You are here

  • Home
  • Archive
  • Volume 9, Issue 4
  • Impact of disease outcomes on the Assessment of SpondyloArthritis International Society Health Index (ASAS HI): a Bayesian network analysis of the DESIR cohort

Article Text

Article menu

Download PDFPDF

Spondyloarthritis
Original research
Impact of disease outcomes on the Assessment of SpondyloArthritis International Society Health Index (ASAS HI): a Bayesian network analysis of the DESIR cohort
  1. Imke Redeker1,2,
  2. Robert Landewé3,4,
  3. Désirée van der Heijde5,
  4. Sofia Ramiro4,5,
  5. Annelies Boonen6,
  6. Maxime Dougados7,
  7. Jürgen Braun8 and
  8. Uta Kiltz1,2
  1. 1Ruhr-Universität Bochum, Bochum, Germany
  2. 2Rheumazentrum Ruhrgebiet, Herne, Germany
  3. 3Clinical Immunology & Rheumatology, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands
  4. 4Department of Rheumatology, Zuyderland Medical Center, Heerlen, The Netherlands
  5. 5Department of Rheumatology, Leiden University Medical Center, Leiden, The Netherlands
  6. 6Department of Internal Medicine, Division of Rheumatology, Maastricht University Medical Center, and the Caphri Research Institute Maastricht University, Maastricht, The Netherlands
  7. 7Hopital Cochin, Rheumatology, Université Paris Descartes Faculté de Médecine, Paris, France
  8. 8Ruhr University Bochum, Bochum, Germany
  1. Correspondence to Dr Uta Kiltz; uta.kiltz{at}rub.de

Abstract

Objective The objective of this study is to build a structural model visualising and quantifying the interrelationships of different disease outcomes with the Assessment of SpondyloArthritis International Society Health Index (ASAS HI) in patients with axial spondyloarthritis (axSpA).

Methods Cross-sectional data collected at month 72 of the Devenir des Spondylarthropathies Indifferénciées Récentes cohort was analysed. Combining prior knowledge and observed data, probabilistic Bayesian network modelling was used to study how the interplay of different disease outcomes affects the ASAS HI, which measures disease-specific overall functioning and health. Disease outcomes comprised, among others, the Ankylosing Spondylitis (AS) Disease Activity Score (ASDAS) and the Bath AS Functional Index (BASFI).

Results Data of 384 patients were analysed. The obtained structure suggests that ASAS HI is determined by both patient-reported physical function (BASFI) and disease activity (ASDAS). The parameters of the structural model show that an increase of ASDAS or BASFI by 1 unit corresponds to an increase of ASAS HI by 0.70 or 1.25 units, respectively. Moreover, the model suggests that disease activity has an indirect impact on ASAS HI via BASFI. No relationship between spinal mobility or structural damage and ASAS HI was found.

Conclusions This is the first structural model developed to better understand the construct and the interplay between clinically relevant outcomes related to ASAS HI in axSpA patients. It shows that disease activity and physical function have a strong impact on ASAS HI, confirming it to be a valid construct of overall functioning and health in axSpA patients.

  • Spondylitis, Ankylosing
  • Patient Reported Outcome Measures
  • Epidemiology

Data availability statement

No data are available. No additional data are available.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

https://doi.org/10.1136/rmdopen-2023-003587

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    WHAT IS ALREADY KNOWN ON THIS TOPIC

    • The Assessment of SpondyloArthritis International Society Health Index (ASAS HI) is a frequently used assessment tool that evaluates overall functioning and health in patients with axial spondyloarthritis (axSpA). Health-related quality of life in axSpA is determined by several factors including disease activity and physical function.

    WHAT THIS STUDY ADDS

    • This is the first structural model visualising and quantifying the interplay between outcomes related to ASAS HI in patients with axSpA from a multidependent perspective.

    HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

    • The ASAS HI was confirmed to be a valid construct of overall functioning and health in patients with axSpA by showing that disease activity and physical function have a strong impact on ASAS HI. This data supports the use of the index as a primary outcome in clinical trials—as in a recent treat-to-target study.

    Introduction

    Patients with axial spondyloarthritis (axSpA) suffer from a variable disease course in which axial involvement, peripheral arthritis, enthesitis and involvement of other organs can occur and may influence the burden of the disease.1 Pain, stiffness and mobility limitations are the most prominent health concerns.2 3 As a consequence, patients face restrictions in working life and social participation.4 Current knowledge on the health status of patients with axSpA mainly focuses on physical function and disease activity. A health model for patients with radiographic axSpA (r-axSpA) proposed that health-related quality of life (HR-QoL) is determined by physical function and disease activity.5 The model has been built on data of patients with r-axSpA including patients with long-standing disease who had been included in the Ankylosing Spondylitis Study for the Evaluation of Recombinant Infliximab Therapy trial.6 Using the 36-Item Short Form (SF-36) health survey questionnaire as a generic measure for HR-QoL, a hierarchical relationship between spinal damage, spinal mobility, disease activity, physical function and HR-QoL has been demonstrated in patients with r-axSpA using traditional regression analyses.5 It is not known to what extent such a relationship holds true when a disease-specific measure for the HR-QoL is used and when patients with non-radiographic axSpA (nr-axSpA) are included.

    Overall functioning and health is a construct introduced in the International Classification of Functioning, Disability and Health and describes an interaction between body functions, body structures and thereof impairments in activities and social participation.7 Disease-specific overall functioning and health can be assessed in patients with axSpA using the Assessment of SpondyloArthritis International Society Health Index (ASAS HI), which encompasses impairments in physical and mental functions, limitations in activities and restrictions in social participation.8 The 17 dichotomous items of the ASAS HI address aspects of pain, emotional functions, sleep, sexual functions, mobility, self-care and community life, representing a wide spectrum of functioning, disability and health in patients with axSpA. The relationship between ASAS HI, disease activity, structural damage and socioeconomic factors is not well known. In fact, most available instruments referred to as HR-QoL actually assess functioning and health but not QoL.

    Overall functioning of patients with axSpA can be influenced by many different factors. The relationship between radiographic damage of the spine and spinal mobility impairment in r-axSpA has been unequivocally demonstrated at the group level. To our knowledge, few studies have examined the interrelationships between ASAS HI and disease outcomes in patients with axSpA.9 10 The Devenir des Spondylarthropathies Indifferénciées Récentes (DESIR) cohort provides an unselected population of patients with axSpA,11 including patients with r-and nr-axSpA. This wide variety of severity and disease activity is an ideal patient sample to assess the relationships of different disease outcomes with ASAS HI from a multidependent perspective. Analysis of observational data is traditionally focused on establishing associations resulting in a list of p values or effect sizes without describing the underlying causal structure among variables. Understanding the causal structure, however, provides a deeper comprehension of how one factor under consideration relates to other factors, which is essential to guide interventions.12 Bayesian networks have become a popular analysis approach to explore complex relationships between input variables.13

    This study aimed to analyse the interplay of different disease outcomes with ASAS HI in patients with axSpA by adopting a Bayesian network modelling approach.

    Patients and methods

    A cross-sectional analysis of data at month 72 of the DESIR cohort (ClinicalTrials.gov identifier: NCT01648907)11 was conducted. The DESIR cohort is a French national prospective multicentre cohort of 708 patients with early inflammatory back pain suggestive of axSpA. Patients were recruited between October 2007 and April 2010. Of those, patients with a definite diagnosis of axSpA at month 72, according to the treating rheumatologist, were selected. Among the axSpA patients, those with non-missing data for the outcome ASAS HI were included in the analysis. A patient inclusion flow chart is provided in online supplemental figure 1. A more detailed description of the DESIR cohort can be found elsewhere (Dougados et al11 and http://www.lacohortedesir.fr/).

    Data were retrieved from the visit performed 72 months after inclusion, which was the first time point of ASAS HI collection within DESIR, except for data on imaging, which were gathered from the visit performed 60 months after inclusion, which was the last time before ASAS HI collection where it was available.

    Participants gave written informed consent.

    Data collection

    The ASAS HI ranges from 0 to 17 with higher values indicating worse disease-specific overall functioning and health.8 Established cut-offs are ≤5 and ≥12 suggesting ‘good’ (ASAS HI≤5), ‘moderate’ (5<ASAS HI<12) and ‘bad’ (ASAS HI≥12) overall functioning and health.14

    Further information included in the analysis comprised disease activity (Ankylosing Spondylitis Disease Activity Score based on C reactive protein—ASDAS15), physical function (Bath Ankylosing Spondylitis Functional Index—BASFI16), spinal mobility (Bath Ankylosing Spondylitis Metrology Index—BASMI17), lateral spinal flexion (a component of BASMI) and structural damage (modified Stoke Ankylosing Spondylitis Spine Score—mSASSS18). These variables were chosen to study a model comparable to the health model for patients with r-axSpA5 while using the disease-specific outcome measure ASAS HI instead of the generic outcome measure SF-36 and including both nr-axSpA and r-axSpA patients. The previously developed health model further included spinal inflammation which could not be included in the current analysis due to a large amount of missing data for that variable (by protocol, at 60 months, MRIs were only performed in participating centres in Paris, which explains the missing data).

    Statistical analysis

    First, for any pair of the five disease-specific variables ASDAS, BASFI, BASMI, mSASSS and ASAS HI the Spearman’s rank correlation coefficient, which assesses a monotonic relationship that does not need to be linear, was computed.

    Second, a Bayesian network analysis approach was used to gain insight into the interplay between ASDAS, BASFI, BASMI, mSASSS and ASAS HI based on their conditional dependencies and independencies. In an additional analysis, the (total) BASMI was replaced by its component lateral spinal flexion, which may be a better reflection of spinal mobility in early disease.19

    Bayesian networks are probabilistic graphical models that consist of two components: a directed acyclic graph (DAG) represented by nodes (random variables) joined by directed edges (the directional relationships between variables) and an accompanying joint probability distribution. Specifically, in a causally interpreted graph, the presence of a directed edge between two nodes A and B indicates a causal link between A and B in the direction of the edge. Conversely, the absence of an edge between A and B indicates that there is no causal effect, which is generally regarded as a stronger assumption.12

    Unlike path models in structural equation modelling, DAGs are entirely non-parametric, enabling the representation of any functional relationship between the variables.12

    One advantage of Bayesian networks is their ability to capture complex relationships between variables. Another is their ability to encode expert knowledge (prior knowledge about relationships between variables from previous studies or scientific theory) by imposing constraints on Bayesian network structures. This aids the algorithm in the correct orientation of edges and may improve power. A well-known example of such a constraint is that no genetic variable can be influenced by a clinical one. Bayesian networks that are well-supported by a given set of data can be used to identify possible causal relationships and have been argued to be competitive with, and in some cases advantageous to, other causal inference methods.20 21

    In this study, the structure of the network was learnt using a modern implementation of the constraint-based Peter-Clark (PC) algorithm,22 which estimates the DAG by conducting a series of conditional independence tests among the variables in the data. In a two-step process it first identifies the undirected edges in the graph (this step is often referred to as ‘skeleton identification’ by starting off with a complete undirected graph, where there is an undirected edge between any pair of variables, and then removing edges if the involved variables are conditionally independent based on the data.12 In the second step, the directions of the edges are determined following specific rules.23 In our study, variables of the network represented by the nodes were continuous and chosen to be modelled with a Gaussian distribution. The following constraints were imposed to aid the algorithm: (1) mSASSS does not depend on BASMI, BASFI or ASAS HI (ie, we assume that these three variables have no effect on structural damage) and (2) mSASSS does not influence ASDAS (ie, we assume that structural damage has no effect on disease activity). Despite of these constraints, it remains challenging to identify the true structure from limited data which gives reason to ask how confident one can be about the existence of an edge identified by the algorithm. To examine the robustness of edges against perturbation of the data, a non-parametric bootstrap to network learning was applied.24 Specifically, the data were resampled 10 000 times using random sampling with replacement and for each of the 10 000 samples, a Bayesian network was learnt. To be included in the final model, an edge (irrespective of its direction) has to appear in more than 90% of the 10 000 bootstrapped networks (strength) and the relative frequency of a directed edge has to be at least 50% (direction).

    After having obtained the final network model, its parameters were fitted using their maximum likelihood estimate. These correspond to the coefficients of a classic linear regression for each (child) node against its parents (nodes with edges directed into the child node). 95% CIs for the regression coefficients were calculated.

    The network was estimated based on complete data under the assumption that the continuous variables follow a Gaussian distribution and imposed constraints must be satisfied. In a sensitivity analysis, the underlying skeleton was identified using an extension of the PC algorithm to non-paranormal data with missing values and no imposed constraints.25

    Data analyses were performed with R V.4.1.2 and additional packages, including bnlearn (V.4.7.1).

    Results

    Characteristics of the study population

    A total of 384 patients from the DESIR cohort with a definite clinical diagnosis of axSpA at month 72, according to the treating rheumatologist, had non-missing data for ASAS HI and were included in the analysis. Table 1 shows these patients’ characteristics and disease outcomes. A total of 46% were male and, 72 months after inclusion, the mean age was 40.3 years with a mean symptom duration of 7.5 years. Patients had, on average, rather low scores for disease activity (ASDAS 2.0), impairment of physical function (BASFI 2.3), impairment of spinal mobility (BASMI 2.4) and structural damage (mSASSS 1.0).

    View this table:
    Table 1

    Patient characteristics and disease outcomes at month 72 if not stated otherwise

    The mean ASAS HI was 5.8 (range: 0–16). Applying existing cut-offs for ASAS HI, 50% had ‘good’ overall functioning (ASAS HI≤5), 40% had ‘moderate’ overall functioning (5<ASAS HI <12) and 10% had ‘bad’ overall functioning (ASAS HI≥12).

    Almost half of the patients (49%) received treatment with tumour necrosis factor inhibitors, about two-thirds (66%) received non-steroidal anti-inflammatory drugs and one-fifth received conventional disease-modifying antirheumatic drugs in the last 12 months.

    Correlation analysis

    Figure 1 shows Spearman’s rank correlation coefficient for any pair of ASDAS, BASFI, BASMI, mSASSS and ASAS HI.

    Figure 1
    Figure 1

    Spearman’s rank correlation coefficient for different disease outcomes in patients with axSpA. axSpA, axial spondyloarthritis; ASAS HI, Assessment of SpondyloArthritis International Society Health Index; ASDAS, Ankylosing Spondylitis Disease Activity Score based on C reactive protein; BASFI, Bath Ankylosing Spondylitis Functional Index; BASMI, Bath Ankylosing Spondylitis Metrology Index; mSASSS, modified Stoke Ankylosing Spondylitis Spine Score.

    The correlation coefficient between BASFI and ASAS HI was highest (0.83), followed by the correlation coefficient between BASFI and ASDAS (0.71) and ASDAS and ASAS HI (0.64), indicating a monotonic relationship between these variables. A weak relationship with mSASSS was found for BASMI (0.24), while other variables had very weak correlation coefficients of at most 0.1 with mSASSS (figure 1).

    Scatter plots and histograms of ASDAS, BASFI, BASMI, mSASSS and ASAS HI are given in online supplemental figure 2.

    Bayesian network analysis

    The structural model that was learnt from combining data and prior expert knowledge (by imposing constraints) is shown in figure 2. It resulted from 10 000 bootstrapped networks, of which the relative frequency with which an edge appeared irrespective of its direction (strength) and the relative frequency of a directed edge (direction) are shown in table 2.

    View this table:
    Table 2

    Relative frequency with which an edge appeared irrespective of its direction (strength) and relative frequency of a directed edge (direction) resulting from non-parametric bootstrap

    Figure 2
    Figure 2

    Structural model on inter-relationships of different disease outcomes with overall functioning and health (ASAS HI) in patients with axSpA. axSpA, axial spondyloarthritis; ASAS HI, Assessment of SpondyloArthritis International Society Health Index; ASDAS, Ankylosing Spondylitis Disease Activity Score based on C reactive protein; BASFI, Bath Ankylosing Spondylitis Functional Index; BASMI, Bath Ankylosing Spondylitis Metrology Index; mSASSS, modified Stoke Ankylosing Spondylitis Spine Score.

    The structural model visualises the relationships between disease-specific variables from a multidependent perspective. It suggests that both disease activity (ASDAS) and physical function (BASFI) have a direct impact on ASAS HI. Moreover, the model indicates that disease activity has an indirect impact on ASAS HI via physical function, which in turn has an impact on spinal mobility (BASMI). In addition to physical function, spinal mobility is determined by structural damage according to the learnt model. On top of that, the model denies a relationship between spinal mobility or structural damage and ASAS HI.

    The fitted parameters of the structural model are given in table 3, according to which an increase of ASDAS by 1 unit (given BASFI) or of BASFI by 1 unit (given ASDAS) corresponds to an increase of ASAS HI by 0.70 units (95% CI 0.30 to 1.11) or 1.25 units (95% CI 1.04 to 1.40), respectively. Furthermore, the model suggests that an increase of ASDAS by 1 unit corresponds to an increase of BASFI by 1.56 units (95% CI 1.39 to 1.78) and that BASMI increases by 0.22 units (95% CI 0.18 to 0.27) when BASFI is increased by 1 unit (given mSASSS) or that BASMI increases by 0.09 units (95% CI 0.07 to 0.12) when mSASSS is increased by 1 unit (given BASFI).

    View this table:
    Table 3

    Parameters of the structural model corresponding to coefficients of linear regression analyses

    The additional analysis with lateral spinal flexion instead of the total BASMI led to similar results (the structural model and its fitted parameters are given in online supplemental figure 3 and table 1, respectively). Moreover, the same underlying skeleton was found in the sensitivity analysis.

    Discussion

    To the best of our knowledge, this is the first large-scale study that used a Bayesian network analysis approach to investigate how the interplay of different disease outcomes affects ASAS HI in patients with axSpA. Our model shows that ASAS HI is determined both by patient-reported physical function (BASFI) and by disease activity (ASDAS).

    Our model agrees with the stratified model proposed by Machado et al5 regarding the association of overall functioning and health with disease activity and physical function, which was also found in previous studies on ASAS HI.9 10 26 The health status in the former study was assessed using the mental and physical component scores of the SF-36 health survey questionnaire. Our study design differs in several ways: First, we used the ASAS HI, which is a disease-specific patient-reported outcome assessing overall functioning and health in patients with axSpA. Second, we adopted a Bayesian network analysis approach combining prior knowledge with measured data to explore and visualise the interrelationships among variables. Third, patients with r-axSpA and nr-axSpA were included in the analysis. Despite the different study design, a further agreement with the stratified model consists of the association of physical function and disease activity. By contrast, our model diverts from the stratified model by disregarding a relationship between spinal mobility or structural damage and ASAS HI. One reason might be the low proportion of patients with spinal mobility impairment or structural damage in our study possibly due to the fact that DESIR is an inception cohort. Nevertheless, in two recent studies, no association between ASAS HI and structural damage was found,9 10 while a third recent study found no association between health status and structural damage but an association between health status and spinal mobility.26 In the latter study, health status was assessed using the Ankylosing Spondylitis Health Assessment Questionnaire.

    Moreover, in contrast to the stratified model, our model suggests that physical function determines spinal mobility. The strength of the edge between BASFI and BASMI exceeded 0.99, which implies that in nearly all of the 10 000 bootstrapped networks, an edge connecting BASFI and BASMI was present. Furthermore, this edge was more often directed from BASFI to BASMI rather than from BASMI to BASFI. Consequently, in the final consensus network, the edge from BASFI to BASMI was incorporated. However, from a clinical perspective, the directionality of this edge is not readily apparent and warrants further investigation to determine if it is potentially misoriented. For this, it may be helpful to include MRI spinal inflammation, which has been shown to contribute to the impairment of spinal mobility in patients with r-axSpA27 but could not be included in the present analysis due to a large amount of missing data (according to protocol, MRI was only performed in all patients at baseline, afterwards only in a small number of centres was MRI performed in all visits). The availability of MRI could also have shed light on the relationship between MRI spinal inflammation and ASDAS, BASFI and ASAS HI.

    The main strength of this study is that, instead of focusing on establishing associations resulting in a list of p values, it provides insight into the causal structure, which is necessary for the identification of causal effects using observational data.28 Furthermore, understanding the causal structure among variables is advantageous for guiding confounder selection since statistical adjustment for the wrong set of covariates may increase confounding bias.12 29 Moreover, our findings support scientific evidence of the association of ASAS HI with disease activity and physical function. Notably, we employed a distinct approach, Bayesian network analysis, to enhance the validity of these findings.

    Notable strengths of this study further include the population-based design and large sample size, both are beneficial for external validity, that is, the bias from applying the study results to other rheumatology healthcare settings (in Europe) is reduced. Furthermore, prior knowledge of interactions between variables was integrated and non-parametric bootstrap was applied to acquire a model with more robust probabilistic relationships.

    This study has several limitations. First, of the 708 initially recruited patients, only 384 could be included in the present analysis at month 72, which might introduce selection bias. However, in terms of patient characteristics and disease outcomes at baseline, these two groups did not differ.

    Second, only disease outcomes were included in the model to keep it comparable to the previously developed health model. Other characteristics, such as socioeconomic status, were absent from the analysis but may ultimately impact ASAS HI. Third, a Bayesian network analysis approach most typically underlies the causal Markov assumption and the causal faithfulness assumption, which could not be verified from the data. Thus, drawing causal conclusions from observational data remains a challenging endeavour,30 31 which causal search algorithms like the PC algorithm do not completely ‘solve’.12 However, estimating the underlying causal structure offers valuable insights that are not attainable through other commonly used methods.12 Further investigations are needed to validate the model in patients with early axSpA (though DESIR is an inception cohort, the model is based on cross-sectional data from month 72) and to study its evolvement over time.

    In conclusion, a Bayesian network analysis approach integrating expert knowledge and observational data were adopted to investigate through graphical probabilistic representation how the interplay of different disease outcomes affects ASAS HI in an axSpA population. This study shows that disease activity and physical function have a strong impact on ASAS HI, confirming it to be a valid construct of overall functioning and health in patients with axSpA. The proposed structure of the network represents a relevant step in better understanding the construct of ASAS HI in patients with axSpA.

    Data availability statement

    No data are available. No additional data are available.

    Ethics statements

    Patient consent for publication

    Ethics approval

    This study involves human participants and the DESIR study was approved by the appropriate local medical ethical committees (Comité de Protection des Personnes Ile de France III). Participants gave informed consent to participate in the study before taking part.

    Acknowledgments

    The DESIR study is conducted as a Programme Hospitalier de Recherche Clinique with Assistance Publique Hopitaux de Paris as the sponsor. The DESIR study is also under the umbrella of the French Society of Rheumatology, which financially supports the cohort. An unrestricted grant from Pfizer has been allocated for the first 10 years. The DESIR cohort is conducted under the control of Assistance publique Hopitaux de Paris via the Clinical Research Unit Paris Centre and under the umbrella of the French Society of Rheumatology and Institut national de la sante et de la recherche medicale (Inserm). Database management is performed within the Department of Epidemiology and Biostatistics (Professeur Jean-Pierre Daures, D.I.M., Nımes, France). We also wish to thank the different regional participating centres: Pr Maxime Dougados (Paris-Cochin B), Pr Andre Kahan (Paris-Cochin A), Pr Philippe Dieudé (Paris-Bichat), Pr Bruno Fautrel (Paris-La Pitie-Salpetriere), Pr Francis Berenbaum (Paris-Saint-Antoine), Pr Pascal Claudepierre (Creteil), Pr Maxime Breban (Boulogne-Billancourt), Dr Bernadette Saint-Marcoux (Aulnay-sous-Bois), Pr Philippe Goupille (Tours), Pr Jean Francis Maillefert (Dijon), Dr Emmanuelle Dernis (Le Mans), Pr Daniel Wendling (Besancon), Pr Bernard Combe (Montpellier), Pr Liana Euller-Ziegler (Nice), Pr Pascal Richette (ParisLariboisiere), Pr Pierre Lafforgue (Marseille), Dr Patrick Boumier (Amiens), Pr Martin Soubrier (ClermontFerrand), Dr Nadia Mehsen (Bordeaux), Pr Damien Loeuille (Nancy), Pr Rene-Marc Flipo (Lille), Pr Alain Saraux (Brest), Pr Xavier Mariette (LeKremlin-Bicetre), Pr Alain Cantagrel (Toulouse), Pr Olivier Vittecoq (Rouen). We wish to thank the research nurses, the staff members of the Clinical Research Unit of Paris Centre, the staff members of the Biological Resource Center of Bichat Hospital, the staff members of the Department of Statistics of Nımes and all the investigators, and in particular Jerome Allain, Emmanuelle Dernis, Salah Ferkal, Clement Prati, Marie-Agnes Timsit, Eric Toussirot for active patient recruitment and monitoring.

    References

    1. ↵
      1. Sieper J,
      2. Poddubnyy D
      . Axial spondyloarthritis. Lancet 2017;390:73–84. doi:10.1016/S0140-6736(16)31591-4
      OpenUrlCrossRefPubMed
    2. ↵
      1. Dagfinrud H,
      2. Kvien TK,
      3. Hagen KB
      . Physiotherapy interventions for ankylosing spondylitis. Cochrane Database Syst Rev 2004:CD002822. doi:10.1002/14651858.CD002822.pub2
    3. ↵
      1. Essers I,
      2. Ramiro S,
      3. Stolwijk C, et al
      . Do extra-articular manifestations influence outcome in ankylosing spondylitis? 12-year results from OASIS. Clin Exp Rheumatol 2016;34:214–21.
      OpenUrlPubMed
    4. ↵
      1. Gordeev VS,
      2. Maksymowych WP,
      3. Evers SMAA, et al
      . Role of contextual factors in health-related quality of life in ankylosing spondylitis. Ann Rheum Dis 2010;69:108–12. doi:10.1136/ard.2008.100164
      OpenUrlAbstract/FREE Full Text
    5. ↵
      1. Machado P,
      2. Landewé R,
      3. Braun J, et al
      . A stratified model for health outcomes in ankylosing spondylitis. Ann Rheum Dis 2011;70:1758–64. doi:10.1136/ard.2011.150037
      OpenUrlAbstract/FREE Full Text
    6. ↵
      1. van der Heijde D,
      2. Dijkmans B,
      3. Geusens P, et al
      . Efficacy and safety of Infliximab in patients with ankylosing spondylitis: results of a randomized, placebo-controlled trial (ASSERT). Arthritis Rheum 2005;52:582–91. doi:10.1002/art.20852
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. World Health Organization
      . International classification of functioning, disability and health: ICF [World Health Organization]. 2001. Available: https://apps.who.int/iris/handle/10665/42407 [Accessed 26 Oct 2022].
    8. ↵
      1. Kiltz U,
      2. van der Heijde D,
      3. Boonen A, et al
      . Development of a health index in patients with ankylosing spondylitis (ASAS HI): final result of a global initiative based on the ICF guided by ASAS. Ann Rheum Dis 2015;74:830–5. doi:10.1136/annrheumdis-2013-203967
      OpenUrlAbstract/FREE Full Text
    9. ↵
      1. Puche Larrubia MÁ,
      2. Castro Villegas MC,
      3. Ortega Castro R, et al
      . ASAS health index in patients with spondyloarthritis and its association with disease activity and disease burden including fibromyalgia. Clin Exp Rheumatol 2021;39 Suppl 130:82–8. doi:10.55563/clinexprheumatol/zr61xv
      OpenUrl
    10. ↵
      1. Kiltz U,
      2. Wiatr T,
      3. Redeker I, et al
      . Effects of patient and disease characteristics on global functioning in patients with axial Spondyloarthritis in routine care. Semin Arthritis Rheum 2022;55:152006. doi:10.1016/j.semarthrit.2022.152006
      OpenUrl
    11. ↵
      1. Dougados M,
      2. d’Agostino M-A,
      3. Benessiano J, et al
      . The DESIR cohort: a 10-year follow-up of early inflammatory back pain in France: study design and baseline characteristics of the 708 recruited patients. Joint Bone Spine 2011;78:598–603. doi:10.1016/j.jbspin.2011.01.013
      OpenUrlCrossRefPubMedWeb of Science
    12. ↵
      1. Quintana R
      . The structure of academic achievement: searching for proximal mechanisms using causal discovery algorithms. Sociol Methods Res 2023;52:85–134. doi:10.1177/0049124120926208
      OpenUrl
    13. ↵
      1. Scanagatta M,
      2. Salmerón A,
      3. Stella F
      . A survey on Bayesian network structure learning from data. Prog Artif Intell 2019;8:425–39. doi:10.1007/s13748-019-00194-y
      OpenUrl
    14. ↵
      1. Kiltz U,
      2. van der Heijde D,
      3. Boonen A, et al
      . Measurement properties of the ASAS health index: results of a global study in patients with axial and peripheral spondyloarthritis. Ann Rheum Dis 2018;77:1311–7. doi:10.1136/annrheumdis-2017-212076
      OpenUrlAbstract/FREE Full Text
    15. ↵
      1. Lukas C,
      2. Landewé R,
      3. Sieper J, et al
      . Development of an ASAS-endorsed disease activity score (ASDAS) in patients with ankylosing spondylitis. Ann Rheum Dis 2009;68:18–24. doi:10.1136/ard.2008.094870
      OpenUrlAbstract/FREE Full Text
    16. ↵
      1. Calin A,
      2. Garrett S,
      3. Whitelock H, et al
      . A new approach to defining functional ability in ankylosing spondylitis: the development of the bath ankylosing spondylitis functional index. J Rheumatol 1994;21:2281–5.
      OpenUrlPubMedWeb of Science
    17. ↵
      1. Jenkinson TR,
      2. Mallorie PA,
      3. Whitelock HC, et al
      . Defining spinal mobility in ankylosing spondylitis (AS). The Bath AS Metrology index. J Rheumatol 1994;21:1694–8.
      OpenUrlPubMedWeb of Science
    18. ↵
      1. Creemers MCW,
      2. Franssen MJAM,
      3. van’t Hof MA, et al
      . Assessment of outcome in ankylosing spondylitis: an extended radiographic scoring system. Ann Rheum Dis 2005;64:127–9. doi:10.1136/ard.2004.020503
      OpenUrlAbstract/FREE Full Text
    19. ↵
      1. Fongen C,
      2. Dagfinrud H,
      3. Berg IJ, et al
      . Frequency of impaired spinal mobility in patients with chronic back pain compared to patients with early axial spondyloarthritis. J Rheumatol 2018;45:1643–50. doi:10.3899/jrheum.170786
      OpenUrlAbstract/FREE Full Text
    20. ↵
      1. Ainsworth HF,
      2. Shin SY,
      3. Cordell HJ
      . A comparison of methods for Inferring causal relationships between genotype and phenotype using additional biological measurements. Genet Epidemiol 2017;41:577–86. doi:10.1002/gepi.22061
      OpenUrl
    21. ↵
      1. Howey R,
      2. Shin S-Y,
      3. Relton C, et al
      . Bayesian network analysis incorporating genetic anchors complements conventional mendelian randomization approaches for exploratory analysis of causal relationships in complex data. PLOS Genet 2020;16:e1008198. doi:10.1371/journal.pgen.1008198
    22. ↵
      1. Schmidt C,
      2. Huegle J,
      3. Uflacker M
      . Order-independent constraint-based causal structure learning for Gaussian distribution models using GPUs. Proceedings of the 30th International Conference on Scientific and Statistical Database Management; New York, NY, USA: Association for Computing Machinery, 2018:1–10 doi:10.1145/3221269.3221292
    23. ↵
      1. Spirtes P,
      2. Glymour C,
      3. Scheines R
      . Causation, Prediction, and Search, 2nd ed. MIT press, 2000.
    24. ↵
      1. Friedman N,
      2. Goldszmidt M,
      3. Wyner A
      . Data analysis with Bayesian networks: a Bootstrap approach. Proceedings of the Fifteenth conference on Uncertainty in artificial intelligence. (UAI’99); San Francisco, CA, USA: Morgan Kaufmann Publishers Inc, 1999:196–205
    25. ↵
      1. Cui R,
      2. Groot P,
      3. Heskes T
      . Learning causal structure from mixed data with missing values using Gaussian Copula models. Stat Comput 2019;29:311–33. doi:10.1007/s11222-018-9810-x
      OpenUrl
    26. ↵
      1. Carvalho PD,
      2. Ruyssen-Witrand A,
      3. Marreiros A, et al
      . Long‐term association between disease activity and disability in early axial spondyloarthritis: results from a prospective observational study of inflammatory back pain. Arthritis Care Res (Hoboken) 2022;74:768–75. doi:10.1002/acr.24515
      OpenUrl
    27. ↵
      1. Machado P,
      2. Landewé R,
      3. Braun J, et al
      . Both structural damage and inflammation of the spine contribute to impairment of spinal mobility in patients with ankylosing spondylitis. Ann Rheum Dis 2010;69:1465–70. doi:10.1136/ard.2009.124206
      OpenUrlAbstract/FREE Full Text
    28. ↵
      1. Pearl J
      . Causality: Models, Reasoning and Inference. Cambridge University Press, 2009. doi:10.1017/CBO9780511803161
    29. ↵
      1. Pearl J
      . Invited commentary: understanding bias amplification. Am J Epidemiol 2011;174:1223–7; doi:10.1093/aje/kwr352
      OpenUrlCrossRefPubMedWeb of Science
    30. ↵
      1. Dawid AP
      . Beware of the DAG. Guyon I, Janzing D, Schölkopf B, eds. NIPS Causality: Objectives and Assessment (JMLR Proceedings; vol. 6); 2010:59–86 Available: http://dblp.uni-trier.de/db/journals/jmlr/jmlrp6.html#Dawid10
    31. ↵
      1. Greenland S
      . Overthrowing the tyranny of null hypotheses hidden in causal diagrams. Heuristics Probab Causality Tribute Jud Pearl 2010:365–82.

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Footnotes

    • Contributors Study concept and design: UK, DvdH, JB, AB, SR and RL. Analysis and interpretation of data: all authors. Writing of the manuscript: all authors. Critical revision of the manuscript for important intellectual content: all authors had access to the data, commented on the report drafts and approved the final submitted version. IR is responsible for the overall content as guarantor.

    • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

    • Competing interests None declared.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.