Abstract
Objective
The development of knee symptoms in OA over time can present with multiple phenotypes. This study aims to identify MRI-defined predictors of trajectories of knee symptoms.
Design
All knees in the Osteoarthritis Initiative (OAI) with at least one baseline score of 0 from one of four knee symptom measurements were included. These measurements included pain severity, knee pain frequency, and the Western Ontario and McMaster Universities Osteoarthritis Index knee pain and knee function scores. The latent class mixed model (LCMM) was employed to categorize knees into different trajectories and to assess their association with various MRI-defined structural features while controlling for important covariates.
Results
A total of 1221 knees were grouped into two classes: 596 with a minimal knee symptom trajectory and 625 with a rapidly progressive trajectory. The class-membership model revealed that the following variables were associated with being in the rapidly progressive phenotype: bone marrow lesions (BMLs) score (Odds Ratio (OR) of 1.12 [95 % Confidence Interval (CI): 1.03, 1.22]) for each unit increase in BML total size scores; females, widespread pain, CES-D, overweight, and obesity. The age group 65 to 79 vs. that of 45 to 55 and highest year of education completed were less likely to be associated with the rapidly progressive knee pain group. Other covariates were not significant.
Conclusion
Two distinct phenotypes/trajectories were consistently identified across all four knee symptom measurements in a multi-trajectory model. Modifiable factors such BMLs, BMI, widespread pain and depression may serve as indicators of higher risk of rapid knee symptom progression.
Keywords: Osteoarthritis, Knee pain, Knee function, Knee symptom phenotypes, Latent class analysis, Imaging, MRI-defined structural features
Introduction
Osteoarthritis (OA) is a degenerative condition that affects all parts of the joint [1]. It is a serious chronic disease that impacted >30 million people in the U.S. and 595 million globally in 2020 [2,3]. The total annual economic burden of OA is 136.8 billion, with direct costs exceeding $10,000 per person [2]. Functional limitations are common consequences of OA, significantly reducing work performance and quality of life. Individuals with OA miss an average of two more workdays per year compared to those without the condition [4].
Despite the significant burden of OA, no Disease-Modifying OA Drugs (DMOADs) have been approved by the Food and Drug Administration (FDA) or the European Medical Agency (EMA) to prevent or reduce symptoms and structural progression of the disease [5]. One major challenge is managing pain, the predominant presenting symptom of OA and the primary reason for hospital and outpatient visits [6]. Pain relief is the main goal of treatment and a key factor in determining cost-effectiveness [7]. However, the origin and pathophysiology of OA-related pain are complex and multifactorial, involving both biological and psychosocial mechanisms [8,9]. In addition, the lack of understanding of OA pain progression has led to treatment and management plans being applied broadly across the entire OA population, without distinguishing between different pain phenotypes and/or trajectories [10,11]. In contrast, successfully identifying distinct pain trajectories can enhance the efficiency and effectiveness of clinical trials by pinpointing patient subgroups with more severe pain trajectories who may benefit from targeted treatment, while distinguishing those with minimal or mild pain trajectories who may require less intensive or no intervention. Therefore, characterizing and understanding differences among various pain trajectories and functional changes over time is crucial for developing customized treatment and management plans, ensuring that individuals most in need who might benefit the most from targeted interventions are accurately identified.
Among all joints, the knee is one of the most commonly affected by OA [12]. Numerous studies have explored knee pain trajectories in patients with knee OA, consistently finding that knee pain profiles are highly heterogeneous [[13], [14], [15], [16], [17]]. A recent systematic review revealed that while half of patients experience unstable daily knee pain trajectories in the short term, about 85 % have steady trajectories in the mid-term [18]. Most studies, however, rely on one single measurement of knee pain to explore its temporal trajectory, which overlooks the complexity of knee pain reporting and provides a less comprehensive understanding of knee pain symptom patterns. Additionally, these studies often select sub-samples based on patients' demographic information or clinical diagnosis, typically including only hundreds of observations [16,[19], [20], [21], [22]]. Moreover, these knee symptom profiles often focus on patients who initially experience moderate to severe pain, thereby offering limited insight into the broader population that begins without symptoms.
Therefore, there is an urgent need for knee symptom phenotyping in large longitudinal cohort studies to better understand the etiology of knee pain in OA, particularly for patients who begin without any prior knee symptoms. To address limitations of previous studies such as simplistic operational definitions of knee symptoms, cross-sectional study design, and small sample sizes, we utilized a statistical model that combines latent classes clustering and their membership predictions. Our objectives are to identify distinct temporal knee symptom phenotypes and to uncover the underlying factors contributing to these phenotypes in OA.
Methods
Data
The Osteoarthritis Initiative (OAI) is a longitudinal cohort study sponsored by the National Institutes of Health (NIH) focused on the natural history of OA and the identification of biomarkers of the development and/or progression of knee OA [23]. The OAI enrolled 4796 participants (58.5 % female, ages 45 to 79) with or at risk of developing symptomatic knee OA across four sites: Ohio State University in Columbus OH; the University of Maryland/Johns Hopkins University in Baltimore MD; Memorial Hospital in Pawtucket RI; and the University of Pittsburgh in Pittsburgh PA. Knee symptom data were obtained at baseline and during annual follow-up visits over nine years, with a high retention rate. Bilateral weight-bearing fixed flexion anterior-posterior knee radiographs and bilateral knee 3T MRI images from identical Siemens MRI scanners were obtained at baseline and at selected annual follow-up visits. Demographic information, and psychosocial and social determinants data were also collected at the annual visits. The data from the OAI are publicly available at https://nda.nih.gov/oai.
Knee symptom measurements
Four key knee symptom measurements were selected from the various outcome assessments in OAI for the phenotype clustering analysis. These measurements include the Numerical Rating Scale (NRS) for knee pain severity, the Knee injury and Osteoarthritis Outcome Score (KOOS) for knee pain frequency (i.e., How often do you experience knee pain? Never, monthly, weekly, or daily), the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) function score, and the WOMAC pain score [24].
Baseline factors associated with pain phenotypes
The primary risk factors considered in this analysis were MRI-defined structural features in OA, chosen for their relevance to pain and the progression of structural damage in OA [25]. These features are graded on a scale of 0 to 3 using the MRI Osteoarthritis Knee Score (MOAKS) grading system and include bone marrow lesions (BMLs), cartilage surface size damage, cartilage depth damage, Hoffa’s synovitis, effusion-synovitis (ES), osteophytes, meniscal damage, and meniscal extrusion. The BMLs are scored based on the total volume of the subregions they occupy. There are 15 subregions around the subchondral plate, and the scores for all subregions are summed in this analysis to reflect the overall BML burden. For articular cartilage, both the extent of surface size damage and depth of cartilage damage are each scored in 14 subregions. We summed all cartilage surface scores and cartilage depth scores separately to represent the overall condition of the cartilage. For osteophytes, meniscal damage, and meniscal extrusion, scores were summed over 12, 6, and 4 subregions, respectively. Age, gender, race, body mass index (BMI), widespread pain, Center for Epidemiological Studies Depression (CES-D), highest grade or level of school completed, Kellgren-Lawrence grading (KLG), comorbidity, Physical Activity Scale for the Elderly (PASE), knee injury, and knee surgery history were included as covariates in the fully adjusted model.
Statistical analysis
The selection criteria were based on four knee symptom measurements (NRS for knee pain severity, KOOS for knee pain frequency (i.e., none, monthly, weekly or daily), WOMAC function, and WOMAC pain). Those knees with at least one knee symptom measurement score of 0 at baseline, along with MRI readings, were selected for further analysis. To this subset, we applied the latent class mixed model (LCMM), a statistical framework based on linear mixed model theory, to identify latent subgroups of knees exhibiting similar 10-year trajectories across the four knee symptom measurements, while simultaneously assessing their associations with preselected risk factors [26]. In some participants, data from both knees were included in the analysis, and the program’s default setting assumed these knees to be independent. The number of latent classes and the order of polynomial terms (which define the shapes of the latent trajectories) were preselected from a set of candidate values (e.g., 2, 3, or 4). Covariates, including demographic characteristics, MRI-defined structural changes, and other risk factors described in the previous subsection, were incorporated into the class membership model to estimate the probability of each knee belonging to each latent group. Group assignment was determined by the highest posterior probability. Model selection from the set of candidate specifications was based on the Bayesian Information Criterion (BIC) and successful model convergence. The size (percentage) of each group was calculated, and the mean and Confidence Intervals (CIs) for the different group knee symptoms measurements were plotted. Baseline characteristics were presented as mean (standard deviation (SD)) for continuous variables and as count (percentage) for categorical variables. Analysis of variance (ANOVA) and Chi-squared tests were used to compare differences among trajectory groups identified by LCMM. A two-sided significance test with a p-value of 0.05 was considered statistically significant.
To assess the association between latent knee pain trajectory subgroups and MRI-defined structural features while adjusting for covariates, we used the class membership model (an adjusted logistic model) output to calculate and plot odds ratios (ORs) with 95 % CIs. To account for potential within-subject correlation, we applied a complementary strategy: for participants contributing both knees, we conducted additional analyses separately, including only the left or only the right knee. To evaluate potential correlations among covariates in the class membership model, particularly among the MRI-defined structural features, we computed a correlation matrix of the MRI-defined structural features. (See Supplemental Table 1).
The LCMM analysis was conducted in R (version 4.2.2) with the package LCMM (version 2.1.0).
Results
Fig. 1 illustrates the inclusion of knees in the analysis. Of the 9592 knees in the OAI cohort, 4751 had at least one score of 0 across the four symptom measurements, and among these, 1221 had MRI MOAKS readings available from publicly accessible data. A total of 1161 participants were included in the analysis, of whom 105 contributed both knees to the analysis. Fig. 2 illustrates the results of the LCMM clustering analysis. Based on the BIC and model convergence, we chose 2 as the number of latent classes and 3 as the order of polynomial terms. The LCMM analysis identified two distinct trajectories over ten years for all four knee symptom measurements: Class 1 (depicted in red, with 596 knees (48.8 %)) is the cluster of minimal progression of knee symptoms; Class 2 (depicted in blue, with 625 knees (51.2 %)) is the cluster of rapid progression of knee symptoms. The average posterior probabilities of membership assignment for Class 1 and 2 were 0.94 and 0.97, respectively. (see Supplemental Table 2)
Fig. 1.
The flow chart of knee inclusion in the analysis.
The flowchart of the study sample based on participant inclusion and exclusion criteria is illustrated. The left column is for the inclusion of knees; the right column shows the number of knees excluded at each step.
Fig. 2.
Knee pain phenotyping results using the selection criteria.
Each cell displays the results for a specific symptom measurement (from top left to bottom right: WOMAC pain, WOMAC function, NRS for pain severity and KOOS for knee pain frequency). The dot denotes the estimated value. The solid lines represent the observed mean trajectory for each latent group, while the dashed lines indicate the 95 % CIs.
Table 1 shows the comparison of the demographic clinical characteristics between the two trajectory clusters. As compared to the minimal symptom progression cluster, the rapid symptom progression cluster is younger, less likely to be Non-Hispanic White, has higher BMI, has more widespread pain, has fewer years of education completed, has more depressive symptoms, has higher radiologic severity, is more likely to have a history of knee surgery and is more likely to have a history of knee injury.
Table 1.
Comparison of the baseline characteristics between the two clusters/trajectories.
| Class 1 | Class 2 | p-value | |
|---|---|---|---|
| (minimal progression) | (rapid progression) | ||
| (N = 596) | (N = 625) | ||
| Age | 0.0323 | ||
| 45 ≤ age < 55 | 161 (27.0 %) | 178 (28.5 %) | |
| 55 ≤ age < 65 | 200 (33.6 %) | 244 (39.0 %) | |
| 65 ≤ age ≤ 79 | 235 (39.4 %) | 203 (32.5 %) | |
| Gender | 0..129 | ||
| Female | 333 (55.9 %) | 377 (60.3 %) | |
| Male | 263 (44.1 %) | 248 (39.7 %) | |
| Race | 0.0636 | ||
| Non-Hispanic White | 533 (89.4 %) | 536 (85.8 %) | |
| Non-Hispanic Black, Asian/Others | 63 (10.6 %) | 89 (14.2 %) | |
| BMI | < 0.001 | ||
| Underweight and Health | 196 (32.9 %) | 106 (17.0 %) | |
| Overweight | 232 (38.9 %) | 260 (41.6 %) | |
| Obese | 168 (28.2 %) | 259 (41.4 %) | |
| Widespread Pain | 0.00486 | ||
| No | 565 (94.8 %) | 565 (90.4 %) | |
| Yes | 31 (5.2 %) | 60 (9.6 %) | |
| CES-D, Mean(SD) | 4.78 (5.95) | 6.76 (6.97) | < 0.001 |
| Highest education | 0.00578 | ||
| Some high school/high school graduate/some college | 170 (28.6 %) | 240 (38.33 %) | |
| College graduate/some graduate school | 209 (35.1 %) | 193 (30.8 %) | |
| Graduate degree | 217 (36.4 %) | 192 (30.7 %) | |
| KLG | < 0.001 | ||
| 0 | 301 (50.5 %) | 228 (36.5 %) | |
| 1 | 230 (38.6 %) | 226 (36.2 %) | |
| 2 | 47 (7.9 %) | 97 (15.5 %) | |
| 3 | 18 (3.0 %) | 74 (11.8 %) | |
| Comorbidity, Mean (SD) | 0.290 (0.750) | 0.360 (0.813) | 0.119 |
| PASE, MEAN (SD) | 169 (79.0) | 167 (79.0) | 0.639 |
| Knee Surgery History | 0.00197 | ||
| No | 570 (95.6 %) | 569 (91.0 %) | |
| Yes | 26 (4.4 %) | 56 (9.0 %) | |
| Knee Injury History | < 0.001 | ||
| No | 499 (83.7 %) | 472 (75.5 %) | |
| Yes | 97 (16.3 %) | 153 (24.5 %) |
Note: BMI Body Mass Index; CES-D Centers for Epidemiologic Studies Depression scale; KLG Kellgren-Lawrence grade; Comorbidity Modified Charlson Comorbidity scale; PASE Physical Activity Scale for the Elderly.
The results of the class membership model within the LCMM framework are summarized in Table 2. Three sets of results are reported. The first is based on the dataset including all available MRI MOAKS readings (sample size is 1221). The other two were obtained from datasets in which, for subjects with available MRI readings for both knees, only the right or only the left knee was included (n = 1116 for each). The findings were highly consistent across the three models. Based on the results of the available MOAKs MRI readings, among MRI-defined structural features, osteophyte and BMLs were significantly associated with a higher risk of being in the rapidly progressive knee symptom cluster, with an OR of 1.07 (95 % CI: 1.02, 1.13), and 1.15 (1.04, 1.26), respectively. Being female, BMI in the overweight and obesity categories, higher CES-D, and widespread pain have significant ORs for being in the rapidly progressive cluster. For age 65–79 and the highest year of education completed (i.e., college graduate, some graduate school, and graduate degree), they all had significant inverse associations with the rapid symptom progression that was only observed in the model including all available knees. This effect may reflect within-subject correlation or potential selection bias. Other MRI-defined features and risk factors were not significantly associated with the prediction of latent class membership. The above results are also shown in Fig. 3, a forest plot summarizing the odds ratios and 95 % confidence intervals for both knees, the right knee alone and the left knee alone from adjusted logistic regression models. The posterior probabilities of class membership for the two classes are summarized in Supplemental Table 2. In Supplemental Table 3, the results for separate adjusted models of the association of each individual MRI-defined structural feature with rapid symptom progression cluster membership are summarized. These models were adjusted for age, gender, race, and BMI.
Table 2.
Results of cluster membership models for datasets with all knees having MRI readings; only the right or left knee from both knees with MRI readings, if applicable.
| Odds ratios (95 % CI) |
|||
|---|---|---|---|
| Both knees | Only left knee | Only right knee | |
| Cartilage depth damage | 0.94 (0.83, 1.06) | 0.96 (0.85, 1.08) | 0.95 (0.84, 1.08) |
| Hoffa's synovitis | 0.98 (0.72, 1.33) | 0.95 (0.71, 1.25) | 1.03 (0.8, 1.32) |
| Effusion-synovitis | 0.95 (0.71, 1.27) | 1.1 (0.83, 1.44) | 0.93 (0.7, 1.25) |
| Meniscal extrusion | 1.01 (0.84, 1.22) | 1.05 (0.89, 1.23) | 1.01 (0.92, 1.09) |
| Cartilage surface damage | 1.04 (0.98, 1.11) | 1.03 (0.97, 1.1) | 1.04 (0.97, 1.1) |
| Meniscus morphology | 1.04 (1, 1.09) | 1.03 (0.98, 1.08) | 1.05 (1, 1.1) |
| Osteophyte | 1.07 (1.02, 1.13) | 1.07 (1.01, 1.13) | 1.07 (1.01, 1.13) |
| BML | 1.15 (1.04, 1.26) | 1.13 (1.03, 1.25) | 1.14 (1.03, 1.25) |
| Age 65 - 79 | 0.67 (0.46, 0.98) | 0.73 (0.48, 1.11) | 0.72 (0.48, 1.09) |
| College graduate/some graduate school | 0.69 (0.49, 0.97) | 0.73 (0.51, 1.04) | 0.73 (0.51, 1.05) |
| Graduate degree | 0.68 (0.48, 0.96) | 0.71 (0.5, 1.02) | 0.8 (0.55, 1.15) |
| Age 55 - 65 | 0.98 (0.73, 1.31) | 1.03 (0.71, 1.5) | 0.9 (0.62, 1.3) |
| Race white | 1.06 (0.71, 1.57) | 0.96 (0.57, 1.64) | 0.98 (0.6, 1.6) |
| PASE | 1 (1, 1) | 1 (1, 1) | 1 (1, 1) |
| Comorbidity | 1.05 (0.89, 1.24) | 1.03 (0.86, 1.24) | 1.04 (0.87, 1.25) |
| CES-D | 1.05 (1.03, 1.08) | 1.06 (1.03, 1.09) | 1.07 (1.04, 1.1) |
| History of knee injury | 1.36 (0.94, 1.96) | 1.29 (0.88, 1.9) | 1.44 (0.97, 2.12) |
| Female gender | 1.44 (1.07, 1.95) | 1.46 (1.07, 1.99) | 1.45 (1.06, 1.98) |
| History of knee surgery | 1.45 (0.74, 2.86) | 1.54 (0.76, 3.12) | 1.6 (0.77, 3.33) |
| BMI Overweight | 2.1 (1.48, 2.99) | 2.12 (1.47, 3.06) | 1.98 (1.37, 2.86) |
| Widespread pain | 2.25 (1.29, 3.92) | 1.87 (1.06, 3.3) | 2.16 (1.21, 3.85) |
| BMI Obese | 2.46 (1.69, 3.57) | 2.42 (1.64, 3.58) | 2.41 (1.63, 3.56) |
Note: BMLs bone marrow lesions; BMI Body Mass Index; CES-D Centers for Epidemiologic Studies Depression scale; KLG Kellgren-Lawrence grade; Comorbidity Modified Charlson Comorbidity scale; PASE Physical Activity Scale for the Elderly.
Fig. 3.
Forest plot summarizing the adjusted odds ratios and 95 % CI of the association between rapid symptom progression and MRI-defined structural features.
PASE Physical Activity Scale for the Elderly; Comorbidity Modified Charlson Comorbidity Scale; CES-D Center for Epidemiologic Studies Depression scale; BML Bone marrow lesions.
Discussion
This study identified two distinct phenotypes of knee symptom progression over nine years: minimally progressive and rapidly progressive. These phenotypes were classified based on the progression of knee symptoms, beginning from the absence of symptoms, which may facilitate the timely matching of appropriate interventions in the early stages of the disease. The rapidly progressive group, in particular, warrants increased attention from both patients and healthcare providers to closely monitor knee symptom progression and implement necessary interventions to prevent further deterioration. Previous studies often focused on specific populations with smaller sample sizes, typically including patients who already had moderate or severe knee pain at baseline [[14], [15], [16],27]. These studies commonly identified two patterns in groups with high baseline knee pain: either a stable pattern or a decrease in pain over time pattern. However, the distinction between these two patterns may not significantly influence treatment decisions, as both groups require immediate pain relief. Additionally, including patients with high baseline knee pain symptoms introduces greater variability and complexity into the model used to identify knee symptom phenotypes, especially for those developing from an initial state of no pain symptoms. To address this limitation, our approach selected patients with no or little baseline knee pain to minimize initial variability and improve the accuracy of phenotype identification. Another key strength of this study is the use of LCMM to classify phenotypes based on four knee symptom measurements simultaneously. The LCMM approach offers a one-step solution for both trajectory clustering and association analysis, in contrast to traditional multi-stage methods such as trajectory clustering followed by logistic regression. This method accounts for the heterogeneous latent structure of multiple trajectory measurements while assessing associations between the identified trajectory clusters and covariates of interest, such as MRI features. As a result, it provides more robust statistical inference. However, the LCMM approach has several limitations. First, performing variable selection in the presence of correlated covariates is challenging. Second, the algorithm is computationally intensive and can be slow to converge. Third, modeling the dependency between two knees from the same subject requires additional parameters, which poses numerical difficulties.
Unlike previous studies that relied on single characteristic measurements for phenotype identification, our method reduces bias associated with self-reported symptom questionnaires and enhances classification accuracy. Furthermore, this study utilized longitudinal data from the OAI, one of the largest OA cohorts with annual assessments over a 9-year period. This extended timeframe exceeds the duration of many previous studies, which often focused on shorter-term or cross-sectional analyses [17,22,28], thus allowing for a more comprehensive assessment of knee symptom patterns and allowing for different patterns of progression in knee OA. The use of this large dataset enhances the generalizability of our findings and strengthens the validity of the identified phenotypes.
In evaluating the association between knee pain phenotypes and risk factors, we identified a gender disparity: females are at a higher risk of experiencing rapid progression of knee symptoms. Although this finding aligns with observations from a previous study [29], a recent systematic review did not report similar results [18], indicating that the gender disparity in the development of knee OA pain requires further investigation in larger and more diverse populations. Additionally, the significantly lower age associated with being in the rapidly progressive group compared to other groups may be attributable to our sample selection criteria, a pattern that has been similarly reported in previous studies [15,16]. Our finding of lower level of educational attainment being associated with worse symptom outcomes is consistent with a systemic review of pain trajectories in knee OA [18].
Another key finding of this study is the significant association between MRI-defined structural features at baseline and knee pain phenotypes. This contrasts with the results of a recent narrative review, which did not identify such an association [30]; the review was based solely on cross-sectional studies, however. The longitudinal data used in our study provides a more robust and comprehensive perspective, revealing a significant link between baseline MRI-defined structural features and knee pain phenotypes. Specifically, our study suggests that a higher burden of BMLs, indicated by higher total size scores, and osteophytes, indicated by total osteophyte score, are associated with a greater likelihood of being in the rapidly progressive group of knee symptoms. This finding is consistent with results from the study by Pan et al. [31], which used a single knee pain measurement and identified three distinct, but generally stable, trajectories over time. Our study strongly suggests that structural abnormalities such as BMLs and osteophytes at baseline may predict rapid knee symptom development. The cluster with rapid symptom progression may represent a subset of patients who might benefit from earlier intervention in their disease course.
Our study did not find significant associations with cartilage, ES, Hoffa's synovitis, meniscus morphology, or meniscus extrusion in the class membership model, which is essentially an adjusted logistic regression analysis. Additionally, our reliance on publicly assessable MOAKS readings led to the exclusion of a large number of the potential observations available for the class membership model. This small number of publicly assessable MOAKS readings limits the generalizability of our findings of the association of MRI-defined features and symptom trajectories. This limitation could be mitigated in the future by increasing the number of MRI readings through expert evaluation or by utilizing deep learning-based methods for more comprehensive analysis.
In conclusion, our study successfully identified two distinct phenotypes of knee symptom trajectories that were consistently identified across all four knee symptom measurements: minimally progressive and rapidly progressive. We also found a significant association between MRI-defined structural abnormalities at baseline and these knee pain phenotypes, with larger BML sizes linked to a higher likelihood of rapid OA symptom progression. Our findings suggest that early detection of structural abnormalities may be crucial in targeting interventions for patients at risk of rapid symptom development, although the limited sample size due to reliance on MRI scan readings should be addressed in future research through improved imaging techniques or advanced deep learning methods.
Role of the funding source
The funding source had no role in the design of the study, the data analyses or the manuscript preparation.
Editor Disclosure
The peer review process did not involve Editorial Board Member Chian Kent Kwoh, and the editorial decision-making was led by editors not involved in the creation of this manuscript.
CRediT authorship contribution statement
Shen Liu: Data curation, Formal analysis, Software, Supervision, Methodology, Validation, Writing – original draft, Writing – review & editing. Xiaoxiao Sun: Conceptualization, Funding acquisition, Supervision, Writing – original draft. Yong Ge: Conceptualization, Funding acquisition, Supervision, Writing – original draft. Thang Nhat Duong: Data curation, Formal analysis, Software, Methodology, Validation, Writing – review & editing. C. Kent Kwoh: Project administration, Conceptualization, Supervision, Funding acquisition, Writing – review & editing.
Declaration of competing interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:
SL: None. XS: None. YG: None. TND: None. CKK is a consultant to Moebius Sun, Novartis, Xalud, GSK, Express Scripts, Formation Bio, AposHealth Avalor, and Pleryon. He is principal investigator of pharma-sponsored clinical trials for Abbive, Artiva, Lilly, BMS, Pfizer, GSK, Cumberland and Galapagos, and on the DSMB for Kolon TissueGene.
Acknowledgments
We acknowledge the contributions of the OAI investigators, the OAI staff at the Clinical sites, the OAI participants, the Pivotal OA MRI Analyses (POMA) investigators, the Biomarkers of Early Arthritis of the Knee (BEAK) investigators, and the Early Knee Osteoarthritis (EKO) investigators.
The OAI is a public-private partnership comprised of five contracts (N01-AR-2-2258; N01-AR-2-2259; N01-AR-2-2260; N01-AR-2-2261; N01-AR-2-2262) funded by the NIH and conducted by the OAI Study Investigators. Private funding partners include Merck Research Laboratories, Novartis Pharmaceuticals Corporation, GlaxoSmithKline, and Pfizer, Inc. Private sector funding for the OAI is managed by the Foundation for the NIH. Funding was also provided in part by HHSN26820100002, R01AR080742, R21AR083495, R01AR080742-02S1, and UA Eighteenth Mile TRIF Funding. This manuscript was prepared using an OAI public use data set (in addition to data obtained within NIH/NIAMS funded ancillary grants) and does not necessarily reflect the opinions or views of the OAI investigators, the NIH, or the private funding partners.
Footnotes
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ostima.2025.100380.
Appendix. Supplementary materials
References
- 1.C. for D. C. and Prevention, “Osteoarthritis (OA) | Basics | Arthritis | CDC,” 2018.
- 2.E. Yelin, “BMUS:The Burden of Musculoskeletal Diseases in the United,” Bone and Joint Burden.
- 3.Steinmetz J.D., et al. Global, regional, and national burden of osteoarthritis, 1990–2020 and projections to 2050: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Rheumatol. 2023;5(9) doi: 10.1016/S2665-9913(23)00163-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Menon J. Osteoarthritis related absenteeism and activity limitations. Osteoarthr. Cartil. 2015;23 doi: 10.1016/j.joca.2015.02.629. [DOI] [Google Scholar]
- 5.Lohmander L.S., Roos E.M. Disease modification in OA — Will we ever get there? Nat Rev Rheumatol. 2019;15(3) doi: 10.1038/s41584-019-0174-1. [DOI] [PubMed] [Google Scholar]
- 6.Neogi T. The epidemiology and impact of pain in osteoarthritis. Osteoarthr. Cartil. 2013;21(9) doi: 10.1016/j.joca.2013.03.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.L.A. Deveza, L. Melo, T.P. Yamato, K. Mills, V. Ravi, and D.J. Hunter, “Knee osteoarthritis phenotypes and their relevance for outcomes: a systematic review,” 2017. doi: 10.1016/j.joca.2017.08.009. [DOI] [PubMed]
- 8.A. Eitner, G.O. Hofmann, and H.G. Schaible, “Mechanisms of osteoarthritic pain. Studies in humans and experimental models,” 2017. doi: 10.3389/fnmol.2017.00349. [DOI] [PMC free article] [PubMed]
- 9.A.M. Malfait and T.J. Schnitzer, “Towards a mechanism-based approach to pain management in osteoarthritis,” 2013. doi: 10.1038/nrrheum.2013.138. [DOI] [PMC free article] [PubMed]
- 10.Zhang W., Robertson J., Jones A.C., Dieppe P.A., Doherty M. The placebo effect and its determinants in osteoarthritis: meta-analysis of randomised controlled trials. Ann Rheum Dis. 2008;67(12) doi: 10.1136/ard.2008.092015. [DOI] [PubMed] [Google Scholar]
- 11.Filardo G., Previtali D., Napoli F., Candrian C., Zaffagnini S., Grassi A. PRP injections for the treatment of knee osteoarthritis: a meta-analysis of randomized controlled trials. Cartilage. 2021;13(1) doi: 10.1177/1947603520931170. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Long H., et al. Prevalence trends of site-specific osteoarthritis from 1990 to 2019: findings from the global burden of disease study 2019. Arthritis Rheumatol. 2022;74(7) doi: 10.1002/art.42089. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Boudreau S.A., et al. Distinct patterns of variation in the distribution of knee pain. Sci Rep. 2018;8(1) doi: 10.1038/s41598-018-34950-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Bastick A.N., et al. Defining knee pain trajectories in early symptomatic knee osteoarthritis in primary care: 5-year results from a nationwide prospective cohort study (CHECK) Br. J. Gen. Pract. 2016;66(642) doi: 10.3399/bjgp15×688129. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Collins J.E., Katz J.N., Dervan E.E., Losina E. Trajectories and risk profiles of pain in persons with radiographic, symptomatic knee osteoarthritis: data from the osteoarthritis initiative. Osteoarthr. Cartil. 2014;22(5) doi: 10.1016/j.joca.2014.03.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Johnson A.J., et al. Knee pain trajectories over 18 months in non-Hispanic black and non-Hispanic white adults with or at risk for knee osteoarthritis. BMC Musculoskelet Disord. 2021;22(1) doi: 10.1186/s12891-021-04284-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Vina E.R., Ran D., Ashbeck E.L., Kwoh C.K. Natural history of pain and disability among African–Americans and Whites with or at risk for knee osteoarthritis: a longitudinal study. Osteoarthr. Cartil. 2018;26(4) doi: 10.1016/j.joca.2018.01.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.D. Previtali et al., “Pain trajectories in knee osteoarthritis—A systematic review and best evidence synthesis on pain predictors,” 2020. doi: 10.3390/jcm9092828. [DOI] [PMC free article] [PubMed]
- 19.Collins J.E., Neogi T., Losina E. Trajectories of structural disease progression in knee osteoarthritis. Arthritis Care Res (Hob.) 2021;73(9) doi: 10.1002/acr.24340. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Halilaj E., Le Y., Hicks J.L., Hastie T.J., Delp S.L. Modeling and predicting osteoarthritis progression: data from the osteoarthritis initiative. Osteoarthr. Cartil. 2018;26(12) doi: 10.1016/j.joca.2018.08.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Trouvin A.P., Marty M., Goupille P., Perrot S. Determinants of daily pain trajectories and relationship with pain acceptability in hip and knee osteoarthritis. A national prospective cohort study on 886 patients. Jt. Bone Spine. 2019;86(2) doi: 10.1016/j.jbspin.2018.06.009. [DOI] [PubMed] [Google Scholar]
- 22.Neelapala Y.V.R., et al. Exploring different models of pain phenotypes and their association with pain worsening in people with early knee osteoarthritis: the MOST cohort study. Osteoarthr. Cartil. 2024;32(2) doi: 10.1016/j.joca.2023.09.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.F. Eckstein, T. Mosher, and D. Hunter, “Imaging of knee osteoarthritis: data beyond the beauty,” 2007. doi: 10.1097/BOR.0b013e328248b4be. [DOI] [PubMed]
- 24.Breivik H., et al. Assessment of pain Assessment of pain intensity and pain relief in acute pain. Br J Anaesth. 2008;101(1) doi: 10.1093/bja/aen103. [DOI] [PubMed] [Google Scholar]
- 25.Hunter D.J., et al. Evolution of semi-quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score) Osteoarthr. Cartil. 2011;19(8) doi: 10.1016/j.joca.2011.05.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Proust-Lima C., Philipps V., Liquet B. Estimation of extended mixed models using latent classes and latent processes: the R package lcmm. J Stat Softw. 2017;78 doi: 10.18637/jss.v078.i02. [DOI] [Google Scholar]
- 27.Radojčić M.R., et al. Pain trajectory defines knee osteoarthritis subgroups: a prospective observational study. Pain. 2020;161(12) doi: 10.1097/j.pain.0000000000001975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Kittelson A.J., Stevens-Lapsley J.E., Schmiege S.J. Determination of pain phenotypes in knee Osteoarthritis: a latent class analysis using data from the Osteoarthritis Initiative. Arthritis Care Res (Hob.) 2016;68(5) doi: 10.1002/acr.22734. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Nicholls E., Thomas E., van der Windt D.A., Croft P.R., Peat G. Pain trajectory groups in persons with, or at high risk of, knee osteoarthritis: findings from the Knee clinical assessment study and the osteoarthritis initiative. Osteoarthr. Cartil. 2014;22(12) doi: 10.1016/j.joca.2014.09.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.F. Saxer et al., “Pain-phenotyping in osteoarthritis: current concepts, evidence, and considerations towards a comprehensive framework for assessment and treatment,” 2024. doi: 10.1016/j.ocarto.2023.100433. [DOI] [PMC free article] [PubMed]
- 31.Pan F., Tian J., Aitken D., Cicuttini F., Jones G. Predictors of pain severity trajectory in older adults: a 10.7-year follow-up study. Osteoarthr. Cartil. 2018;26(12) doi: 10.1016/j.joca.2018.08.002. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.