Original Article

*Correspondence to J. N. Belo, Erasmus Medical Center Rotterdam, Department of General Practice, PO Box 2040, 3000 CA, Rotterdam, The Netherlands

Received: 28 June 2005; Accepted: 4 April 2006

10.1002/art.22475  About DOI

INTRODUCTION

Radiographic osteoarthritis (OA) of the knee affects more than 33% of persons age 60 years and older, whereas 10-15% of persons age 60 years and older have symptomatic knee OA in which symptoms are defined as knee pain on most days ([1]). Because of aging of the population, the prevalence of OA is expected to increase in the next decades ([2]); in western countries, the increase in prevalence in the next 20 years is expected to be around 40%, making OA the fourth leading cause of disability ([2][3]).

To optimize the management of OA, it is important to increase our knowledge regarding the predictors of progression of OA. If certain prognostic factors are modifiable, they may enhance our ability to reduce OA progression. Even if these prognostic factors are not modifiable, they can still be used to identify high-risk groups, which may have implications for patient information and the perspective of medical treatment ([4]). Knowledge about modifiable factors and high-risk groups is also relevant for clinical research, such as for evaluating therapeutic interventions including disease-modifying therapies.

Several original studies have reported on prognostic factors of knee OA progression. Until now, however, a systematic overview of the determinants of progression of knee OA has not been available. To gain insight into the prognostic factors of progression of knee OA, we systematically reviewed the available studies on this topic using modern methods of identifying, assessing, and summarizing the available evidence.

MATERIALS AND METHODS

Identification of studies

To identify the observational studies on this subject, a search for relevant studies published up to December 2003 was performed in Medline and Embase. Key words used were knee, osteoarthritis (or arthritis, or arthrosis), prognostic (or progressive, or predictive, or precipitate), and case-control (or cohort, or retrospective, or prospective, or longitudinal, or followup). Screening the list of references of all identified relevant articles extended the search. A study was included when it fulfilled all of the following criteria: patients in the study had clinical or radiographic evidence of knee OA; the study investigated factors associated with the radiologic and/or clinical progression of knee OA; the followup period was at least 1 year; the study design was a prospective cohort or a nested case-control study; the article was written in English, Dutch, German, or French; full text was available for the article; and the study population had no underlying pathology (e.g., rheumatoid arthritis, bacterial infection) of the joint.

Methodologic quality

To assess the methodologic quality of the included articles, a scoring list was used (Table 1) based on the scoring lists used by Lievense et al ([5]) and Scholten-Peeters et al ([6]) and on the framework for assessing internal validity of articles dealing with prognosis as described by Altman ([7]). All studies were scored independently by 2 of the reviewers (JNB and MYB or MR). For each study, a total quality score was computed by counting all positively rated items (maximum score 13 points). In case of disagreement, both reviewers tried to achieve consensus. When consensus was not achieved, a third reviewer (BWK) was asked to give a final judgment.

Data extraction

Study characteristics (followup duration, study population characteristics) were extracted and, when possible, the odds ratio (OR) or relative risk was provided or calculated. Otherwise, other measures of associations (hazard ratio, correlations) or values for statistical significance (P value) of the reported association were given.

Evidence synthesis

ORs or relative risks were statistically pooled when clinical and statistical homogeneity in several studies was assumed. In the absence of homogeneity, a best evidence synthesis was used to summarize the data.

The level of evidence was based on the guidelines of van Tulder et al ([8]) and was divided into the following levels: 1) strong evidence (consistent [>75%] findings among multiple [ 2] high-quality studies); 2) moderate evidence (findings in 1 high-quality study and consistent findings in multiple low-quality studies); 3) limited evidence (findings in 1 high-quality study or consistent findings in multiple low-quality studies); and 4) conflicting evidence (provided by conflicting findings [<75% of the studies reported consistent findings]). When strong evidence was provided only by studies with a small sample size, we decided to judge those studies as not strongly associated. Articles were judged as high quality when they had a quality score >8 (>60% of the maximal attainable score). Only statistically significant associations were considered as associated prognostic factors in the best evidence synthesis. When several radiologic outcomes for the progression of OA were provided, the best evidence synthesis was based on the increase in the Kellgren and Lawrence (K/L) score or the decrease in joint space width (JSW).

RESULTS

Studies included

Of the 1,004 articles identified using our search strategy, 37 articles met the criteria for inclusion. In the methodologic quality assessment, the 3 reviewers scored 520 items in total and agreed on 402 items (77%;   = 0.40). The 118 disagreements were resolved in a single consensus meeting. Almost all studies were of high quality and scored in the range of 9 to 12 (maximum score 13). One study (Sahlström et al [[9]]) scored 6 (<50%); because this was the only study of low quality, we excluded it from the analysis. An overview of the characteristics of the included studies (n = 36) is presented in Table 2.

All studies had a prospective research design. The studied populations were either population or hospital based. For the definition of OA, most studies used the Kellgren and Lawrence criteria or JSW. One study ([12]) used the Osteoarthritis Research Society International Atlas for the definition of OA, and another study ([19]) used the presence of clinical and radiologic OA. All studied populations, except 2 ([12][34]), contained more women than men. Progression was defined as radiologic progression in almost all studies. Only 4 studies ([11][19][23][24][45]) (15%) also reported on a clinical outcome (total knee replacement).

Study results

An overview of the potential prognostic factors and their relationship to the progression of radiologic OA is presented in Tables 3, 4, 5, and 6 and is summarized below.

Systemic factors (Table 3).

Schouten et al ([37]) found a significant association between age and progression of knee OA only for the comparison of the fourth quartile (higher age) versus the first quartile (lower age). In their study, there was no significant association reported for the second and third quartile versus the first quartile. Miyazaki et al ([22]) also found a significant association between age and progression of knee OA. Bagge et al ([17]), Dieppe et al ([19]), and Felson et al ([31]) reported no significant association, but an OR or P value was not provided. Wolfe and Lane ([45]) also found no significant association.

Although no ORs or P values were given, Dieppe et al ([19]) and Felson et al ([31]) reported no significant association between sex and progression of knee OA. Ledingham et al ([21]) reported an association between sex and the change in cyst number but not between sex and change in K/L grade or joint space narrowing (JSN). Miyazaki et al ([22]), Schouten et al ([37]), Spector et al ([39]), and Wolfe and Lane ([45]) found no significant association between sex and progression of knee OA.

Hart et al ([33]) reported no statistically significant difference in bone density between the nonprogressive group and the progressive group (OR or P value not provided). In contrast, Zhang et al ([27]) found an association between high versus low bone density and progression of knee OA (fourth versus first quartile OR 0.1, 95% confidence interval [95% CI] 0.03-0.3). The difference in progression of knee OA for the change in bone density of the lowest versus the second lowest group was not significant.

For insulin-like growth factor 1 (IGF-1), Schouten et al ([37]) found a significant association between progression of knee OA in the third versus the first tertile only. No significant association was found for the second versus the first tertile. Fraenkel et al ([32]) found no association between IGF-1 and progression of knee OA in the comparison of the third tertile versus the first; they also found no significant association for the comparison between men and women of the second tertile versus the first tertile.

Zhang et al ([26]) investigated the relationship between estrogen use and radiologic progression of knee OA. No significant association was found between current use, past use, or never use of estrogen and progression of knee OA. Schouten et al ([37]) found no association between uric acid concentration and progression of knee OA.

Disease characteristics (Table 4).

Only the study by Wolfe and Lane ([45]) found a significant relationship between knee pain at baseline and progression of knee OA. Dieppe et al ([19]) also investigated the relationship between knee pain at baseline and a subsequent operation of the knee, and found a significant association (P   0.001).

For markers of bone or cartilage turnover, Bettica et al ([28]) found a relationship between the level of type I collagen telopeptides in urine and progression of knee OA (P value not provided). One study by Bruyere et al ([42]) and 2 studies by Sharif et al ([23][25]) found a significant association between the level of hyaluronic acid in serum and progression of knee OA. Conflicting associations were found for the level of keratan sulfate and the level of cartilage oligomeric matrix protein (COMP). With regard to COMP, Bruyere et al ([42]) found no significant association (OR or P value not provided) in contrast to Sharif et al ([24]) and Vilim et al ([16]). Bruyere et al ([42]) found no significant associations between osteocalcin, pyridinoline, or deoxypyridinoline and progression of knee OA (OR or P values not provided). Doherty et al ([30]) found a statistically significant association for the level of inorganic pyrophosphate in the synovial fluid (OR 0.97, 95% CI 0.95-0.99). Ledingham et al ([21]) found a significant association between the presence of calcium pyrophosphate crystals and change in attrition of the knee joint (OR 2.41, 95% CI 1.33-4.39); the relationship of this determinant to progression in the K/L score or JSN was not provided. Sugiyama et al ([41]) found a significant association between the level of type II procollagen propeptide and progression of knee OA (P = 0.001).

Concerning severity of OA, only Wolfe and Lane ([45]) found a significant association between the initial JSW score and progression. Ledingham et al ([21]) found a significant association with the change in attrition; in their study, no association was found between radiologic severity and change in the K/L score or JSN. Contradictory associations were found in the relationship between clinical severity and progression of knee OA ([11][45]).

Whereas Schouten et al ([37]) reported a significant association with Heberden's nodes, Cooper et al ([29]) found no significant association.

Ledingham et al ([21]) and Schouten et al ([37]) both reported a significant positive association between the presence of generalized OA (radiologically and/or clinically determined) and progression of knee OA. The clinical diagnosis of local OA made by a physical examination was not related to radiologic progression of knee OA in the study by Schouten et al ([37]).

Dieppe et al ([19]) found no association with duration of symptoms (OR or P value not provided) and Wolfe and Lane ([45]) found a borderline significant association. In contrast to the significant relationship between C-reactive protein level (CRP) and progression of knee OA found by Spector et al ([40]), Sharif et al ([25]) did not find a significant association.

Ledingham et al ([21]) described synovial fluid volume and nodal warmth in relation to progression of knee OA. For synovial fluid volume as a continuous variable, a significant relationship was found (change in K/L score OR 1.03, 95% CI 1.01-1.05; change in attrition OR 1.80, 95% CI 1.00-1.05). With regard to nodal warmth in relation to change in K/L score, a significant relationship was found (OR 1.80, 95% CI 1.02-3.17).

Felson et al ([12]) found an association between medial bone marrow edema lesions versus no medial lesions in relation to progression of knee OA (OR 5.6, 95% CI 2.1-14.8). No association was found between lateral bone marrow edema lesions versus no lateral lesions in relation to progression.

Intrinsic factors (Table 5).

Three studies ([10][14][22]) reported a statistically significant association between varus alignment and progression of OA measured by a decrease in JSW. A nonsignificant relationship between the varus alignment and progression of OA was only found in the analysis of the K/L grade 0-1 group in the study by Cerejo et al ([10]). In the study by Miyazaki et al ([22]), a statistically significant OR was found for the univariate analysis of varus alignment and progression of knee OA (OR 3.10, 95% CI 1.07-9.12), but not in the multivariate analysis. Cerejo et al ([10]) and Sharma et al ([14]) also investigated the relationship between valgus alignment and progression of lateral knee OA. Both studies found a statistically significant relationship with progression of OA (valgus versus nonvalgus OR 10.44 and 4.89, respectively). Sharma et al ([14]) also compared varus with neutral/mild valgus and valgus with neutral/mild varus. In both comparisons, a statistically significant association was found (OR 2.98 and 3.42, respectively).

Miyazaki et al ([22]) also investigated the association between the adduction moment and progression of OA. For the adduction moment, a statistically significant association was reported.

Two studies reported the relationship between injury of the knee joint and progression of OA. Both studies found no statistically significant relationship. In the study by Cooper et al ([29]), the relationship between previous knee injury and progression of OA was investigated. In the study by Schouten et al ([37]), knee injury was assessed at followup. Schouten et al ([37]) also investigated the relationship between sport injury and progression of OA, but no statistically significant association was found.

For the relationship between meniscectomy and progression of OA, Schouten et al ([37]) found no statistically significant association. In the same study, the evaluated relationship between chondrocalcinosis and progression of knee OA was not statistically significant. Hart et al ([33]) investigated the relationship between the history of a fracture and the progression of OA, but no statistically significant relationship was found (OR or P value not provided).

Extrinsic factors (Table 6).

In the study by Cooper et al ([29]), a significant relationship with body mass index (BMI) was only found in the comparison of the highest tertile versus the lowest tertile in the group with baseline K/L grade 2 or higher. No statistically significant relationship was found in the comparison of the middle tertile versus the lowest tertile in the group with baseline K/L grade 1 or higher or grade 2 or higher (OR 2.3, 95% CI 0.8-6.4 and OR 1.8, 95% CI 0.4-8.2, respectively) and in the comparison of the highest versus the lowest tertile in the group with baseline K/L grade 1 or higher. The studies by Dieppe et al ([19]), Miyazaki et al ([22]), and Spector et al ([15]) found no statistically significant relationship between BMI and the progression of OA. Schouten et al ([37]) found a significant association in the 2 highest quartiles versus the lowest quartile. In the comparison of the second quartile versus the first quartile, no statistically significant association was found. Ledingham et al ([21]) only found a significant association for the relationship of BMI with JSN. In that study, a borderline significant association was found for the relationship with the change in osteophytes and no statistical association was found in the relationship between change in K/L grade. Wolfe and Lane ([45]) also found a borderline significant relationship between BMI and progression of OA. Spector et al ([15]) also investigated the relationship between change in BMI and progression of OA, but no statistically significant association was found (OR or P value not provided).

Two studies ([18][38]) investigated the relationship between quadriceps strength at baseline and progression of OA. Both studies found no statistically significant association.

Lane et al ([34]) and Schouten et al ([37]) investigated the relationship between running and progression of OA. In both studies, no statistically significant association was reported. In the study by Lane et al ([34]), running activities occurred during the followup period. In the study by Schouten et al ([37]), running was only assessed at baseline and no information about running during the followup period was presented.

Cooper et al ([29]) found no statistically significant association between regular sport activities assessed during followup and progression of OA (OR 0.7, 95% CI 0.4-1.6). Schouten et al ([37]) analyzed different types of activities. All of these activities were assessed at followup. For physical activity in general, no statistical association was found. For walking, squatting/kneeling/crawling, knee knocking, and lifting heavy objects, no statistically significant association was found. For standing, a significant association was only found in the comparison of the medium versus the lowest level. A statistically significant association was also found between bowing legs or knocking knees and the progression of OA.

McAlindon et al ([13][35]) investigated the relationship between nutrition variables and progression of OA. For vitamin D, a low dietary intake and low serum level were statistically and significantly associated with progression of knee OA (OR 4.05, 95% CI 1.40-11.6 and OR 2.89, 95% CI 1.01-8.25 for lowest versus highest tertile of dietary intake and serum levels, respectively). Also for vitamin C intake, a significant association was found (highest versus lowest tertile OR 0.26, 95% CI 0.11-0.61). A significant association was reported for  -carotene intake only in the comparison of the highest versus the lowest tertile (middle versus lowest OR 1.42, 95% CI 0.68-3.00). For vitamin E intake, a borderline significant association was found in the comparison of the middle versus the lowest tertile, but no statistically significant association was found for the highest versus the lowest tertile. For the relationship between progression of OA and the intake of vitamin B1, vitamin B6, niacin, and folate, no statistically significant relationship was found.

Schouten et al ([37]) reported no statistical association between smoking and progression of OA. Wolfe and Lane ([45]) investigated the relationship between depression and anxiety and progression of knee OA. For both factors, no statistically significant association was found. The same study also investigated the relationship between being a high school graduate and progression of knee OA, but no statistically significant association was found. In another study, Dieppe et al ([20]) reported a statistically significant relationship between the use of drugs (nonsteroidal antiinflammatory drugs [NSAIDs]) and the overall measure of change of OA (i.e., clinical improvement or worsening; P = 0.017 for improvement among users versus nonusers). Dieppe et al ([11]) also reported a statistically significant association between the use of walking aids and the clinical improvement of OA (P < 0.001).

Best evidence synthesis

Because of the heterogeneity of the reviewed studies, statistical pooling of the extracted data was not feasible. Therefore, we applied a best evidence synthesis. The best evidence synthesis demonstrated that, based on multiple high-quality studies, there seems to be strong evidence that the level of hyaluronic acid in serum and generalized OA are associated with radiologic progression of knee OA. Also based on multiple high-quality studies, there seems to be strong evidence that sex, knee injury, quadriceps strength, and regular sport activities are not associated, and that knee pain at baseline and radiologic severity of OA at baseline are not strongly associated with the radiologic progression of knee OA.

There is, as yet, limited evidence that there is a relationship between progression of knee OA and synovial fluid volume, nodal warmth, medial bone marrow edema lesions, adduction moment, alignment of the joint (varus/valgus), low serum levels and dietary intake of vitamin D, low intake of vitamin C, use of walking aids, and the use of drugs (NSAIDs). There also is limited evidence that there is no strong association between progression of knee OA and estrogen, uric acid concentration, clinical diagnosis of localized OA, the other markers of bone or cartilage turnover, lateral bone marrow edema lesions, meniscectomy, chondrocalcinosis, running, niacin, folate, smoking, depression or anxiety, being a high school graduate, and low intake of  -carotene, vitamin E, vitamin B1, and vitamin B6. Conflicting evidence is found in the relationship between progression of knee OA and age, bone density, IGF-1, Heberden's nodes, keratan sulfate, COMP, duration of symptoms, clinical severity, CRP level at baseline, and BMI.

DISCUSSION

There seems to be strong evidence that the presence of generalized OA and the level of hyaluronic acid in serum are predictors for radiologic progression of knee OA. There also seems to be strong evidence that sex, knee pain, radiologic severity at baseline, quadriceps strength, knee injury, and regular sport activities are not predictive. For the other studied factors, the evidence is limited or conflicting.

In this review, only significant associations were considered as associated prognostic factors in the best evidence synthesis. Several studies included small sample sizes, which implied low statistical power. For the factors with strong evidence for not being a predictor of OA progression, no associations were found in studies with both small and large sample sizes. For age, IGF-1, COMP, bone density, Heberden's nodes, and BMI, the conflicting associations could not be explained by the difference in sample size; however, the sample size might be an explanation for the conflicting findings for keratan sulfate, duration of symptoms, and the level of CRP.

Especially for the studies with small sample sizes, pooling of the data would have provided a more precise estimate of the association with the outcome. However, because of the heterogeneity of the setting of the studied populations and the differences in disease definition of the included studies, pooling of the data was not possible. Because of small sample sizes, we report instead that there seems to be strong evidence for no association rather than stating that there is no such association at all.

In the presented studies, OA at baseline was defined in different ways, including using the K/L scale, JSW, or radiologic and clinical characteristics of OA. In all determinants (except knee pain at baseline and the duration of symptoms) the conflicting findings are not likely to be explained by the difference in the definition of OA used.

In contrast to the review on prognostic factors of progression of hip OA by Lievense et al ([5]), we found only 4 studies (15%) that also used a clinical outcome for progression of knee OA ([11][19][23][24][45]). The study by Dieppe et al ([11]) did not include persons with a total knee replacement in the analysis; the other 4 studies used either a radiologic change or a total knee replacement in their definition for progression of knee OA ([19][23][24][45]). The numbers of patients with a total knee replacement in another study by Dieppe et al ([11]) (n = 415) and 2 studies by Sharif et al ([23][24]) (n = 75) were 12 and 14, respectively. In the study by Wolfe and Lane ([45]), the number of persons with a total knee replacement was not provided. Therefore, the present review mainly dealt with radiologic progression of knee OA and the conclusions cannot directly be applied to clinical progression of knee OA, whereas for clinical use it is of major interest to know the relationship between these factors and clinical progression of knee OA.

In contrast to the findings of radiologic severity as a predictor for progression of hip OA ([5]), we found evidence that radiologic severity hardly has predictive value for radiologic progression of knee OA. The reason for this difference may simply be that in the present review none of the studies reporting the relationship between radiologic severity and progression of knee OA used total knee replacement as an outcome measure, whereas in the review by Lievense et al ([5]) a total hip replacement was often used as an outcome, and radiologic severity was one of the indicators for a subsequent total hip or knee replacement ([46]).

Future research on the potential relationship between prognostic factors of radiologic progression of knee OA is needed. The factors where conflicting associations were found (e.g., BMI, age) especially need to be clarified. Furthermore, several factors were investigated in a single study only and provided limited evidence; replicated studies of these factors are needed.

Future study on clinical progression of knee OA is of major importance because of its implications for patient information and appropriate medical treatment. In the best evidence synthesis, in case of strong evidence for either the presence or the absence of an association, future scientific consensus is needed on how to summarize the evidence provided by studies with a small sample size. In summary, this review provides the currently available evidence, but also identifies the lack of data with respect to prognostic factors of progression of knee OA.

AUTHOR CONTRIBUTIONS

Dr. Belo 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 design. Drs. Belo, Berger, Reijman, and Bierma-Zeinstra.

Acquisition of data. Drs. Belo and Bierma-Zeinstra.

Analysis and interpretation of data. Drs. Belo, Berger, Reijman, Koes, and Bierma-Zeinstra.

Manuscript preparation. Drs. Belo, Berger, Reijman, Koes, and Bierma-Zeinstra.

Statistical analysis. Dr. Belo.