In summary, increasing bone volume fraction, trabecular number, trabecular thickness, and decreasing trabecular spacing are associated with structural progression and severity of OA of the knee.
Peri-articular bone mineral density DXA and CT (three cohort, four cross-sectional, onecase control study): two prospective cohorts with welladjusted but below-average-quality analyses reported that increasing tibial subchondral BMD is associated with structural progression of OA [42,68]. In one prospective cohort with an unadjusted below-average-quality analysis, the medial-to-lateral ratio of tibial peri-articular BMD was associated with structural progression [53]. All crosssectional analyses [31,52,54,55], including two that were well-adjusted [52,55], reported increasing BMD with greater structural severity. One well-adjusted analysis using quantitative CT (qCT) reported higher and lower BMD in the anterior and posterior tibial plateau, respectively, in knees of patients with moderate OA relative to asymptomatic controls. In summary, increasing peri-articular radiographic BMD is associated with structural progression and severity of OA.

Scintigraphy
Scintigraphy (three cohort, two cross-sectional studies): prospective cohorts with high quality analyses found greater late-phase bone signal was associated with structural progression of OA, with no or inadequate covariate adjustment [37,38], but not after adequate covariate adjustment [37]. A prospective cohort, with below-averagequality, unadjusted analysis found greater bone signal was associated with structural progression of OA [58]. Bone signal was associated with structural severity in well-adjusted and unadjusted cross-sectional analyses [59,62]. In summary, bone scintigraphy signal is associated, but not independently so, with structural progression of OA.

2D Knee bone shape
One cross-sectional, well-adjusted analysis identified an association between greater femoral and tibial bone width and elevating tibial plateau, and greater structural severity [36]. In summary, 2D bone shape is associated with structural severity of OA.

Relationship between knee bone feature and pain
The association between bone features and pain is described in Tables 2 and 5. In all types of study, bone features were compared with the presence, chronicity and severity of pain. In longitudinal studies, bone features were also compared with change in the presence or severity of pain (e.g., change in Western Ontario and McMaster Universities arthritis index (WOMAC) pain score). Change in the presence of pain included developing new frequent pain, [49], or the resolution of existing pain.

Bone marrow lesions
MRI (9 cohort, 18 cross-sectional, 5 case-control studies): in 3 prospective cohort, well-adjusted, high quality analyses the baseline or longitudinal increase in size of BMLs was associated with longitudinally increasing knee WOMAC pain severity [21,72,124]. This association was observed in one [28] but not two [89,95] similar prospective-cohort, unadjusted, lower quality analyses. Baseline BML size in the lateral but not the medial tibiofemoral joint was associated with incident frequent knee pain in a prospective-cohort, well-adjusted, high quality analysis [125]. Longitudinally increasing BML size was associated with incident frequent knee pain in a similar but inadequately adjusted analysis of below average quality [126]. In cross-sectional studies the size or presence of BMLs was inconsistently associated with the presence of a heterogenous range of pain measures, irrespective of adequate covariate adjustment [22,29,48,96,97,99,101,103,125,[127][128][129][130][131][132][133][134][135]. In summary, BMLs are independently associated with longitudinally increasing pain severity and are associated with incident frequent knee pain.
Osteophytes MRI (one cohort, eight cross-sectional, one case-control study): one prospective cohort, unadjusted, below-averagequality analysis reported increasing baseline osteophyte size was associated with increasing WOMAC pain severity score [28]. In well-adjusted cross-sectional analyses, osteophyte size was associated with the presence [130] but not severity of pain [136]. In unadjusted cross-sectional analyses osteophytes were inconsistently associated with a heterogenous range of pain measures [22,97,101,103,127,137]. In summary, osteophytes are associated with longitudinally increasing pain severity and the crosssectional presence of pain.

Bone attrition
MRI (no cohort, two cross-sectional, one case-control study); cross-sectional analyses found greater attrition was associated with greater pain severity, without covariate adjustment [103,138], but not after adequate covariate adjustment [138]. An unadjusted case-control analysis found an association between attrition and prevalent pain [139]. In summary, bone attrition is associated, but not independently so, with severity of pain.
Bone shape/dimension MRI (one cohort, one cross-sectional study): one prospective, well-adjusted, high quality analysis found the femorotibial articulating surface mismatch was associated with incident frequent knee pain [114]. One unadjusted crosssectional analysis found the irregularity of the femoral condyle surface was associated with severity of knee pain [47]. In summary, specific features of bone shape are independently associated with incident frequent knee pain and severity of pain.
Bone cyst MRI (one cohort, five cross-sectional, two case-control studies): one prospective cohort, unadjusted, low quality analysis found no association between bone cyst size and increasing WOMAC pain score [28]. In mostly unadjusted cross-sectional [22,101,103,130,137] and case control analyses [139,140] of heterogenous cyst measures and pain measures, an association between cysts and pain was inconsistently found. In summary, bone cysts may not be associated with longitudinal severity of pain and crosssectional association with pain is uncertain.     Positive correlation reported between bone feature and outcome measure (+); negative correlation reported between bone feature and outcome measure (−). BMI body mass index, BML bone marrow lesion, C a feature or outcome described in cross-section, knee pain on most days for at least the last month (chronic pain) confidence interval (CI), KL Kellgren-Lawrence, L a feature or outcome described longitudinally, NA no association, NC no conclusion could be found for an association between bone feature and outcome measure, NR not reported, OA osteoarthritis, OAI Osteoarthritis Initiative, OR odds ratio, PFJ patellofemoral joint, ROA radiographic osteoarthritis, SSR subchondral surface ratio VAS visual analogue scale, WOMAC Western Ontario and McMaster Universities arthritis index, qCT quantitative computed tomography 2D Knee bone shape One inadequately adjusted cross-sectional analysis found an association between the elevation of the lateral tibial plateau and severity of pain [36]. In summary, 2D lateral tibial bone shape is associated with cross-sectional severity of pain.

Relationship between hand bone feature and structural progression
The association between bone features and structural progression is described in Tables 3 and 5.

Bone marrow lesions
MRI (one case series, two cross-sectional studies): one well-adjusted, high quality analysis of a prospective OA case series, found that increasing BML number and size in the interphalangeal joints at baseline conferred greater odds of structural progression of OA [141]. Two adjusted cross-sectional analyses found increasing BML number and size scores were associated with increasing severity of structural progression [142,143]. In summary, BMLs are independently associated with structural progression of hand OA.

Osteophyte attrition and cysts
One cross-sectional, adjusted analysis found greater MRI attrition or MRI osteophyte number and size was associated with greater structural severity [142]. However, greater presence of cysts observed on MRI was not associated with greater structural severity of OA [142]. In summary, osteophytes and attrition, but not cysts, are associated with structural severity of hand OA.

Relationship between hand bone feature and pain
The association between bone features and pain is described in Tables 4 and 5.

Bone marrow lesions
MRI (one case series, one cross-sectional study): one welladjusted, high quality analysis of a prospective OA case series, found that BML number and size at baseline was not associated with longitudinal change in hand pain [144]. One adjusted cross-sectional analysis found no association of BMLs with severity of pain [145]. In summary, BMLs are not independently associated with longitudinal or cross-sectional severity of pain.

Osteophyte attrition and cysts
One cross-sectional, adjusted analysis found no association between bone features, osteophytes, attrition or cysts observed on MRI, and pain severity [145]. In summary, osteophytes, attrition and cysts are not associated with severity of hand pain.

Scintigraphy
Scintigraphy (one cross-sectional study): one cross-sectional unadjusted analysis found no significant association between bone signal in the hands and severity of pain. In summary, bone scintigraphy signal is not associated with severity of pain in hand OA.

Relationship between hip bone feature and structural progression
The association between bone features, and structural progression and joint replacement is described in Tables 3  and 5.
Bone marrow lesions MRI (two cross-sectional studies): one well-adjusted [69] and one unadjusted [46] cross-sectional analysis both found that BMLs were associated with greater structural severity. In summary, BMLs are associated with structural severity of hip OA.

Trabecular bone morphometry
One unadjusted cross-sectional analysis found greater MRI bone volume fraction, trabecular thickening, trabecular number and lower trabecular spacing were associated with greater structural severity of OA [33]. In summary, bone volume fraction, trabecular thickening, number and spacing are associated with structural severity in hip OA.
Peri-articular bone mineral density DXA (two cross-sectional studies): one well-adjusted [30] and one adjusted [32] cross-sectional analysis found greater BMD was associated with greater structural severity. In summary, BMD is associated with structural severity of hip OA.

2D and 3D hip bone shape
Hip bone shape (three cohort, two cross-sectional, three case-control studies): in two prospective cohort, welladjusted, high quality analyses increasing asphericity of the femoral head (measured as an elevated alpha angle, or in shape modes 11 and 15) was associated with total hip replacement (THR) [146] or with structural progression and THR [147] respectively. In one prospective cohort, well-adjusted, high quality analysis, acetabular undercoverage of the femoral head (a low centre-edge angle) was associated with structural progression or THR [148]. In one well-adjusted cross-sectional analysis, 2D asphericity deformity of the femoral head (cam-type deformity) was associated with structural severity [149]. In one well-adjusted cross-sectional analysis of MRI-determined femoral head asphericity in asymptomatic young men, there was a significantly lower cartilage thickness in those with than those without any detectable asphericity. This became insignificant after covariate adjustment [27]. Case-control analyses    (57) Positive correlation was reported between bone feature and outcome measure (+); negative correlation reported between bone feature and outcome measure (−).BMD bone mineral density, BML bone marrow lesion, C a feature or outcome described in cross-section, CT computed tomography, DXA dual-energy x-ray absorptiometry, HOAMS Hip osteoarthritis MRI scoring system, IPJ interphalangeal joint, JSN joint space narrowing, JSW joint space width, KL Kellgren-Lawrence, L a feature or outcome described longitudinally, NA no association, NC no conclusion could be found for an association between bone feature and outcome measure, MRI magnetic resonance imaging, PFJ patellofemoral joint, ROA radiographic osteoarthritis, OA osteoarthritis, OARSI Osteoarthritis Research Society International, OR odds ratio, RR relative risk, TFJ tibiofemoral joint, THR total hip replacement, TKR total knee replacement, VAS visual analogue scale, WOMAC Western Ontario and McMaster Universities arthritis index, WORMS whole-organ magnetic resonance imaging score Table 4 Hand and hip pain associations by feature and quality score identified the same associations as the cohort analyses [39,43,150]. In summary, asphericity of the femoral head and acetabular undercoverage of the femoral head are independently associated with structural progression and THR.

Relationship between hip bone feature and pain
The association between bone features and pain is described in Tables 4 and 5.
Bone marrow lesions MRI (two cross-sectional studies): two cross-sectional, unadjusted analyses found that increasing semi-quantitative BML scores were associated with greater severity of pain [151,152]. In summary, BMLs are associated with severity of pain in hip OA.

Bone cyst
One cross-sectional, unadjusted analysis found that increasing semi-quantitative cyst scores on MRI were associated with greater severity of pain [151]. In summary: cysts are associated with severity of pain in hip OA.

Relationship between ankle bone features and structural progression
The association between bone features and structure is described in Table 3 and 5.

Scintigraphy
Scintigraphy (two cross-sectional studies): one well-adjusted [59] and one unadjusted [62] cross-sectional analysis found the presence or semi-quantitative scoring of late-phase bone signal in the tibiotalar joint was associated with greater structural severity. In summary, bone scintigraphy signal is associated with ankle structural severity.

Discussion
This systematic review is the first to have incorporated quality scoring alongside statistical adjustment in the comprehensive examination of the relationship of subchondral bone pathology with both structural progression of OA and pain for all non-conventional types of radiographic imaging of peripheral joints with OA. This systematic review has concluded that there are independent associations between imaging-assessed bone pathology and structural progression and pain in the knee, hand, and hip. Subchondral bone pathology may lead to cartilage degeneration by altering the biomechanical force distribution across joint cartilage, or disruption of the osteochondral junction and release of soluble biomediators influencing the cartilage [153,154]. In OA the homeostatic process of subchondral bone remodeling fails, leading to increased bone turnover, volume and change in stiffness and shockabsorbing capacity [155][156][157]. BMLs histologically represent increased bone turnover [158]. Cartilage overlying CT computed tomography, dual-energy DXA x-ray absorptiometry, (i) independent association, IFP incident frequent pain, (n) association with no or inadequate covariate adjustment, TKR total knee replacement, THR total hip replacement, LPS mean change in longitudinal pain severity, (w) well-adjusted association, (0) association insignificant after covariate adjustment altered bone has been observed to have greater damage than healthy bone in knees from human cadavers [159]. That study, and an excluded study [160], concur with the independent association between BMLs, and structural progression of OA in knees and hands and total knee replacement, as concluded by this analysis. Although randomised control trials were not excluded from this review, several such trials were excluded on the basis of failure to formally quantify any correlation between BMLs and structural progression outcomes. These include the strontium [161], intensive weight-loss therapy [162] and glucosamine [163] trials, and some of these describe a concordant reduction in BML size and cartilage volume loss. Osteophytes represent subchondral bone hypertrophy typical of OA. They represent endochondral and direct bone formation and create a circumferential increase in bone area around each knee cartilage plate, particularly on the medial side in OA [118], which concurs with the independent association between osteophytes demonstrated on MRI and structural progression as observed in this analysis.
In terms of bone morphology, knee OA is associated with shallow trochlear patellar grooves in multiple epiphyseal dysplasia [164]. These findings concur with the findings of Stefanik and Kalichman, and colleagues in studies of knee OA in this review [115,116,165]. Anteriorcruciate ligament (ACL) rupture represents a risk factor for developing knee OA. In cases of ACL tear in previously normal knees of young healthy adults, the 3D shape of the femur, tibia and patella expands more rapidly than in controls without radiographic evidence of knee OA in the subsequent 5 years [166]. The 3D shape of the same knee bones has also been associated with the outcome of joint replacement [167]. This highlights the importance of bone shape and concurs with our conclusion that 3D knee shape and 2D hip shape are independently associated with structural progression of OA and total joint replacement.
We found that bone attrition and cysts were associated with structural progression or severity, but not after covariate adjustment, which included other OA subchondral bone features. This suggests these bone features are an epiphenomenon of the pathogenic process of structural progression rather than a primary cause. This hypothesis is supported by bone cysts and attrition frequently occurring synchronously with BMLs [88,138] and incident bone attrition has been strongly associated with the presence of BMLs within the same compartment [168].
Increasing bone volume fraction, trabecular number and thickness, but decreasing trabecular spacing on CT and MRI studies were associated with structural progression. These specific associations concur with numerous histological analyses of peripheral joint OA [169][170][171].
Subchondral bone, particularly BMLs, have been found to be associated with pain in knee, hip and hand OA. However, some analyses, in which pain was measured using heterogenous pain outcomes, report an absence of longitudinal or cross-sectional association with BMLs [101,172,173]. Furthermore, previous systematic reviews have concluded moderate association at the most between BMLs and knee pain [7,14]. With the benefit of incorporating more well-adjusted analyses in this systematic review, we have highlighted that BMLs are independently longitudinally associated with change in severity of pain, but are only associated with incident frequent knee pain. In analyses excluded from the current review, incident knee BMLs predicted incident knee pain in healthy community-based adults at risk of OA [174]. Concurrent trends in reduction of pain and BML size were observed in the zoledronic acid trial [175] and the intensive diet and exercise for arthritis trial [176]. These were not included because they did not make a formal comparison of pain and BMLs. The mechanism by which BMLs may cause pain is unknown but may include subchondral microfractures, angina from a decreased blood supply causing ischaemia, and raised intraosseous pressure [177][178][179].
The independent association between a mismatch of the femoral and tibial articulating surface areas and incident frequent knee symptoms indicates that bone shape may predict not only the incidence of radiographic knee OA [120], but also symptomatic OA.
In terms of limitations, stratifying observational studies by quality may artificially create relatively high quality studies from a collection of generally low quality studies. However the distribution and summary statistics of quality scores indicate a suitably broad range of quality, particularly in the influential cohort studies with a mean of 54 % and range of 22−83 %. The decision to exclude articles reporting analysis of association that included fewer than 20 patients with OA may seem arbitrary. However, several papers report associations with the presence or absence of pain or structural progression based upon small numbers of patients. Our threshold decision reflects the absence of specific guidelines on how to exclude such papers, with inherent risk of imprecision, in the context of heterogenous populations and statistical analyses. Had these papers been included there would have been no change in any of the conclusions in Table 5 (data not shown). The use of joint replacement as an outcome measure has a number of limitations including the effect of patient willingness, variation in orthopaedic opinion, availability of health services and health insurance, and therefore may be influenced depending upon the country and context in which the study is performed.
Publication bias could not be assessed with a funnel plot as there were insufficient results for odds and relative risk ratios. The heterogenous nature of the measures of bone features and structural or pain outcomes precluded a meta-analysis or calculation of an effect size. This was because there were insufficient analyses describing the same association between the same bone features and outcome measure pair.

Conclusions
In conclusion subchondral bone plays an integral role in the pathogenesis of OA. BMLs, osteophytes identified on MRI and tibial bone area are independently associated with structural progression of knee OA. BMLs and tibial bone area are independently associated with TKR. BMLs are independently associated with structural progression of hand OA and 2D hip bone shape is associated with progression of structural hip OA and THR. BMLs are independently associated with longitudinal change in severity of pain and femorotibial articulating area mismatch is independently associated with incident frequent knee pain. These bone features may be used in the future for targeting treatment, stratifying patients into those most in need of OA modification and measuring treatment response. Abbreviations 2D: two-dimensional; 3D: three-dimensional; ACL: anterior cruciate ligament; BLOKS: Boston-Leeds osteoarthritis knee score; BMD: bone mineral density; BMI: body mass index; BML: bone marrow lesion; BOKS: Boston osteoarthritis of the knee study; BVF: bone volume fraction; Cam: a resemblance to a camshaft; CRP: C-reactive protein; CT: computed tomography; DMOAD: disease-modifying osteoarthritis drug; DXA: dual-energy x-ray absorptiometry; EMBASE: Excerpta Medica database; GARP: Genetics, osteoarthritis and progression study; HOAMS: Hip osteoarthritis MRI scoring system; IPJ: interphalangeal joint; JSN: joint space narrowing; JSW: joint space width; KL: Kellgren-Lawrence; KOSS: knee osteoarthritis scoring system; MOST: multicentre osteoarthritis study; MRI: magnetic resonance imaging; NA: no association; NC: no conclusion; OA: osteoarthritis; OAI: Osteoarthritis Initiative; OARSI: Osteoarthritis Research Society International; OR: odds ratio; PET: positron emission tomography; PFJ: patellofemoral joint; PRISMA: Preferred reporting items for systematic reviews and meta-analyses; qCT: quantitative computed tomography; ROA: radiographic osteoarthritis; RR: relative risk ratio; SSR: subchondral surface ratio; THR: total hip replacement; TFJ: tibiofemoral joint; TKR: total knee replacement; VAS: visual analogue scale; WOMAC: Western Ontario and McMaster Universities arthritis index; WORMS: whole-organ magnetic resonance imaging score.

Competing interests
Dr Bowes is an employee and shareholder of Imorphics Ltd. Professor Conaghan, Sarah Kingsbury, Andrew Barr, Devan Hopkinson and Thomas Mark Campbell have nothing to disclose.
Authors' contributions AB carried out conception and design, eligibility assessment, extraction of data, quality assessment, along with drafting and revising of the manuscript content. DH carried out design, eligibility assessment and extraction of data. TC carried out conception and design, quality assessment and revising manuscript for content. MB carried out conception and design along with revising the manuscript for content. SK carried out conception and design, eligibility assessment and revising the manuscript for content. PC carried out conception and design, quality assessment and revising the manuscript for content. All authors read and approved the final manuscript version for publication.