Dr. Bingham has received consultant fees (less than $10,000) from, serves as an investigator for, and has received research grant support from Procter & Gamble.
Identifying common trajectories of joint space narrowing over two years in knee osteoarthritis†
Article first published online: 29 NOV 2011
Copyright © 2011 by the American College of Rheumatology
Arthritis Care & Research
Volume 63, Issue 12, pages 1722–1728, December 2011
How to Cite
Bartlett, S. J., Ling, S. M., Mayo, N. E., Scott, S. C. and Bingham, C. O. (2011), Identifying common trajectories of joint space narrowing over two years in knee osteoarthritis. Arthritis Care Res, 63: 1722–1728. doi: 10.1002/acr.20614
Unrestricted access to data was provided by Procter & Gamble. Drs. Bartlett and Bingham's work was supported by research grants from the Johns Hopkins Arthritis Center Discovery Fund.
- Issue published online: 29 NOV 2011
- Article first published online: 29 NOV 2011
- Accepted manuscript online: 8 SEP 2011 03:39PM EST
- Manuscript Accepted: 24 AUG 2011
- Manuscript Received: 28 MAR 2011
- Procter & Gamble
- Johns Hopkins Arthritis Center Discovery Fund
Little is known about the natural history of knee osteoarthritis (OA). We sought to identify common patterns of joint space narrowing (JSN) in well-characterized knee OA patients in the placebo arm of a 2-year international study.
We performed secondary data analyses of 622 adults ages 39–80 years in North America (n = 310) and Europe (n = 312) with symptomatic knee OA. Fluoroscopically positioned semiflexed anteroposterior radiographs were obtained at 0, 12, and 24 months. Group-based trajectory modeling was used to identify distinctive groups of individuals with similar trajectories of JSN, taking into account sex, age, and body mass index.
Seven groups were identified. Four exhibited joint space width (JSW) stability over 2 years representing the most common trajectory (71%), which was unrelated to initial JSW. Atypical courses included slow, rapid, and moderate progressors; most had significant JSN at study entry. Slow progressors (20%) had a mean JSN of 0.2 mm over 2 years. Only 2% of the sample demonstrated rapid JSN (2.1 mm), while 7% had JSN of 0.7 mm. Rapid progressors tended to be men, while slow and moderate progressors were older and heavier.
Most (70%) people with OA demonstrated no significant JSN over 2 years; 20% showed slow progression, 7% had moderate, and 2% had rapid JSN. Progressors tended to have less JSW at study entry and were older and heavier; rapid progressors were more likely to be men. Understanding common patterns of the course of knee OA may offer new opportunities to target those at greatest risk of disability.
Knee osteoarthritis (OA) is a major public health challenge as the population ages (1). In 2000, 25 million people in the US and Canada had knee OA, and that number is expected to double by 2020 (2). Knee OA, a common medical condition, is responsible for the largest proportion of disability in adults (3).
The diagnosis and severity of knee OA are commonly confirmed by evaluating radiographs for joint space narrowing (JSN) along with the presence of osteophytes (2). Serial radiographs typically have been used to monitor progression of JSN as an outcome for clinical trials. JSN is thought to reflect reductions in cartilage thickness as well as meniscus integrity, and can indicate the need for clinical intervention. For example, complete loss of joint space (e.g., bone on bone) with concomitant symptoms is a primary indicator for joint replacement surgery (4, 5). While magnetic resonance imaging (MRI) offers several advantages over plain radiographs (e.g., visualization, ability to discriminate among joint structures), in clinical and research applications, measurement of joint space width (JSW) using radiographs remains the current gold standard and the most common method of evaluating knee OA. Regulatory requirements for the development of disease-modifying drugs in OA still consider the measurement of JSN as a primary end point to demonstrate efficacy. As a result, large clinical trials and observational studies generally assess mean group change in JSW as the primary end point or report the proportion of subjects meeting a specified criterion (e.g., any increase in Kellgren/Lawrence score or JSW score or change beyond the standard detectable difference in the measurement) (5, 6).
Despite this, remarkably little is known about the natural history of knee OA. Identifying typical and atypical courses of change in the knee structure over time would offer important information for patients and clinicians about those at greatest risk of disability. Understanding the individual variability in response to interventions is necessary to move therapeutic efforts forward in randomized clinical trials. Also, a major barrier to identifying new drugs that may alter the disease course in OA is the inability to identify patients with an increased risk of progressive disease in whom a treatment effect can potentially be demonstrated.
Therefore, the goal of this study was to identify common patterns of change in JSW over time in a large group of well-characterized people with knee OA who were participating in the placebo arm of a 2-year international study. We applied a new statistical method known as group-based trajectory modeling (GBTM), which offers several advantages (7, 8). GBTM is able to incorporate all data collected on individuals over time to identify distinctive groups of individuals with similar trajectories of JSN over 2 years. Real differences across individuals can be distinguished from chance variation. By profiling key characteristics and behaviors of individuals within distinctive trajectories, additional information can be obtained about the association of these variables with the outcome (7). We hypothesized that the majority of individuals would demonstrate no change in JSW; a second group would show evidence of modest JSN, while relatively few individuals would show substantial changes over time. This study was reviewed and approved by the original institutions that enrolled participants.
SUBJECTS AND METHODS
Secondary data analyses were conducted on subjects enrolled in the placebo arm of a 2-year multinational study of the effects of a bisphosphonate, risedronate, on OA symptoms and radiographic progression (Knee OA Structural Arthritis [KOSTAR]), which has been described elsewhere (9). The original data set consisted of 622 men and women ages 39–80 years in North America (n = 310) and Europe (n = 312) who were randomized to the placebo arm. Patients had signal knee pain due to OA on most days during at least 1 month in a 3-month period prior to screening, plus at least 1 of the following: age >50 years, morning knee stiffness lasting for <30 minutes, or knee crepitus according to the American College of Rheumatology criteria for knee OA (10). Major exclusions included: known inflammatory arthritis, body mass index (BMI) >40 kg/m2, cancer within 10 years, tetracycline use within 6 months, corticosteroid or hyaluronan injections within 3 months, calcitonin or fluoride use within 6 months, and prior use of bisphosphonates within 12 months or for >60 days ever.
All patients underwent standardized radiography with fluoroscopically-positioned semiflexed anteroposterior radiographs to confirm the presence of OA. JSW was evaluated at the narrowest point in the medial tibiofemoral compartment using standardized radiographs with a semiflexed view of the knee, aided by fluoroscopy, and by attaching a metal sphere to the fibula head to correct magnification effects. Minimum medial compartment JSW was measured with a semiautomated computerized method. The SD for this technique was ∼0.2 mm for radiographs obtained 2 days apart, based on repeat measurements (previously reported) (11). At least 1 osteophyte and minimum JSW of 2 mm inclusive in the medial tibiofemoral compartment and a medial compartment that was narrower than the lateral compartment were required. (If both knees qualified, the signal knee was defined as the knee with the smaller JSW.) Knee JSW was evaluated at baseline and after 1 and 2 years of followup. Patients who withdrew were asked to return for the 24-month assessment, which included radiography. Most subjects (76%) had 3 JSW measures available.
GBTM, a form of latent class analysis for continuous outcomes (8), was used to identify distinctive groups of individuals with similar trajectories of JSN. JSN was modeled using all data points available over 2 years with a censored normal model, taking into account sex, age, and BMI and having considered cohort (North America versus Europe). Comparisons were made between models allowing for varying numbers of trajectory groups, as well as models including time as a linear and intercept-only factor. Model fit was assessed using the Bayesian and Akaike's information criteria (7, 8). A posterior probability of membership in each group was provided for each individual, with good fit arising from a high probability of membership into a single group. Analyses were conducted using the SAS-based TRAJ procedure developed by Jones and Nagin with SAS, version 9.1 (12).
Subjects were mostly women and white. A total of 77 subjects had 0 or 1 JSW measurements only and were removed from further analysis. Most (87%) of the remaining 549 subjects had at least 3 JSW measures available within the 2-year followup period. Patient characteristics at study entry for the full cohort of 626 and the 549 analyzed are shown in Table 1. No differences were noted between the entire cohort and the subset analyzed.
|All subjects (n = 626)||≥2 JSW measures (n = 549)||<2 JSW measures (n = 73)|
|Age, mean ± SD years||61.9 ± 8.9||62.0 ± 8.7||61.0 ± 9.5|
|39–49||50 (7.9)||41 (7.5)||7 (9.6)|
|50–59||197 (31.5)||167 (30.4)||29 (39.7)|
|60–69||247 (39.5)||223 (40.6)||24 (32.9)|
|≥70||132 (21.1)||118 (21.5)||13 (17.8)|
|Female sex||439 (70.1)||390 (71.0)||47 (64.4)|
|North American sites||313 (50)||269 (49)||41 (56)|
|White||565 (90.3)||494 (90.0)||69 (94.5)|
|Body mass index, mean ± SD kg/m2||30.0 ± 4.6||29.8 ± 4.6||31.0 ± 4.6|
|WOMAC, mean ± SD|
|Pain||40.0 ± 21.8||39.3 ± 22.0||45.5 ± 19.5|
|Stiffness||46.6 ± 26.0||46.1 ± 26.3||50.7 ± 24.0|
|Function||43.9 ± 23.0||43.2 ± 23.2||49.8 ± 20.8|
|Total||43.4 ± 21.8||42.7 ± 22.0||49.0 ± 19.7|
|Patient global assessment, mean ± SD||54.8 ± 23.3||54.5 ± 23.2||56.6 ± 24.6|
|Glucosamine/chondroitin use||106 (16.9)||93 (16.9)||13 (17.8)|
|CTX-II/Cr, mean ± SD ng/nmole Cr||336.02 ± 271.74||333.11 ± 271.27||361.68 ± 280.24|
|NTX-I/Cr, mean ± SD nmole BCE/nmole Cr||43.42 ± 29.75||42.84 ± 22.24||48.14 ± 62.13|
The optimal number of trajectory groups identified through GBTM was 7. Group assignment was associated with sex, age, and BMI but not geographic site (North America versus Europe). Mean posterior probabilities of group assignment (the group with the largest posterior probability) were high, ranging from 0.84–0.96, supporting good fit of the model. The difference between the theoretical proportion in each trajectory group (based on the sum of all posterior probabilities for that group) and calculated group size (based on the group to which an individual was assigned), an indication of classification error, was small, ranging from 0.1–0.7% across the groups.
JSW stability over 2 years was observed in 4 groups, representing the most common trajectory shape (71%). JSW stability appeared to be largely independent of initial JSW (except at the lowest levels), as shown in Figure 1. Three groups showed statistically significant (P < 0.001) narrowing of JSW: 2 groups with the least joint space at study entry (2.1 and 2.3 mm JSW) comprised 7% and 20% of the sample and showed a mean narrowing of 0.7 mm and 0.2 mm, respectively, over the 2-year period. A small group (2%) had a mean decrease of 2.1 mm, indicating significant JSN.
As compared to participants in the group labeled “flat 2,” comprising 22% of the sample with a high mean JSW at study entry of 3.7 mm and no decline over 2 years, participants with minimal and moderate declines in JSN were more likely to be older and heavier at study entry; those in the small group with the greatest decline were much more likely to be men (Table 2). The small group of individuals (3%) with the greatest JSW at study entry appeared to have a small increase in JSW, although this may have been due to measurement error; as a result, this group was interpreted as being essentially flat.
|Flat 1||Flat 2 (reference)||Flat 3||Flat 4||Minimal decline||Moderate decline||Greatest decline|
|No. (%)†||17 (3)||126 (23)||134 (24)||113 (21)||109 (20)||39 (7)||11 (2)|
|Joint space width, mean mm‡||4.2||3.6||3.2||2.7||2.3||2.2||3.2|
|Joint space narrowing over 2 years, mean mm||0||0||0||0||0.2||0.7||2.1|
|Age, mean years||54.6§||61.6||60.8||63.3||62.8¶||64.8§||63.8|
|Body mass index, mean kg/m2||28.8||29.3||29.1||29.7||31.4§||31.1§||27.9|
|WOMAC pain||30.3 ± 18.5||35.6 ± 21.9||38.9 ± 22.9||37.8 ± 22.0||44.0 ± 21.2§||45.9 ± 17.1§||49.4 ± 27.6§|
|WOMAC stiffness||36.9 ± 29.7||42.1 ± 26.3||45.6 ± 26.4||43.3 ± 25.6||52.5 ± 26.0§||53.0 ± 23.3§||53.9 ± 28.5§|
|WOMAC function||40.5 ± 26.1||39.8 ± 24.5||43.2 ± 23.8||40.4 ± 22.2||48.0 ± 22.2**||48.8 ± 18.9||45.7 ± 22.8|
|WOMAC total||37.6 ± 23.2||38.9 ± 23.2||42.7 ± 22.6||40.2 ± 21.3||47.6 ± 21.1||48.5 ± 17.4||46.5 ± 24.0|
|Patient global||50.0 ± 27.0||53.0 ± 22.3||53.8 ± 23.2||54.4 ± 22.9||56.6 ± 23.0||57.3 ± 24.8||58.0 ± 28.9|
|CTX-II/Cr, ng/nmole Cr||265.88 ± 218.69||277.40 ± 224.61||318.97 ± 282.56||326.41 ± 201.02||414.67 ± 348.21§||384.54 ± 307.66¶||349.45 ± 151.89|
|NTX-I/Cr, nmole BCE/ nmole Cr||36.31 ± 15.99||39.54 ± 19.11||43.78 ± 21.87||42.23 ± 24.01||47.07 ± 24.16§||42.45 ± 22.73||46.22 ± 25.76**|
We evaluated JSW changes over 2 years of a diverse, well-characterized, placebo-treated cohort of individuals with medial compartment knee OA using plain radiography, the current gold standard in research and clinical practice. Using GBTM to identify and summarize patterns of change in the medial compartment of the knee, we found that in most (71%) men and women, JSW remained stable over 2 years, irrespective of study entry levels of JSW. Subjects with the least JSW at study entry were the most likely to deteriorate.
Our results also provide new information about the shape of the most common trajectory of JSN and the proportion of patients represented within different groups. These data suggest that in mild to moderate knee OA, the usual course of events over 2 years is relative stability of JSW. Atypical courses include groups of “slow” and “rapid” progressors. The 3 progressor groups had among the least JSW at study entry. The “slow progressors” included approximately 20% of the sample, with an average JSN of 0.2 mm over the 2 years, an amount equal to the smallest detectable difference in radiographs in this study. Conversely, only 2% of the sample demonstrated a substantial level of JSN (2.1 mm), while 7% had JSN of 0.7 mm. Overall, rapid progressors were more likely to be men, while slow and moderate progressors tended to be older with a significantly higher BMI. In a small group (3%) with the greatest JSW at study entry, JSW appeared to increase slightly. This may have reflected either measurement error or perhaps in some cases hypertrophic repair of articular cartilage as has been shown on radiographs (13) and confirmed by MRI (14).
GBTM was chosen over standard growth curve modeling for the analysis of JSN because it assumes that the population consists of clusters of distinctive trajectories. While standard growth curve modeling can also be used to measure and explain developmental processes, it models average growth and assumes that trajectories vary continuously across individuals according to a multivariate normal distribution. This was not a reasonable assumption for JSN since we had hypothesized that most people would experience no change, and that changes that did occur would be in the direction of a narrowing, not a widening. Therefore, random slopes would not vary about mean growth (or, in this case, decline) in a multivariate normal manner.
Our results add to the growing body of evidence suggesting that while JSN is highly variable in people with mild to moderate knee OA, only a small percentage show evidence of radiographic disease progression over several years. In the Glucosamine/Chondroitin Arthritis Intervention Trial, over 2 years only 19–22% of participants had JSN >0.48 mm (15). Over 3 years, in an older group of 68 individuals in Japan, JSN was evident in 32% of knees (16); this was similar to the proportion (31%) of those exhibiting JSN >2 mm over 3 years in the Bristol OA500 study (17). Sahlstrom et al reevaluated 110 (of 256 individuals) who had radiographs for knee pain between 1970 and 1973 at a university hospital and who remained untreated or conservatively treated. Twenty years later, among those classified as having milder disease (defined as Albäck class I: ≥50% JSN at study entry; n = 47), 39% showed no radiographic progression (18). A 2008 systematic review of 27 observational and randomized controlled trial (RCT) studies of knee OA JSN concluded that estimates ranged from −0.10 mm/year (indicating an increase in JSW over time) to 0.70 mm/year, with a mean ± SD loss of 0.13 ± 0.15 mm/year (19). In the original parent study for our analysis, radiographic progression of the entire group was only 0.088 mm over 2 years in the European cohort and 0.13 mm in the North American cohort, which was significantly less than an anticipated rate of 0.20 mm per year. When analyzed using a stringent criterion (>0.6 mm or 3 × smallest detectable difference), only a small proportion of individuals (13%) would be classified as having significant JSN, or “progressors” (9).
Understanding the radiographic course of knee OA may have important implications for research and clinical practice. Current studies of potential disease-modifying agents are often limited by the low number of patients experiencing significant changes in JSW over the time of the study, with only a few patients driving mean changes. Recently, Osteoarthritis Research Society International (OARSI) and Outcome Measures in Rheumatology (OMERACT) recommended that OA clinical trials evaluate the proportion of progressors as an outcome measure using a cutoff of the smallest detectable difference of the measurement (5). In 2004, OARSI and OMERACT established a task force to define criteria for knee and hip replacement. These criteria, which reflected the domains of pain, impairment in function, and structural damage, would serve as end points in clinical trials evaluating potential disease-modifying OA drugs. The most recent report of the task force (2009) noted that there was insufficient evidence to identify a specific cut point for JSN.
To our knowledge, this is the first study to use GBTM to identify trajectories of disease course in OA. GBTM and similar statistical modeling methods are increasingly used to describe atypical courses of diverse outcomes in clinical and observational trials, including disability in the last year of life (20), apathy after stroke (21), and adherence to therapy in human immunodeficiency virus (22). Understanding the assumptions associated with this approach is required. For example, the basic assumption that is made a priori is that there are qualitatively distinct trajectories within a large sample. Trajectory groups are assembled using a formal statistical methodology rather than subjectively derived classification rules, significantly increasing the likelihood that individual change observed reflects true change versus random variation. No assumptions are made about the number or shape of the trajectories (although more measures per unit are required to model more complex shapes). Each trajectory is thought to characterize individuals following a common pathway and to offer a good representation for all members of the group. The purpose of GBTM is not to infer there are literal entities of distinct groups among the population with OA. Rather, GBTM can be used to provide methodologic support to test theories that would predict that different clusters of individuals follow similar trajectories within the population, the outcomes associated with differing trajectories, and subject characteristics associated with group membership. This method is suitable for use in analyzing future clinical trials of disease progression and interventions (i.e., disease-modifying OA drug).
Our sample was well suited to describe the clinical course of knee OA in individuals enrolled in the placebo arm of a multinational RCT, and for use of GBTM. Subjects were persons with mild to moderate knee OA who were recruited into the largest multinational drug development trial to date (n = >2,400) and had been assigned to the placebo arm with well-characterized symptoms over 2 years (9). Changes in JSW and symptoms served as primary end points. (Findings from the main trial showed no treatment effect of varying doses of risedronate.) The data set was relatively complete, with 87% having at least 3 measures available; discontinuations and dropouts were similar across treatment groups in the overall study.
There are limitations to this study. First, subjects were volunteers in a clinical trial and were judged to have mild to moderate knee OA with JSW in the medial compartment between 2 and 4 mm at entry, and their date of diagnosis was not known. Changes in mean JSW in the entire group were much smaller than anticipated. The study was only conducted in patients with medial compartment OA; therefore, it is not possible to generalize to patients with lateral compartment OA or patellofemoral disease. There was no assessment of varus–valgus alignment in this study, which has been recognized as a risk factor for radiographic progression (23, 24). Recent studies using MRI have indicated that meniscal pathology is common in patients with knee OA, and degeneration and extrusion of the meniscus may also be reflected in loss of JSW using plain radiography (25). Because the KOSTAR study only used a plain radiographic outcome, the contribution of meniscal damage cannot be evaluated. These data provide a snapshot of the overall pattern of JSN over 2 years but cannot provide information about the nature of cartilage breakdown; studies of potential biomarkers suggest that cartilage breakdown may occur in a phasic rather than linear manner (26, 27). It will be important in future studies to evaluate the impact of these additional radiographic and other clinical and biochemical features in determining the factors that are associated with assignment of subjects to the different groups of rapid and slow progressors and nonprogressors. It is not certain whether the 7 groups identified represent clinically meaningful differences in trajectories. These data may not represent the prevalence of different natural courses of progression as people with OA that were likely to progress rapidly may not be represented in this clinical trial. Further, both the severity and progression of structural damage often correlate poorly with clinical symptoms and in people with OA (2).
In summary, this study used GBTM to identify typical and atypical profiles of joint JSN over 2 years in a group of well-characterized patients with mild to moderate knee OA in the placebo arm of a large clinical trial. Most (71%) demonstrated no significant JSN. Progressors tended to have the least JSW at study entry. Approximately 20% had a slight decrease of JSW of 0.2 mm over 2 years; 7% had JSN of 0.7 mm and tended to be older with a higher BMI. Only 2% demonstrated a substantial level of JSW loss (2.1 mm); these individuals were more likely to be men. GBTM is a statistical approach that offers information about typical and atypical patterns of the disease course. Within the context of knee OA, a common condition among an aging population, knowledge about the typical patterns of radiographic changes along with information about clinical and patient-reported parameters may offer new insights for patients, clinicians, public health officials, and policymakers about those at greatest risk of disability.
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. Dr. Bartlett 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. Bartlett, Ling, Mayo, Bingham.
Acquisition of data. Bingham.
Analysis and interpretation of data. Bartlett, Ling, Mayo, Scott, Bingham.
- 3Centers for Disease Control and Prevention (CDC). Prevalence and most common causes of disability among adults: United States, 2005. MMWR Morb Mortal Wkly Rep 2009; 58: 421–6.
- 8Group-based modeling of development. Cambridge (MA): Harvard Press; 2005..
- 9Risedronate decreases biochemical markers of cartilage degradation but does not decrease symptoms or slow radiographic progression in patients with medial compartment osteoarthritis of the knee: results of the two-year multinational knee osteoarthritis structural arthritis study. Arthritis Rheum 2006; 54: 3494–507., , , , , , et al.