Effectiveness of second-line biologics in juvenile idiopathic arthritis: the IIBIOJIA Study

Abstract

Objectives

To describe the prescription patterns of second-line biological disease-modifying antirheumatic drugs (bDMARDs) in non-systemic juvenile idiopathic arthritis (JIA) and to assess the effectiveness of TNFα inhibitors (TNFi) versus non-TNFi as second-line bDMARDs.

Methods

This retrospective cohort study included patients with non-systemic JIA treated with at least two bDMARDs. Multivariate analyses were performed using logistic generalised estimating equations (GEE) models with robust SEs to assess factors associated with the presence of clinically inactive disease (CID) during the second course of bDMARD treatment. Propensity scores, stratified into quintiles, were included in the analysis.

Results

A total of 127 patients were recruited at two paediatric rheumatology tertiary centres. In 25% of cases, a non-TNFi was chosen as second-line bDMARD, with tocilizumab being the most common (66%). The choice of second-line bDMARD was influenced by the calendar year of prescription. In median, non-TNFi treatments were initiated in 2021 (IQR 2017–2023), compared with 2019 (IQR 2013–2022) for TNFi treatments (p=0.03). In the multivariate GEE model adjusted for propensity score quintiles, switching to a second TNFi over a non-TNFi was significantly associated with achieving CID (OR 4.19, 95% CI 1.66 to 10.61, p<0.001). Time on the first TNFi did not significantly impact the likelihood of achieving CID during the second TNFi (OR 0.98, CI 0.79 to 1.21, p=0.83).

Conclusion

Our findings suggest that TNFi remains a viable and effective option as second-line bDMARDs in non-systemic JIA, regardless of the duration of prior TNFi exposure.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Current recommendations for second-line biological disease-modifying anti-rheumatic drugs in non-systemic juvenile idiopathic arthritis (JIA) are limited and based on very low-quality evidence, largely extrapolated from rheumatoid arthritis.

• Switching to a non-TNFi after primary TNFi failure or cycling to another TNFi after secondary failure is suggested, but supporting data are weak.

• For JIA-associated uveitis, changing to a different mechanism of action after failure of two TNFi agents is only conditionally recommended. As a result, treatment selection remains largely empirical, highlighting the need for stronger evidence.

WHAT THIS STUDY ADDS

• This study provides real-world evidence from over two decades of clinical practice suggesting that switching to a second TNFi after first-line failure in non-systemic JIA may be associated with higher rates of clinically inactive disease (CID) compared with switching to a non-TNFi.

• It is the first study in JIA to report a potential advantage of TNFi cycling, with more than a fourfold increased likelihood of achieving CID, independent of uveitis status.

• Additionally, it shows that outcomes and drug retention with second-line TNFi are not influenced by the duration of the initial TNFi.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings support TNFi cycling as a valid and potentially preferable second-line strategy in clinical practice, including in patients switching due to adverse events or uveitis.

• They may contribute to refining current recommendations, which are presently based on limited evidence, and help reduce reliance on empirical treatment decisions.

Introduction

The recent advancements in our understanding of the pathophysiology of juvenile idiopathic arthritis (JIA), the most frequent chronic paediatric rheumatic disease, have allowed the development of targeted therapies, which have significantly improved patient outcomes over the past two decades.^1–3 The ultimate goal of treatment is to achieve disease inactivity, prevent joint damage and minimise potential iatrogenic effects. Traditionally, a step-up approach has been followed, where biological disease-modifying antirheumatic drugs (bDMARDs) are introduced only after the failure of synthetic DMARDs, mainly methotrexate (MTX). However, bDMARDs are now being used earlier in the disease course, particularly following the adoption of a treat-to-target strategy.⁴ Tumor Necrosis Factor-α inhibitors (TNFi) are the most frequently prescribed first-line bDMARDs for patients with non-systemic JIA.^5–10 In some cases, treatment with first-line TNFi has to be discontinued because of primary non-response, secondary non-response or side effects: in cohort studies, 23%–35% of patients with JIA require a second-line bDMARD.^5–10 Currently, there are few and weak recommendations supporting the choice of the second-line bDMARDs for non-systemic JIA. Indeed, in cases of active arthritis, it is currently recommended to switch to a non-TNFi if the initial TNFi fails primarily, whereas switching to another TNFi may be considered in cases of secondary failure.¹¹ However, this recommendation is based on very low-quality evidence, mostly derived from studies in patients with rheumatoid arthritis (RA).^{12 13} In contrast, for JIA-associated uveitis, guidelines recommend switching to a different mechanism of action after the failure of two distinct TNFi agents (excluding etanercept).¹⁴ However, this recommendation remains conditional due to the very low quality of supporting evidence. As a result of these uncertainties, selecting a second-line bDMARD often relies on a ‘trial-and-error’ approach, where different therapies are attempted until clinically inactive disease (CID) or, at minimum, low disease activity is achieved. Therefore this study aims to:

1. Describe prescription patterns of second-line bDMARDs in non-systemic JIA across two tertiary centres;

2. Compare the effectiveness of TNFi as first-line and second-line bDMARD and identify predictor of response of second course of TNFi;

3. Compare the effectiveness of TNFi versus non-TNFi as second-line bDMARDs.

Methods

This retrospective cohort study included patients with non-systemic JIA who were treated with at least two bDMARDs.

Patients were included in the present study in case of (1) A diagnosis of JIA according to International League of Associations for Rheumatology (ILAR) criteria,¹⁵ (2) Treatment with at least two bDMARDs, and (3) Having received a TNFi as the first course of bDMARD before 18 years. Patients were recruited at two rheumatology paediatric tertiary centres in Milan, Italy. Patients’ charts were reviewed over a period spanning from 2003 to 2024.

Exclusion criteria enlisted the diagnosis of systemic JIA and the presence of pre-existing conditions or comorbidities that could influence the choice of treatment or medications’ effectiveness such as immunodeficiency, malignancy, severe hepatic or renal disease, or other autoimmune or inflammatory conditions requiring systemic immunosuppressive therapy.

Demographic and clinical data were recorded, including age at disease onset and diagnosis, sex, antinuclear antibody status, ILAR category, prior DMARD treatments, and details of the first and second bDMARDs (including start and stop dates and reasons for drug discontinuation). Given the broad timeframe of patient recruitment, all bDMARDs available for the treatment of inflammatory arthritis were considered, regardless of their licensure status for JIA.

Outcome measures were assessed at 6 months and 12 months following the initiation of both the first and second bDMARDs. Given the retrospective nature of the study, visits occurring within ±1 month of each scheduled timepoint were considered for the corresponding assessment. They included the presence of active arthritis, active uveitis, active joint count, raised inflammatory markers (either erythrocyte sedimentation rate (ESR) and C reactive protein (CRP)), and attainment of the CID status.

Uveitis was classified as active in the presence of at least one of the following: anterior chamber cell grade ≥1, one active inflammatory chorioretinal or retinal vascular lesion, or vitreous haze grade ≥1.¹⁶ CID was defined as absence of joints with active arthritis, absence of active uveitis, absence of enthesitis, normal ESR or CRP level, and no disease activity indicated in the physician global assessment (Visual Analogue Scale =0). This definition was adapted from the Wallace criteria for inactive disease,¹⁷ with specific modifications. Morning stiffness was excluded due to the lack of a clear definition and potential age-related self-reporting biases. Systemic manifestations (fever, rash, serositis, splenomegaly or generalised lymphadenopathy) were not considered as patients with systemic JIA were excluded from the study.

Drug retention of the second-line bDMARD was assessed both as a dichotomous variable (1 year survival rate) and as a continuous variable (years on treatment), with censoring at drug discontinuation or last follow-up.

Compliance with the General Data Protection Regulation (Regulation (EU) 2016/679) and applicable Italian data protection legislation, as well as the principles of the Declaration of Helsinki, was maintained throughout the study. Informed consent for medical chart data collection was obtained from all patients or their legal guardians.

Statistical analysis

Quantitative variables were summarised as medians with range, or IQR, while categorical variables were expressed as absolute and relative frequencies. The Shapiro-Wilk test was used to assess the normality of distributions. Differences in quantitative variables were analysed using the non-parametrical Mann-Whitney U test. Associations between categorical variables were evaluated with either the χ² test or Fisher’s exact test, as appropriate. Paired comparisons between the first and second TNFi treatment courses were performed using McNemar’s test.

The propensity quintile-adjusted Cox regression was applied to compare the treatment duration between patients starting a second TNFi and those receiving a non-TNFi as second-line treatment. Patients who did not discontinue the second-line bDMARD were censored at their latest follow-up visit.

To account for repeated measurements from the same subjects at these two time points, generalised estimating equations (GEE) models with robust SEs and an exchangeable correlation structure were adopted. This approach allowed each time point to contribute separately to the analysis while adjusting for within-subject correlation. The primary outcome was the presence of CID at each time point (6 months and 12 months) during the second course of bDMARD treatment. Patients who discontinued treatment before a scheduled visit were classified as non-responders for the corresponding effectiveness analysis. Importantly, patients were not required to have CID at both time points to be classified as reaching the outcome since each observation (CID status at 6 months and at 12 months) was treated as an individual data point in the model. This analysis allowed us to compare the effectiveness of first-line TNFi versus (1) Second-line TNFi and (2) Non-TNFi as the second bDMARD. In addition, predictors of CID following initiation of second-line TNFi were evaluated. ORs and 95% CIs were calculated. Propensity scores were calculated considering the following covariates: sex, JIA subtype, disease duration at the onset of the first bDMARD, disease duration at the onset of the second bDMARD, presence of active arthritis at the time of the switch, history of uveitis, year of second bDMARD start. Propensity scores stratified into quintiles were included as an indicator variable in the regression models. The distribution of patients across propensity score quintiles by treatment group is reported in online supplemental table 1. The propensity score was included in the GEE model as quintiles. To avoid redundant adjustment and potential collinearity, variables included in the propensity score model were not entered again into the outcome model. The model included only clinically relevant covariates not included in the propensity score. The propensity score-adjusted GEE model included eight parameters (treatment group, age at second bDMARD initiation, concomitant MTX use, timepoint and propensity score quintiles). The longitudinal data set included 61 CID events across the 6-month and 12 month assessments, corresponding to an events-per-variable ratio of 7.6. In addition, a sensitivity analysis was performed in patients with a polyarticular course of JIA without uveitis using the same propensity score-adjusted GEE model.

Missing data were assumed to be missing at random. Analyses were performed using complete-case analysis, including all available data for each variable without imputation. A significance threshold of p≤0.05 was applied. All analyses were performed using RStudio (V.2021.09.2+382 for macOS).

Results

Prescription patterns of second bDMARDs

127 patients (mainly women, 79%) fulfilling the inclusion criteria were enrolled in the present study. The main clinical features of included subjects stratified on the mechanism of action of second-line bDMARD (TNFi vs non-TNFi) are summarised in table 1. The most frequently used first-line TNFis were etanercept (67 patients; 53%) and adalimumab (50 patients; 39%) (figure 1A). In our cohort, 118 subjects out of 127 (94%) initiated the first bDMARD in combination with MTX. The overall median follow-up period was 11 years (IQR 6–16), and the median duration of the treatment with the first bDMARD was 1.52 years (IQR 0.64–3.88). The main reason for switching to a second bDMARD was ineffectiveness, reported in 109 out of 127 patients (86%), while adverse events accounted for 18 cases (14%). Active uveitis was the reason for switching in 25 out of 109 patients (23%). Adalimumab was the most frequently selected second-line TNFi (56 out of 95; 59%). A non-TNFi was chosen as the second-line bDMARD in 32 out of 127 patients (25%); in this case, tocilizumab was the most common choice (21 cases; 66%) (table 1).

Table 1. General features of the cohort and comparison between the TNFi and non-TNFi groups as the second bDMARDs.

Characteristic	Overall n=127*	TNFi n=95*	Non-TNFi n=32*	P value†
Sex				0.7
Female	100 (79%)	74 (78%)	26 (81%)
Age at diagnosis, years				0.09
Median (Q1–Q3)	4.7 (2.3–9.0)	5.0 (2.3–10.4)	3.0 (1.5–7.8)
JIA subtype				0.6
OligoJIA‡	64 (50%)	46 (48%)	18 (56%)
PolyJIA	53 (42%)	40 (42%)	13 (41%)
ERA/PsA	10 (7.9%)	9 (9.5%)	1 (3.1%)
Uveitis	56 (44%)	39 (41%)	17 (53%)	0.2
Follow-up overall, years				0.4
Median (Q1–Q3)	11 (6–16)	11 (6–17)	10 (6–14)
Age at onset of IbDMARD, years				0.06
Median (Q1–Q3)	8.8 (5.2–15.4)	9.3 (6.0–15.8)	6.4 (4.4–13.5)
Disease duration at onset of IbDMARD, years				0.3
Median (Q1–Q3)	2.1 (0.8–5.0)	2.0 (0.6–6.0)	2.3 (1.4–4.5)
Duration of IbDMARD, years				0.08
Median (Q1–Q3)	1.52 (0.64–3.88)	1.74 (0.67–4.00)	0.99 (0.49–2.67)
Year of IbDMARD onset				0.08
Median (Q1–Q3)	2016 (2010–2019)	2016 (2010–2019)	2016 (2013–2020)
Reason for switch				0.068
Arthritis	76 (60%)	56 (59%)	20 (63%)
Uveitis	25 (20%)	21 (22%)	4 (13%)
Arthritis and uveitis	8 (6.3%)	3 (3.2%)	5 (16%)
Adverse event	18 (14%)	15 (16%)	3 (9.4%)
Age at onset of IIbDMARD, years				0.06
Median (Q1–Q3)	13 (9–18)	14 (9–18)	12 (6–15)
Disease duration at onset of IIbDMARD, years				0.8
Median (Q1–Q3)	6.0 (2.7–10.2)	6.0 (2.5–10.5)	5.2 (3.0–9.6)
Follow-up after the onset of the IIbDMARD, years				0.1
Median (Q1–Q3)	3.6 (1.1–7.1)	4.1 (1.2–7.3)	2.2 (0.8–5.0)
Year of IIbDMARD onset				0.03
Median (Q1–Q3)	2019 (2014–2022)	2019 (2013–2022)	2021 (2017–2023)
IIbDMARD ongoing at last follow-up	69 (54%)	49 (52%)	20 (63%)	0.3
IIbDMARD				<0.001
Abatacept	9 (7.1%)	0 (0%)	9 (28%)
Adalimumab	56 (44%)	56 (59%)	0 (0%)
Certolizumab	2 (1.6%)	2 (2.1%)	0 (0%)
Etanercept	16 (13%)	16 (17%)	0 (0%)
Golimumab	12 (9.4%)	12 (13%)	0 (0%)
Infliximab	9 (7.1%)	9 (9.5%)	0 (0%)
Ixekizumab	2 (1.6%)	0 (0%)	2 (6.3%)
Tocilizumab	21 (17%)	0 (0%)	21 (66%)

^*n (%).

^†Pearson’s χ2 test; Fisher’s exact test; Wilcoxon rank-sum test.

^‡20 patients with extended oligoarthritis.

bDMARD, biological disease-modifying antirheumatic drug; ERA/PsA, enthesitis related arthritis/psoriatic arthritis; IbDMARD, first-line bDMARD; IIbDMARD, second-line bDMARD; JIA, juvenile idiopathic arthritis; OligoJIA, oligoarticular JIA; PolyJIA, polyarticular JIA; Q1, first quartile; Q3, third quartile; TNFi, TNF inhibitor.

Figure 1. Sankey diagrams illustrating: (A) TNFi-to-TNFi switches; (B) TNFi-to-non-TNFi switches. ABA, abatacept; ADA, adalimumab; ETA, etanercept; IFX, infliximab; O_TNFi, golimumab and certolizumab; non-TNFi, non-TNFα inhibitors; NN_TNFi: ixekizumab; TCZ, tocilizumab; TNFi, TNFα inhibitors.

Sankey diagrams illustrating TNFi-to-TNFi (panel A) and TNFi-to-non-TNFi switches (panel B) are shown in figure 1. Interestingly, at initiation of the second-line bDMARD concomitant MTX was less commonly prescribed as compared with first-line treatment (94% vs 53%; p<0.01). This difference was not affected by the second-line mechanism of action, as the rate was similar across the two subgroups (52% in the TNFi group vs 57% in the non-TNFi group received also MTX, p=0.6).

A non-TNFi was more frequently chosen when the switch was due to the concomitant activity of both arthritis and uveitis (3.2% for TNFi vs 16% for non-TNFi; p=0.02). The choice of the second-line bDMARD was also affected by the calendar year of prescription. As shown in figure 2, non-TNFi treatments were in median initiated in 2021 (IQR 2017–2023), compared with 2019 (IQR 2013–2022) for TNFi treatments (p=0.03). No further significant differences were observed in the selection of the second bDMARD. Among the 127 enrolled patients, clinical assessments at 6 months and 12 months were available for 108/127 (85.0%) and 79/127 (62.2%) patients receiving the first TNFi, for 78/95 (82.1%) and 69/95 (72.6%) patients receiving the second TNFi, and for 23/32 (71.8%) and 15/32 (46.9%) patients receiving a non-TNFi, respectively.

Figure 2. Violin plot showing the distribution of the calendar year of initiation for the second biologic DMARD (IIbDMARD), stratified by the specific bDMARD used. bDMARD, biological disease-modifying anti-rheumatic drug; TNFi, tumour necrosis factor alpha inhibitors.

Effectiveness of the second bDMARD: first-line TNFi versus second-line TNFi

A subanalysis on the effectiveness of the second-line TNFi after the failure of the first course was performed on 95 patients. At the initiation of the second-line TNFi, significantly fewer patients received MTX compared with the first TNFi course (79% vs 52%; p<0.01).

The rates of active arthritis, uveitis and elevated inflammatory markers at 6 months and 12 months were comparable between the first and second TNFi courses, as illustrated in figure 3 (panels A and B). Similarly, there were no significant differences in the achievement of CID between the first and second TNFi courses at both 6 months (66% vs 58%; p=0.5) and 12 months (49% vs 52%; p=0.75). The retention rate on TNFi at 1 year was higher with the second-line TNFi, even though the difference did not achieve statistical significance (66% vs 76%; p=0.2).

Figure 3. Comparison of outcome parameters. Panels (A) and (B): first-line versus second-line TNFi at 6 months and 12 months, respectively. Panels (C) and (D): TNFi versus non-TNFi at 6 months and 12 months, respectively. Percentages are based on patients with available clinical assessments at each follow-up. All individual variables were available for these patients. Clinical assessments at 6 months and 12 months were available for 108/127 (85.0%) and 79/127 (62.2%) patients receiving the first TNFi, 78/95 (82.1%) and 69/95 (72.6%) receiving the second TNFi, and 23/32 (71.8%) and 15/32 (46.9%) receiving a non-TNFi, respectively. Paired comparisons between the first and second TNFi treatment courses were performed using McNemar’s test. Comparisons between second-line TNFi and non-TNFi groups were performed using Pearson’s Chi-squared test and Fisher’s exact test. CRP, C reactive protein; ESR, erythrocyte sedimentation rate; NA, McNemar’s test not applicable (not discordant pairs); TNFi, TNF alpha inhibitor.

Multivariate GEE analysis, using CID as the dependent variable, identified a significant association with the reason for switching (table 2). The potential impact of the duration of the first TNFi treatment on the effectiveness of the second-line TNFi was also evaluated. Interestingly, the duration of the first TNFi did not influence the likelihood of achieving CID (table 2). Furthermore, the probability of achieving CID was similar at 6 months and 12 months.

Table 2. Multivariate logistic regression model with generalised estimating equations (GEE) of factors associated with CID after second-line TNFi therapy.

Variable	OR (95% CI)	P value
Sex Male	2.28 (0.73 to 7.14)	0.16
Age at II TNFi onset	1.01 (0.93 to 1.09)	0.83
JIA subtype PolyJIA*	1.32 (0.49 to 3.56)	0.59
JIA subtype ERA/PsA*	1.35 (0.26 to 3.56)	0.73
Reason switch uveitis†	6.00 (1.71 to 21.08)	0.01
Reason switch arthritis and uveitis†	2.01 (0.14 to 29.37)	0.61
Reason switch adverse event†	4.10 (1.03 to 16.37)	0.05
Duration I TNFi (years)	0.98 (0.79 to 1.21)	0.83
MTX with II TNFi yes	0.69 (0.27 to 1.73)	0.43
Timepoint CID at 12 months‡	0.97 (0.90 to 1.04)	0.34

^*Reference oligoarticular JIA.

^†Reference arthritis.

^‡Reference CID at 6 months.

CID, clinically inactive disease; ERA/PsA, enthesitis related arthritis/psoriatic arthritis; JIA, juvenile idiopathic arthritis; MTX, methotrexate; OligoJIA, oligoarticular JIA; PolyJIA, polyarticular JIA; TNFi, TNF inhibitor.

In a subanalysis of 71 subjects which excluded individuals with active uveitis at the time of the initiation of the second TNFi, the switch due to an adverse event tended to be associated with a more than threefold higher odds of reaching CID compared with the arthritis-driven switch in the adjusted model (OR 3.82, CI 0.95 to 15.33; p=0.06) (see online supplemental table 2).

Effectiveness of the second bDMARD: TNFi versus non-TNFi

The effectiveness of the second-line bDMARDs was compared between patients receiving a non-TNFi and those receiving another TNFi (figure 3C,D).

Although the median follow-up duration after the initiation of the second bDMARD course was not significantly different between the two groups, the TNFi group had a longer median follow-up (4.1 years (IQR 1.2–7.3) for TNFi vs 2.2 years (IQR 0.8–5) for non-TNFi; p=0.1). Interestingly, the median duration of the second bDMARD (censored at discontinuation or last follow-up) was significantly higher in the TNFi group compared with the non-TNFi group (2.2 years (IQR 0.7–5.3) vs 0.8 years (IQR 0.5–3.6); p=0.02). Conversely, the multivariable Cox regression analysis adjusted for propensity score quintiles did not confirm this association (see online supplemental table 3). The 1-year survival rate of the second bDMARD was significantly higher in the TNFi group than in the non-TNFi group (76% vs 48%; p=0.01).

At 6 months, CID rate was significantly higher in the TNFi group compared with the non-TNFi group (58% vs 26%; p<0.01) but this difference narrowed at 12 months (48% vs 29%; p=0.06) for patients still on treatment (69 patients on TNFi and 15 patients on non-TNFi).

The impact of the mechanism of action of second-line bDMARD on CID was also assessed by applying a multivariate logistic regression GEE analysis adjusted for propensity quintiles.

In this model, switching to a second TNFi was significantly associated with CID (table 3). In the subanalysis of 94 subjects without active uveitis at the time of the switch, the multivariable propensity quintile-adjusted GEE model confirmed that switching to a second TNFi was associated with CID (OR 3.59, CI 1.09 to 11.84; p=0.04) (see online supplemental table 4).

Table 3. Multivariate logistic regression model with generalised estimating equations (GEE), adjusted for propensity score quintiles, comparing TNFi versus non-TNFi as the second bDMARD and using CID as the dependent variable.

Variable	OR (95% CI)	P value	PS-adjusted OR (95% CI)	PS-adjusted P value
Switch type TNFi to TNFi*	4.65 (1.87 to 11.59)	<0.001	4.19 (1.66 to 10.61)	<0.001
Age at II bDMARD onset	0.98 (0.92 to 1.04)	0.48	0.98 (0.92 to 1.04)	0.43
MTX with II bDMARD yes	0.87 (0.40 to 1.87)	0.71	0.88 (0.41 to 1.90)	0.74
Timepoint CID at 12 months†	0.87 (0.63 to 1.22)	0.43	0.98 (0.93 to 1.03)	0.44

Given the small number of patients in the non-TNFi group, we repeated the analysis using four propensity score strata instead of five (quartiles). The results are reported in online supplemental tables 6 and 7.

^*Reference is TNFi to non-TNFi switch.

^†Reference CID at 6 months.

bDMARD, disease-modifying antirheumatic drugs; CID, clinically inactive disease; MTX, methotrexate; PS adjusted, propensity score quintile adjusted; TNFi, TNF inhibitor.

Furthermore, a sensitivity analysis was conducted in patients with a polyarticular course of JIA (including extended oligoarthritis) without a history of uveitis (n=62). Using the propensity quintile-adjusted GEE model, patients receiving a second TNFi remained more likely to achieve CID than those switching to a non-TNFi biologic (OR 6.48, 95% CI 1.09 to 38.72; p=0.04) (online supplemental table 5). Although the CI widened because of the reduced sample size, the direction and magnitude of the association were consistent with the primary analysis.

Discussion

The revision of bDMARD switching patterns over more than two decades in the clinical practice of two tertiary paediatric rheumatology centres in Italy provided several valuable insights into a crucial question in JIA clinical management: which is the optimal approach when the first-line TNFi fails?

According to our data, opting for a second-line TNFi could provide a better therapeutic option. Indeed, the comparison between the TNFi and non-TNFi groups revealed a higher CID rate in the TNFi group. In particular, switching to a TNFi was associated with more than a fourfold increase in the odds of reaching CID in the multivariate propensity quintile-adjusted GEE model. Importantly, this finding remained significant in the sensitivity analysis restricted to patients with a polyarticular course of JIA without a history of uveitis. The present study provides the first description in JIA literature of a possible superiority of second-line TNFi over non-TNFi, even though it should be acknowledged that previously available evidence is limited to two studies. An analysis from two UK registries between 2004 and 2019 focused on polyarticular-course patients without a history of active uveitis. The authors found no significant differences in 6-month outcomes or overall treatment duration between patients receiving a second TNFi and those switching to a different biologic class, although the rates of minimally active disease and American College of Rheumatology Pediatric response (ACR ped) 90% tended to be higher in the TNFi group.⁶ Additionally, a monocentric US study comparing TNFi and non-TNFi effectiveness reported significantly longer treatment persistence and greater clinical Juvenile Arthritis Disease Activity Score (cJADAS) reduction at 6 months in TNFi-treated patients receiving their first biologic.⁷ However, survival analysis of second-line therapy showed no significant differences between TNFi and non-TNFi users.⁷ Studies in RA comparing non-TNFi with TNFi as second-line therapy after TNFi failure have yielded conflicting results.^{12 13 18 19}

Importantly, in our cohort the clinical outcomes and retention rates with the second-line TNFi were similar to those observed with first-line TNFi. Patients who switched to a second TNFi due to active uveitis or adverse events were more likely to achieve CID in the adjusted model. Another consideration with important clinical implications relates to the lack of association between the duration of the first TNFi course and the achievement of CID with the second-line TNFi.

As expected, in our cohorts TNFis were the most frequently prescribed second-line bDMARDs; however, a shift towards non-TNFi agents was observed in recent years, consistent with findings from other studies.^{6 7} The primary reason for switching therapy was an inadequate response, as reported in other cohorts.^{6 8} Furthermore, patients with concomitant active arthritis and active uveitis were more prone to receive non-TNFi, a finding that may reflect the tendency for patients with higher disease activity to receive a non-TNFi agent.⁷ Regardless of the mechanism of action of the second-line bDMARD, we observed a significant reduction in concomitant MTX use compared with first-line biologic therapy (94% vs 53%; p<0.01). This finding is consistent with a study on adalimumab as either first or second bDMARD, showing a decrease in MTX co-administration during second-line course (79% vs 62%; p<0.01).²⁰

Several limitations of this study should be acknowledged, primarily its retrospective design. The heterogeneity of the study population, including the inclusion of patients with uveitis and enthesitis related arthritis/psoriatic arthritis, together with changes in treatment practices over the two-decade study period, represents an additional limitation that should be considered when interpreting the findings. Another limitation is the use of a modified Wallace CID definition and the lack of continuous disease activity measures. Real world data analysis presents challenges, including potential confounding by indication. To mitigate this, we used propensity scores stratified into quintiles and performed a subgroup analysis excluding patients with uveitis. Nonetheless, it is worth noting that the proportion of patients initially treated with etanercept who later switched due to uveitis was low (10 out of 67 patients receiving etanercept as first bDMARD). Additionally, due to limited sample sizes, we could not compare single bDMARDs as first-line and second-line treatments. We also lacked data on factors that may influence second-line bDMARD selection, such as antidrug antibody development or treatment adherence. Another limitation is the relatively small size of the non-TNFi group. Another limitation is the relatively small size of the non-TNFi group. Although the primary and sensitivity analyses yielded consistent results, the small sample size and substantial loss to follow-up, particularly at 12 months (15/32 subjects; 46.9%), may have reduced the precision and robustness of the comparative estimates, as reflected by the relatively wide CIs. Importantly, while all individual variables were available for patients with a follow-up clinical assessment, bias related to missing longitudinal data cannot be excluded. Therefore, these findings should be interpreted with appropriate caution and confirmed in larger cohorts with more complete longitudinal follow-up. The participation of two paediatric rheumatology tertiary centres, and the resulting patient selection, may impinge the generalisability of our findings. Nevertheless, patients with non-systemic JIA requiring a second-line bDMARD are typically managed in referral centres, supporting the relevance of our results to similar settings.

As a whole, our study highlights that a TNFi is a viable option even as second-line bDMARD, especially in cases of switching due to adverse events or active uveitis, regardless of the duration of the first TNFi course. In this cohort, TNFi agents were the most commonly chosen second-line biologic therapy despite the increasing use of non-TNFi agents in recent years, particularly in patients with more severe disease.

Surely our data need to be confirmed in larger prospective studies: the Trial of Sequential Medications After TNF Failure in JIA will provide crucial insights to guide optimal therapy sequencing in children with JIA who do not respond to their first TNFi (NCT06654882). Hopefully, in the future the identification of ‘theragnostic biomarkers’ will enable a shift from the current ‘trial-and-error approach’ to precision medicine in JIA.

Data availability statement

Data are available upon reasonable request.