Introduction

The prevalence of osteoarthritis of the knee (OA) is increasing worldwide and this burden will continue to increase owing to aging of the general population [95]. Consequent to an increase in incidence is the rise in the number of patients with knee OA who are prone to further deterioration of the knee. It therefore is important to better understand, control, and attempt to prevent further progression of disease in patients with knee OA.

In 2007, Belo et al. [4] published the first systematic review on prognostic factors for progression of knee OA. They found that generalized OA and hyaluronic acid levels were associated with progression of knee OA. Knee pain, baseline radiographic severity, sex, quadriceps strength, knee injury, and regular sport activities were not associated. For the remaining factors the evidence was limited or conflicting. Their literature search had been performed up to December 2003; however, many articles studying radiographic progression of knee OA have been published in the decade since that review. Therefore, we performed an update of the systematic review of observational studies by Belo et al. [4] to determine the currently available evidence on prognostic factors for radiographic progression of knee OA.

Search Strategy and Criteria

Literature Search

In the review by Belo et al. [4], the search of the literature had been performed in MEDLINE and EMBASE for all available observational studies up to December 2003. We searched in MEDLINE and EMBASE from December 2003 up to February 2013. Key words were: knee, osteoarthritis (or arthritis, or arthrosis, or degenerative joint disease), progression (or prognosis, or precipitate, or predictive), and case-control (or cohort, or longitudinal, or follow-up). Articles were reviewed for inclusion independently by two authors (ANB and JNB or JR). The following inclusion criteria were used: 85% or more of participants in the analyses for OA progression had radiographic evidence of knee OA at baseline; the study investigated determinants associated with radiographic knee OA progression; radiographic progression was the outcome measure; the study had a case-control or cohort design with a minimal 1-year followup; full text of the article was available; the study was in English, Dutch, German, or French. Studies that observed the incidence of knee OA were excluded. A detailed description of our search strategy is available online (Appendix 1. Supplemental materials are available with the online version of CORR®). All articles were reviewed for inclusion independently by two authors (ANB and JNB or JR). Studies that used MRI features to define OA progression were excluded. However, studies determining MRI features as prognostic factors were included.

Methodologic Quality

The same methodologic quality assessment criteria as in the original review by Belo et al. [4] were used for this review (Table 1). These criteria were based on established criteria used in systematic reviews of prognostic factors for patients with musculoskeletal disorders and were described by Lievense et al. [49], Scholten-Peeters et al. [69], and Altman [1]. The criteria cover the internal validity and the informativeness of the study. All included articles were scored independently by two authors (ANB and JNB or JR). Cohen’s kappa coefficient (κ) was calculated to indicate the interrater agreement.

Table 1 Methodologic quality assessment criteria

Data Extraction

Study population characteristics, observed risk factors, definitions of knee OA progression, and measures of association were extracted.

Evidence Synthesis

Odds ratios (ORs), relative risks (RRs), or hazard ratios (HRs) were pooled when there was consistency in definition of study population, measured determinants, and assessed outcome (using Review Manager [RevMan], Version 5.3; Copenhagen, Denmark: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). We tested for heterogeneity with the chi-square and I-square tests. If heterogeneity was absent, a fixed effects model was applied to calculate pooled OR through the Mantel Haenszel test. In the absence of consistency among definitions for OA, a best-evidence synthesis was used to summarize the data. The level of evidence was based on the updated guidelines by Furlan et al. [34] and was divided into the following levels: (A) strong, ie, consistent (> 75%) findings among two or more high-quality studies; (B) moderate, ie, findings in one high-quality study and consistent findings in two or more low-quality studies; (C) limited, ie, findings in one high-quality study or consistent findings in three or more low-quality studies; and (D) conflicting or inconclusive evidence, ie, less than 75% of the studies reported consistent findings, or the results were based on only one study. High quality was defined as a quality score of 9 or greater (> 65% of the maximal attainable score). When performing the best-evidence synthesis, we only differentiated between high- and low-quality studies.

Studies Included

Of the 1912 articles identified using our search strategy, 43 met the inclusion criteria [2, 5, 7, 11, 13, 19, 20, 2528, 30, 35, 3844, 46, 48, 5052, 55, 5762, 6466, 73, 74, 78, 85, 88, 9193]. Belo et al. reviewed 36 articles [3, 8, 12, 1416, 18, 2124, 29, 31, 32, 37, 45, 47, 53, 54, 56, 63, 7072, 7577, 7983, 87, 89, 94, 96]; therefore the total number of included studies was 79, studying 59 different determinants for the progression of knee OA (Table 2). Three reviewers scored 559 items for the methodologic quality assessment of the 43 newly included articles and agreed on 519 items (93%; κ = 0.79). The 53 disagreements were resolved in a single consensus meeting. Seventy-two of the 79 included articles were scored as high quality (score, 9–13), and only one article had the maximum attainable score. The remaining seven were scored as low quality, however no article was scored less than 6. Six different criteria were used for the inclusion of participants with OA and 13 definitions were applied to define radiographic OA progression. Furthermore, there were differences in how the determinants under study were measured, ie, continuous, dichotomous, or categorical with varying cut-off points.

Table 2 Study characteristics of the reviewed manuscripts (n = 79)

Study Results

Because of the large number of studied determinants (n = 59), we pragmatically grouped our findings into five different categories: systemic factors (Table 3); disease characteristics (Table 4); intrinsic factors (Table 5); extrinsic factors (Table 6); and markers (Table 7). Some authors presented statistically significant associations to OA progression, but used p values or regression coefficients as measures of association [3, 5, 12, 14, 20, 21, 23, 31, 37, 41, 42, 44, 45, 47, 48, 52, 62, 63, 72, 74, 77, 80, 82, 85, 87, 93]. We chose to present only OR, RR, or HR as measures of associations; however, we have tabulated whether there was a significant association with OA progression in an article.

Table 3 Systemic factors discussed in the reviewed studies
Table 4 Disease characteristics discussed in the reviewed studies
Table 5 Intrinsic factors discussed in the reviewed studies
Table 6 Extrinsic factors discussed in the reviewed studies
Table 7 Markers discussed in the reviewed studies

Sensitivity Analysis

For factors in which we were forced to use a best-evidence synthesis, we conducted a sensitivity analysis to check whether differences in sample size could have altered our conclusions. Additionally we checked whether large variances in followup could have led to different conclusions.

Results

Summaries of the results for systemic factors, disease characteristics, intrinsic factors, extrinsic factors, and markers are available (Appendix 2. Supplemental material is available with the online version of CORR®.).

Pooled Results

The presence of knee pain at baseline and Heberden nodes were associated with the progression of knee OA. The pooled ORs based on pools of studies with consistency among the definitions for OA inclusion, OA progression, and the determinant under study, were 2.38 for knee pain at baseline (95% CI,1.74–3.27; I2 = 52%) (Fig. 1) and 2.66 for the presence of Heberden nodes (95% CI, 1.46–8.84); I2 = 0%) (Fig. 2). Because of the large number of determinants with only a restricted number of studies per determinant and owing to lack of consistency between the reviewed studies regarding inclusion criteria, outcome measures, and measures of association, statistical pooling was not possible for the majority of the determinants.

Fig. 1
figure 1

A forest plot for the pooled odds ratio (OR) shows the association between the presence of knee pain at baseline and radiographic progression of knee osteoarthritis (OA). The OR can deviate from the OR in Table 4 because pooled ORs were obtained through crude ORs, as opposed to the adjusted OR in Table 4. The results from Dieppe and Wolfe for pooling were not available and were not included in this analysis. The results from the chi-square and I2 tests indicate homogeneity between the studies. M–H = Mantel Haenszel test; Fixed = fixed effects model; df = degrees of freedom.

Fig. 2
figure 2

A forest plot for the pooled odds ratio (OR) shows the association between the presence of Heberden nodes at baseline and radiographic progression of knee osteoarthritis (OA). The OR can deviate from that in Table 4 because pooled ORs were obtained through crude ORs, as opposed to the adjusted OR in Table 4. The results from the chi-square and I2 tests indicate homogeneity between the studies. M–H = Mantel Haenszel test; Fixed = fixed effects model; df = degrees of freedom.

Best-evidence Synthesis

For the remaining determinants, we applied a best-evidence synthesis, which showed that based on consistent findings in multiple high-quality studies, there seems to be strong evidence that varus alignment, serum TNFα level, and serum hyaluronic acid level are associated with radiographic progression of knee OA. There also is strong evidence that sex (female), former knee injury, quadriceps strength, smoking, running, and regular performance of sports are not associated with progression of knee OA.

There was moderate evidence showing that a higher dietary intake of vitamin D is inversely associated with progression of knee OA. Thus far, there is limited evidence that ethnicity, metabolic syndrome, genetic components adduction moment, meniscal damage, knee ROM, general vitamin and β-carotene intake, serum levels IL-10 and N-propeptide of type II collagen, synovial levels aggrecan neoepitope amino acid sequence and IL-18, and fractal dimension progression on radiographic fractal signature analysis are associated with progression of knee OA. There also is limited evidence that knee OA progression is not associated with osteoporosis; past or present estrogen use; uric acid concentrations; depression or anxiety; hand grip (muscle) strength; bone marrow lesions or edema; meniscectomy; chondrocalcinosis; MRI-detected subchondral bone cysts, cartilage loss, or joint effusion; AP knee laxity; vitamin E intake; serum levels IL-1Ra and transforming growth factor-β1; and 99mTc-MDP uptake on bone scintigraphy.

Conflicting evidence was found for the associations between knee OA progression and age; low bone density; serum insulin growth factor-1 level; baseline radiographic or clinical OA severity; generalized osteoarthritis; duration of symptoms; valgus alignment or malalignment in general; past knee injury; the presence of tibiofemoral osteophytes; BMI; leg length inequality; serum vitamin D level; dietary intake of vitamin C; serum C-reactive protein, IL-1β, keratan sulfate, and serum cartilage oligometric matrix protein levels, and urinary crosslinked C-telopeptide level. Inconclusive evidence was found for the determined associations between knee OA progression and the single nucleotide polymorphisms CILP_395 (cartilage intermediate-layer proteins) and rs3740199, patellofemoral alignment, and serum pentosidine levels. There also was inconclusive evidence for no associations found between knee OA progression and the single nucleotide polymorphisms rs1871054, ADAM12_48 (A disintegrin and matrix metalloproteinase domain 12), and TNA_106 (tetranectin plasminogen-binding protein), and serum levels of YKL-40 (chitinase-3-like protein 1), MMP-9 (matrix metalloproteinase-9); and TIMP-9 (tissue inhibitors of metalloproteinase).

Sensitivity Analysis

In this analysis, we tested whether conclusions from relatively small studies (less than 200) incorrectly influenced conclusions drawn from larger studies with more statistical power studying the same determinant, or that results from studies with a relatively short followup (cutoff 24 months) altered conclusions from studies with a longer followup. Our sensitivity analysis found that our conclusions did not change across the range of clinically plausible differences in followup duration or sample size regarding the strong, moderate, or conflicting evidence we found for the various presented determinants.

Discussion

We performed an updated systematic review of available evidence regarding prognostic factors for radiographic knee OA progression. We found that there is strong evidence that baseline knee pain and Heberden nodes, varus alignment, and high baseline serum levels of hyaluronic acid and TNFα are predictive for knee OA progression. There also seems to be strong evidence that sex (female), former knee injury, quadriceps strength, smoking, running, and regular performance of sports are not predictive for progression of knee OA. For all other studied factors in our review, the evidence is limited, conflicting, or inconclusive. In the best-evidence synthesis, we considered only significant associations as associated prognostic factors. However, several of the included articles had small sample sizes, which consequently can lead to lower statistical power and more often to failure to detect differences that might be present.

A possible limitation to our inclusion criteria was addressed by Zhang et al. [97]. They reported that, unlike randomized trials, observational studies of patients with preexisting disease are subject to various biases that may account for discrepancies found between risk factors for disease incidence and progression. They hypothesized that risk factors actually might exist for progressive knee OA but that flaws in study design and the measure of disease progression may prevent us from detecting risk factors [97]. Having cited their article, it seems reasonable that there is the possibility that we have not determined all risk factors for progression of knee OA, because some factors might not have achieved significance in multivariable analyses in a study and thus were not included in our evidence synthesis. Nonetheless, we believe we have summarized all presently known risk factors of which a possible association with knee OA progression has been studied.

We acknowledge that when applying a best-evidence synthesis, one might unjustly conclude that there may be conflicting or strong evidence for or against an association of the determinant under study with knee OA. We would have preferred to pool the data of all included studies. However, because of large variation in criteria used in the articles for defining disease, or disease progression, pooling of the data generally was not possible. We encountered six different criteria that were used for the inclusion of OA (Table 2). Another approximately 13 different definitions were applied for OA progression (Tables 37). Furthermore, there were differences in how the determinants under study were measured, (continuous, dichotomous, or categorical), and varying cutoff points were used. As previously described, we pooled the results for “knee pain” and “Heberden nodes” for which both results showed associations with the progression of knee OA. This is different from the conclusions we would have drawn from a best-evidence synthesis, which would show conflicting evidence for both determinants. In our opinion, it is likely that more of the conflicting associations we presented are attributable to the differences in definitions of knee OA or knee OA progression. For example, the conflicting evidence for BMI probably would be altered if statistical pooling was feasible; given that all 11 significant risk estimates (OR/RR/HR) regarding BMI were positive associations and that six of the 12 nonsignificant associations also were positive associations, it seems likely that if pooled, the combined overall association between BMI and knee OA would be a positive, significant one. In addition, the conflicting evidence for age, seven of the 10 presented analyses (70%) showed no significant association, falling just short for the criteria for ascertaining strong evidence (> 75%) for no association between age and OA progression.

In the original review by Belo et al. [4] and in a review by van Dijk et al. [86], the evidence for association between varus alignment and OA progression was limited. However, a couple studies have been performed since these reviews were published that have determined significant associations with varus alignment, which enabled us to conclude that there is strong evidence for this finding. The latter is in accordance with results published in later systematic reviews by Tanamas et al. [84] and Chapple et al. [17]. Except for the original review by Belo et al., there are to our knowledge no other reviews available that have determined the predictive value of serum hyaluronic acid levels and OA progression [9]. In addition, to our knowledge, no reviews have been published assessing the predictive value of serum level TNFα for knee OA progression.

We found strong evidence that sex was not associated with knee OA progression, as did Belo et al. [4]. This is in contrast to the earlier reviews published by van Dijk et al. [86] and Chapple et al. [17]. van Dijk et al. found limited evidence for the absence of an association with sex, but they included articles that used physical functioning as an outcome measure. Chapple et al. found conflicting evidence; however, their evidence was based on four analyses of three studies, which also are included in our review [21, 47, 70]. Three of the four analyses were consistent (no association); one was conflicting (significant association) [47]. Our evidence synthesis was based on 10 analyses, of which nine analyses were consistent (no association), consequently outweighing the one conflicting finding. van Dijk et al. and Chapple et al. reported limited evidence for the absence of an association between quadriceps strength and knee OA progression. This is consistent with our finding; however, our conclusion is based on more evidence. Consistent results also were found for regular performance of sports, in which van Dijk et al. reported limited and Chapple et al. reported strong evidence for absence of an association. However, in articles by Fransen and McConnell [33] and Bennell and Hinman [6] reviewing the effect of exercise therapy in patients with knee OA, the authors reported that exercise has a short-term benefit in patients with knee OA, although the magnitude of the reported benefit is small. This highlights the importance of the need to understand the working mechanism of exercise therapy.

A topic of considerable interest is the potential association between BMI and knee OA progression. Previous reviewers have established a positive association between BMI and incident knee OA [10, 95]. However, the evidence for an association between BMI and progression of knee OA remains conflicting in our review, which is consistent with the findings by Belo et al. [4] and Chapple et al. [17].

Noteworthy is the lack of overlap in evidence for prognostic factors for hip and knee OA progression. In two large reviews studying prognostic factors for hip OA, Lievense et al. [49] provided strong evidence for an association between hip OA progression with type of hip migration and with atrophic bone response. They also presented strong evidence for the absence of an association with BMI. Wright et al. [90] reported strong evidence for association of hip OA progression with age, joint space width at entry, femoral head migration, femoral osteophytes, bony sclerosis, baseline hip pain, and certain hip OA severity indexes. They also provided strong evidence for the absence of an association with acetabular osteophytes. The discrepancy between the findings for hip and knee OA is unclear but could be attributable to the difference in the number of studies available determining risk factors for progression of hip or knee OA [9].

Future research on the true relationship between prognostic factors for radiographic progression of knee OA is needed, mainly on the factors where conflicting evidence was presented (eg, age, baseline OA severity, BMI). Furthermore, we presented limited, inconclusive, or conflicting evidence on many factors with potential associations with OA progression. It would be important to investigate determinants that can be influenced or modified to reduce the risk of OA progression, perhaps including metabolic syndrome, bone marrow lesions, or osteoporosis. Moreover, there would be obvious advantages to testing the effect of new or existing disease-modifying pharmacologic or surgical interventions in patients with an established increased risk of OA progression.

We found strong evidence that baseline knee pain and Heberden nodes, varus alignment, and high baseline serum levels of hyaluronic acid and TNFα are predictive for knee OA progression. Sex (female), former knee injury, quadriceps strength, smoking, running, and regular performance of sports are not predictive for progression of knee OA. Many studies have been performed and are being performed determining risk factors for knee OA progression, but the variability in how OA and OA progression are defined across the relevant studies remains an impediment to pooling the available evidence. We strongly recommend future researchers use uniform definitions of determinants, disease, and disease progression; it would enable more precise determination of possible risk factors for knee OA progression through meta-analyses. The majority of the included studies used the Kellgren-Lawrence classification as definition of disease and disease progression. This classification has been criticized because the criteria have been described and interpreted differently in various studies [67]. However, the Kellgren-Lawrence criteria provide a reliable classification of knee OA and OA progression, given that the original description of the criteria are applied [67, 68]. We therefore recommend that future researchers use the Kellgren-Lawrence classification to define radiographic OA and OA progression. Furthermore, considering that some MRI scoring systems have been and currently are being developed to define knee OA progression [36], it seems preferable that the same MRI scoring system would be used universally in future studies on prognostic factors for knee OA progression. We would like to call on expert committees, such as the Osteoarthritis Research Society International (OARSI) for OA Imaging to announce their recommendations on this important topic.