Safety of JAK inhibitors versus other biologic therapies following anti-TNF failure in patients with rheumatoid arthritis Authors: Lorenzo Di Luozzo, Mariangela Salvato, Francesca Frizzera, Kiren Khalid, Margherita Zen, Andrea Doria, Roberta Ramonda, Alessandro Giollo Published: 2025 ABSTRACT AbstractBackgroundTreatment of rheumatoid arthritis (RA) frequently requires Janus kinase inhibitors (JAKi) and biologic DMARDs (bDMARDs) with other mechanisms of action (OMA) after tumour necrosis factor inhibitors (TNFi) failure. Comparative safety data between JAKis and OMA in TNFi-inadequate responders (TNFi-IR) remain limited.ObjectivesTo evaluate and compare safety profiles of JAKis versus OMA in TNFi-IR RA patients.MethodsA retrospective cohort study followed 305 RA patients (163 JAKis; 142 OMA) previously treated with TNFi for up to 24 months. Incidence rates (IR) of adverse events (AEs) per 100 patient-years (PY) and incidence rate ratios (IRR) adjusted for disease activity and previous bDMARD exposure were calculated using Poisson regression.ResultsAt baseline, OMA patients had higher disease activity (DAS28-CRP = 4.5 vs. 3.8), whereas comorbidities and other baseline characteristics were similar. Overall, AE incidence was higher with OMA (IR 111.6/100 PY) compared to JAKis (IR 65.3/100 PY; adjusted IRR 1.67, 95% CI 1.22–2.31). Serious AEs occurred more frequently with OMA (IR 11.7/100 PY) versus JAKis (IR 7.9/100 PY; adjusted IRR 0.41, 95% CI 0.18–0.95). Infection rates were comparable; herpes zoster occurred exclusively in the JAKi group during the initial year. Venous thromboembolism, cardiovascular events, and malignancies were rare and similar between groups.ConclusionIn RA patients with inadequate response to TNFis, JAKis and OMA showed comparable safety profiles over 24 months. Serious AEs were numerically higher with OMA; herpes zoster was more frequent with JAKis. No new significant safety signals emerged.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13075-025-03686-6.Keywords: Rheumatoid arthritis, JAKi, BDMARDs, Safety, Infections Introduction The management of rheumatoid arthritis (RA) may require the utilisation of biologic and targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs), with tumour necrosis factor inhibitors (TNFi) as the first-line treatment for most patients. However, TNFi may fail and be discontinued within 3–5 years in up to 50–60% of RA patients [1–3]. The latest update of the EULAR recommendations for the management of RA does not prioritise any specific class of b/tsDMARDs following the failure of the first bDMARD, including cycling to another TNFi or switching to a bDMARD with other mechanisms of action (OMA) than TNF-blockade or to a Janus kinase inhibitor (JAKi) [4]. However, in the latest 15 years, the treatment strategy after TNFi failure has evolved to cycling to JAKi or OMA rather than trialling a second TNFi [5], with evidence in favour of switching for better outcomes [6–9]. JAKi appear to be associated with an increased risk of venous thromboembolism (VTE), cardiovascular events (CVE), and cancer compared to TNFi [10]. The EULAR guidelines emphasise that JAKi should be used with caution in patients with certain risk factors, including those aged 65 years or older, current or past long-term smokers, individuals with a history of atherosclerotic cardiovascular disease, and those with other malignancy risk factors [4, 11]. In such cases, JAKi should only be prescribed if no suitable alternative treatments are available. Aside from the indirect comparisons coming from trials of JAKi/OMA that included TNFi-IR patients, comparative safety data from RCTs between JAKi and OMA in RA patients who are TNFi-IR are limited [12]. Observational safety data from large international registries and population studies are difficult to interpret due to the heterogeneity of cohorts comprising patients who initiated treatment with JAKi or OMA following several mixed lines of b/tsDMARDs [13–17]. While several studies have addressed the efficacy and safety of individual b/tsDMARDs, less is known about how different post-TNFi strategies compare directly in real-world practice. The objective of this study was not to test whether failure of a TNFi biologically modifies subsequent safety outcomes, but rather to perform a comparative safety analysis of the two main strategies commonly adopted in routine care after TNFi failure: JAKi and OMA. These agents are associated with distinct theoretical safety concerns—such as herpes zoster reactivation, thromboembolic risk, and malignancy with JAKi, versus serious infections and infusion-related complications with OMA—which directly influence therapeutic decisions in TNFi-inadequate responders. Methods Study aim and design This monocentric retrospective study aimed to assess the safety profile of JAKi (tofacitinib, baricitinib, upadacitinib, or filgotinib) compared to OMA (abatacept, tocilizumab, and rituximab) in a cohort of RA patients with inadequate response to at least one TNFi followed at a tertiary care centre. The duration of the study was 24 months. Data collection Patients who initiated b/tsDMARDs within the study period were identified from the local registry for b/tsDMARDs prescriptions. The following data were extracted from the registry and electronic medical records: demographics (e.g., sex, age at baseline, and body mass index); clinical and serological characteristics (e.g., symptom duration, presence of joint erosions, extra-articular manifestations, seropositivity for rheumatoid factor and anti-citrullinated protein antibodies, and baseline C-reactive protein levels); and treatment history (number of prior bDMARD classes used, prior or concomitant use of csDMARDs, and reasons for discontinuation of previous therapies). Extra-articular manifestations included cutaneous, pulmonary, ocular, haematological, cardiac, neurological, and systemic manifestations. Comorbidities were assessed at baseline and included cardiovascular disease (myocardial infarction, cerebrovascular disease, heart failure, peripheral artery disease), diabetes, osteoporosis, chronic pulmonary disease (asthma, chronic obstructive pulmonary disease), solid and haematological neoplasms, hypertension, and dyslipidaemia. Disease activity and therapy (use of glucocorticoids, csDMARDs, and/or JAKi) were assessed at follow-up visits, as well as drug persistence, and the main reason for drug discontinuation was recorded at each follow-up visit and categorised as primary or secondary failure, adverse events, intolerance, loss to follow-up, remission, or other reasons. The patients who experienced AEs or treatment failure discontinued treatment according to clinical practice guidelines. Inclusion and exclusion criteria Inclusion criteria were age ≥ 18 years old; a confirmed diagnosis of RA based on the 2010 ACR/EULAR classification criteria; previous failure of at least one TNFi; initiated therapy with either a JAKi (first course) or OMA between 1 March 2018 and 1 March 2024. Per national reimbursement policies [18], JAKi treatment is approved only for patients with an inadequate response to conventional synthetic DMARDs (csDMARDs) and at least one prior TNFi. Therefore, bDMARD-naïve patients were excluded from the analysis. Additional exclusion criteria were use of a non-TNFi bDMARD as the first biologic therapy, as this choice is often dictated by comorbidities or extra-articular manifestations; overlap syndromes; and follow-up of less than six months. Moreover, patients treated with anakinra were not included, as anakinra is used mainly for patients with RA and adult-onset Still’s disease in our centre. Study outcomes The primary outcome of the study was the incidence rate (IR) of any adverse event (AE) per 100 person-years (100 PY) exposure at 24 months. The secondary outcomes were the IR of specific AEs at 24 months and the IR between 0 and 6, 6–12, and 12–24 months for all AEs and each AE of interest. Study definitions The main exposure was defined as the period from the initiation of JAKi or OMA therapy until the earliest occurrence of treatment discontinuation, loss to follow-up, death, or the end of the study period. We defined AEs as any unintended medical occurrence associated with the study drugs. Two independent assessors (AD, RR) systematically adjudicated AEs at each follow-up visit in both JAKi- and OMA-treated patient groups. Assessed events included serious adverse events (SAEs), infections (including COVID-19), cardiovascular events (CVE), venous thromboembolism (VTE), gastrointestinal disorders, major adverse cardiovascular events (MACE), and malignancies (excluding non-melanoma skin cancers, NMSC). We defined MACE as myocardial infarction, stroke, or death due to cardiovascular disease. The definition of CVE encompassed both the onset of conditions predisposing to cardiovascular disease (e.g., hypertension, diabetes, and hypercholesterolaemia) and the worsening of pre-existing risk factors, as well as the occurrence of cardiovascular disease and MACE. We defined SAEs as any undesirable medical occurrence in a patient that results in death, is life-threatening, requires hospitalisation or prolongation of hospitalisation, leads to significant disability or permanent damage, or causes a congenital anomaly or birth defect. It also includes any other medically important condition that, if left untreated, could undermine the patient’s health or require medical intervention to prevent serious outcomes. Statistical analysis Continuous variables were expressed as means with standard deviations (SD) or medians with interquartile ranges (IQR), as appropriate based on data distribution. Categorical variables were expressed as absolute frequencies and percentages. Comparisons between treatment groups were performed using the Student’s t-test or Mann-Whitney U test for continuous variables, depending on normality assessed by the Shapiro-Wilk test. Categorical variables were compared using the Chi-square test or Fisher’s exact test, as appropriate. Incidence rates (IR) were expressed as the number of patients experiencing ≥ 1 event divided by the total patient-years (PY) of exposure and were reported per 100 PY. For each patient, follow-up time was defined as the cumulative duration (in years) from first drug administration until the earliest of (i) last study visit, (ii) treatment discontinuation, or (iii) death. Total PY were obtained by summing individual follow-up times within each treatment cohort (JAKi vs. OMA). Exact 95% confidence intervals (CI) for IR were calculated with the Byar’s approximation to the Poisson distribution. Crude incidence rate ratios (IRR) between JAKi and OMA (reference: JAKi) were derived as the ratio of the respective IR; 95% CI for IRR were computed on the logarithmic scale using standard Poisson variance and exponentiated. To adjust for potential confounding, multivariable Poisson regression models were fitted for each adverse event. The dependent variable was the count of patients with ≥ 1 event, the primary exposure was treatment group (JAKi = reference), and an offset term equal to the log of individual PY was included to account for differential follow-up. Covariates entered a priori were baseline age, sex (female = reference), disease duration, DAS28-CRP (continuous), and the number of prior biologic classes (ordinal). Robust (sandwich) variance estimators were used to guard against mild overdispersion. Adjusted IRRs with 95% Wald CI and corresponding two-sided P-values were reported. Time to treatment discontinuation due to AEs was analyzed using Kaplan–Meier survival estimates, with time defined as the cumulative number of observation days until AE-related discontinuation or censoring at the last recorded visit for patients who remained on therapy. Differences between the JAKi and OMA groups were compared using the log-rank test. All analyses were performed with Python 3.11 (pandas 2.2, statsmodels 0.14). Power analysis A post-hoc power analysis based on the Wald test for the log-incidence-rate ratio indicated that the current cohort (124 patients with ≥ 1 AE in the JAKi arm and 181 in the OMA arm, corresponding to 190.3 and 162.6 patient-years, respectively) was more than adequately powered to detect the observed difference in overall AE incidence. The standard error of the log-IRR was 0.12, yielding a test statistic of Z = 4.61 for the observed IRR of 1.71. This translates into an achieved two-sided power of approximately 99% at α = 0.05. Conversely, with the same number of events, the study would retain 80% power to detect an IRR of ≈ 1.39 or greater (or ≤ 0.72), but power would fall below 60% for differences smaller than ~ 30%. Thus, the sample size and accumulated follow-up are sufficient to identify clinically meaningful differences in total AE rates, although the study remains underpowered to discern very modest (≤ 30%) rate changes between the two treatment groups. Statistical significance was defined as p < 0.05. Results Disease characteristics Overall, we enrolled 305 patients with RA who initiated treatment with JAKis (n = 163) or OMA (n = 142). Compared to patients treated with OMA (Table 1), those receiving JAKis had slightly lower mean (SD) disease activity (DAS28 3.81 (1.00) vs. 4.45 (1.19); p < 0.001), fewer erosions (n = 129 (80.1%) vs. 122 (88.4%); p = 0.038), were less seropositive (n = 118 (73.3%) vs. 115 (83.3%); p = 0.023), and were less refractory to two or more classes of bDMARDs (p < 0.001). CPR levels at baseline were similar between the two groups. Table 1Disease characteristics of the study populationMean (SD) or n (%)All(n = 305)JAKis(n = 163)OMA(n = 142)p-valueDisease duration, mean (SD)16.33 (9.87)17.14 (10.13)15.38 (9.51)0.124Erosions, n (%)251 (83.9)129 (80.1)122 (88.4)0.038Extra-articular manifestations, n (%)55 (18.4)35 (21.7)20 (14.5)0.144RF or ACPA positive, n (%)233 (77.9)118 (73.3)115 (83.3)0.023Failed bDMARD classes 2208 (69.6)77 (47.8)131 (94.9)< 0.001 ≥ 391 (30.5)84 (52.2)7 (5.1)DAS28CRP, mean (SD)4.10 (1.13)3.81 (1.00)4.45 (1.19)< 0.001CRP mg/L baseline, mean (SD)15.12 (22.68)13.96 (21.93)16.48 (23.53)0.341Use of csDMARD at baseline, n (%)95 (31.8)46 (28.6)49 (35.5)0.246PDN mg/d, mean (SD)4.64 (5.07)4.39 (5.49)4.93 (4.54)0.355RF Rheumatoid factor, ACPA Anti−citrullinated peptide antibodies, CRP C−reactive protein, DAS28CRP Disease activity score in 28 joints, PDN Prednisone Disease characteristics of the study population RF Rheumatoid factor, ACPA Anti−citrullinated peptide antibodies, CRP C−reactive protein, DAS28CRP Disease activity score in 28 joints, PDN Prednisone Overall, the commonest reason for the failure of previous bDMARD treatments was inadequate disease control (77.3%), while 22.7% of patients discontinued therapy due to adverse events. A greater proportion of patients in the OMA group were treated with glucocorticoids (76.1%) during the study period, compared to the JAKi group (66.5%), although this difference was not statistically significant (p = 0.089). The use of concomitant csDMARD was similar across the two groups (n = 49 (35.5%) and 46 (28.6%), respectively). Baseline safety risk assessment Our cohort was predominantly female in both groups (Table 2), with a mean age of 58 years and a BMI in the upper normal range (mean (SD) 24.6 (4.2) kg). The two groups were comparable as it pertains to baseline CVD risk, with a third of patients 65 years old or older and half being current or previous smokers; a small number had prior MACE, with higher numbers in the OMA group (p = 0.050). Given the study period, we included several patients who initiated JAKi or OMA before the European Medicines Agency (EMA) recommended measures to minimise the risk of serious side effects with JAKi for chronic inflammatory disorders; only 12 patients (four JAKi and eight OMA) were enrolled afterwards. About 70% of patients in both groups fulfilled at least one EMA criterion for minimising the risk of adverse events with JAKi, with no significant differences between treatment groups. Table 2Safety assessment at baseline in JAKi- and OMA-treated groupsRisk assessmentOverallJAKiOMAp-valueFemales, n (%)260 (87.0)145 (90.1)115 (83.3)0.121Age, mean (SD)58.96 (13.03)57.85 (12.23)60.25 (13.84)0.113 ≥ 65 years, n (%)114 (37.4%)54 (33.1%)60 (42.3%)0.127Smoking (ever), n (%)144 (47.2%)70 (42.9%)74 (52.1%)0.138Diabetes mellitus, n (%)23 (7.7)14 (8.7)9 (6.5)0.627Hypertension, n (%)79 (26.4)40 (24.8)39 (28.3)0.234Dyslipidaemia, n (%)45 (15.1)20 (12.4)25 (18.1)0.112MACE, n (%)13 (4.3%)3 (1.8%)10 (7.0%)0.050ASCVD, n (%)20 (6.7)8 (5.0)12 (8.7)0.292BMI, mean (SD)24.61 (4.18)24.37 (4.16)24.89 (4.21)0.294 ≥ 35 kg/m2, n (%)7 (2.3%)3 (1.8%)4 (2.8%)0.709COPD or ILD, n (%)31 (10.2%)19 (11.7%)12 (8.5%)0.463Prior cancer excluding NMSC, n (%)17 (5.6%)11 (6.7%)6 (4.2%)0.479EMA criteria, n (%)*211 (69.2%)106 (65.0%)105 (73.9%)0.119ASCVD Atherosclerotic cardiovascular disease, BMI Body mass index, COPD Chronic obstructive pulmonary disease, CVD Cardiovascular disease, EMA European medicines agency, ILD Interstitial lung disease, OP Osteoporosis, MACE Major adverse cardiovascular events, NMSC Non−melanoma skin cancer* EMA measures to minimise the risk of serious side effects with JAKi suggest that JAKi should be used in the following patients only if no suitable treatment alternatives are available: those having one or more of the following: age ≥65 years old, long−time smoker, cardiovascular risk factors (e.g., diabetes mellitus, hypertension, dyslipidaemia), prior cardiovascular disease or prior history of solid or haematological malignancy excluding non−melanoma skin cancer Safety assessment at baseline in JAKi- and OMA-treated groups ASCVD Atherosclerotic cardiovascular disease, BMI Body mass index, COPD Chronic obstructive pulmonary disease, CVD Cardiovascular disease, EMA European medicines agency, ILD Interstitial lung disease, OP Osteoporosis, MACE Major adverse cardiovascular events, NMSC Non−melanoma skin cancer * EMA measures to minimise the risk of serious side effects with JAKi suggest that JAKi should be used in the following patients only if no suitable treatment alternatives are available: those having one or more of the following: age ≥65 years old, long−time smoker, cardiovascular risk factors (e.g., diabetes mellitus, hypertension, dyslipidaemia), prior cardiovascular disease or prior history of solid or haematological malignancy excluding non−melanoma skin cancer Overall safety profile of JAKi and OMA Over the entire observation period, 124 of 163 patients treated with JAKi experienced ≥ 1 AE, and a total of 181 adverse events were recorded among the 142 patients treated with OMA, corresponding to crude AE incidence rates of 65.3 and 111.6 per 100 PY, respectively (Fig. 1). When the difference was modelled with adjusted Poisson regression, OMA retained a significantly higher risk, yielding an adjusted IRR of 1.67 (95% CI, 1.22–2.31, p = 0.002). Inspection of the time-split data (Supplementary table S1) showed that the divergence between groups was already apparent in the first six months of therapy, with rates of 38.7 versus 24.6 per 100 patient-years for OMA and JAK inhibitors, respectively; it widened modestly during the second semester (92.6 vs. 84.8) and became numerically greatest beyond the first treatment year. Due to the very small amount of residual follow-up accumulated in later time frames (< 2% of total person-time within 18–24 months), IRs beyond 12 months were unstable and should not be interpreted as a true increase in event frequency. Incidence rates (IRs) of adverse events (AEs) in JAKi and OMA group Kaplan–Meier analysis of treatment persistence showed longer time to AE-related discontinuation for JAKi-treated patients compared to OMA-treated patients (log-rank p = 0.0125), with higher proportions of JAKi users remaining on therapy without AE-related cessation over follow-up (Fig. 2). Kaplan–Meier survival curves for time to treatment discontinuation due to adverse events (AEs) in patients treated with JAK inhibitors (JAKi) or biologicals with other mechanism of action (OMA) Serious adverse events Rates of SAE were 7.9 for JAKi vs. 11.7/100 PY for OMA, yielding an adjusted IRR of 0.41 (95% CI 0.18–0.95; p = 0.038). Temporal IRs (JAKi vs. OMA,/100 PY; Supplementary table S1) were 5.2 vs. 8.9 (0–6 months), 6.3 vs. 11.3 (6–12 months), 12.5 vs. 11.4 (12–18 months), and 33.2 vs. 61.8 (18–24 months). One death occurred in a patient treated with tocilizumab (OMA group, period 0–6 months; IR = 1.49/100 PY); none were recorded under JAKi. SAEs were more common among older patients (Supplementary table S2). We recorded 15 SAE in 125.3 patient-years among patients ≥ 65 years old versus 19 SAE in 227.7 patient-years among those < 65 years old, yielding an age IRR of 2.30 (95% CI 1.17–4.53; p = 0.016). Composite CVEs were numerically more frequent in the older subgroup (IRR = 1.48), but the 95% CI crossed unity. Ever-smokers showed a significant excess in overall infections: 42 events in 68.7 patient-years among smokers versus 157 events in 284.2 PY among non-smokers, corresponding to an IRR of 1.76 (95% CI 1.08–2.87; p = 0.023). The increase was attenuated and lost significance when COVID-19 episodes were excluded (IRR = 1.38, 95% CI 0.77–2.47). Infections Infections were the most frequently reported AEs in our cohort (Table 3). The most common infections were upper respiratory tract and Coronavirus Disease 2019 (COVID-19), followed by urinary tract infections, gastroenteritis, and others. The overall incidence of infections was comparable (21.0 vs. 22.8/100 PY; adjusted IRR = 0.68). Time-stratified IR (JAKi vs. OMA,/100 PY): 13.0 vs. 19.3 (0–6 months), 20.3 vs. 24.5 (6–12 months), 25.0 vs. 22.7 (12–18 months), and 116.2 vs. 41.2 (18–24 months). Rates of infections not COVID-19-related followed a pattern similar to total infections: 9.1 vs. 16.4, 17.2 vs. 20.7, 20.0 vs. 19.9, and 49.8 vs. 41.2/100 PY across successive half-year windows. The incidence of HZ infection was confined to JAKi in the first year (1.3 and 3.1/100 PY for 0–6 months and 6–12 months; no HZ infection under OMA), then 5.0 vs. 2.8/100 PY in months 12–18; no cases were reported thereafter. Table 3Incidence rate ratios (IRR) with 95% CI and Poisson regression resultsAEEvents JAKiIR JAKi (95% CI)Events OMAIR OMA (95% CI)Crude IRR (95% CI)Adj. IRR (95% CI) p Death00.00 (0.00–1.94)10.61 (0.02–3.43)NANA0.638Infections4021.02 (15.01–28.62)3722.75 (16.02–31.36)0.92 (0.59–1.44)0.68 (0.41–1.13)0.140Herpes Zoster52.63 (0.85–6.13)10.61 (0.02–3.43)4.27 (0.50–36.57)6.42 (0.62–66.05)0.118Infections excluding COVID-192915.24 (10.20–21.88)3119.06 (12.95–27.06)0.80 (0.48–1.33)0.64 (0.36–1.14)0.132CVE178.93 (5.20–14.30)127.38 (3.81–12.89)1.21 (0.58–2.53)1.64 (0.70–3.80)0.254Gastrointestinal AEs42.10 (0.57–5.38)10.61 (0.02–3.43)3.42 (0.38–30.58)6.23 (0.60–64.74)0.125MACE10.53 (0.01–2.93)21.23 (0.15–4.44)0.43 (0.04–4.71)0.30 (0.02–5.30)0.412VTE31.58 (0.33–4.61)10.61 (0.02–3.43)2.56 (0.27–24.64)5.18 (0.42–63.65)0.199Cancer21.05 (0.13–3.80)10.61 (0.02–3.43)1.71 (0.15–18.85)0.18 (0.00–9.10)0.393SAE157.88 (4.41–13.00)1911.68 (7.03–18.25)0.67 (0.34–1.33)0.41 (0.18–0.95)0.038Adjusted for age, sex, DAS28, disease duration, and number of previous bDMARDs Incidence rate ratios (IRR) with 95% CI and Poisson regression results Adjusted for age, sex, DAS28, disease duration, and number of previous bDMARDs Cardiovascular events Composite CVE occurred at 8.9 vs. 7.4/100PY overall (adjusted IRR = 1.64). MACE were rare: none with JAKi during the first year, one OMA event in each semester (1.5–1.9/100 PY), then 2.5/100 PY with JAKi and none with OMA at 12–18 months; no MACE were recorded thereafter. Venous thromboembolism Three VTEs occurred under JAKi (4.7/100 PY, months 6–12) and one under OMA (1.5/100 PY, 0–6 months); no events were registered later, giving a non-significant adjusted IRR of 0.20 (0.02–2.64). Malignancy Two cancers were observed with JAKi and one with OMA. The cancers were solid tumours with no haematological malignancies recorded. The very low counts produced an unstable adjusted IRR (54.0, 4.6–631); no reliable treatment effect can be inferred. Gastrointestinal AEs Four GI events occurred in the JAKi arm (2.6/100 PY in the first semester; 2.5 and 16.6/100 PY in the third and fourth windows) versus one in OMA (1.5/100 PY, first semester). The adjusted IRR was 0.06 (0.00–2.19; p = 0.12). No diverticulitis or bowel perforation was observed in the study period. Sensitivity analyses A multivariable Poisson regression was performed to determine whether the previously observed difference in AEs between treatment groups persisted after additional adjustment for age, smoking status, or both (Supplementary table S3). No high-risk subgroup (patients aged ≥ 65 years old, smokers, or both) exhibited a statistically significant difference in overall or specific adverse-event rates between OMA and JAKi therapy. Among patients aged ≥ 65 years old, the adjusted IRR for any-grade AE was 1.18 (95% CI 0.81–1.71); infections showed an IRR of 2.08 (0.94–4.57) and SAEs 2.21 (0.78–6.30), but all confidence intervals spanned unity. In current or former smokers the pooled IRR for all AEs was 1.13 (0.80–1.60); infections 1.39 (0.63–3.04); and SAE 2.70 (0.64–11.42). In the composite subgroup “≥ 65 years old or smoker,” the IRR for all AEs was 1.09 (0.83–1.45); infections 1.60 (0.86–2.97); and SAE 2.13 (0.83–5.49). For rarer outcomes (death, cancer, VTE, MACE, HZ, CVEs), the models were dominated by sparse counts and produced highly unstable point estimates with very wide intervals, precluding reliable inference. Among patients who had failed only one TNFi (n = 152; 96 OMA and 63 JAKi), AEs were numerically more frequent during OMA treatment, although the difference did not reach statistical significance. Specifically, 96 events occurred over 194.4 patient-years in the OMA group (IR 49.4 per 100 PY, 95% CI 40.0–60.3) compared with 47 events over 117.5 PY in the JAKi group (IR 40.0 per 100 PY, 95% CI 29.4–53.2). The corresponding crude incidence-rate ratio (IRR) was 1.24 (95% CI 0.87–1.75; p = 0.24), and the adjusted IRR—accounting for age, disease duration, and DAS28-CRP—was 1.17 (95% CI 0.91–1.50; p = 0.23). For serious adverse events (SAE), 17 episodes were reported in the OMA arm and 6 in the JAKi arm, corresponding to IRs of 8.7 and 5.1 per 100 PY, respectively. The crude IRR was 1.71 (95% CI 0.68–4.34; p = 0.26), while the adjusted model yielded an IRR of 1.60 (95% CI 0.64–3.96; p = 0.31). Discussion In our cohort of TNFi-IR RA patients, both the JAKi and OMA treatment regimens were associated with a similar overall rate of AEs. While there were some differences in the timing and frequency of AEs between the two groups, with a higher early incidence of infections observed in the OMA group, HZ in the JAKi group, and increased MACE, VTE, and malignancies after 12 months of treatment with JAKis, most CIs substantially overlapped between groups. Safety data comparing JAKi to OMA in TNFi-IR RA patients are lacking. A large prospective cohort study found no evidence of a difference between rituximab, abatacept, or tocilizumab for the mean duration of survival without death, the presence of cancer or serious infections, or MACE among TNFI-IR RA patients [19]. Our data indicate that TNFi-IR RA patients have more risk factors for CVD, comprising a subgroup potentially at high risk for AEs with both JAKi and OMA. A closer look at high-risk groups (those older than 65 years or smokers) showed that JAKi did not confer a statistically significant excess or reduction in overall, infectious, or SAEs within elderly, smoker, or combined high-risk subsets. The incidence rates of MACE and CVD were similarly low in both the JAKi and OMA treatment groups, even in patients at highest risk (elderly and smokers). It is well established that high disease activity in patients with RA, which is associated with elevated systemic inflammation, contributes to an increased risk of MACE irrespective of changes in lipid profile during treatment [20]. However, bDMARDs treatment may help mitigate said risk vs. those receiving csDMARDs only or remaining untreated for long periods. For instance, a higher DAS28-CRP was associated with increased MACE risk in bDMARD non-users (HR 1.21, 95% CI 1.07 to 1.37) but not in users (HR 0.69, 95% CI 0.40 to 1.20) [21]. Despite the potent anti-inflammatory effects of JAKi, their use has been associated with increased risk of MACE and cancer compared to TNFi, mainly in patients with a history of ASCVD [22]. Evidence for an increased risk of CVD with JAKi when compared to OMA is mixed [14, 23–27], with most of them showing no effect. Importantly, for most such studies, it is difficult to overcome the indication bias, as JAKis are commonly used for more refractory RA or after failure of other bDMARDs (typically a TNFi). For instance, one study using data from the FDA Adverse Event Reporting System (FAERS) [28] compared the risk of adverse events between a JAKi group (baricitinib, tofacitinib, upadacitinib) and a bDMARD group (including OMA rituximab and tocilizumab but also TNFi). Compared to the bDMARD group, JAKis were associated with an increased reporting of VTE [n = 1,393, adj.ROR = 2.11 (1.97–2.25)], stroke [n = 973, adj.ROR = 1.25 (1.16–1.34)], ischaemic heart disease [IHD, n = 999, adj.ROR = 1.23 (1.13–1.33)], peripheral oedema [n = 2699, adj.ROR = 1.22 (1.17–1.28)], and tachyarrhythmias [n = 370, adj.ROR = 1.15 (1.00–1.33.00.33)]. Most events occurred in the first year after treatment initiation. When different JAKis were compared, VTE, stroke, and ischaemic heart disease were more frequently reported with upadacitinib and baricitinib than tofacitinib. When stratified by age category, all safety signals were statistically significant in patients aged ≤ 65 years old. An increased risk of VTE has been observed in patients with RA and high disease activity when compared to those in remission (RR 2.03, 95% CI 1.73–2.38) [29]. In our study, IRRs were corrected for baseline DAS28 to account for the effect of disease activity on the risk of VTE and were similar between JAKi and OMA. Whilst pre-existing VTE risk factors were comparable between OMA and JAKi, we did not observe differences in IR between the two groups, even in the high-risk patient subgroups. Several studies have reported an increased dose-dependent incidence of VTE in patients with pre-existing thromboembolic risk factors treated with JAKi, particularly among users of tofacitinib [10, 29, 30] and baricitinib [29], compared to TNFi or placebo. Similar findings for tofacitinib and baricitinib have been replicated in large population-based studies [29, 31, 32]. Conversely, data from the US CORRONA registry [30] have shown similar rates of VTE in patients treated with tofacitinib and those receiving TNFi. It remains unclear whether JAKis have a direct causal role in the development of thromboembolic events or whether these events are attributable to the elevated baseline risk associated with chronic inflammation or patient comorbidities. Moreover, data on VTE for OMA are scarce [12, 26]. There is an increased risk of malignancies — particularly lymphoma and lung cancer — in patients with RA [33]. Our study found a low number of malignancies (excluding NMSC) with similar IR between JAKi and OMA, which probably hinders the ability to draw any definitive conclusions. Nevertheless, mounting evidence indicates that the risk of malignancy could be higher in RA patients exposed to JAKi for several months or years [10, 22, 34], also compared to TNFi [35]. Regarding OMA, there is conflicting evidence for abatacept [36–41], whereas rituximab and tocilizumab are usually seen as safe as it pertains to their malignancy risk [42, 43]. The ORAL Surveillance trial found an HR for tofacitinib of 1.93 (1.22–3.06) from month 18 of treatment onwards compared to TNFi for malignancies, excluding NMSC [44]. Indeed, there appears to be a subgroup of RA patients at increased risk of cancer, specifically those who are older than 65, heavy smokers, or those with a prior history of ASCVD [22]. We found no differences between JAKi and OMA in the IRs of malignancies among higher-risk subgroups, such as the elderly, smokers, or individuals with multiple comorbidities; however, estimating cancer risk can be a challenging task in a retrospective study, and the short follow-up period (24 months) might be insufficient to detect a safety signal for cancer [44]. Moreover, since our study population was not comparable in treatment exposure characteristics to that of the ORAL Surveillance trial, this finding should be interpreted with caution and highlights the need for further studies. The risk of infections and serious infections appears to be comparable across the various b/tsDMARDs, yet higher than csDMARDs [45, 46]. However, infections — particularly HZ and gastrointestinal — in patients with RA tend to be more frequent with JAKi vs. TNFi [25]. Importantly, the risk of infections with b/tsDMARD use increases with subsequent treatment lines, making it difficult to ascertain the actual infection risk associated solely with JAKi compared to OMA. Our study revealed that among TNFi-IR patients, the rates of infections were similar between JAKi and OMA, albeit with some important distinctions according to specific infectious agents. We found a higher IR for HZ infections among JAKi recipients compared to OMA, as previously reported by several studies. In fact, a higher incidence of HZ reactivation has been consistently reported in association with all JAKis in all age groups [11]. Nonetheless, SAEs—which were mainly driven by serious infections in our study—occurred significantly more commonly in OMA-treated patients compared to JAKi. Although an increased risk of infections has been observed in patients over 75 years of age treated with JAKi [11], we found an increased risk of SAEs among patients older than 65 undergoing OMA. Several factors could explain this difference, which deserves further investigation in larger and more diverse cohorts. The point estimates for infections and SAEs across OMA should be interpreted cautiously in light of the limited number of events and resultant imprecision. Study strengths and limitations Strengths include the comprehensive data collection with systematic adjudication of AEs, adjustment for key confounders in multivariable Poisson models, and time-stratified incidence rate calculations over a 24-month follow-up. The inclusion of a homogeneous population previously treated with TNFi enhances comparability between treatment groups, and the use of robust sensitivity analyses further supports the validity of the findings. Limitations include the retrospective and monocentric design, which may have limited the generalisability of our findings and introduced a selection bias. Our study was underpowered to detect differences in rare events such as MACE, VTE, or malignancy, as reflected by the wide confidence intervals around point estimates. While the overall adjusted comparisons should be interpreted as reliable, the late-interval estimates should be seen as unstable exploratory descriptors. Unmeasured confounding, including frailty, prior infections, and lifestyle factors, cannot be fully excluded and may have influenced treatment allocation and AE risk. A further limitation is that our dataset did not allow systematic linkage of AEs across treatment lines, and we therefore could not evaluate whether an event observed during TNFi therapy predisposed to recurrence under JAKi or OMA. Addressing this clinically relevant question will require larger multicentre cohorts with detailed longitudinal annotation of AEs, which would enable exploration of patient-level susceptibility patterns beyond the population-level incidence rates reported here. Conclusion RA patients previously exposed to TNFi and subsequently treated with JAKis or OMA have a low overall risk of AEs, at least when observed for up to 24 months. Supplementary Information Supplementary Material 1. Acknowledgements None. Authors’ contributions Lorenzo Di Luozzo: Conceptualization, Methodology, Investigation, Writing - Original Draft; Writing - Review & Editing; Mariangela Salvato: Conceptualization, Methodology, Investigation, Writing - Original Draft; Writing - Review & Editing; Francesca Frizzera: Investigation; Kiren Khalid: Investigation; Margherita Zen: Investigation, Writing - Review & Editing; Roberta Ramonda, Andrea Doria: Supervision, Writing - Review & Editing; Alessandro Giollo: Conceptualization, Methodology, Investigation, Writing - Original Draft; Writing - Review & Editing, Supervision, Project administration. Funding Alessandro Giollo and Mariangela Salvato have received financial support from the Ministry of Health POS-T3-AN for the publication of this work. Data availability The data supporting this study cannot be shared, as they were collected with consent for use only in the current study or for predefined research purposes. Declarations Ethics approval and consent to participate This study was approved by the Ethics Committee of the Central - Eastern Veneto Area (CET-ACEV) with number 6007/AO/24. Consent for publication Not applicable. Competing interests Lorenzo Di Luozzo has no known competing financial interests or personal relationships. Mariangela Salvato has received grants from the Ministry of Health. Francesca Frizzera has no known competing financial interests or personal relationships. Kiren Khalid has no known competing financial interests or personal relationships. Margherita Zen has received consulting and speaking fees from AbbVie, Eli Lilly, GSK, Pfizer, UCB, and AstraZeneca. Andrea Doria has received travel grants, consulting and speaking fees, and advisory board fees from Eli Lilly, GSK, Pfizer, Celgene, Baxalta, and AstraZeneca. Roberta Ramonda has received travel grants, consulting and speaking fees, and advisory board fees from AbbVie, Novartis, Johnson & Johnson, Eli Lilly, MSD, Pfizer, UCB, Amgen, Celltrion, and AstraZeneca. Alessandro Giollo has received consulting and speaking fees from AbbVie, Novartis, Eli Lilly, UCB, Medac, Boehringer, and Galapagos, as well as funding from the Ministry of Health. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Lorenzo Di Luozzo and Mariangela Salvato shared first co-authorship. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The data supporting this study cannot be shared, as they were collected with consent for use only in the current study or for predefined research purposes.