Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohort

Sara Altamirano; Mylène P Jansen; Daniel L Oberski; Marinus J C Eijkemans; Simon C Mastbergen; Floris P J G Lafeber; Willem E van Spil; Paco M J Welsing

doi:10.1371/journal.pone.0283717

. 2023 Jul 14;18(7):e0283717. doi: 10.1371/journal.pone.0283717

Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohort

Sara Altamirano ¹, Mylène P Jansen ^1,^*, Daniel L Oberski ^2,³, Marinus J C Eijkemans ³, Simon C Mastbergen ¹, Floris P J G Lafeber ¹, Willem E van Spil ^1,⁴, Paco M J Welsing ¹

Editor: Aqeel M Alenazi⁵

PMCID: PMC10348540 PMID: 37450467

Abstract

Objective

To gain better understanding of osteoarthritis (OA) heterogeneity and its predictors for distinguishing OA phenotypes. This could provide the opportunity to tailor prevention and treatment strategies and thus improve care.

Design

Ten year follow-up data from CHECK (1002 early-OA subjects with first general practitioner visit for complaints ≤6 months before inclusion) was used. Data were collected on WOMAC (pain, function, stiffness), quantitative radiographic tibiofemoral (TF) OA characteristics, and semi-quantitative radiographic patellofemoral (PF) OA characteristics. Using functional data analysis, distinctive sets of trajectories were identified for WOMAC, TF and PF characteristics, based on model fit and clinical interpretation. The probabilities of knee membership to each trajectory were used in hierarchical cluster analyses to derive knee OA phenotypes. The number and composition of potential phenotypes was selected again based on model fit (silhouette score) and clinical interpretation.

Results

Five trajectories representing different constant levels or changing WOMAC scores were identified. For TF and PF OA, eight and six trajectories respectively were identified based on (changes in) joint space narrowing, osteophytes and sclerosis. Combining the probabilities of knees belonging to these different trajectories resulted in six clusters (‘phenotypes’) of knees with different degrees of functional (WOMAC) and radiographic (PF) parameters; TF parameters were found not to significantly contribute to clustering. Including baseline characteristics as well resulted in eight clusters of knees, dominated by sex, menopausal status and WOMAC scores, with only limited contribution of PF features.

Conclusions

Several stable and progressive trajectories of OA symptoms and radiographic features were identified, resulting in phenotypes with relatively independent symptomatic and radiographic features. Sex and menopausal status may be especially important when phenotyping knee OA patients, while radiographic features contributed less. Possible phenotypes were identified that, after validation, could aid personalized treatments and patients selection.

Introduction

Osteoarthritis (OA) is the most common form of arthritis worldwide and the knee is among the most affected joints [1]. Its high prevalence is further increasing due to the aging population as well as increasingly widespread risk factors, especially obesity and a sedentary lifestyle [2]. It is estimated that more than 300 million people around the world currently suffer from OA and that the prevalence in the population aged ≥45 will increase from 26.6% in 2012 to 29.5% in 2032, partially due to increasing obesity [3, 4].

On average, people start experiencing OA symptoms around the age of 55 years and live 26 years with the condition [5]. OA can typically be described by a variety of symptoms and structural and biochemical joint changes in multiple tissues including bone, cartilage and synovium. The main symptoms of knee OA are pain, stiffness, and loss of function, leading to reduced mobility and quality of life. Conservative treatment options are focused on symptom reduction and are usually modestly effective at population level. Disease-modifying treatments with the ability to positively affect both the symptomatic and structural disease course are still not available [6].

As OA is hypothesized to be a heterogeneous disease that consists of different phenotypes, it is now believed that personalized diagnostics and treatment are required to optimize intervention [7]. It is plausible that disease heterogeneity is most perceptible and relevant in early-stage OA, as different etiologic processes may accumulate and coalesce over time to a final common pathway [8]. Moreover, as symptoms are generally more manageable and less advanced in early-stage OA, this is likely the most opportune moment to achieve modification of the disease course.

OA phenotypes have been defined as "subtypes of OA that share distinct underlying pathobiological and pain mechanisms and their structural and functional consequences" [9]. The optimal way to identify phenotypes of OA patients and their clinical value is still under investigation and is considered fundamental for the advancement of OA research [7]. Knowledge on potential subgroups of knee OA would allow further research into relevant etiological mechanisms and, in due course, provide insights for more effective prevention and treatment strategies. Such strategies are highly anticipated as they would help improve pain, functional disability, and quality of life for patients, as well as reduce societal burden and healthcare costs related to the disease.

In this study, we aimed to identify knee OA phenotypes using data from Cohort Hip and Cohort Knee (CHECK), consisting of 1002 individuals with symptoms and/or signs of early-stage knee and/or hip OA, having visited their general practitioner for the first time for these symptoms ≤6 months before inclusion. This large group of individuals was followed for 10 years and clinical data and radiographic data of both knees was collected, allowing analysis of longitudinal trajectories and interrelations of OA symptoms and radiographic features over time to identify phenotypes of knee OA.

Methods

Study population

CHECK, an initiative of the Dutch Arthritis Foundation, is a Dutch multi-center, 10-year prospective cohort study of 1002 individuals with symptoms and/or signs of early symptomatic OA of hip and/or knee, aged 45–65 years at the time of inclusion, and without a previous healthcare consultation or with a first consultation no longer than six months ago for these complaints to their primary care physician. Participants were evaluated clinically through regular physical examinations and questionnaires, radiographically through knee and hip radiographs, and biochemically through assessment of markers in plasma, serum, and urine samples [10]. Participants were examined and provided radiographs and samples at baseline (year 0) and, 2, 5, 8 and 10 years thereafter. For patients with more severe complaints at baseline (having 2 of pain, morning stiffness <30 min, crepitus and bony tenderness for the knee and/or 2 of pain, morning stiffness <60min pain on hip internal rotation or internal rotation <15°; most patients), yearly questionnaire data was collected [11]. As a result, compared to these patients with yearly visits, a small group of patients with less severe complaints (9%) missed three or more visits and where excluded from all analyses. The CHECK study was approved by the medical ethics committees of all participating centers, and all participants gave their written informed consent before entering the study. Data collected in the CHECK study is publically available upon request [11, 12].

Functional data analysis

To identify possible phenotypes, an unsupervised learning approach was used. Functional data analysis allows for division/clustering of subjects based on the course over time of multiple separate characteristics. For clinical signs and symptoms, we used the data of the WOMAC [13] subscales for pain, function, and stiffness in the knees. For structural signs, continuous measures of radiographic OA features were used for the tibiofemoral (TF) joint: joint space width (JSW), subchondral bone density, and osteophyte area, all measured medially and laterally using the Knee Images Digital Analysis (KIDA) method [14, 15]. Moreover, for the patellofemoral (PF) joint, semi-quantitative grades (0–3) of radiographic OA features were used: joint space narrowing (JSN), sclerosis and osteophytes, as scored on skyline radiographs [16]. For all features, data from all time points was used. We removed records of knees with missing data for three or more time points (of eleven possible time points for clinical features and five possible time points for radiographic features) and used linear interpolation to input remaining missing values. Baseline values of all parameters used for the functional data analysis and hierarchical cluster analysis were compared between included and removed records using Mann-Whitney U tests and Chi-square tests for continuous and categorical variables, respectively. Thereafter, to classify knees regarding these outcome trajectories in the course of OA, model-based clustering of functional data was performed on the included records using R packages funHDDC [17] and fda [18, 19]. This method aims to cluster the observed functional curves into K homogeneous trajectory groups [20]. We selected the number of trajectory clusters (K) based on the Bayesian Information Criterion (BIC) [21] and evaluation of graphical representations of trajectory groups with different K’s by experts in the subject matter, also taking the number of individual knees in each trajectory group into account (without a formal minimum).

Hierarchical cluster analysis

To define possibly relevant phenotypes, patients should be clustered based on both their characteristics (non-functional data) and their (clinical and structural) progression over time (functional data), but this is not (directly) possible in a single analysis. To combine the functional data (expressed as probabilistic membership of a subject/knee to belong to a specific trajectory group found with the functional data analysis) and non-functional data (i.e. subject characteristics possibly relevant to define phenotypes (see below)) we used a hierarchical cluster analysis with Euclidean distance and Ward linkage method [22]. We used a stepwise approach to hierarchical cluster analysis starting with the trajectory groups (for WOMAC, TF and PF features) and then adding possibly relevant patient characteristics. The following characteristics were investigated based on possibly being related to experience of complaints and biomechanical or tissue processes: sex, menopausal status (if menopausal state was missing in females (22%) age > 50 was used), age, body mass index (BMI), biochemical markers in blood and urine, and pain-coping inventory (PCI) measures [23]. In explorative iterations we inspected clustering patterns using heatmaps created based on selected variables with the R packages heatmaply [24] and dendextend [25]. The specific data to include in the final model and number of phenotypes was again selected based on statistical consideration (average silhouette method (ASM) [26]), clinical considerations, interpretability and size of the clusters.

Differences in subjects’ baseline characteristics between clusters were explored and tested using chi-square for categorical variables and Kruskal-Wallis Rank sum test for continuous variables. We followed reporting recommendations from the consensus-based framework for conducting and reporting osteoarthritis phenotype research [9].

Results

Functional data analysis

From the 1,002 subjects (2,004 knees) available in CHECK, 819 subjects (82%) and 1,788 knees (89%) had sufficient longitudinal data on all clinical and radiographic features for trajectory analysis (S1 Table). Baseline values did not differ between included and excluded knees (all p>0.05), except for TF medial JSW (median and interquartile range (IQR): included 4.66 (4.16–5.16); excluded 4.76 (4.26–5.26); p = 0.038) and PF JSN (included 0.00 (0.00–0.00); excluded 0.00 (0.00–0.00); p = 0.003; % JSN 0/1/2/3 for included 85.9/11.2/2.5/0.4, for excluded 85.8/8.8/5.1/0.3). During several iterations of data exploration no clear patterns could be obtained using all features in a combined analysis. Therefore, it was decided to perform separate analyses for 1) WOMAC pain, 2) WOMAC function, and 3) WOMAC stiffness, for 4) radiographic TF OA features and for 5) radiographic PF OA features, respectively. This was also deemed more appropriate because WOMAC scores were assessed at patient level as opposed to the radiographic features that were assessed at knee level, and because radiographic features were measured continuously for TF OA but discretely for PF OA. The finally derived trajectory groups for the different outcome sets are shown in Figs 1–3.

Fig 1 — Average values within trajectory groups are shown.

Fig 3 — Average values within trajectory groups are shown.

Trajectories for WOMAC scores were defined at patient level, for the TF OA and PF OA scores, analysis was performed and presented at the knee level as each knee in a patient may be differently affected by OA, and it was also considered possible that joints differed regarding their phenotype.

Regarding WOMAC pain, function and stiffness, five trajectories were discovered which could be named as (i) increasing, (ii) decreasing, (iii) high-stable, (iv) moderate-stable, and (v) low-stable (see Fig 1). Most patients were part of the moderate-stable (38–40%) or low-stable (23–39%) trajectories; the other trajectories contained 5–15% of patients.

For radiographic TF OA features, we identified eight trajectories. All trajectories showed decreasing medial JSW and increasing lateral JSW suggesting a population of primarily medial knee OA (although the most affected compartment was not determined). The derived TF OA trajectories were named as: (i) low osteophytes, moderate bone density, (ii) increasing bone density, (iii) low bone density, (iv) increasing lateral osteophytes, moderate bone density, (v) increasing osteophytes, low medial joint space width, increasing bone density, (vi) moderate-high bone density, (vii) high bone density, (viii) moderate-increasing bone density. Fig 2 shows the TF OA trajectories for medial and lateral TF OA features separately. The most pronounced changes occurred in osteophytes, although only for trajectories (iv) and (v), while JSW showed relatively little variation and change.

Fig 2 — Average values within trajectory groups are shown.

For radiographic PF OA characteristics, we found six trajectories and named them as: (i) low joint space narrowing, moderate osteophytes (ii) moderate-increasing OA features, (iii) low OA features, (iv) low joint space narrowing, low increasing osteophytes, (v) high-increasing OA features, (vi) high osteophytes (Fig 3). All three PF features showed relatively high change over time for some trajectories and variation between trajectories.

Hierarchical Cluster Analysis (HCA)

Performing HCA with WOMAC pain, function and stiffness trajectories resulted in four clusters as the optimal solution that could be defined as low clinical impact, high clinical impact, increasing clinical impact, and decreasing clinical impact in line with the individual WOMAC trajectories. Subsequently, the PF OA characteristics were added. Resulting heatmaps and ASM suggested four clusters as statistically optimal, but PF OA damage was inconsistent within the clusters so after taking into account clinical considerations, a solution of six clusters was chosen as optimal (see below). Thereafter, we combined WOMAC and TF OA trajectories. However, inclusion of TF OA characteristics did not prove useful, since none of the cluster groups seemed to be driven by these TF OA characteristics, as determined graphically based on the heatmaps and changes in cluster division within the dendogram. The four groups resulting from clustering based on WOMAC and TF OA characteristics were very similar to the solution using WOMAC scores only, as 76% of knees were clustered in comparable groups. When combining WOMAC, PF OA characteristics, and TF OA characteristics, again TF OA did not add to the clustering solution.

To continue exploring potential phenotypes, the other selected patient characteristics (sex, menopausal status, age, BMI, biochemical markers (see S2 Table), and PCI) were added to the WOMAC and PF OA trajectories. Most of the added characteristics did not show any variability between derived clusters upon visual inspection of heatmaps. Only sex and menopause (categorized as men, premenopausal women, postmenopausal women) clearly differentiated between clusters, actually dominating the clusters. Therefore, we decided to remove the other baseline characteristics that mainly appeared to add noise.

Adding the TF OA trajectories again to the latter solution, or to a solution without the sex/menopause and including the other relevant characteristics, again confirmed that TF OA did not add to clustering solutions (results not shown).

Ultimately, two solutions were selected as clinically and statistically optimal. First, combining WOMAC pain, function, and stiffness with PF OA features which showed the following six clusters that were described as: (i) Low JSN, low WOMAC pain and function, (ii) Low WOMAC pain and function, (iii) high WOMAC pain and function, (iv) Moderate WOMAC features, low JSN, (v) Moderate WOMAC features, and (vi) Low increasing osteophytes, low JSN, moderate WOMAC features (see Fig 4). S2 Table describes patient characteristics for these final ‘phenotype’ clusters. Clusters differed statistically significantly on smoking, hip endo- and exorotation active range of motion, lateral and medial bone density. For example, the cluster with high WOMAC pain and functional complaints (vi) was less likely to smoke and had better hip ROM, while the cluster with moderate WOMAC and low JSN (iv) had somewhat higher TF bone density.

Fig 4 — The different WOMAC (underscore) and PF (bold) trajectories are indicated on the x-axis. WOMAC pain, function and stiffness trajectories are: (i) increasing, (ii) decreasing, (iii) high-stable, (iv) moderate-stable, and (v) low-stable. PF trajectories are: (i) low joint space narrowing, moderate osteophytes (ii) moderate-increasing OA features, (iii) low OA features, (iv) low joint space narrowing, low increasing osteophytes, (v) high-increasing OA features, (vi) high osteophytes.

The other solution added sex and menopausal status as a feature which resulted in sex-dominated clusters. The resulting clusters were named: (i) Premenopausal females with low and moderate pain and function, (ii) Males with low pain and function, (iii) Males with moderate pain and function, (iv) Postmenopausal females with low pain and function, (v) Postmenopausal females with moderate pain and function and low joint space narrowing, (vi) Postmenopausal females with moderate pain and function and varying PF OA characteristics, (vii) Postmenopausal females with high pain and function, and (viii) Postmenopausal females with increasing WOMAC pain and function (Fig 5). These eight clusters differed statistically significantly on almost all characteristics without a clear trend, as shown in S3 Table.

Fig 5 — Sex is split into male, female pre-menopause, female post-menopause. The different WOMAC (underscore) and PF (bold) trajectories and sex (italics) are indicated on the x-axis. WOMAC pain, function and stiffness trajectories are: (i) increasing, (ii) decreasing, (iii) high-stable, (iv) moderate-stable, and (v) low-stable. PF trajectories are: (i) low joint space narrowing, moderate osteophytes (ii) moderate-increasing OA features, (iii) low OA features, (iv) low joint space narrowing, low increasing osteophytes, (v) high-increasing OA features, (vi) high osteophytes.

Discussion

In this study we identified trajectories of patient reported knee symptoms as well as tibiofemoral and patellofemoral radiographic characteristics of knee OA. Trajectories could be classified based on the absolute level over time of specific radiographic features (i.e. JSW, osteophytes, sclerosis) as well as their course over time. Some of these trajectories might be classified as progressive subtypes (i.e. for TF OA increasing lateral and medial osteophytes, low medial JSW, increasing bone density, and for PF OA increasing/high sclerosis and JSN, and high osteophytes) or non-progressive (i.e. for TF low osteophytes, stable bone density, or for PF OA, low sclerosis, JSN, and osteophytes). Likewise, for symptoms, groups with low, moderate and high levels of symptoms over time as well as increasing and decreasing trends over time were identified.

As our primary aim was to identify knee OA phenotypes based on radiographic features as well as symptoms and possible etiologically relevant patient characteristics, we combined the trajectory groupings for each of these features in a cluster analysis. Results indicated that these features (and their course over time) are relatively independent. Specifically, the radiographic TF OA characteristics did not seem to add to subgrouping/phenotyping knees and radiographic PF OA features did contribute only to a limited extent. This is, however, in line with previous findings [27–29] suggesting that TF OA features are not strongly related to clinical features. The fact that PF OA was (somewhat) more related to complaints may indicate that PF OA is typically an early sign of the development of knee OA and the TF OA characteristics follow later, and thus may not influence phenotypes in this cohort of early OA. This has been suggested before [30]. Alternatively, the absence of TF features in the clustering might be the result of the continuous measurements used for TF features, sensitive to inter-subject variation, instead of semi-quantitative measurement of PF features. While this does directly not matter for the clustering, since only trajectory probabilities were used and no measurement values, it may have influenced the identified trajectories. Additionally, general patient characteristics were of limited help in subgrouping. Only sex and menopausal status seemed to influence subgrouping, which resulted in an almost complete separation between males and pre- and post-menopausal women, with further subgrouping based on reported knee complaints. Although, the analyses seemed to be somewhat dominated by sex and menopausal state, it may also suggest that taking these factors into account in defining phenotypes may indeed be important from a pathobiological view. Also, it is not clear whether importance of sex in the phenotyping is caused by biological differences (e.g. hormones, which are influenced by menopause as well) or social differences, or both.

There is a chance findings do not point to different phenotypes, but to different disease stages. However, there is no obvious chronologic line in the clusters (i.e. you cannot simply sort them from less to more advanced OA phase). Also, all patients have had their first complaints within 6 months of the first visit, meaning they are clinically all in the same phase of the disease at baseline, i.e. have the same starting point. Although this starting point is only based on clinical outcome, and not on radiographic features, it is very relevant in clinical practice. Further, the baseline KL grade and most tibiofemoral parameters were comparable and not significantly different between the found clusters (S2 Table), so they likely did not start at very different structural stages either. Even if we derived trajectories represent different disease phases, groups would (at least partly) represent patients that progress more or less quickly through different phases, which may be regarded in itself a phenotype.

While some of our results were anticipated, we expected more trajectories showing discordant courses between radiographic features and symptoms (e.g. a subgroup of patients with severe symptoms and only limited radiographic signs, reflecting central pain sensitization). Instead, radiographic features and symptoms mainly showed concordant courses and only the PF features ‘low-increasing osteophytes, low JSN’ and the clusters containing ‘moderate WOMAC features, low JSN’ could be considered somewhat discordant. On the other hand, the different clusters based on WOMAC severity did not show significant differences in tibiofemoral features (e.g. JSW, osteophytes, KL grade), as shown in S2 Table, indicating that there is some discordance between symptoms and radiographic TF features, but this is not expressed in the clustering since TF features did not contribute. Also, we expected to see a distinction between lateral and medial OA, potentially with different results for patients with predominantly lateral or medial OA, but this was not the case. As medial TF knee OA is generally more prevalent, the number of patients with predominant lateral TF knee OA might have been too low to be represented in separate trajectories. Since medial vs lateral OA was not officially determined, it is unknown how many participants had medial or lateral OA. Increasing the number of trajectories and/or clusters might have resulted in finding a lateral OA group, but even with defining up to 10 clusters no clear lateral OA group emerged, indicating that lateral OA is not an important subgroup of early OA.

Other phenotype analyses have been performed in OA cohorts before [31]. Two studies used data from the CHECK cohort as well and identified distinct trajectories in the Numeric (Pain) Rating Scale (NRS) over five years. While one showed similar results as we did for the ten-year WOMAC trajectories [32], the other showed trajectories based only on the level of pain and not the change in pain over time [33]. Both studies, as well as another CHECK study identifying WOMAC function over five years, showed significant differences between trajectory subgroups in BMI, education level, and comorbidity [34]. In our clusters based on WOMAC and PF features, we did not find significant differences in these characteristics (all p>0.07; S2 Table). However, when comparing patients from different WOMAC trajectory subgroups, we found similar differences for the number of comorbidities (p = 0.001) but not BMI or education level (both p>0.14). Between clusters that were also based on sex and menopausal status all these three characteristics differed significantly, but these differences may be driven by the sex and menopausal status differences. The fact that we did not find differences between groups for BMI or education level could be because we did not compare characteristics between trajectories as the previous studies did, but between clusters based on more than only clinical outcome, and possibly because of the longer follow-up of ten instead of five years. Other multivariate clustering studies, using different patient cohorts, have been performed as well, although clustering was based on baseline data and not on changes over time in those studies, and PF characteristics were not taken into account [35, 36]. Like in our analyses, those clusters showed differences in OA structural damage or pain, but unlike in our study, they differed significantly in BMI and, in one study, results showed separate clusters for medial or lateral OA. The fact that previous studies found significant differences in BMI and we did not might be because previous studies clustered based only on pain, and BMI is known to be linked to musculoskeletal pain, while we looked at other factors as well.

This study naturally had limitations. Some data was missing (4%, 13% and 6% of visits for WOMAC, TF and PF parameters, respectively), which was imputed using linear interpolation to enable functional data analysis. While we think the impact of this missing data was minimal, it may still have influenced subgrouping results. Also, excluding patients with ≥3 visits means patients with less severe complaints and therefore without yearly visits were excluded. This might have caused selection bias, although the baseline complaints (clinical outcome) were not different between included and excluded patients, and looking at the WOMAC trajectories, the population clearly still included patients with less severe complaints. JSW/JSN did show a small but significant difference between included and excluded patients, with the excluded patients showing a somewhat less affected JS. Furthermore, the fact that only radiographs were available for evaluation of structural TF and PF parameters was a major limitation. Using other imaging modalities such as MRI or CT may have allowed for analysis of additional relevant structural parameters (e.g. bone marrow lesions) or for better evaluation of the current parameters (e.g. directly evaluating cartilage thickness). Further, while the results were based on statistical and expert-knowledge based considerations, final results are sensitive to the expert knowledge of the researchers as well as the data source (as is always the case).

Because of the nature of the CHECK cohort, patients with predominantly hip OA as well as patients without (knee) OA could have been included. As such, while most patients likely had early knee OA, patients without knee OA were included in these results as well, which might have influenced and be represented in the trajectories and clustering with few progression and complaints. Also, while the WOMAC questionnaire is specific to the knee, having hip OA might influence knee function as well, as could the contralateral knee and hip, since the WOMAC was available only on patient level and not on joint level. An important next step in this research would be to validate our results in another cohort with early-stage knee OA or another cohort with similar measurements like the OsteoArthritis Initiative (OAI). Moreover, for some of the identified trajectories/phenotypes, it may be of value to predict if a patient will likely belong to that group in an early phase of the disease. For example, more progressive groups (e.g. the ‘increasing osteophytes, low medial JSW, increasing bone density’ trajectory or ‘postmenopausal females with increasing pain and function’ phenotype) may be most likely to be destined for joint replacement, although this did not clearly seem to be the case during the 10-year follow-up of CHECK (S2 and S3 Tables). This would have to be evaluated with prediction algorithms and validation in a future study with long follow-up.

In conclusion, based on our results, patient trajectories in pain and radiographic features seem to be largely independent, and sex and menopausal status need to be considered when phenotyping knee OA patients. While further validation is required as a next step, interrelations of OA symptoms and radiographic features were analyzed to identify phenotypes in a large group of early knee OA patients, which in the future could be used for personalized treatments and patient selection.

Supporting information

S1 Table. Final list of variables included in the functional analysis.

*We removed records of knees with missing data for three or more time points and used linear interpolation to input remaining missing values.

(DOCX)

Click here for additional data file.^{(13.1KB, docx)}

S2 Table. Descriptive statistics of combined scenario with WOMAC and PF OA features.

ROM: range of motion; CTX-I: C-terminal telopeptide of collagen I; CTX-II: C-terminal telopeptide of type II collagen; C1,2C: collagen of types I and II; COMP: cartilage oligomeric matrix protein; PIIANP: collagen N-propeptide of type IIA; CS846: chondroitin sulphate 846; NTX-I: N-terminal telopeptide of collagen I; OC: osteocalcin; PINP: aminoterminal propeptide of type I procollagen; HA: hyaluronic acid; PIIIANP: N-terminal propeptide of type III procollagen; hsCRP: high-sensitivity C-reactive protein; BSE: erythrocyte sedimentation rate. * P-values calculated with chi-square test. **for missing values, mean of all SOS items was taken (n = 6). Median and interquartile range (Q1-Q3) given for continuous variables unless otherwise indicated.

(DOCX)

Click here for additional data file.^{(27.4KB, docx)}

S3 Table. Descriptive statistics of combined scenario with WOMAC, PF OA features and sex.

(DOCX)

Click here for additional data file.^{(30.2KB, docx)}

Data Availability

Restrictions on sharing the data for this study are imposed by the Institutional Review Board of the University Medical Center Utrecht, Utrecht, The Netherlands. Data sharing is restricted because the dataset contains possible identifying information. All relevant data are available upon request by sending an email to the Rheumatology department of the UMC Utrecht (urrci@umcutrecht.nl). This is a non-author email address that allows for maintenance of long-term data accessibility. The full CHECK dataset, of which we used data for the current manuscript, is publicly available through the ‘THEMATIC COLLECTION: CHECK (COHORT HIP & COHORT KNEE)’, accessible through the DANS database (https://doi.org/10.17026/dans-252-qw2n).

Funding Statement

This project was funded by ReumaNederland (project number 18-2-202). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1.Murphy SL, Lyden AK, Phillips K, Clauw DJ, Williams DA. Subgroups of older adults with osteoarthritis based upon differing comorbid symptom presentations and potential underlying pain mechanisms. Arthritis Res Ther. 2011;13: R135. doi: 10.1186/ar3449 [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Palazzo C, Nguyen C, Lefevre-Colau MM, Rannou F, Poiraudeau S. Risk factors and burden of osteoarthritis. Ann Phys Rehabil Med. 2016;59: 134–138. doi: 10.1016/j.rehab.2016.01.006 [DOI] [PubMed] [Google Scholar]
3.Safiri S, Kolahi AA, Smith E, Hill C, Bettampadi D, Mansournia MA, et al. Global, regional and national burden of osteoarthritis 1990–2017: A systematic analysis of the Global Burden of Disease Study 2017. Ann Rheum Dis. 2020;79. doi: 10.1136/annrheumdis-2019-216515 [DOI] [PubMed] [Google Scholar]
4.Turkiewicz A, Petersson IF, Björk J, Hawker G, Dahlberg LE, Lohmander LS, et al. Current and future impact of osteoarthritis on health care: A population-based study with projections to year 2032. Osteoarthr Cartil. 2014;22: 1826–1832. doi: 10.1016/j.joca.2014.07.015 [DOI] [PubMed] [Google Scholar]
5.Losina E, Weinstein AM, Reichmann WM, Burbine SA, Solomon DH, Daigle ME, et al. Lifetime risk and age at diagnosis of symptomatic knee osteoarthritis in the US. Arthritis Care Res. 2013;65: 703–711. doi: 10.1002/acr.21898 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Karsdal MA, Michaelis M, Ladel C, Siebuhr AS, Bihlet AR, Andersen JR, et al. Disease-modifying treatments for osteoarthritis (DMOADs) of the knee and hip: lessons learned from failures and opportunities for the future. Osteoarthr Cartil. 2016;24: 2013–2021. doi: 10.1016/j.joca.2016.07.017 [DOI] [PubMed] [Google Scholar]
7.Deveza LA, Nelson AE, Loeser RF. Phenotypes of osteoarthritis: current state and future implications. Clin Exp Rheumatol. 2019;37: 64–72. Available: /pmc/articles/PMC6936212/ doi: 10.1016/j.joca.2019.06.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Driban JB, Sitler MR, Barbe MF, Balasubramanian E. Is osteoarthritis a heterogeneous disease that can be stratified into subsets? Clin Rheumatol. 2010;29: 123–131. doi: 10.1007/s10067-009-1301-1 [DOI] [PubMed] [Google Scholar]
9.Van Spil WE, Bierma-Zeinstra SMA, Deveza LA, Arden NK, Bay-Jensen AC, Kraus VB, et al. A consensus-based framework for conducting and reporting osteoarthritis phenotype research. Arthritis Res Ther. 2020;22: 54. doi: 10.1186/s13075-020-2143-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Van Spil WE, Jansen NWD, Bijlsma JWJ, Reijman M, DeGroot J, Welsing PMJ, et al. Clusters within a wide spectrum of biochemical markers for osteoarthritis: Data from CHECK, a large cohort of individuals with very early symptomatic osteoarthritis. Osteoarthr Cartil. 2012;20: 745–754. doi: 10.1016/j.joca.2012.04.004 [DOI] [PubMed] [Google Scholar]
11.Wesseling J, Boers M, Viergever MA, Hilberdink WK, Lafeber FP, Dekker J, et al. Cohort profile: Cohort Hip and Cohort Knee (CHECK) study. Int J Epidemiol. 2016;45: 36–44. doi: 10.1093/ije/dyu177 [DOI] [PubMed] [Google Scholar]
12.Bijlsma JWJ, Wesseling J. Thematic collection: CHECK (Cohort Hip & Cohort Knee). DANS. 2015. 10.17026/dans-252-qw2n [DOI] [Google Scholar]
13.Bellamy N, Buchanan W, Goldsmith C, Campbell J, Stitt L. Validation study of WOMAC: a health status instrument for measuring clinically important patient relevant outcomes to antirheumatic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol. 1988. pp. 1883–40. [PubMed] [Google Scholar]
14.Marijnissen ACA, Vincken KL, Vos PAJM, Saris DBF, Viergever MA, Bijlsma JWJ, et al. Knee Images Digital Analysis (KIDA): a novel method to quantify individual radiographic features of knee osteoarthritis in detail. Osteoarthr Cartil. 2008;16: 234–243. doi: 10.1016/J.JOCA.2007.06.009 [DOI] [PubMed] [Google Scholar]
15.Jansen MP, Welsing PMJ, Vincken KL, Mastbergen SC. Performance of knee image digital analysis of radiographs of patients with end-stage knee osteoarthritis. Osteoarthr Cartil. 2021;29: 1530–1539. doi: 10.1016/j.joca.2021.07.013 [DOI] [PubMed] [Google Scholar]
16.Burnett SJHD, Hart DJ, Cooper C, Spector TD. A radiographic atlas of osteoarthritis. 1994. [Google Scholar]
17.Schmutz A, Jacques J, Bouveyron C. funHDDC: Univariate and Multivariate Model-Based Clustering in Group-Specific Functional Subspaces. R package version 2.3.0. 2021. [cited 6 Oct 2021]. Available: https://cran.r-project.org/package=funHDDC [Google Scholar]
18.Ramsay JO, Graves S, Hooker G. fda: Functional Data Analysis. R package version 2.4.8. 2021. https://CRAN.R-project.org/package=fda [Google Scholar]
19.Core Team R. R: A language and environment for statistical computing. 2013. [Google Scholar]
20.Schmutz A, Jacques J, Bouveyron C, Chèze L, Martin P. Clustering multivariate functional data in group-specific functional subspaces. Comput Stat. 2020;35: 1101–1131. doi: 10.1007/s00180-020-00958-4 [DOI] [Google Scholar]
21.Schwarz G. Estimating the Dimension of a Model. Ann Stat. 2007;6: 461–464. doi: 10.1214/aos/1176344136 [DOI] [Google Scholar]
22.James G, Witten D, Hastie T, Tibshirani R. An introduction to Statistical Learning. Current medicinal chemistry. 2013. doi: 10.1007/978-1-4614-7138-7 [DOI] [Google Scholar]
23.Kraaimaat FW, Evers AWM. Pain-Coping Strategies in Chronic Pain Patients: Psychometric Characteristics of the Pain-Coping Inventory (PCI). Int J Behav Med. 2003;10: 343–363. doi: 10.1207/s15327558ijbm1004_5 [DOI] [PubMed] [Google Scholar]
24.Galili T, O’Callaghan A, Sidi J, Sievert C. Heatmaply: An R package for creating interactive cluster heatmaps for online publishing. Bioinformatics. 2018;34: 1600–1602. doi: 10.1093/bioinformatics/btx657 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Galili T. dendextend: An R package for visualizing, adjusting and comparing trees of hierarchical clustering. Bioinformatics. 2015;31: 3718–3720. doi: 10.1093/bioinformatics/btv428 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Kaufman L, Rousseeuw PJ. Finding Groups in Data: An Introduction to Cluster Analysis. 1990. doi: 10.1002/9780470316801 [DOI] [Google Scholar]
27.Son KM, Hong JI, Kim D-H, Jang D-G, Crema MD, Kim HA. Absence of pain in subjects with advanced radiographic knee osteoarthritis. BMC Musculoskelet Disord 2020 211. 2020;21: 1–9. doi: 10.1186/s12891-020-03647-x [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Creamer P, Lethbridge‐Cejku M, Hochberg MC. Factors associated with functional impairment in symptomatic knee osteoarthritis. Rheumatology. 2000;39: 490–496. doi: 10.1093/rheumatology/39.5.490 [DOI] [PubMed] [Google Scholar]
29.Bedson J, Croft PR. The discordance between clinical and radiographic knee osteoarthritis: A systematic search and summary of the literature. BMC Musculoskelet Disord 2008 91. 2008;9: 1–11. doi: 10.1186/1471-2474-9-116 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Eijkenboom JFA, Waarsing JH, Oei EHG, Bierma-Zeinstra SMA, Van Middelkoop M. Is patellofemoral pain a precursor to osteoarthritis? Bone Jt Res. 2018;7: 541–547. doi: 10.1302/2046-3758.79.BJR-2018-0112.R1 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Deveza LA, Melo L, Yamato TP, Mills K, Ravi V, Hunter DJ. Knee osteoarthritis phenotypes and their relevance for outcomes: a systematic review. Osteoarthr Cartil. 2017;25: 1926–1941. doi: 10.1016/J.JOCA.2017.08.009 [DOI] [PubMed] [Google Scholar]
32.Bastick AN, Wesseling J, Damen J, Verkleij SP, Emans PJ, Bindels PJ, 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: e32. doi: 10.3399/bjgp15X688129 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Wesseling J, Bastick AN, Ten Wolde S, Kloppenburg M, Lafeber FPJG, Bierma-Zeinstra SMA, et al. Identifying trajectories of pain severity in early symptomatic knee osteoarthritis: A 5-year followup of the cohort hip and cohort knee (CHECK) study. J Rheumatol. 2015;42: 1470–1477. doi: 10.3899/jrheum.141036 [DOI] [PubMed] [Google Scholar]
34.Holla JFM, Leeden M van der, Heymans MW, Roorda LD, Bierma-Zeinstra SMA, Boers M, et al. Three trajectories of activity limitations in early symptomatic knee osteoarthritis: a 5-year follow-up study. Ann Rheum Dis. 2014;73: 1369–1375. doi: 10.1136/annrheumdis-2012-202984 [DOI] [PubMed] [Google Scholar]
35.Waarsing JH, Bierma-Zeinstra SMA, Weinans H. Distinct subtypes of knee osteoarthritis: data from the Osteoarthritis Initiative. Rheumatology. 2015;54: 1650–1658. doi: 10.1093/rheumatology/kev100 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Pan F, Tian J, Cicuttini F, Jones G, Aitken D. Differentiating knee pain phenotypes in older adults: a prospective cohort study. Rheumatology. 2019;58: 274–283. doi: 10.1093/rheumatology/key299 [DOI] [PubMed] [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0283717.r001

Decision Letter 0

Aqeel M Alenazi

19 Dec 2022

PONE-D-22-19232Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohortPLOS ONE

Dear Dr. Jansen,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

==============================

ACADEMIC EDITOR: Revisions are required for this manuscript to improve clarity and details on the analysis and methods.

==============================

Please submit your revised manuscript by Jan 29 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Aqeel M Alenazi

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In the Methods section of your revised manuscript, please include the full name of the institutional review board or ethics committee that approved the protocol, the approval or permit number that was issued, and the date that approval was granted.

3. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

Additional Editor Comments (if provided):

Revisions are required for this manuscript to improve clarity and details on the analysis and methods.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: I Don't Know

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is an interesting and complex manuscript about describing the longitudinal progression of people with knee OA. My understanding of trajectories and clusters analyses are that they are descriptive analyses, retrospectively classifying people into groups.

My main comment is that I am unable to provide any meaningful review of the statistical analysis and the decision-making process. My knowledge of these techniques is too limited. I am also not certain that others could replicate the analysis process of this paper. I think there were multiple, iterative analyses which lead to results-driven decisions about the next stage. I believe that these investigator decisions are likely to be valid, but that possible limitations of this approach should be discussed. Is this liable to something similar to "over-fitting", or the usual limitations of post-hoc analysis?

I believe that all data from all time points were used in this paper. However, I didn't find it easy to confirm this opinion when I read the methods. Maybe make this clearer?

Surely the data collection method will have a strong influence on the final analyses? So the unusual decision to measure different "severities" at different rates needs a discussion. The authors mention that there was a lack of discordant findings in their clusterings. Could this have been influenced by the data collection frequencies?

I tried to check the trajectories and I think they look ok. I couldn't see any issues with their names/descriptions.

I think figures 4 and 5 could benefit from detailed long legends. I am not certain what the PFi, Pain iv etc x-axis labels describe.

In the supplements, the authors report different clusters based upon WOMAC severity, and they often have similar radiographic findings. This might be evidence of the discordance that the authors were looking for in their clusters? Is it surprising how often KL grade was similar across many clusters (supplement tables)?

In general I found the Discussion to be conservative. The impact of this paper will probably be due to classification of people into different outcomes in their future. Do the clusters/trajectories yield something that appears to be clinically-meaningful? Which groups of people seem most likely to be destined for joint replacement etc? One major finding was the separation of males from pre- and post-menapase females; did the final clinical outcomes of these groups differ in a meaningful way? Also, if the authors choose to increase their sex-based discussion, then a greater discussion of possible sex/gender-related confounding should be included.

Multiple clusters had relative small percentages of people in them. Is the failure to identify a lateral OA cluster a serious limitation?

Reviewer #2: Major comments:

As this is a study in a supposedly “early” OA cohort, it is a great pity that the characterization at a structural level was limited to radiography, as using several MRI features (cartilage and meniscus damage, meniscal extrusion, bone marrow lesions, synovitis) would have been much more informative, with radiography missing all these tissue specific pathologies Even JSW (and JSN) in radiography has been shown to not be specific to cartilage thickness, but is confounded by meniscus extrusion. The authors should specify this as a major limitation of the study.

This being said, the current analysis is of course still very useful, since MRI is not currently used in the clinical routine of managing knee OA.

Yet, I think the authors need to provide more evidence that they are examining really phenotypes, and not just different stages of the disease. There appear to be patients with and without radiographic signs of TF and PF disease, and exhibiting different grades. How can the authors ensure that entering these variables, they are not just differentiating stages of the same disease (same path of progression), rather than different phenotypes (different paths). Similar considerations apply to the WOMAC scores.

Minor comments:

ABSTRACT

Design: Please give a short hint (maybe in brackets) how early OA is defined.

Results:

How is the sentence: Combining….and radiographic (PF) parameters….

It is unclear from my perspective whether the six clusters are only based on PF radiographic OA, and why TF radiographic parameters are not included.

When you say “gender”, I think this should be termed “sex”, since you are talking about biological and not social attributes. This comment also applies to the main text.

When including baseline parameters, does none of the radiographic features contributes to any of the clusters ?

Conclusions:

Although several…., they were shown…. Why are the first and second part of the sentence contradictions. The sentence is pretty general and could be a bit more specific.

Are really “gender” and “menopausal status” the main components making up 6-8 clusters (including men?), when phenotyping knee OA patients. From this sentence it appears that functional and radiographic (PF and TF) make not contribution. If this is so, please state this in the conclusion. Again, the conclusion should, in my view, be more concrete and more specific.

INTRODUCTION

P9, line 50: Please omit “already”

P9, line 56: What is meant by “biochemical joint changes”. Please name the tissue you are referring to.

P9, line 59/60: … treatment….are available: Either “treatments” or “is available”

P9, line 61: I suggest to say: disease “that consists of different phenotypes”

P9, line 64: suggest to add: “coalesce over time to a final common pathway. (delete “in patients”)

P10, line76: Please state in the introduction how you define “early” OA, as this term is very fashionable and it appears every researcher understands something different when using this term.

P 10, line 78: Please omit “for the first time”

METHODS

P10, line 84. As this study appears to focus on knee OA, but some patients also have hip OA, did you consider to include hip OA in your analysis and phenotypic characterization, as lower limb function may be impacted by the presence / absence of hip OA as well as knee OA.

P10, lines 90/91: How can the patients have “severe complaints at baseline (most patients”, if this is a cohort of “early” disease?

P11, lines 95/96: Did you use WOMAC scores from one knee only (which one), or from both knees, as did the OAI. How did you account for contralateral knee (and hip) functional status in your analyses?

P 11, line 98: Why did you use an absolute measure of JSW that varies substantially between patients of different sex and/or different body height, instead of using the common JSN grading (comparing the JSW with the less affected compartment and/or contralateral knee), as a real measure of pathology.

P 11, line 98: Why did you use osteophyte “area” rather than the common semi-quantitative grading system?

P 11, line 99/100: Why did you use a quantitative system for the TF joint, and a semi-quantitative one for the PF joint? As you appear to find a greater contribution of the PF than TF joint to the clusters, may this be due to PF radiographic grades being used semi-quantitatively (as a sign of pathology), whereas the TF joint is evaluated quantitatively, where absolute measures are not really a sign of radiographic pathology, but are strongly confounded by inter-subject variation.

P 11, line 106: Thereafter…. Is this then referring only to the group included. This is a bit confusing.

P 11, line 111: domain “experts”. Is this word used correctly here. May be due to my unfamiliarity with this particular method.

P 12, lines 115-117: Can you give the non-specialist reader a sense why the non-functional data is possibly relevant to define phenotypes (and not the functional analysis)?

P12, lines 120-123: It appears that the PF and TP radiographic features were not included at this level; why not?

P12, lines 115 ff: Could you try to give the reader a sense of what these various methods do conceptionally, as not every reader may be familiar with theses statistical methods.

RESULTS

P13, lines 138-141: Why do you think specifically the TF JSW and PF JSN differ between included and excluded participants, but none of the other parameters?

P13, lines 141 ff: it was decided…. based on which criteria

P13, lines 149 ff: Why was the WOMAC score not determined at the knee level (as was done in OAI). Which knee was selected for the analysis regarding WOMAC in each patient? How do the authors expect the contralateral knee to confound the scores?

P13, lines 160 ff: How did you define medial knee OA in the absence of radiographic JSN scores?

P14, line 184: Adding WOMAC…. Adding to what ?

P15, lines 190 ff: I wonder whether the lack of relevance of TF OA is due to the lack of use of JSN scores, as JSW is not a measure of pathology in a cross sectional context?

Can you provide results on what the JSW change was over 10 years in the medial and in the lateral compartment, respectively?

P15, line 198 : Please do not say “unfortunate”, the presentation of results should be neural

P15, lines 200 ff: When sex and menopause were added (how did you classify men in this context), did really all the WOMAC and radiographic PF OA data became irrelevant (Therefore we decided to remove the other characteristics……)

P15, lines 206ff: In my view, these phenotypes appear to represent different stages of the disease.

P 15/16: The listing of the different clusters in the text is a bit difficult to follow. Maybe these could be put into Tables, cross-tabling the different features. This will also display visually which combinations were not in the set of selected phenotypes.

DISCUSSION:

P 16, lines 238 ff.: It would be beneficial if the structure of presentation of results could differ more clearly between cross sectional and longitudinal (change) distributions of the various features.

P 17: When talking about TF and PF OA here, please add “radiographic” to the description.

P 18, lines 268/269: How many of your participants had medial vs. lateral TF radiographic OA?

P 19, line 291: Why do the authors think the results are so discordant between studies, given that they have examined a relatively large cohort.

P 19, line 293: Instead or “some missing data was present”, please say “ some data was missing”.

P 19, lines 308 ff: Shouldn’t the Osteoarthritis Initiative (OAI) data be used for validation?

P 20, conclusions: There is more text about what was done than about what was found. Please focus more on concrete and specific findings in the conclusion section (also in the abstract)

P 20, conclusions: Please do not state “for the first time”, since the study is one in line of such analyses, and of course every study has its peculiarities, but these to not necessarily qualify for stating “for the first time”, if similar approaches have been taken before.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Daniel McWilliams

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2023 Jul 14;18(7):e0283717. doi: 10.1371/journal.pone.0283717.r002

Author response to Decision Letter 0

26 Jan 2023

We thank the reviewers for the helpful comments. We addressed these comments and feel the manuscript has improved. Please find an itemized response to all questions of the reviewers below. Hopefully we answered all questions and addressed all comments to the intention and satisfaction of the reviewers.

Journal requirements

- Journal Comment:

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

- Author response:

The manuscript’s style and file naming have been updated to meet PLOS ONE’s style requirements (this has been done without tracked changes, for readability).

- Journal comment:

In the Methods section of your revised manuscript, please include the full name of the institutional review board or ethics committee that approved the protocol, the approval or permit number that was issued, and the date that approval was granted.

- Author response:

For this study, we used publically available data originally collected in the CHECK cohort. As such, we do not have a name and number for the ethical approval that was granted. However, we have added the following to the Methods section of the revised manuscript [line 101-103], with references to the original study and the data: “The CHECK study was approved by the medical ethics committees of all participating centers, and all participants gave their written informed consent before entering the study. Data collected in the CHECK study is publicly available upon request.[11,12]”

- Journal comment:

Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

- Author response:

Supporting Information file names and captions have been updated as required (without tracked changes, for readability).

Reviewer 1

- Reviewer comment:

This is an interesting and complex manuscript about describing the longitudinal progression of people with knee OA. My understanding of trajectories and clusters analyses are that they are descriptive analyses, retrospectively classifying people into groups. My main comment is that I am unable to provide any meaningful review of the statistical analysis and the decision-making process. My knowledge of these techniques is too limited. I am also not certain that others could replicate the analysis process of this paper. I think there were multiple, iterative analyses which lead to results-driven decisions about the next stage. I believe that these investigator decisions are likely to be valid, but that possible limitations of this approach should be discussed. Is this liable to something similar to "over-fitting", or the usual limitations of post-hoc analysis?

- Author response: Indeed we use trajectory and cluster analysis to identify possible relevant subgroups in progression of knee OA. It is important to realize that we use these statistical techniques as a tool to enable clinical evaluation and definition of (clinically) relevant subgroups. The analysis is thus merely used for identifying and illustrating possible subgroups for clinical validation. This validation is to our opinion key and the advanced analyses helps to perform this tasks more sensitively. Overfitting is not an important issue here to our opinion, because results are a combination of statistical as well as expert knowledge-based feasibility. This may actually be better generalizable than either the statistically most feasible or expert knowledge-based solutions only. However, final results are, of course, sensitive to expert knowledge of the researchers as well as the data source (as is always the case). This has been added to the limitations in the Discussion section of the revised manuscript [line 354-356].

- Reviewer comment:

I believe that all data from all time points were used in this paper. However, I didn't find it easy to confirm this opinion when I read the methods. Maybe make this clearer?

- Author response:

Indeed data from all time points was used. We have clarified this in the Methods section of the revised manuscript [line 114].

- Reviewer comment:

- Author response:

Unfortunately the decision to measure different severities at different rates was already taken, as we used previously collected data from the CHECK cohort. However, the lack of discordant findings in our clustering was likely not the result of the difference in data collection frequencies, since patients with less frequent visits (because of less severe complaints) missed ≥3 visits and were therefore excluded from all analyses in this manuscript. As such our population consisted of early OA with clinically relevant complaints. This is discussed in the Discussion section [line 345-350 of the revised manuscript with tracked changes].

- Reviewer comment:

I tried to check the trajectories and I think they look ok. I couldn't see any issues with their names/descriptions.

- Author response:

We are happy to hear that.

- Reviewer comment:

I think figures 4 and 5 could benefit from detailed long legends. I am not certain what the PFi, Pain iv etc x-axis labels describe.

- Author response:

These labels describe the different WOMAC pain, function, stiffness trajectories and patellofemoral (PF) trajectories resulting from the functional data analysis. We should have made this more clear in the legends of Fig4 and Fig5, and have now done so in the revised manuscript [line 236-240 and 253-257].

- Reviewer comment:

- Author response:

This is a very good point and indeed is indeed evidence that there is some discordance between radiographic (TF) features and symptoms, it just is not expressed in the clustering since TF features did not contribute to that. We have added this consideration in the Discussion section of the revised manuscript [line 307-310].

- Reviewer comment:

- Author response:

While development of prediction algorithms and validation is outside of the scope of the current paper, it is indeed important for confirming the clinical relevance of the found clusters/trajectories and we should indeed discuss this in the current paper. As mentioned by the reviewer, it is relevant to discuss that especially the more progressive groups (e.g. the ‘increasing osteophytes, low medial JSW, increasing bone density’ trajectory or ‘postmenopausal females with increasing pain and function’ phenotype) may be most likely to be destined for joint replacement, although this did not clearly seem to be the case during the 10-year follow-up of CHECK (Table S2/S3). We have now included this in the Discussion section of the revised manuscript [line 366-371].

- One major finding was the separation of males from pre- and post-menapase females; did the final clinical outcomes of these groups differ in a meaningful way? Also, if the authors choose to increase their sex-based discussion, then a greater discussion of possible sex/gender-related confounding should be included.

- Author response:

Especially progression of clinical features (i.e. WOMAC pain) was quite different between males and –pre- and post-menopausal females, which is why the phenotypes (Fig 5) are so clearly separated between sexes. It is not clear if the differences between sexes and menopausal status is because of biological differences (e.g. hormones, which are influenced by menopause as well) or social differences, or both. We have added this consideration to the Discussion section of the revised manuscript [line 288-290].

- Reviewer comment:

Multiple clusters had relative small percentages of people in them. Is the failure to identify a lateral OA cluster a serious limitation?

- Author response:

Indeed there were three clusters with <10% of patients, with the smallest cluster consisting of 78 knees (5%), which is small but not negligible. Cluster size and trajectory group size was one of our considerations in deciding the number of clusters and trajectories, as mentioned in the Methods section [line 122-125 and 141-143]. Increasing the number of trajectories/clusters further may have resulted in us finding a lateral OA group but even when increasing the number of clusters up to 10 no clear lateral OA group emerged, and group sizes would likely have been much smaller, making results less applicable. As such indeed missing a lateral OA cluster may indeed not be an important limitation. We have included this in the Discussion section of the revised manuscript [line 313-317]. Failing to identify a lateral OA cluster is not considered a serious limitation, solely an unexpected observation. These are early OA patients, so perhaps there are simply not many patients with a clearly more severely affected lateral compartment.

Reviewer 2

- Reviewer comment:

As this is a study in a supposedly “early” OA cohort, it is a great pity that the characterization at a structural level was limited to radiography, as using several MRI features (cartilage and meniscus damage, meniscal extrusion, bone marrow lesions, synovitis) would have been much more informative, with radiography missing all these tissue specific pathologies. Even JSW (and JSN) in radiography has been shown to not be specific to cartilage thickness, but is confounded by meniscus extrusion. The authors should specify this as a major limitation of the study. This being said, the current analysis is of course still very useful, since MRI is not currently used in the clinical routine of managing knee OA.

- Author response:

We agree that the fact that structural evaluation was limited to radiography is a limitation. This was mentioned in the Discussion section, and we have now specified further that it is a major limitation in the Discussion section of the revised manuscript [line 350-351].

- Reviewer comment:

- Author response:

This is a good point and we cannot be certain that we are not looking at different disease stages. However, there are several reasons why we think that we are not simply looking at different disease stages. First, there is no obvious chronologic line in the clusters (i.e. you cannot simply sort them from less to more advanced OA phase). Second, all patients have had their first complaints within 6 months of the first visit, meaning they are clinically all in the same phase of the disease at baseline, i.e. have the same starting point. Although this starting point is only based on clinical outcome, and not on radiographic features, it is very relevant in clinical practice. Further, the baseline KL grade and most tibiofemoral parameters were comparable and not significantly different between the found clusters/phenotypes (Table S4), so they likely did not start at very different structural stages either. Even if we derived trajectories represent different disease phases, groups would (at least partly) represent patients that progress more or less quickly through different phases, which may be regarded in itself a phenotype. We have added these considerations in the Discussion section of the revised manuscript [line 291-300].

ABSTRACT

- Reviewer comment:

Design: Please give a short hint (maybe in brackets) how early OA is defined

- Author response:

Early OA was defined as at or within 6 months of first visit to the general practitioner for these symptoms. We have added this to the Abstract of the revised manuscript [line 27-28].

- Reviewer comment:

How is the sentence: Combining….and radiographic (PF) parameters….

- Author response:

This sentence has been reworded to clarify our meaning [line 37-40].

- Reviewer comment:

It is unclear from my perspective whether the six clusters are only based on PF radiographic OA, and why TF radiographic parameters are not included.

- Author response:

This is because, when adding TF radiographic parameters, they were found to not significantly contribute to the clustering. We have clarified this in the Abstract of the revised manuscript [line 38-41].

- Reviewer comment:

When you say “gender”, I think this should be termed “sex”, since you are talking about biological and not social attributes. This comment also applies to the main text.

- Author response:

Yes this is correct, it should be termed ‘sex’, and have changed ‘gender’ to ‘sex’ throughout the entire manuscript.

- Reviewer comment:

When including baseline parameters, does none of the radiographic features contributes to any of the clusters ?

- Author response:

‘Baseline characteristics’ would be a better way to describe this. We do not mean baseline parameters of WOMAC/TF/PF features, as these are already incorporated in the trajectories, and the subjects’ probabilities of belonging to each trajectory was used for clustering. The baseline characteristics that we used were: sex, menopausal status, age, BMI, biochemical markers in blood and urine, and pain-coping inventory (PCI) measures (described in the Methods section [line 136-139]). We have changed ‘baseline parameters’ to ‘baseline characteristics’ in the Abstract of the revised manuscript [line 41-43].

- Reviewer comment:

Conclusions: Although several…., they were shown…. Why are the first and second part of the sentence contradictions. The sentence is pretty general and could be a bit more specific.

- Author response:

These parts are indeed not clear contradictions. We have rewritten this sentence and specified it more as suggested in the Abstract of the revised manuscript [line 44-46].

- Reviewer comment:

- Author response:

Indeed TF features did not contribute to the clustering, we did not make this clear in the abstract but have now done so in the Abstract of the revised manuscript in response to your earlier comment [line 40-41]. PF features did contribute, but their contribution was minor compared to sex, menopausal status, and WOMAC. We clarified this in the Abstract of the revised manuscript [line 41-43 and 46-48].

INTRODUCTION

- Reviewer comment:

P9, line 50: Please omit “already”

- Author response:

We have omitted ‘already’ from the revised manuscript [line 54].

- Reviewer comment:

P9, line 56: What is meant by “biochemical joint changes”. Please name the tissue you are referring to.

- Author response:

We have specified this in the revised manuscript by adding ‘in multiple tissues including bone, cartilage and synovium’ [line 61].

- Reviewer comment:

P9, line 59/60: … treatment….are available: Either “treatments” or “is available”

- Author response:

Thank you for noticing this, we have changed ‘treatment’ to ‘treatments in the revised manuscript [line 64].

- Reviewer comment:

P9, line 61: I suggest to say: disease “that consists of different phenotypes”

- Author response:

We have made the suggested change in the revised manuscript [line 66].

- Reviewer comment:

P9, line 64: suggest to add: “coalesce over time to a final common pathway. (delete “in patients”)

- Author response:

We have made the suggested change in the revised manuscript [line 69].

- Reviewer comment:

P10, line76: Please state in the introduction how you define “early” OA, as this term is very fashionable and it appears every researcher understands something different when using this term.

- Author response:

We have included this definition (first general practitioner visit ≤6 months before inclusion) in the Introduction of the revised manuscript as suggested [line 80-82].

- Reviewer comment:

P 10, line 78: Please omit “for the first time”

- Author response:

We have made the suggested change in the revised manuscript [line 84].

METHODS

- Reviewer comment:

- Author response:

Unfortunately there was no official diagnosis of whether patients had hip OA (or knee OA) or not, which makes it difficult to consistently control for hip OA. The fact that there is no significant difference in hip pain or stiffness between the clusters would indicate that hip OA likely did not have a significant influence on the clustering, but it could still have had an influence. For example, the WOMAC, although knee-specific, could have been influenced by patients having hip OA complaints as well. This consideration has been added to the Discussion section of the revised manuscript [line 360-362].

- Reviewer comment:

P10, lines 90/91: How can the patients have “severe complaints at baseline (most patients”, if this is a cohort of “early” disease?

- Author response:

More severe in this case is defined relative with this population of early OA patients. Indeed the criteria patients need to satisfy are still very mild: having 2 of pain, morning stiffness <30 min, crepitus and bony tenderness for the knee and/or 2 of pain, morning stiffness <60min pain on hip internal rotation or internal rotation <15°. We have now specified this in the revised manuscript [line 97-98].

- Reviewer comment:

P11, lines 95/96: Did you use WOMAC scores from one knee only (which one), or from both knees, as did the OAI. How did you account for contralateral knee (and hip) functional status in your analyses?

- Author response:

The WOMAC questionnaire was filled out on subject level instead of on knee level. The trajectory analyses were therefore also performed on subject level for the WOMAC, while they were performed on knee level for the radiographic features (see S1 Table). We have included this in the limitations [line 360-362].

- Reviewer comment:

- Author response:

We used JSW because it is a continuous measure and as such more sensitive to change, allowing us to identify trajectories more specifically. For PF features we used JSN, as continuous JSW was not available.

- Reviewer comment:

P 11, line 98: Why did you use osteophyte “area” rather than the common semi-quantitative grading system?

- Author response:

Similar to the JSW, we used a continuous measure for osteophyte size as it is more sensitive to change, allowing us to identify trajectories more specifically. For PF features there was no continuous measure for osteophytes available, so we used semi-quantitative scores.

- Reviewer comment:

- Author response:

For the clustering, only the trajectory probabilities were included, so in this sense the direct influence of parameters being measured continuously or semi-quantitatively is limited. However, it may indeed have influenced the identification of trajectories themselves. For example part of the TF trajectories could be inter-subject variation, meaning the found trajectories are not a full reflection of the actual progression, or trajectories might have been missed using grading scores. This is a good point and we have added this to the Discussion of the revised manuscript [line 279-283].

- Reviewer comment:

P 11, line 106: Thereafter…. Is this then referring only to the group included. This is a bit confusing.

- Author response:

Indeed this is only referring to the included records, we clarified this in the revised manuscript [line 121].

- Reviewer comment:

P 11, line 111: domain “experts”. Is this word used correctly here. May be due to my unfamiliarity with this particular method.

- Author response:

‘Experts’ here merely refers to graphical evaluation and exploration of population differences in clusters by subject matter experts, it is not a statistical term. We understand the confusion and have rephrased this in the revised manuscript [line 124].

- Reviewer comment:

P 12, lines 115-117: Can you give the non-specialist reader a sense why the non-functional data is possibly relevant to define phenotypes (and not the functional analysis)?

- Author response:

Both the non-functional and functional data are important to define phenotypes and both were used. The only difference is that it is not possible to use both types of data in one analysis, so we first performed the functional data analysis and used the probabilities of belonging to the identified trajectories as a way to use both types of data in the same analysis for defining subgroups. We have clarified this (also based on the comment below on explaining better what the various methods do conceptionally) in the revised manuscript [line 128-134].

- Reviewer comment:

P12, lines 120-123: It appears that the PF and TP radiographic features were not included at this level; why not?

- Author response:

PF and TF radiographic features were included at this level. We should have clarified this better and did so in the revised manuscript [line 135].

- Reviewer comment:

P12, lines 115 ff: Could you try to give the reader a sense of what these various methods do conceptionally, as not every reader may be familiar with theses statistical methods.

- Author response:

Functional data analysis allows for division/clustering of subjects based on the course over time of multiple separate characteristics. To define possibly relevant phenotypes, patients should be clustered based on both their characteristics (non-functional data) and their (clinical and structural) progression over time (functional data), but this is not (directly) possible in a single analysis. To combine the functional data and non-functional data we firsts ‘reduced’ the functional data to a probability of membership of a subject/knee to belong to a specific trajectory group. Expressed in this way it can be regarded a characteristic of a patient/knee and can be combined with the non-functional data (i.e. subject characteristics possibly relevant to define phenotypes) in a hierarchical cluster analysis to define subgroups. We have included this explanation in the revised manuscript [line 106-108 and 128-134].

RESULTS

- Reviewer comment:

P13, lines 138-141: Why do you think specifically the TF JSW and PF JSN differ between included and excluded participants, but none of the other parameters?

- Author response:

We think this is because all patients that were not part of the more ‘severe’ group were not followed yearly, and thus missed 3 or more visits and were excluded. As such, it makes sense that the excluded patients have (slightly) less severe OA, although this was not the case when comparing WOMAC. We highlight this in the Discussion section of the revised manuscript [line 345-350].

- Reviewer comment:

P13, lines 141 ff: it was decided…. based on which criteria

- Author response:

This was because no clear patterns could be obtained using all features in combined analyses. We reworded this part in the revised manuscript to clarify [line 157-161].

- Reviewer comment:

- Author response:

Unfortunately there was no WOMAC on knee level available in the CHECK cohort. The WOMAC trajectory analysis was done on patient level and not on knee level for this reason, but in the cluster analysis this could indeed have confounded the scores. We have included this in the limitations in the Discussion section of the revised manuscript [line 360-362].

- Reviewer comment:

P13, lines 160 ff: How did you define medial knee OA in the absence of radiographic JSN scores?

- Author response:

We meant that the fact that all found trajectories shows decreasing medial JSW and increasing lateral JSW suggests patients are primarily affected on the medial side, since it could be expected that the more affected compartment decreases over a period of 10 years (and the less affected compartment might increase as a result of wedging). However we cannot say for certain, since we did not officially determine the most affected compartment, and have added this in the revised manuscript [line 178].

- Reviewer comment:

P14, line 184: Adding WOMAC…. Adding to what ?

- Author response:

‘Adding’ was not chosen well, we meant to say that we performed Hierarchical Cluster Analysis with the WOMAC trajectories. We have reworded this sentence in the revised manuscript [line 201-202].

- Reviewer comment:

P15, lines 190 ff: I wonder whether the lack of relevance of TF OA is due to the lack of use of JSN scores, as JSW is not a measure of pathology in a cross sectional context?

- Author response:

Yes this is a good point in line with an earlier comment on the difference between semi-quantitative and continuous measures. We have added this to the Discussion of the revised manuscript [line 279-283].

- Reviewer comment:

Can you provide results on what the JSW change was over 10 years in the medial and in the lateral compartment, respectively?

- Author response:

The JSW change over time in the medial and lateral compartment can be seen in Fig 1. (In numbers, the lateral compartment showed an increase of 1.15 mm (95%CI 1.07-1.23) and the medial compartment a decrease of 0.45 mm (95%CI -0.49 – -0.41)).

- Reviewer comment:

P15, line 198 : Please do not say “unfortunate”, the presentation of results should be neural

- Author response:

We removed ‘unfortunately’ from the results of the revised manuscript [line 216].

- Reviewer comment:

- Author response:

Sex and menopause were one variable categorized as men, premenopausal women, postmenopausal women (see line 218). WOMAC and PF OA data did not become irrelevant when adding sex and menopause, we meant that the other added baseline characteristics became irrelevant. We have clarified this in the revised manuscript [line 219].

- Reviewer comment:

P15, lines 206ff: In my view, these phenotypes appear to represent different stages of the disease.

- Author response:

- Reviewer comment:

- Author response:

We understand that listing the different clusters is a bit difficult to follow. A Table cross-tabling all different features and their contribution would be difficult to execute, since there are no clear cut-off points when features do and do not contribute and the grouping/naming was performed on clinical interpretation as well. However, we have clarified the legends of Fig 4 and Fig 5 to more clearly present the different clusters, and hope this makes the listing of the clusters easier to follow [line 236-240 and 253-257].

DISCUSSION:

- Reviewer comment:

- Author response:

In the trajectory analyses we never look at cross-sectional distributions. We realize that the word ‘absolute’ here is unclear, we meant that trajectories could be classified based on absolute levels over time (e.g. constant high level or low level over 10 years) and the course over time. We have clarified this in the Discussion of the revised manuscript [line 262].

- Reviewer comment:

P 17: When talking about TF and PF OA here, please add “radiographic” to the description.

- Author response:

We have added ‘radiographic’ to ‘TF OA characteristics’ and ‘PF OA characteristics’ the first time they are mentioned in the Discussion section [line 273-275].

- Reviewer comment:

P 18, lines 268/269: How many of your participants had medial vs. lateral TF radiographic OA?

- Author response:

We do not know, as this was not officially determined. We have added this to the discussion of the revised manuscript [line 313-315].

- Reviewer comment:

P 19, line 291: Why do the authors think the results are so discordant between studies, given that they have examined a relatively large cohort.

- Author response:

The one clear difference between our study and previous studies is that previous studies showed significant differences based on BMI and we did not. We think this could be because the other studies clustered based only on pain, while we looked at other factors as well, and BMI is known to be linked to pain. We have included this in the Discussion section [line 338-340].

- Reviewer comment:

P 19, line 293: Instead or “some missing data was present”, please say “ some data was missing”.

- Author response:

We have made the suggested change in the revised manuscript [line 342].

- Reviewer comment:

P 19, lines 308 ff: Shouldn’t the Osteoarthritis Initiative (OAI) data be used for validation?

- Author response:

Yes it is a good suggestion to add this and we have done so in the revised manuscript [line 363-364].

- Reviewer comment:

P 20, conclusions: There is more text about what was done than about what was found. Please focus more on concrete and specific findings in the conclusion section (also in the abstract)

- Author response:

We have changed the conclusion of the abstract [line 44-49] and the manuscript [line 373-382] to focus more on our findings.

- Reviewer comment:

- Author response:

We removed ‘for the first time’ from the revised manuscript as suggested [line 381].

Attachment

Submitted filename: Response to reviewers v2.docx

Click here for additional data file.^{(34KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0283717.r003

Decision Letter 1

Aqeel M Alenazi

15 Mar 2023

Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohort

PONE-D-22-19232R1

Dear Dr. Jansen,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Aqeel M Alenazi

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: I Don't Know

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #2: Yes

**********

6. Review Comments to the Author

Reviewer #2: Thank you for the detailed response and author action to the comments made.

Thank you for the detailed response and author action to the comments made.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

**********

PLoS One. doi: 10.1371/journal.pone.0283717.r004

Acceptance letter

Aqeel M Alenazi

24 Mar 2023

PONE-D-22-19232R1

Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohort

Dear Dr. Jansen:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Aqeel M Alenazi

Academic Editor

PLOS ONE

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 Table. Final list of variables included in the functional analysis.

*We removed records of knees with missing data for three or more time points and used linear interpolation to input remaining missing values.

(DOCX)

Click here for additional data file.^{(13.1KB, docx)}

S2 Table. Descriptive statistics of combined scenario with WOMAC and PF OA features.

(DOCX)

Click here for additional data file.^{(27.4KB, docx)}

S3 Table. Descriptive statistics of combined scenario with WOMAC, PF OA features and sex.

(DOCX)

Click here for additional data file.^{(30.2KB, docx)}

Attachment

Submitted filename: Response to reviewers v2.docx

Click here for additional data file.^{(34KB, docx)}

Data Availability Statement

[pone.0283717.ref001] 1.Murphy SL, Lyden AK, Phillips K, Clauw DJ, Williams DA. Subgroups of older adults with osteoarthritis based upon differing comorbid symptom presentations and potential underlying pain mechanisms. Arthritis Res Ther. 2011;13: R135. doi: 10.1186/ar3449 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref002] 2.Palazzo C, Nguyen C, Lefevre-Colau MM, Rannou F, Poiraudeau S. Risk factors and burden of osteoarthritis. Ann Phys Rehabil Med. 2016;59: 134–138. doi: 10.1016/j.rehab.2016.01.006 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref003] 3.Safiri S, Kolahi AA, Smith E, Hill C, Bettampadi D, Mansournia MA, et al. Global, regional and national burden of osteoarthritis 1990–2017: A systematic analysis of the Global Burden of Disease Study 2017. Ann Rheum Dis. 2020;79. doi: 10.1136/annrheumdis-2019-216515 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref004] 4.Turkiewicz A, Petersson IF, Björk J, Hawker G, Dahlberg LE, Lohmander LS, et al. Current and future impact of osteoarthritis on health care: A population-based study with projections to year 2032. Osteoarthr Cartil. 2014;22: 1826–1832. doi: 10.1016/j.joca.2014.07.015 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref005] 5.Losina E, Weinstein AM, Reichmann WM, Burbine SA, Solomon DH, Daigle ME, et al. Lifetime risk and age at diagnosis of symptomatic knee osteoarthritis in the US. Arthritis Care Res. 2013;65: 703–711. doi: 10.1002/acr.21898 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref006] 6.Karsdal MA, Michaelis M, Ladel C, Siebuhr AS, Bihlet AR, Andersen JR, et al. Disease-modifying treatments for osteoarthritis (DMOADs) of the knee and hip: lessons learned from failures and opportunities for the future. Osteoarthr Cartil. 2016;24: 2013–2021. doi: 10.1016/j.joca.2016.07.017 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref007] 7.Deveza LA, Nelson AE, Loeser RF. Phenotypes of osteoarthritis: current state and future implications. Clin Exp Rheumatol. 2019;37: 64–72. Available: /pmc/articles/PMC6936212/ doi: 10.1016/j.joca.2019.06.011 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref008] 8.Driban JB, Sitler MR, Barbe MF, Balasubramanian E. Is osteoarthritis a heterogeneous disease that can be stratified into subsets? Clin Rheumatol. 2010;29: 123–131. doi: 10.1007/s10067-009-1301-1 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref009] 9.Van Spil WE, Bierma-Zeinstra SMA, Deveza LA, Arden NK, Bay-Jensen AC, Kraus VB, et al. A consensus-based framework for conducting and reporting osteoarthritis phenotype research. Arthritis Res Ther. 2020;22: 54. doi: 10.1186/s13075-020-2143-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref010] 10.Van Spil WE, Jansen NWD, Bijlsma JWJ, Reijman M, DeGroot J, Welsing PMJ, et al. Clusters within a wide spectrum of biochemical markers for osteoarthritis: Data from CHECK, a large cohort of individuals with very early symptomatic osteoarthritis. Osteoarthr Cartil. 2012;20: 745–754. doi: 10.1016/j.joca.2012.04.004 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref011] 11.Wesseling J, Boers M, Viergever MA, Hilberdink WK, Lafeber FP, Dekker J, et al. Cohort profile: Cohort Hip and Cohort Knee (CHECK) study. Int J Epidemiol. 2016;45: 36–44. doi: 10.1093/ije/dyu177 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref012] 12.Bijlsma JWJ, Wesseling J. Thematic collection: CHECK (Cohort Hip & Cohort Knee). DANS. 2015. 10.17026/dans-252-qw2n [DOI] [Google Scholar]

[pone.0283717.ref013] 13.Bellamy N, Buchanan W, Goldsmith C, Campbell J, Stitt L. Validation study of WOMAC: a health status instrument for measuring clinically important patient relevant outcomes to antirheumatic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol. 1988. pp. 1883–40. [PubMed] [Google Scholar]

[pone.0283717.ref014] 14.Marijnissen ACA, Vincken KL, Vos PAJM, Saris DBF, Viergever MA, Bijlsma JWJ, et al. Knee Images Digital Analysis (KIDA): a novel method to quantify individual radiographic features of knee osteoarthritis in detail. Osteoarthr Cartil. 2008;16: 234–243. doi: 10.1016/J.JOCA.2007.06.009 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref015] 15.Jansen MP, Welsing PMJ, Vincken KL, Mastbergen SC. Performance of knee image digital analysis of radiographs of patients with end-stage knee osteoarthritis. Osteoarthr Cartil. 2021;29: 1530–1539. doi: 10.1016/j.joca.2021.07.013 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref016] 16.Burnett SJHD, Hart DJ, Cooper C, Spector TD. A radiographic atlas of osteoarthritis. 1994. [Google Scholar]

[pone.0283717.ref017] 17.Schmutz A, Jacques J, Bouveyron C. funHDDC: Univariate and Multivariate Model-Based Clustering in Group-Specific Functional Subspaces. R package version 2.3.0. 2021. [cited 6 Oct 2021]. Available: https://cran.r-project.org/package=funHDDC [Google Scholar]

[pone.0283717.ref018] 18.Ramsay JO, Graves S, Hooker G. fda: Functional Data Analysis. R package version 2.4.8. 2021. https://CRAN.R-project.org/package=fda [Google Scholar]

[pone.0283717.ref019] 19.Core Team R. R: A language and environment for statistical computing. 2013. [Google Scholar]

[pone.0283717.ref020] 20.Schmutz A, Jacques J, Bouveyron C, Chèze L, Martin P. Clustering multivariate functional data in group-specific functional subspaces. Comput Stat. 2020;35: 1101–1131. doi: 10.1007/s00180-020-00958-4 [DOI] [Google Scholar]

[pone.0283717.ref021] 21.Schwarz G. Estimating the Dimension of a Model. Ann Stat. 2007;6: 461–464. doi: 10.1214/aos/1176344136 [DOI] [Google Scholar]

[pone.0283717.ref022] 22.James G, Witten D, Hastie T, Tibshirani R. An introduction to Statistical Learning. Current medicinal chemistry. 2013. doi: 10.1007/978-1-4614-7138-7 [DOI] [Google Scholar]

[pone.0283717.ref023] 23.Kraaimaat FW, Evers AWM. Pain-Coping Strategies in Chronic Pain Patients: Psychometric Characteristics of the Pain-Coping Inventory (PCI). Int J Behav Med. 2003;10: 343–363. doi: 10.1207/s15327558ijbm1004_5 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref024] 24.Galili T, O’Callaghan A, Sidi J, Sievert C. Heatmaply: An R package for creating interactive cluster heatmaps for online publishing. Bioinformatics. 2018;34: 1600–1602. doi: 10.1093/bioinformatics/btx657 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref025] 25.Galili T. dendextend: An R package for visualizing, adjusting and comparing trees of hierarchical clustering. Bioinformatics. 2015;31: 3718–3720. doi: 10.1093/bioinformatics/btv428 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref026] 26.Kaufman L, Rousseeuw PJ. Finding Groups in Data: An Introduction to Cluster Analysis. 1990. doi: 10.1002/9780470316801 [DOI] [Google Scholar]

[pone.0283717.ref027] 27.Son KM, Hong JI, Kim D-H, Jang D-G, Crema MD, Kim HA. Absence of pain in subjects with advanced radiographic knee osteoarthritis. BMC Musculoskelet Disord 2020 211. 2020;21: 1–9. doi: 10.1186/s12891-020-03647-x [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref028] 28.Creamer P, Lethbridge‐Cejku M, Hochberg MC. Factors associated with functional impairment in symptomatic knee osteoarthritis. Rheumatology. 2000;39: 490–496. doi: 10.1093/rheumatology/39.5.490 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref029] 29.Bedson J, Croft PR. The discordance between clinical and radiographic knee osteoarthritis: A systematic search and summary of the literature. BMC Musculoskelet Disord 2008 91. 2008;9: 1–11. doi: 10.1186/1471-2474-9-116 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref030] 30.Eijkenboom JFA, Waarsing JH, Oei EHG, Bierma-Zeinstra SMA, Van Middelkoop M. Is patellofemoral pain a precursor to osteoarthritis? Bone Jt Res. 2018;7: 541–547. doi: 10.1302/2046-3758.79.BJR-2018-0112.R1 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref031] 31.Deveza LA, Melo L, Yamato TP, Mills K, Ravi V, Hunter DJ. Knee osteoarthritis phenotypes and their relevance for outcomes: a systematic review. Osteoarthr Cartil. 2017;25: 1926–1941. doi: 10.1016/J.JOCA.2017.08.009 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref032] 32.Bastick AN, Wesseling J, Damen J, Verkleij SP, Emans PJ, Bindels PJ, 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: e32. doi: 10.3399/bjgp15X688129 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref033] 33.Wesseling J, Bastick AN, Ten Wolde S, Kloppenburg M, Lafeber FPJG, Bierma-Zeinstra SMA, et al. Identifying trajectories of pain severity in early symptomatic knee osteoarthritis: A 5-year followup of the cohort hip and cohort knee (CHECK) study. J Rheumatol. 2015;42: 1470–1477. doi: 10.3899/jrheum.141036 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref034] 34.Holla JFM, Leeden M van der, Heymans MW, Roorda LD, Bierma-Zeinstra SMA, Boers M, et al. Three trajectories of activity limitations in early symptomatic knee osteoarthritis: a 5-year follow-up study. Ann Rheum Dis. 2014;73: 1369–1375. doi: 10.1136/annrheumdis-2012-202984 [DOI] [PubMed] [Google Scholar]

[pone.0283717.ref035] 35.Waarsing JH, Bierma-Zeinstra SMA, Weinans H. Distinct subtypes of knee osteoarthritis: data from the Osteoarthritis Initiative. Rheumatology. 2015;54: 1650–1658. doi: 10.1093/rheumatology/kev100 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0283717.ref036] 36.Pan F, Tian J, Cicuttini F, Jones G, Aitken D. Differentiating knee pain phenotypes in older adults: a prospective cohort study. Rheumatology. 2019;58: 274–283. doi: 10.1093/rheumatology/key299 [DOI] [PubMed] [Google Scholar]

PERMALINK

Identifying multivariate disease trajectories and potential phenotypes of early knee osteoarthritis in the CHECK cohort

Sara Altamirano

Mylène P Jansen

Daniel L Oberski

Marinus J C Eijkemans

Simon C Mastbergen

Floris P J G Lafeber

Willem E van Spil

Paco M J Welsing

Roles

Abstract

Objective

Design

Results

Conclusions

Introduction

Methods

Study population

Functional data analysis

Hierarchical cluster analysis

Results

Functional data analysis

Fig 1. Functional data analysis-based trajectories for WOMAC scores for pain, function, stiffness.

Fig 3. Functional data analysis-based trajectories for OA patellofemoral features: Sclerosis, joint space narrowing, osteophytes.

Fig 2. Functional data analysis-based trajectories for quantitative tibiofemoral OA features: Osteophytes, joint space width, bone density.

Hierarchical Cluster Analysis (HCA)

Fig 4. Clustering based on WOMAC features (pain, function, stiffness) and patellofemoral (PF) osteoarthritis without sex.

Fig 5. Clustering based on WOMAC features (pain, function, stiffness) and patellofemoral (PF) osteoarthritis with sex.

Discussion

Supporting information

Data Availability

Funding Statement

References

Decision Letter 0

Aqeel M Alenazi

Roles

Author response to Decision Letter 0

Decision Letter 1

Aqeel M Alenazi

Roles

Acceptance letter

Aqeel M Alenazi

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases