1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information


To identify, by systematic review, patient characteristics that can be used by health care practitioners to predict the likelihood of knee osteoarthritis (OA) progression.


A search was conducted of the electronic databases Medline, EMBase, CINAHL, AMED, and Web of Science in November 2010. Two reviewers screened articles using inclusion/exclusion criteria. Study participants were adults with established knee OA. Outcome measures for disease progression were change in pain or function or deterioration in radiographic features. Included studies identified clinically relevant prognostic factors at baseline and reported a statistical association with outcome. Minimum followup was 1 year. Articles were assessed for bias, and strength of evidence was summarized for potential predictors of progression.


Thirty studies were included, of which 26 were of high quality. Age, varus knee alignment, presence of OA in multiple joints, and radiographic features had strong evidence as predictors of knee OA progression. Body mass index was a strong predictor for long-term progression (>3 years). Moderate participation in physical activity was not associated with progression. Numerous variables had limited or conflicting evidence.


Relatively few predictive variables have strong supporting evidence; numerous variables have limited or conflicting evidence. All variables with strong evidence can be easily evaluated and utilized in clinical practice. Existing knowledge should be developed in future research, particularly in cases where study numbers are low or findings are limited or conflicting. Standardized measurement of potential predictors and outcome measures is recommended.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Osteoarthritis (OA) of the knee is a common disease, with a prevalence of 12.5% in populations ages >45 years (1). The lifetime risk of developing symptomatic knee OA is estimated as 44.7% (2), and the annual rate of progression in subjects diagnosed with knee OA has been reported as approximately 4% per year, indicating slow evolution of the disease (3). Heterogeneity of knee OA as a disease results in a wide range of clinical presentations and varying rates of progression. Identifying those patients most likely to progress or those at risk of rapid progression is important for optimal allocation of health care resources and research into therapeutic interventions (4). Additionally, the ability to identify prognostic indicators from patient history and examination would be beneficial at an individual level, allowing health care practitioners to more accurately predict the likelihood of disease progression and direct patients to appropriate interventions.

Previous systematic reviews have sought to identify predictors of knee OA progression using radiographic change (5, 6) or functional decline and change in pain (7) as outcome measures. Usual clinical practice is to gather information from both radiographs and patient-reported symptoms. Therefore, to enhance clinical relevance, the approach of this review was to include any validated clinical information for both potential predictor variables and measures of outcome. Furthermore, advances in search strategies for prognostic and nonrandomized studies, as well as more studies investigating progression, are likely to have led to additional evidence being available in the field of knee OA progression, which makes a new review both timely and relevant (8).

The aim of this study was to identify patient characteristics that can be used by health care practitioners to predict the likelihood of knee OA progression. A systematic review of the literature was performed to meet this objective.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Search and identification of studies.

A protocol for conducting this review was developed with reference to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (9). A comprehensive search strategy was developed to detect prognostic studies, using a PICOS (populations, interventions, comparators, outcomes, study design) framework to guide selection of terms (10–12). An example of the search strategy for Medline is included in Supplementary Appendix A (available in the online version of this article at; this was adapted for EMBase, CINAHL, AMED, and Web of Science. Reference lists of identified reviews were mined for further studies (5–7). Neither study type nor language was limited (10, 13). An initial search was conducted in August 2009 and updated in November 2010.

Inclusion/exclusion criteria were determined a priori. Study participants were adults (ages >18 years, any sex, any duration of symptoms) with knee OA classified by clinical or radiographic reference standards (1). Cohort studies had to describe any intended exposure; included randomized control trials (RCTs) had to describe their interventions. Comparison was between progression and nonprogression of knee OA. Outcome was progression of disease defined as either deterioration in functional status or pain report as evaluated with recognized instruments such as the Western Ontario and McMaster Universities Osteoarthritis Index or the visual analog scale, or radiographic change defined as an increase in Kellgren/Lawrence (K/L) grade or joint space narrowing (JSN) score, increase in osteophytes, or decrease in joint space width (JSW). As the study design was not limited, RCTs with potential for inclusion had to demonstrate no effect of intervention or had to report the placebo group separately. All studies had to identify prognostic factors at baseline and report a statistical association (or lack of association) with outcome. To ensure clinical relevance, baseline variables had to include patient characteristics such as pain or functional limitation, psychosocial factors, demographic information, body mass index (BMI), dietary intake, or gait abnormalities. All measures should be in routine clinical use and not require sophisticated equipment or complex analysis. Minimum time for followup was set at 1 year to allow sufficient time for progression to occur.

Studies were excluded if participants had comorbidities such as rheumatoid arthritis, cancer, osteoporosis, and joint infection; generalized OA where knee OA results were not reported separately; or previous surgery for knee OA. Duplicate reports from the same study reporting the same data were excluded. To ensure applicability of review findings to routine clinical practice, any study that reported exclusively on radiographic variables or laboratory tests with no reference to patient presentation was excluded.

The search strategy and inclusion/exclusion criteria were piloted to ensure good coverage of the field, key articles were identified and retained, and irrelevant articles were discarded.

Screening of articles using inclusion/exclusion criteria was completed independently by 2 reviewers initially focusing on titles and abstracts, and then full-text versions of retrieved articles. Reasons for exclusion were documented. Consensus on final inclusion of articles was reached following discussion; a third reviewer (GDB) was available to resolve any outstanding disagreements. The process is summarized in Figure 1.

thumbnail image

Figure 1. Identification and screening process. ‡ = reasons for exclusion: study design, not knee osteoarthritis (OA) progression, lack of a definition of OA knee at baseline, no baseline clinical predictors (magnetic resonance imaging/biomarkers/radiographs only); † = reasons for exclusion from synthesis of evidence: 4 low-quality studies (<12) and 6 incorrect analysis/unadjusted analysis.

Download figure to PowerPoint

Assessment of bias.

Selected articles were assessed for bias using a scoring tool (Table 1) adapted from Wright et al (14), and based upon a similar tool used in previous studies of musculoskeletal problems (5–7, 15). The tool was specifically aimed at defining study quality in terms of how potential sources of bias were addressed, rather than solely on the quality of reporting. The revised tool addressed 6 potential sources of bias in prognostic studies: study participation (items A–D), study attrition (items E–G), prognostic factor measurement (items H and J–L), outcome measurement (items M–O), confounding measurement and statistical adjustment (multiple items, I, and S), and reporting and analysis of results and conclusions (items P–T) (16, 17). Confounding and reporting of conclusions are considered particularly relevant in observational studies (16). Detailed operational definitions for the assessment of bias tool are included in Supplementary Appendix B (available in the online version of this article at For each criterion, articles scored 1 if they met the operational definition or 0 if they did not meet it, if it was not applicable, or if there was insufficient information to make a decision. The maximum score was 20. An arbitrary cutoff of ≥60% of criteria met (i.e., a raw score of ≥12) defined studies of high quality in line with similar studies (5, 7, 15).

Table 1. Criteria for assessment of bias*
  • *

    Adapted, with permission, from ref.14.

Study participation 
 AInception cohort
 BDescription of study population
 CDescription of inclusion/exclusion criteria
 DResponse of ≥75% for cohorts and controls
Study attrition 
 EFollowup of at least 12 months
 FDropouts/loss to followup of ≤20%
 GInformation completers vs. loss to followup
Measurement and data presentation 
 HProspective data collection
 ITreatments or exposure of cohort are fully described and standardized
 JClinically relevant prognostic factors
 KStandardized or valid measures of prognostic factors
 LData presentation of the most important prognostic factors
 MClinically relevant outcome measures
 NStandardized or valid outcome measures
 OData presentation of the most important outcome measures
Analysis and presentation of results 
 PAppropriate analysis techniques
 QPrognostic model is presented
 RSufficient numbers

Two reviewers (CMC and/or JHA) used the tool to independently score included articles, unblinded to authors. The level of agreement beyond chance is reported using the kappa statistic. Discussion was used to reach consensus on any scoring disagreements; a third reviewer (GDB) was available to resolve any lack of agreement.

Data extraction.

Study details and results were extracted from included articles by 2 unblinded reviewers. A form to guide and document a systematic data extraction process was developed and tested. Results included odds ratios, relative risks, hazard ratios, or P values. Adjusted results were extracted where possible to address the problem of confounding (13). Sources of funding were noted.

Synthesis of evidence.

Options for synthesis of evidence included meta-analysis. However, heterogeneity is a common problem in prognostic studies and provides evidence against pooling data for a formal meta-analysis (17, 18). An alternative was considered based on 3 well-established domains of quality, quantity, and consistency of findings from included studies (19). Studies had to be of high quality, scoring ≥60% (≥12) with the assessment of bias tool. Additionally, univariate analyses adjusted for confounders, multivariate analyses, or stratified analyses were considered as appropriate analyses for identification of prognostic factors (17). The quantity of findings was addressed by reporting the number of studies investigating a particular variable, combined sample size, and magnitude of effect. Consistency was the extent to which similar findings were found in studies of similar design (19). A ranking system proposed by Lievense et al (15) and utilized in subsequent systematic reviews (6, 7) was modified for this purpose. Strong evidence required generally consistent findings in multiple (≥2) high-quality cohort studies, limited evidence required generally consistent findings in a single high-quality cohort study, and conflicting evidence resulted from conflicting findings in high-quality studies (<75% of studies reported consistent findings). The heterogeneity of included studies was examined prior to deciding on the method for synthesis of evidence.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Search results.

Results of the search strategy and screening process are shown in Figure 1. Consensus was achieved in all cases and the third reviewer was not required. Thirty studies were eligible for full review, 25 from the initial search and 5 from the update (20–49).

Study details.

Details of included studies are available in Supplementary Table A (available in the online version of this article at Participants were recruited from community settings, rheumatology or orthopedic clinics, as well as previous studies. Sample sizes ranged from 54 (48) to 2,964 (31). Four studies recruited women only (32, 35, 45, 46) and 1 study recruited men only (22), with remaining studies reporting between 41% (29) and 84% female participants (34). The mean age ranged from 44.8 (48) to 70.3 years (36, 37). K/L grade was the most frequently used set of criteria for classification of knee OA. The American College of Rheumatology criteria and JSN score were also used. Followup times varied from 1 year (23, 28, 32) to 14 years (45).

Assessment of bias.

Results of the bias assessment are shown in Table 2. Three reviewers assessed 600 items, agreeing on 503 (percent agreement = 84%; κ = 0.59). Disagreements were resolved by discussion. The majority of studies (26 of 30) scored ≥60% (≥12) and could be considered of high quality. Although few studies provided sample size calculation (item R), sample size was considered adequate for analysis in 28 of the 30 included studies, based on a ratio of 10 subjects per prognostic variable analyzed. The majority of studies were longitudinal cohorts, except 4 studies that used subjects recruited either partly or wholly from RCTs (24, 34, 35, 47). This resulted in a good overall score for item I. Overall, criteria relating to study participation (items A–D) were not well met. Reporting of response rates was lacking in many studies or confused due to lack of clarity about numbers approached, numbers eligible, and final numbers that participated (item D). Several studies presented only P values to describe an association between the predictive factor(s) and outcome (item P). Not all studies presented adjusted results for potential confounders, the minimum requirements for which were age and sex (item S). It should be noted that studies that scored the highest addressed all aspects of bias to some extent; low-quality studies failed to adequately address all areas of bias.

Table 2. Results of the assessment of bias
Author, year (ref.)ABCDEFGHIJKLMNOPQRSTScore
Sharma et al, 2010 (41)1110111111111111111119
Benichou et al, 2010 (23)1011110111111111111118
Sharma et al, 2003 (42)1110111111111111011118
Sharma et al, 2003 (43)1110110111111111111118
Sharma et al, 2001 (44)0110111111111111111118
Golightly et al, 2010 (30)1110100111111111111117
Harvey et al, 2010 (31)1010110111111111111117
Amin et al, 2008 (21)0111111111111010011116
Felson et al, 2004 (29)0110110111111111011116
Le Graverand et al, 2009 (32)1010111111111110011116
Mazzuca et al, 2006 (34)1000110111111111111116
McAlindon et al, 1996 (36)0100101111111111111116
McAlindon et al, 1996 (37)0100101111111111111116
Niu et al, 2009 (38)1100110111111101111116
Spector et al, 1994 (46)0101111111111111011016
Amin et al, 2009 (20)0011111111111010011115
Cooper et al, 2000 (25)1100101111101111011115
Ledingham et al, 1995 (33)0110100101111111111115
Reijman et al, 2007 (39)0101100111111111011115
Schouten et al, 1992 (40)1100100111111111011115
Thorstensson et al, 2004 (48)0101110111111111001115
Wolfe et al, 2002 (49)1100100111111111110115
Amin et al, 2007 (22)0010111111111010011114
Mazzuca et al, 2005 (35)1100110111111110010114
Shiozaki et al, 1999 (45)1100100111111101010113
Dieppe et al, 1997 (26)0001111111101110010113
Bruyere et al, 2003 (24)0000101101111100110010
Dieppe et al, 1993 (27)1100101101111100000010
Dougados et al, 1992 (28)0000001111111010010110
Spector et al, 1992 (47)000010010110111001008

Sources of funding for the studies were noted but not included in the scoring tool; 5 studies did not report this (24, 34, 39, 45, 47). Results from all studies are included in Supplementary Tables B and C (available in the online version of this article at

Synthesis of evidence.

Included studies exhibited marked heterogeneity in terms of study design, descriptions of patient characteristics, followup periods, outcome measures, statistical analysis, and reporting of results. Consequently, meta-analysis was not possible and the alternative method was used.

Ten studies were excluded from the synthesis of evidence: 4 were of low quality (24, 27, 28, 47) and 6 did not perform an appropriate analysis to identify prognostic factors or did not adjust for confounding (20–22, 32, 35, 43). Removal of studies from the synthesis of evidence component is shown in Figure 1.

Results from the remaining 20 studies are shown in Tables 3 and 4. Table 3 includes characteristics typically obtained from a patient history or self-administered questionnaires. Table 4 includes clinical variables that can be observed or measured by a health practitioner. The level of significance was set at P < 0.05 or confidence intervals that did not cross 1.0. Significant predictors of progression are identified as well as significant predictors of nonprogression (which could be considered protective factors). Variables with no significant association with progression are also shown. Strength of evidence based on the adapted ranking system is stated for each variable. Number of studies, combined sample size, and study results give an indication of the magnitude of effect size and the quantity of information from which the strength of evidence is obtained.

Table 3. Evidence from high-quality studies for predictors of progression of knee osteoarthritis: demographics and patient history (self-report measures)*
Baseline variableLevel of evidenceNumber of studiesCombined sample sizeAssociation with progression (95% CI) [ref.]Analysis/results
  • *

    Results are ordered by number of studies, effect size, and significance. High-quality studies are those with a bias score of ≥60% (≥12). 95% CI = 95% confidence interval; OR = odds ratio; uni = univariate results/adjusted; multi = multivariate results; [UPWARDS ARROW] = increased; [DOWNWARDS ARROW] = decreased; NG = data not given for adjusted results or multivariate models (but reported as the same); NS = nonsignificant result; HR = hazard ratio.

  • Adapted from Lievense et al (15).

 AgeStrong61,006OR 3.84 (1.10, 13.4) [40]Uni
    OR 1.34 (1.15, 1.57) [42]Multi
    OR 1.07 (1.02, 1.13) [33]Multi
    [UPWARDS ARROW] age and [DOWNWARDS ARROW] function more likely to progress [26]Multi, NG
    β = 0.01 ± 0.006 mm/year [23]Multi
    OR 1.13 (0.87, 1.48) [34]Uni, NS
 Sex (F:M)Conflicting3475OR 2.17 (1.13, 4.15) [33]Multi
    Male/[DOWNWARDS ARROW] pain more likely to improve than female/[UPWARDS ARROW] pain [26]Multi, NG
    Female/[UPWARDS ARROW] rest, pain/[DOWNWARDS ARROW] joint pain more likely to progress [26]Multi, NG, NS
    OR 0.5 (0.22, 1.11) [40]Uni
 PainConflicting3735Male/[DOWNWARDS ARROW] pain more likely to improve than female/[UPWARDS ARROW] pain [26]Multi, NG
    OR 1.12 (0.90, 1.40) [42]Multi, NS
    Female/[UPWARDS ARROW] rest, pain/[DOWNWARDS ARROW] joint pain more likely to progress [26]Multi, NG, NS
    OR 0.8 (0.4, 1.7) toUni
    OR 2.4 (0.7, 8.0) [25]Uni
 FunctionConflicting3487[UPWARDS ARROW] age and [DOWNWARDS ARROW] function more likely to progress [26]Multi, NG
    OR 1.16 (0.92, 1.47) [34]Uni, NS
    OR 0.98 (0.32, 3.02) toUni, NG, NS
    OR 1.56 (0.53, 4.60) [48]Uni, NG, NS
 Joint stiffnessLimited1288OR 1.39 (1.09, 1.77) [34]Uni
 Disease severityLimited11,232HR 1.01 (1.00, 1.02) [49]Multi
 Symptom durationLimited11,232HR 1.3 (1.00, 1.05) [49]Multi
 Vitamin D intakeLimited1126OR 2.99 (1.06, 8.49) toMulti
    OR 4.05 (1.4, 11.6) [36]Multi
 Antioxidant intake     
  [UPWARDS ARROW] vitamin CLimited1187OR 0.06 (0.01, 0.67) toMulti
    OR 0.31 (0.12, 0.79) [37]Multi
  [UPWARDS ARROW] vitamin EConflicting1187OR 0.07 (0.01, 0.61) toMulti
    OR 0.48 (0.40, 2.90) [37]Multi, NS
 Psychosocial factors     
  Mental health scoreLimited1236OR 0.58 (0.39, 0.86) [42]Multi
  Social supportLimited1236OR 0.85 (0.73, 0.98) [42]Multi
  Self-efficacyConflicting1236OR 0.80 (0.65, 0.98) toMulti
    OR 0.86 (0.68, 1.09) [42]Multi, NS
 Physical activity/sports participationStrong3732OR 0.84 (0.69, 1.02) toMulti, NS
OR 0.86 (0.71, 1.05) [42]Multi, NS
OR 0.7 (0.4, 1.6) toUni, NS
OR 0.9 (0.3, 2.5) [25]Uni, NS
OR 0.53 (0.17, 1.68) [40] 
 Antioxidant intake: β-caroteneLimited1187OR 0.26 (0.04, 1.57) toMulti, NS
OR 1.75 (0.74, 4.11) [37]Multi, NS
 Psychosocial factors     
  Role functioningLimited1236OR 0.99 (0.75, 1.32) [42]Multi, NS
  Previous knee injuryLimited1354OR 1.2 (0.5, 3.0) to OR 1.1 (0.3, 4.4) [25]Uni, NS Uni, NS
  SmokingLimited1142OR 1.07 (0.38, 3.04) toUni, NS
    OR 0.96 (0.34, 2.75) [40]Uni, NS
Table 4. Evidence from high-quality studies for predictors of progression of knee OA: physical examination and clinical tests*
Baseline variableLevel of evidenceNumber of studiesCombined sample sizeAssociation with progression (95% CI) [ref.]Analysis/results
  • *

    Results are ordered by number of studies, effect size, and significance. High-quality studies are those with a bias score of ≥60% (≥12). OA = osteoarthritis; 95% CI = 95% confidence interval; BMI = body mass index; OR = odds ratio; uni = univariate results/adjusted; RR = relative risk ratio; multi = multivariate results; NG = data not given for adjusted results or multivariate models (but reported as the same); HR = hazard ratio; NS = nonsignificant results; [UPWARDS ARROW] = increased; JSW = joint space width; JSN = joint space narrowing; K/L = Kellgren/Lawrence; PF = patellofemoral; CPPD = calcium pyrophosphate dihydrate crystals.

  • Adapted from Lievense et al (15).

 BMI, kg/m2Conflicting1110,084  
  >27.3   OR 11.1 (3.28, 37.3) [40]Uni
  25.97–27.73   OR 5.28 (1.54, 18.1) [40]Uni
  >27.5   OR 2.1 (1.2, 3.7) toUni
    OR 3.2 (1.1, 9.7) [39]Uni
  >25.4   OR 2.6 (1.0, 6.8) [25]Uni
  >35   RR 1.8 (1.0, 3.2) [38]Multi
  26.7–34.7   RR 1.51 (1.22, 1.87) [45]Uni, NG
  25.0–26.6   RR 1.38 (1.10, 1.73) [45]Uni, NG
    OR 1.06 (1.00, 1.12) toMulti
    OR 1.07 (1.02, 1.14) [33]Multi
    HR 1.03 (1.00, 1.06) [49]Multi
    β = −0.03 ± 0.01 mm per kg/m2 [23]Multi
    RR 4.69 (0.63, 34.75) [46]Uni, NS
  25–34.9   RR 1.2 (0.7, 2.1) toMulti, NS
    RR 1.2 (0.7, 2.2) [38]Multi, NS
    OR 1.14 (0.89, 1.46) [42]Multi, NS
    OR 0.96 (0.72, 1.26) toMulti, NS
    OR 1.00 (0.83, 1.20) [29]Multi, NS
 BMI: stratified by alignment     
  ModerateConflicting1227OR 1.39 (1.07, 1.80) toMulti
    OR 1.14 (0.92, 1.40) [29]Multi, NS
  SevereLimited1227OR 0.92 (0.71, 1.18) toMulti, NS
    OR 0.96 (0.68, 1.36) [29]Multi, NS
  VarusLimited12,623RR 0.9 (0.7, 1.1) [38]Multi, NS
  ValgusLimited12,623RR 0.8 (0.6, 1.4) toMulti, NS
    RR 1.4 (0.9, 2.1) [38]Multi, NS
Combined radiographic featuresStrong42,240  
 [UPWARDS ARROW] osteophyte score 1288OR 0.47 (0.33, 0.66) [34]Uni
 JSW 1288OR 0.63 (0.47, 0.86) [34]Multi
 JSN 11,232HR 2.53 (1.94, 3.31) [49]Multi
 Severity (K/L grade) 1188OR 1.29 (1.07, 1.57) toMulti
    OR 1.72 (1.36, 2.19) [33]Multi
 Chondrocalcinosis 1532OR 2.01 (0.55, 7.42) [40]Uni, NS
Heberden's nodes/bony swelling/Bouchard's nodesConflicting3684OR 5.97 (1.54, 23.1) [40]Uni
OR 1.80 (1.02, 3.17) [33]Multi
OR 0.7 (0.4, 1.6) toUni, NS
OR 2.0 (0.7, 5.7) [25]Uni, NS
 Multiple joint OAStrong2720OR 3.28 (1.30, 8.27) [40]Uni
    OR 2.39 (1.16, 4.93) toMulti
    OR 2.42 (1.02, 5.77) [33]Multi
  Varus (with medial JSN)Strong21,180OR 2.98 (1.51, 5.89) toUni
    OR 4.09 (2.20, 7.62) [44]Uni
    OR 3.59 (2.62, 4.92) [41]Multi
  Valgus (with lateral JSN or [UPWARDS ARROW] K/L grade)Conflicting21,180OR 2.51 (0.91, 6.89) toUni, NS
   OR 4.89 (2.13, 11.20) [44]Uni
    OR 4.85 (3.17, 7.42) [41]Multi
  BilateralConflicting1230OR 2.23 (0.05, 4.41) toUni, NS
    OR 3.22 (1.28, 8.12) [44]Uni
  UnilateralLimited1230OR 0.17 (−1.66, 2.01) toUni, NS
    OR 2.33 (0.97, 5.62) [44]Uni, NS
 Leg length inequalityConflicting24,547HR 1.13 (0.53, 2.39) toMulti, NS
    HR 1.83 (1.10, 3.05) [30]Multi
    OR 1.0 (0.4, 2.5) toMulti, NS
    OR 1.3 (1.0, 1.7) [31]Multi
 Ipsilateral PF OALimited1288OR 2.31 (1.37, 3.88) toUni
    OR 3.36 (1.83, 6.18) [34]Multi
 Serum vitamin DLimited1126OR 2.83 (1.02, 7.85) toMulti
    OR 2.89 (1.01, 8.25) [36]Multi
 WarmthLimited1188OR 2.14 (1.30, 3.52) [33]Multi
 CPPDLimited1188OR 1.85 (1.04, 3.29) toMulti
    OR 2.41 (1.33, 4.39) [33]Multi
 [UPWARDS ARROW] volume synovial fluidLimited1188OR 1.02 (1.00, 1.05) toMulti
OR 1.03 (1.01, 1.05) [33]Multi
  Varus (with lateral JSN)Limited1950OR 0.12 (0.07, 0.21) [41]Multi
  Valgus (with medial JSN)Limited1950OR 0.34 (0.21, 0.55) [41]Multi
 Contralateral OALimited1288OR 1.53 (0.82, 2.85) [34]Uni, NS
 Localized OALimited1532OR 1.17 (0.51, 2.72) [40]Uni, NS
 Uric acid concentrationLimited1532OR 1.05 (0.36, 3.00) toUni, NS
    OR 1.36 (0.46, 4.02) [40]Uni, NS
 Muscle strengthLimited1263  
  Quadriceps   OR 0.88 (0.70, 1.11) [42]Multi, NS
  Hamstrings   OR 0.86 (0.60, 1.23) [42]Multi, NS

Patient characteristics with strong evidence for being predictors of knee OA progression are older age, presence of OA in multiple joints, and varus malalignment of the knee. Evidence combined from studies using different radiographic measures suggests that radiographic features at baseline are strongly associated with progression. There was strong evidence that physical activity or moderate participation in sport is not significantly associated with progression. All other variables had either a limited or conflicting level of evidence.

Sensitivity analysis.

Variation in length of followup in included studies has been noted. As knee OA has been recognized as a slowly progressing disease (3, 7), shorter followup times for studies may not allow sufficient time for progression to occur, resulting in smaller numbers for analysis of prognostic factors, a problem exacerbated in small cohort studies. Therefore, a sensitivity analysis was performed synthesizing evidence from studies with a followup period of 3 years or greater, a cutoff used in a previous review (7). Eleven studies met this criterion (25, 26, 30, 36, 37, 39, 40, 42, 45, 48, 49). BMI became a strong predictor of progression; evidence became limited for multiple joint OA and conflicting for radiographic baseline features. The level of evidence for a number of baseline variables remained the same, while for others (including alignment) no evidence is available for the longer time period.

A sensitivity analysis was also performed based on quality. Findings of the review did not change if the quality threshold was reduced (<60%), as no lower-quality studies used appropriate analysis to identify prognostic variables. However, when the threshold for high-quality studies was raised to ≥70% (i.e., ≥15 for the current study), 4 articles were reclassified as low quality (22, 26, 35, 45). Only 2 of these were previously included in the synthesis of evidence (26, 45). At the higher threshold of quality, there is strong evidence that baseline pain and function are not significantly associated with progression. Other findings are unaltered.

Results can also be synthesized according to the type of outcome measure used. Age remains a strong predictor of progression when investigated using function as an outcome measure (26, 33, 42), while evidence becomes conflicting if radiographic outcome is used (23, 34, 40). Other findings remain unchanged except baseline function, which changes to strong evidence of no association with progression when radiographic outcome is used (34, 48).


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Baseline patient characteristics with strong evidence for being predictors of knee OA progression consist of age, presence of OA in multiple joints (clinical observation), varus deformity of the knee (radiographic), and radiographic features (JSN/JSW, chondrocalcinosis, severity of OA as measured by K/L grade, and osteophyte score). Additionally, findings from sensitivity analyses suggest that higher BMI is a strong predictor of progression over a longer time period (>3 years), and pain and function have strong evidence of no association with progression (quality score of ≥15). These variables are easily assessed in clinical practice and may assist health care practitioners with providing their patients with appropriate advice and interventions for managing their disease. Patients can also be reassured that moderate participation in physical activity (patient report) is unlikely to have any effect on progression. Clinicians can have confidence in these findings, as the information is drawn from high-quality studies and exhibits consistency between studies. Additionally, study samples were drawn from a variety of settings, i.e., orthopedic and rheumatology clinics as well as community-based cohorts, making findings of the review generalizable. However, the quantity of information is somewhat limited for varus alignment, multiple joint OA, radiographic features, physical activity, knee pain, and function due to low study numbers for evidence synthesis. Additionally, for some variables effect sizes are small or inconsistent. These factors should be considered before applying review findings to clinical practice (19).

Table 5 compares findings of the current and previous reviews for variables with strong evidence only. Similar findings from multiple reviews increase confidence in the conclusions. For example, strong evidence from 2 reviews, plus limited evidence from another review, identifies varus alignment as a predictor of progression. This also highlights the importance of biomechanical influences on radiographic progression of knee OA.

Table 5. Comparison of findings from systematic reviews that identify variables with strong evidence for progression of knee OA*
Baseline variablesCurrent studyBelo et al, 2007 (5)Van Dijk et al, 2006 (7)Tanamas et al, 2009 (6)
  • *

    OA = osteoarthritis; JSN = joint space narrowing.

  • Other results limited/conflicting.

  • Not associated.

 Knee malalignmentStrong (varus alignment with medial JSN)LimitedLimitedStrong
 Multiple joint/generalized OAStrongStrong
 Radiographic featuresStrongStrong
 Hyaluronic acidStrong
Not associated with progression    
 Participation in physical activityStrongStrongLimited
 Muscle strengthLimitedStrongLimited (progression less likely)
 Previous knee injuryLimitedStrong
Unclear association with progression    
 Knee painConflictingStrongLimited (predictor)

There is strong evidence from 2 reviews that generalized or multiple joint OA (assessed by clinical observation) is a predictor of progression. While the precise reasons for this association are not entirely clear, it is possible that generalized OA may reflect an underlying systemic or genetic influence on cartilage that contributes to an increased likelihood of disease progression (40).

Similarly, 2 reviews found strong evidence that participation in physical activity is not significantly associated with progression, with limited evidence from another review. Physical activity included aerobic exercise, jogging, or being a member of a sports club (25, 40, 42). This lack of association with disease progression is helpful, given the many recognized health benefits associated with regular exercise.

There are inconsistent findings for the remaining variables in Table 5. For example, in the current review, combined radiographic features have strong evidence of being predictive of progression. In contrast, Belo et al (5) report strong evidence that radiologic severity is not associated with progression, having arrived at a different interpretation from similar studies.

Dissimilar review objectives may have contributed to disparity in findings. Belo et al (5) included articles with progression defined by radiographic measures only. Unlike the current review, they did not require inclusion of clinical features at baseline. Tanamas et al (6) conducted a review focused solely on the role of alignment as a predictor; van Dijk et al (7) concentrated on studies with changes in functional status or pain. Additionally, the pool of articles in each review was affected by decisions such as date of search, search strategies, and inclusion/exclusion criteria. Inclusion of different source studies contributes to the diversity of conclusions in the reviews. For example, age was found to be a strong predictor in the current review and conflicting in another. However, only 1 study contributed results on age to both reviews (40).

As previously mentioned, the number of studies will also affect the strength of evidence and may have contributed to the disparity in findings between reviews. Relatively few studies were included in the current review for variables such as function and pain. This makes synthesis of evidence susceptible to change with the addition or deletion of articles from the analysis. This point is reinforced by results of a sensitivity analysis that raised the threshold for high-quality studies to ≥70%. The exclusion of 1 study (26) changed the evidence from conflicting to strong so that pain and function were not significantly associated with progression. Tables 3 and 4 highlight the low number of studies from which the level of evidence is calculated.

There was a wide range of outcome measures used in included studies. Results of a sensitivity analysis revealed that age remained a strong predictor when assessed with change in function (26, 33, 42); however, evidence became conflicting when using radiographic outcome measures (23, 34, 40), exemplifying how the choice of outcome measure for knee OA progression can influence results. However, there are no universally accepted outcome measures and most have recognized limitations. For radiographs, these include the slow rate of progression and the fact that early disease may not be manifested in radiographic features (4, 50). Outcome measures describing change in pain or function lack responsiveness, with limited evidence that they deteriorate over a 3-year period (7). Levels of pain or function may reflect short-term fluctuations in disease activity rather than progression. Alternatives such as magnetic resonance imaging (MRI) changes or biomarkers were not utilized in this review due to uncertainty about their responsiveness in detecting progression, lack of standardized or universally accepted scoring methods, and their lack of a clear relationship to clinical symptoms (4, 50). In addition, they may lack clinical utility due to cost and technical requirements, being more suited to use in the research environment.

The focus of this systematic review was to identify predictive variables that could be easily assessed in routine clinical practice. This resulted in exclusion of some studies that used sophisticated equipment or analytical techniques, making it possible that other predictors of knee OA progression have been missed. Other limitations of the current review relate to shortcomings of included studies. Prognostic studies have been criticized for poor quality and variable methodology (18). The potential impact on the current findings was addressed with the assessment of bias tool. Results in Table 2 suggest that conclusions from this review could be at risk of selection bias and attrition bias due to weaknesses in included studies. Improving selection and reporting of study participants, especially response rates at different stages, and reasons for dropouts would address these biases and should be incorporated into future research. Two-thirds of reviewed studies addressed confounding by age and sex, although there may be other potential confounders not investigated in these studies, so it is not possible to conclude definitively that confounding was absent (13). However, the requirement for adjusted or multivariate analyses in this review should have minimized the effect of confounding.

The lack of standardized methods of assessment for baseline variables in addition to the variety of outcome measures already outlined are areas that should be addressed in future research. Consistent measures would make pooling of results and meta-analysis more likely and thereby strengthen the body of evidence and confidence in the findings. This approach may need to be tempered if new methods for defining progression are validated and adopted as best practice (e.g., MRI changes or biomarkers) (4, 50).

Identification of predictors of long-term progression is an area requiring further research. Existing, well-designed, and adequately powered studies can indicate known variables worthy of further investigation, rather than searching for new predictors (17).

This systematic review has summarized current evidence for baseline characteristics that predict progression of knee OA. It has also highlighted areas where methodology is lacking and possible directions for future research. All of the variables identified with strong evidence as predictors (varus alignment, presence of OA in multiple joints, age, radiographic features, and BMI) can be easily evaluated and utilized in clinical practice. Additionally, knowledge that participation in physical activity is not associated with progression should be used to encourage patients to remain active. Patients can also be reassured that the presence of baseline pain and limitation of function is not associated with progression. There are numerous other potential predictors where current evidence is limited or conflicting. They provide a direction for future research, which should be undertaken using existing standardized and validated methods of assessment and outcome measurement.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Ms Chapple had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Chapple, Nicholson, Baxter, Abbott.

Acquisition of data. Chapple.

Analysis and interpretation of data. Chapple, Nicholson, Baxter, Abbott.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

We thank Brigid Ryan, research assistant, for screening articles for inclusion in the review, scoring articles, abstracting, and checking data; Jan Wyllie, medical librarian, for assisting with developing the search strategy; and Alexis Wright, PhD, for assistance in performing the review and for reviewing the manuscript.


  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  8. Acknowledgements
  10. Supporting Information

Additional Supporting Information may be found in the online version of this article.

ACR_20492_sm_appendix.doc200KSupplementary Data

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.