Trajectories of disease evolution upon treatment initiation in systemic lupus erythematosus: results from four clinical trials of belimumab

Abstract

Objective

Upon commencement of therapy for active disease, patients with systemic lupus erythematosus (SLE) show varying evolution regarding disease activity measures and patient-reported outcomes (PROs). Our objective was to identify disease evolution trajectories to gain a deeper understanding of SLE progression, ultimately improving future trial design.

Methods

Patients with ≥2 visits and available data on Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K), British Isles Lupus Assessment Group (BILAG), Physician Global Assessment (PGA), Functional Assessment of Chronic Illness Therapy—Fatigue (FACIT-F), and glucocorticoid use were included in a post hoc analysis of four randomized controlled trials of belimumab (BLISS-52, BLISS-76, BLISS-SC, EMBRACE). Growth mixture modelling identified latent classes.

Results

Among 2868 patients analysed, baseline median disease duration was 4.5 (interquartile range: 1.5–9.7) years and mean (±standard deviation) Systemic Lupus International Collaborating Clinics/American College of Rheumatology (SLICC/ACR) Damage Index (SDI) 0.7 (±2.0), SLEDAI-2K 10.2 (±3.6), BILAG 17.0 (±7.8), PGA 1.5 (±0.5), FACIT-F 30.6 (±11.9) and prednisone dose 11.0 (±8.9) mg/day. In the initial model, glucocorticoid use and dose yielded high standard errors, indicating a weak link with the latent process. A refined model considered only clinical measures and FACIT-F, corrected for intervention and SDI; no other covariates improved the fit. Four classes best described disease evolution: highly active, responders; highly active, non-responders; moderately active, responders; moderately active, non-responders. Lupus Low Disease Activity State and Definitions of Remission in SLE remission attainment associated with latent classes.

Conclusion

By linking disease activity measures with PROs, we identified four distinct trajectories describing SLE evolution following the initiation of therapy. This classification could be valuable for personalizing treatment and guiding biological studies aimed at distinguishing patients with varying anticipated treatment responses, as no single clinical variable alone can predict disease progression.

Rheumatology key messages.

• Four SLE patient classes with distinct longitudinal disease trajectories were identified using growth mixture modeling.

• Classes were defined by disease activity (SLEDAI-2K, BILAG, PGA) and patient-reported (FACIT-F) data.

• The study identifies PROs as crucial for SLE patient characterisation, monitoring, and management goal determination.

Introduction

Systemic lupus erythematosus (SLE) is an autoimmune disease that is characterized by a flaring-remitting pattern and a marked heterogeneity of clinical manifestations and immunological aberrancies [1, 2]. People with SLE report impaired health-related quality of life (HRQoL) [3, 4], even despite adequate clinical response to therapies [5], and fatigue is often described as the most debilitating symptom [6, 7]. A major issue contributing to hurdles in management is the unpredictability of disease evolution, e.g. how the disease responds to therapeutic interventions for active disease, when a flare is about to occur, or which organ systems are likely to be afflicted.

Management is mainly based on current organ affliction, and pharmacotherapy comprises glucocorticoids, antimalarial agents, non-biological immunosuppressants, and, since a couple of decades, targeted therapies. The monoclonal antibody belimumab that targets the soluble form of B cell activating factor has been approved for treating SLE since 2011, and its efficacy has been demonstrated in several clinical trials and observational studies. More recently, the type I interferon receptor blocker anifrolumab was approved for SLE. Several novel agents are in current trial programmes, and one may foresee an increasing enrichment of our therapeutic armamentarium in the future. However, to better understand which therapy is suitable for which patient and when during a patient’s disease course, we still need to gain insights into how the disease evolves over time. Trajectory modelling is a valuable tool in this context, increasingly employed to predict disease patterns and study long-term outcomes across patient subsets with shared characteristics and disease evolution. This approach aids in tailoring interventions to meet the specific needs of these groups [8, 9].

Goals with therapy in SLE include low disease activity (LDA) and, when possible, remission [10]. Commonly used sets of criteria for LDA and remission are those of Lupus Low Disease Activity State (LLDAS) [11, 12] and the prevailing definition among the Definitions of Remission in SLE (DORIS) sets [13, 14], respectively. Notably, despite the importance of HRQoL aspects for people living with the disease [15, 16] and explicit recommendations from the Outcome Measures in Rheumatology (OMERACT) SLE working group [17, 18], these definitions of LDA and remission do not incorporate patient-reported outcomes (PROs).

To partially address this issue, we herein aimed to identify trajectories of disease evolution based on clinical, routine laboratory, and patient-reported parameters based on an analysis of integrated data from large clinical trial settings.

Patients and methods

Patients

Data from patients with active SLE who participated in three phase III, multicentre, randomized, double-blind, placebo-controlled trials comparing belimumab with placebo were considered. These trials included BLISS-52 (NCT00424476, n = 867) [19], BLISS-76 (NCT00410384, n = 797) [20], and BLISS-SC (NCT01484496, n = 822) [21]. Additionally, data from a phase III/IV randomized trial comparing belimumab with placebo in patients of Black African ancestry (NCT01632241, n = 448) [22] were included.

From the overall pool of patients, subjects who met the following criteria were then selected: (i) complete data for at least two consecutive visits, starting from that of baseline; (ii) availability of Functional Assessment of Chronic Illness Therapy—Fatigue (FACIT-F) data [23] (scale: 0–52, with higher values representing lower fatigue levels, and with the general healthy population reporting a mean of 43 [24, 25]); (iii) availability of Safety of Estrogens in Lupus Erythematosus National Assessment—Systemic Lupus Erythematosus Disease Activity Index (SELENA-SLEDAI) Physician Global Assessment (PGA) data [26] (scale: 0–3, with lower values representing better health status); (iv) availability of classic British Isles Lupus Assessment Group (BILAG) data [27]; (v) availability of Systemic Lupus Erythematosus Disease Activity Index 2000 (SLEDAI-2K) data [28]; (vi) availability of information about the daily prednisone (or equivalents) dose at study visits. For analysis purposes, classic BILAG domain scores were transformed into a numerical total score as described previously (A = 12; B = 8; C = 1; D = 0; E = 0) [29].

Clinical and routine laboratory test data were used to calculate the LLDAS [11, 12] and DORIS remission [13, 14] at each study visit. Additional measures of interest collected in the belimumab trials included the 36-item Short Form health survey questionnaire (SF-36), summarized in its Physical Component Summary (PCS) and Mental Component Summary (MCS) [30] (scale: 0–100, with values over 50 representing better than average and below 50 poorer than average health perception), the 3-level version of EQ-5D (EQ-5D-3L) index (scale: −0.59 to 1, with higher values representing better health perception and values below 0 representing health perception that is considered worse than death), and the EQ-5D visual analogue scale (EQ-VAS) (scale: 0–100, with higher values representing better health perception) [31].

The minimal clinically important difference (MCID) for BILAG was set at a ≥ 7-point reduction compared with baseline and for SLEDAI-2K at a ≥ 6-point reduction compared with baseline, as recommended [32]. Similarly, the MCID for FACIT-F was set at a ≥ 4-point increase compared with baseline, as recommended [33, 34].

Trajectory analysis

Due to differences in study design and time points for collection of PROs, the following time points were considered in the analysis: week (w)0 (baseline), w4, w12, w24, w36, and w52. Intermediate time points (w16, w20, w28, w40, w44, w48) were only used for plotting results of available data.

Disease trajectories were modelled via growth mixture algorithms as implemented in the R package lcmm [8]. The analysis was optimized according to a previously described framework [35] and according to the vignettes on the authors’ website (https://cecileproust-lima.github.io/lcmm/index.html). The latent process characterizing the burden of SLE was determined using multiple disease activity indices and PROs (SLEDAI-2K, classic BILAG, PGA, FACIT-F, SF-36, EQ-5D-3L), as well as glucocorticoid use and dose. To describe the common latent process measured simultaneously by these quantities, we next applied the multivariate multlcmm function. Briefly, the following steps were performed to optimize the analysis: (i) normalization (an empty lcmm model was applied to each variable, and the corresponding predicted values were used as normalized values); (ii) rescaling (normalized values were rescaled to the same unit, according to the formula: scaled value = 100 × [raw value − min value]/[max value − min value]; for FACIT-F, since the direction of the scores is opposite to that of the other measures [i.e. higher scores correspond to lower fatigue], a gap statistic approach was followed; gap values = 100 − scaled values); (iii) evaluation of variables of interest [first, a scoping multlcmm model including all variables was built, and residual standard errors (RSE) of each variable were calculated; next, variables with high RSE, indicative of a poor link to the latent process, were discarded from the model]; (iv) evaluation of covariates [baseline covariates, including age, sex, disease duration, current use of antimalarial agents, current use of non-biological immunosuppressants, and Systemic Lupus International Collaborating Clinics/American College of Rheumatology (SLICC/ACR) Damage Index (SDI) scores] [36] were included in the scoping model in a stepwise manner, and the fit of the model was evaluated via the Bayesian information criterion [BIC] [37]; as data were generated from controlled trials, to account for differences owing to the trial interventions [belimumab and placebo regimens], the study arm was forced as a covariate into the model); (v) inclusion of random intercepts (the scoping model with covariates was next retested to include random intercepts, and its fit was evaluated via the BIC criterion); (vi) selection of the best number of trajectories (the number of classes that best described the latent process was finally selected by building multlcmm models following the pipeline described above, with k = 1–6 classes, and the optimal number was selected via the BIC criterion; the selected model was considered the best final model).

Other statistics

The distribution of categorical variables in relation to the latent classes of disease progression was evaluated via the chi-squared (χ²) test. Analysis of variance with Tukey’s post hoc correction was used to determine pairwise differences in continuous covariates. Data are described as the mean ± standard deviation (s.d.) or the median and interquartile range (IQR).

Ethics

Data from the belimumab trials were made available by GlaxoSmithKline (Uxbridge, UK) through the Clinical Study Data Request (CSDR) consortium. The trial protocols were approved by regional ethics review boards for all participating centres and complied with the ethical principles of the Declaration of Helsinki. Written informed consent was obtained from all study participants prior to enrolment. This study was approved by the Swedish Ethical Review Authority (registration number: 2019–05498).

Patient and public involvement

Patient research partners (K.B.S., Y.E.) were involved in the design and reporting of this research. The public was not involved in the design, conduct, reporting or dissemination plans of this research.

Results

Clinical and demographic characteristics

A total of 2868 patients met the inclusion criteria. Their baseline characteristics are summarized in Table 1. The majority of study participants were women (n = 2713, 98%) with active disease as assessed by SLEDAI-2K (10.2 ± 3.6), BILAG (17.0 ± 7.8), and PGA (1.5 ± 0.5). BILAG characterization is provided in Supplementary Table S1, available at Rheumatology online. PROs indicated a poor HRQoL perception at baseline, with a mean FACIT-F score of 30.6 ± 11.9, SF-36 PCS 39.1 ± 9.6, SF-36 MCS 40.9 ± 11.2, and EQ-5D-3L index score 0.73 ± 0.20.

Table 1.

Demographics and clinical data of the study population

Variable	SLE patients (n = 2868)
Female sex; n (%)	2713 (98)
Age, years; mean ± s.d.	38.3 ± 11.8
Disease duration, years; median (IQR)	4.5 (1.5–9.7)
SLEDAI-2K; mean ± s.d.	10.2 ± 3.6
BILAG scores; mean ± s.d.	17.0 ± 7.8
PGA; mean ± s.d.	1.5 ± 0.5
SDI; mean ± s.d.	0.7 ± 2.0
FACIT-F; mean ± s.d.	30.6 ± 11.9
SF-36 PCSa; mean ± s.d.	39.1 ± 9.6
SF-36 MCSa; mean ± s.d.	40.9 ± 11.2
EQ-5D-3Lb; mean ± s.d.	0.73 ± 0.20
EQ-VASb; mean ± s.d.	63.1 ± 20.0
Prednisone equivalent dose, mg/day; mean ± s.d.	11.0 ± 8.9
Antimalarial agents; n (%)	1970 (68.7)
Immunosuppressants; n (%)	1420 (49.5)

Data are presented as numbers (percentage) or means (standard deviation). In case of non-normal distributions, the medians (IQR) are indicated. In case of missing values, the total number of patients with available data is indicated. The total number of visits was 34 039.

BILAG: British Isles Lupus Assessment Group; EQ-5D-3L: 3-level version of EQ-5D health survey; FACIT-F: Functional Assessment of Chronic Illness Therapy—Fatigue; IQR: interquartile range; MCS: Mental Component Summary; PCS: Physical Component Summary; PGA: Physician Global Assessment; s.d.: standard deviation; SDI: Systemic Lupus International Collaborating Clinics (SLICC)/American College of Rheumatology (ACR) Damage Index; SF-36: 36-item Short Form health survey questionnaire; SLEDAI-2K: Systemic Lupus Erythematosus Disease Activity Index 2000; VAS: visual analogic scale.

^aData from 1599 patients.

^bData from 1601 patients.

Steroid use and latent processes in SLE disease evolution

To determine the variables that contributed to the latent process characterizing disease severity and SLE burden, we first analysed the RSE of the variables included in the initial multlcmm scoping model. The resulting values were as follows: SLEDAI-2K score 1.139; BILAG score 1.225; PGA score 1.387; FACIT-F score 4.005; prednisone equivalent dose 12.809 mg/day. The high value of daily prednisone doses suggests that glucocorticoids are poorly linked to the latent process and do not share much information with the other variables. Similarly, we calculated the coefficient of determination (R²) of these variables, with values at the end of the observation period as follows: SLEDAI-2K score 63.779; BILAG score 60.366; PGA score 54.287; FACIT-F score 12.464; prednisone equivalent dose 1.373 mg/day. Overall, these results suggest that glucocorticoid use is poorly linked to the process describing disease progression, and thus, this variable can be safely removed when modelling latent disease trajectories.

Baseline organ damage and SLE disease evolution

Several baseline variables were included in the scoping multlcmm model to determine relevant covariates for characterizing disease evolution. The baseline multlcmm model that included treatment arm as a covariate (belimumab or placebo) had a baseline BIC of 564169 and was improved by the inclusion of baseline SDI scores (BIC = 564103). A stepwise addition of other clinical variables, including sex, disease duration, use of antimalarial agents or use of non-biological immunosuppressants did not improve the model and, thus, these variables were considered negligible for the modelling of SLE disease trajectories.

SLE progression is recapitulated by four distinct trajectories (latent classes)

The investigation of the optimal number of trajectories to describe the latent process simultaneously linked to patient-reported (FACIT-F) and disease activity measures (SLEDAI-2K, BILAG, PGA), while correcting for treatment arm and baseline SDI scores, identified four latent classes as the best fit for describing SLE disease trajectories within the 52-week timeframe of the belimumab trials. Model adequacy assessment, measured by relative entropy (E_K), a metric of class separation, showed an E_K of 0.55, which exceeds the desirable threshold of 0.5 [35]. Similarly, the average maximum posterior probability of assignments for all classes was above the desirable threshold of 70% [35].

The proportions of patients assigned to the different classes were 13.9%, 12.4%, 13.1% and 60.6%. Baseline clinical characteristics of patients in these classes are reported in Table 2. The classes were automatically ordered from the most severe (class 1) to the least severe (class 4) class in terms of disease activity and PROs. On average, classes 1 and 2 encompassed patients with highly active disease, while classes 3 and 4 comprised patients with moderately active disease. Notably, although the algorithm was optimized to capture FACIT-F trends, the resulting subdivision also applied well to the other PROs.

Table 2.

Demographics and clinical characteristics of patients assigned to the different classes

Variable	Class 1 (n = 398)	Class 2 (n = 355)	Class 3 (n = 376)	Class 4 (n = 1739)
Female sex; n (%)	380 (95.5%)	331 (93.2%)	354 (94.1%)	1648 (94.8%)
Age, years; mean ± s.d.	37.2 ± 11.1	37.1 ± 11.5	39.1 ± 11.8	38.6 ± 11.9
Disease duration, years; median (IQR)	5.2 (1.6–10.1)	4.37 (1.6–9.3)	3.2 (1.0–8.0)	4.5 (1.6–9.5)
BILAG scores; mean ± s.d.	22.5 ± 7.4	21.2 ± 8.3	16.7 ± 7.6*	15.1 ± 6.9*,  **
SLEDAI-2K; mean ± s.d.	13.1 ± 4.1	12.4 ± 4.1***	9.6 ± 2.9*	9.3 ± 3.0*
PGA; mean ± s.d.	1.7 ± 0.4	1.7 ± 0.5***	1.5 ± 0.5*	1.4 ± 0.5*,****
SDI; mean ± s.d.	0.7 ± 1.1	0.7 ± 1.1	0.6 ± 1.1	0.8 ± 1.2
FACIT-F; mean ± s.d.	26.2 ± 12.1	30.1 ± 11.8	31.3 ± 12.2	31.6 ± 11.5
SF-36 PCSa; mean ± s.d.	35.3 ± 9.8	38.4 ± 9.0*****	39.7 ± 10.1******	39.9 ± 9.5******
SF-36 MCSa; mean ± s.d.	37.3 ± 11.0	38.8 ± 11.0	41.5 ± 10.5*	41.8 ± 11.2*
EQ-5D-3Lb; mean ± s.d.	0.67 ± 0.21	0.73 ± 0.17*****	0.73 ± 0.18*****	0.76 ± 0.18******
EQ-VASb; mean ± s.d.	56.8 ± 19.4	63.1 ± 18.8*****	64.7 ± 20.9******	64.9 ± 18.6******
Prednisone equivalent dose, mg/day; mean (s.d.)	10.8 ± 8.4	11.5 ± 9.1	11.2 ± 9.1	10.9 ± 8.9

Data are presented as numbers (percentage) or means (standard deviation). In case of non-normal distributions, the medians (IQR) are indicated. In case of missing values, the total number of patients with available data is indicated.

BILAG: British Isles Lupus Assessment Group; EQ-5D-3L: 3-level version of EQ-5D health survey; FACIT-F: Functional Assessment of Chronic Illness Therapy—Fatigue; IQR: interquartile range; MCS: Mental Component Summary; PCS: Physical Component Summary; PGA: Physician Global Assessment; s.d.: standard deviation; SDI: Systemic Lupus International Collaborating Clinics (SLICC)/American College of Rheumatology (ACR) Damage Index; SF-36: 36-item Short Form health survey questionnaire; SLEDAI-2K: Systemic Lupus Erythematosus Disease Activity Index 2000; VAS: visual analogic scale.

^aData from 1599 patients.

^bData from 1601 patients.

^* P < 0.001 vs class 1 and class 2;.

^** P < 0.05 vs class 3;.

^*** P < 0.05 vs class 1.

^**** P < 0.001 vs class 3;.

^***** P < 0.05 vs class 1;.

^****** P < 0.001 vs class 1.

Despite differences among latent classes, no single baseline variable, or combination of variables in multinomial or machine learning models, was capable of accurately predicting class membership (results not shown). This suggests that other, herein unexplored features may be relevant for class composition and SLE disease trajectory modelling.

When plotting the trends of SLEDAI-2K, BILAG, PGA and FACIT-F scores across the different classes, both highly and moderately active groups could be further subdivided into responders (class 2 and 3) and non-responders (class 1 and 4) (Fig. 1, left column). Specifically, when considering the ratio of visit scores to baseline scores, it became evident that the curves for moderately active patients were shifted to the left, indicating a higher likelihood of achieving a lower level of disease activity within a shorter period of time compared with patients with highly active disease at baseline (Fig. 1, right column). Trends of PROs across the different classes yielded similar patterns but with less pronounced differences, as shown in Supplementary Fig. S1, available at Rheumatology online.

Figure 1.

Trajectories of disease activity and patient-reported data across latent classes. Lines link circles denoting mean values. Dotted lines denote standard errors of the mean. BILAG: British Isles Lupus Assessment Group; FACIT-F: Functional Assessment of Chronic Illness Therapy—Fatigue; PGA: Physician Global Assessment; SLEDAI-2K: Systemic Lupus Erythematosus Disease Activity Index 2000

Latent classes in relation to clinically meaningful outcomes

Upon determination of the association between latent classes and disease activity measures, we next analysed the proportion of patients reaching clinically meaningful outcomes relative to the different disease trajectories. A clear distinction was observed between patients labelled as ‘responders’ and those labelled as ‘non-responders’, yielding a greater likelihood of attaining LLDAS or DORIS remission among moderately active patients at baseline (Fig. 2). Considering the MCIDs for SLEDAI-2K, BILAG and FACIT-F scores, similar trends were observed, though with less pronounced separations for FACIT-F scores (Fig. 3).

Figure 2.

Proportions of patients attaining LLDAS or DORIS remission over time on treatment across latent classes. Lines link circles denoting mean values. Dotted lines denote standard errors of the mean. DORIS: Definitions of Remission in SLE; LLDAS: Lupus Low Disease Activity Status; SLE: systemic lupus erythematosus

Figure 3.

Proportions of patients attaining MCIDs in BILAG, SLEDAI-2K or FACIT-F scores over time on treatment across latent classes. Lines link circles denoting mean values. Dotted lines denote standard errors of the mean. BILAG, British Isles Lupus Assessment Group; FACIT-F: Functional Assessment of Chronic Illness Therapy—Fatigue; MCID: minimal clinically important difference; SLEDAI-2K: Systemic Lupus Erythematosus Disease Activity Index 2000

Discussion

The primary aim of this study was to identify disease trajectories in SLE based on an integrated analysis of clinical, routine laboratory, and patient-reported parameters from large clinical trial settings. Our results revealed four distinct latent classes representing different disease trajectories within a 52-week timeframe, which were characterized by varying degrees of disease activity and patient-reported HRQoL, particularly the degree of fatigue. These classes effectively captured the heterogeneous nature of SLE and provided insights into the patterns of disease evolution in response to treatment.

Trajectory modelling has gained endorsement in the study of chronic diseases such as SLE both from epidemiological and biological perspectives. It enables to anticipate the evolution of disease over time and differentiate across subgroups that follow distinct patterns of disease progression, thus disentangling disease heterogeneity and facilitating the introduction of targeted therapies and personalized management approaches [9]. The methodology employed in this study involved growth mixture modelling to analyse data from multiple disease activity indices and PROs, an approach that is considered suitable for the determination of disease trajectories as it allows for the identification of subgroups within a patient population that follow distinct paths of disease over time [35]. The contributing features were chosen based on data availability and clinical expertise, ensuring that variables with clinically meaningful information were included. Subsequently, clinical parameters were added to the equation but did not improve the fit and were, therefore, discarded in a stepwise approach, and some features were deemed important to correct the models for. This careful selection enhances the relevance and applicability of our findings to real-world clinical practice.

Current definitions of targets of management in SLE, such as those of LLDAS [11, 12] and DORIS remission [13, 14], typically do not incorporate PROs. In comparison with previous studies attempting identification of disease trajectories of SLE solely based on physician-based measures [38, 39], our approach emphasized the significance of PROs in determining disease trajectories, thereby placing a high value on the patients’ health experiences. Including PROs, such as FACIT-F, in the characterization of an SLE population appears essential based on our data. Fatigue is in fact a major complaint among SLE patients and significantly impacts their quality of life [6]. Our previous work has emphasized that patients with SLE may experience severe fatigue despite adequate clinical response to therapy [5]. Furthermore, we and others have showed that PROs are not fully explained by physician-based measures, and may depend on variables which are not captured by currently used disease-specific activity scores [40–42]. This study highlights the need for PROs to be an integral part of clinical evaluation and incorporated in definitions of management targets in SLE, ensuring that patient-centred outcomes are prioritized [43]. Importantly, this aligns with recommendations from the OMERACT SLE working group [17, 18] and the multi-stakeholder consortium Treatment Response Measure for SLE (TRM-SLE) [44], which aim to develop novel core outcome sets and measures for assessing drug efficacy in a way that is meaningful and integrated for both patients and clinicians.

Interestingly, the analysis showed that glucocorticoids had a negligible contribution to the determination of latent classes. This finding may be due to several reasons. Glucocorticoids are typically used for their broad anti-inflammatory effects rather than targeting specific disease mechanisms. Additionally, their dosing and use can be highly variable and influenced by factors such as physician preference and patient tolerance [45, 46], which might dilute their apparent impact on disease trajectories when analysed alongside other variables that carry important clinical information.

While this study has several strengths, including the use of comprehensive data from multiple large-scale clinical trials and the robust statistical methods employed, it also has limitations. One limitation is the potential for selection bias imposed by the inclusion criteria of the trials and the selection of patients for analysis based on data availability. Additionally, the post hoc nature of the study limits the ability to establish causality between identified trajectories and specific treatments. The strengths of the study lie in its large sample size, the inclusion of patient populations of diverse ethnic backgrounds, and the integration of both clinical parameters and PROs, providing a comprehensive view of SLE disease evolution.

In conclusion, this study identified four distinct disease trajectories in SLE patients undergoing treatment for active disease, emphasizing the importance of incorporating PROs into disease activity assessments. The negligible role of glucocorticoids in defining these trajectories suggests that other factors may be more critical in influencing disease evolution in response to therapies. These findings underscore the need for a holistic approach to SLE management that includes both clinical measures and patient-reported experiences, ultimately aiming to improve patient outcomes and quality of life. Future research should focus on validating these trajectories in independent cohorts and exploring additional variables that may further refine our understanding of SLE disease evolution.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

The datasets used and analysed during this study are available from the corresponding author upon reasonable request.

Contribution statement

Ioannis Parodis and Lorenzo Beretta were involved in the conceptualization of the study. Ioannis Parodis, Julius Lindblom, Alexander Tsoi, Dionysis Nikolopoulos and Lorenzo Beretta were involved in data curation. Ioannis Parodis, Julius Lindblom, Alexander Tsoi, Dionysis Nikolopoulos and Lorenzo Beretta were involved in data analysis. All authors were involved in interpretation of results. Ioannis Parodis and Lorenzo Beretta were involved in the visualization. Ioannis Parodis, Leonardo Palazzo and Lorenzo Beretta were involved in manuscript drafting. All authors were involved in critical review of manuscript drafts, and approval of the final draft prior to submission.

Funding

I.P. has received grants from the Swedish Rheumatism Association (R-995882), King Gustaf V’s 80-year Foundation (FAI-2023–1055), Swedish Society of Medicine (SLS-974449), Nyckelfonden (OLL-1000881), Professor Nanna Svartz Foundation (2021–00436), Ulla and Roland Gustafsson Foundation (2024–43), Region Stockholm (FoUI-1004114) and Karolinska Institutet. D.N. has received grants from the Swedish Rheumatism Association (R-995557), King Gustaf V’s 80-year Foundation (FAI-2023–1006), Ulla and Roland Gustafsson Foundation (2024–49), Ulla and Gustaf af Uggla Foundation (2023–025029) and Karolinska Institutet.

Disclosure statement: I.P. has received research funding and/or honoraria from Amgen, AstraZeneca, Aurinia, BMS, Elli Lilly, Gilead, GSK, Janssen, Novartis, Otsuka and Roche. The other authors declare no conflicts of interest. The funders had no role in the design of the study, in the collection, analyses or interpretation of data, in the writing of the manuscript or in the decision to publish the results.