Real-World Safety of JAK Inhibitors Tracks Closer to TNF Inhibitors Than Expected
A large Spanish registry analysis found that Janus kinase inhibitors and tumour necrosis factor inhibitors had broadly similar drug persistence across rheumatoid arthritis, psoriatic arthritis and axial spondyloarthritis, while safety signals differed by diagnosis and treatment context. The BIOBADASER 3.0 cohort included 6,826 patients treated between 2015 and 2023, offering a long clinical window into how these targeted therapies perform outside the rigid structure of randomized trials.
The comparison matters because JAK inhibitors have entered routine care with both promise and caution attached to them. They can be effective after previous treatment failures and are often used in patients with more advanced disease, but concerns about infections and other adverse events have shaped prescribing habits. This registry analysis places those concerns into a real-world setting where older age, comorbidity burden, prior treatment exposure and background glucocorticoid use all influence outcomes at the bedside.
Mixed Diagnostic Cohorts, Unequal Baseline Risk
The study population was not evenly balanced across disease types. Roughly 52% had rheumatoid arthritis, 25% had psoriatic arthritis and 23% had axial spondyloarthritis. Treatment exposure was dominated by TNF inhibition, which accounted for 86% of use, while JAK inhibitors were more often reserved for older patients with longer disease duration, more comorbid conditions, later treatment lines and more frequent monotherapy.
Those baseline differences are clinically important. Patients moved to JAK inhibition later in their disease course often arrive with a heavier burden of prior failure and accumulated risk, including steroid exposure and the consequences of persistent inflammation. In that setting, crude comparisons can mislead; a safer-looking therapy may simply be reaching sicker patients, or a riskier-looking therapy may be carrying the weight of more complex disease. The analysis therefore depended on multivariate models to separate treatment effect from case mix as much as the data allowed.
The mean treatment duration was 2.2 ± 2.0 years for TNF inhibitors and 1.8 ± 1.5 years for JAK inhibitors. Even before formal persistence modeling, those numbers suggest that both drug classes remained in use for meaningful periods in day-to-day care, with neither class showing an obvious collapse in real-world continuation.
Infection and Gastrointestinal Events Favored the TNF Comparator in Rheumatoid Arthritis
Safety signals were not uniform across diagnoses. In patients with rheumatoid arthritis, the incidence rate ratio for all infections and gastrointestinal events was higher among those treated with JAK inhibitors. Herpes zoster also appeared more frequent with JAK inhibition in this group, fitting the biologic expectation that interruption of intracellular cytokine signaling can alter antiviral immune surveillance.
The mechanism is plausible. JAK pathways transmit signals from multiple cytokines that coordinate innate and adaptive immunity, so inhibition can shift host defense in ways that may favor viral reactivation and some infectious complications. TNF blockade carries its own infectious risk, but in this dataset the rheumatoid arthritis population treated with JAK inhibitors showed the more conspicuous burden of infection-related and gastrointestinal adverse events. In clinical practice, that should sharpen attention to age, prior infections, steroid exposure and vaccination history before starting therapy.
A single sentence captures the practical problem: the same class that can simplify dosing and improve convenience may also expose vulnerable patients to a different adverse-event profile.
Persistence Stayed Similar Despite Different Prescribing Patterns
Drug persistence was nearly superimposable between the two classes. At 1, 2 and 3 years, persistence was 69%, 55% and 45% for TNF inhibitors, compared with 68%, 54% and 45% for JAK inhibitors. Those curves suggest that, in routine practice, both treatment strategies can remain viable over time despite the different patient profiles that led to their use.
In multivariate analysis, however, JAK inhibitors were associated with a lower probability of discontinuation overall, with a hazard ratio of 0.85 and a 95% confidence interval of 0.78 to 0.92. Concomitant conventional synthetic disease-modifying antirheumatic drugs also reduced discontinuation risk, with a hazard ratio of 0.90 and a 95% confidence interval of 0.84 to 0.96. Persistence worsened with glucocorticoid use, comorbidities, higher disease activity and later treatment lines.
Persistence is not a cosmetic endpoint. In chronic inflammatory arthritis, remaining on therapy often reflects a balance of efficacy, tolerability, convenience and patient preference, filtered through the realities of access and follow-up. A drug can look successful on paper while being abandoned because of recurrent infections, inadequate symptom relief or the cumulative burden of adverse effects. Conversely, a therapy that is deployed more selectively may appear to persist well simply because clinicians already anticipate who is most likely to tolerate it.
Why the Registry Signal Matters for Daily Prescribing
The results support a cautious, diagnosis-specific view of JAK inhibitor use rather than a broad class warning applied indiscriminately to every patient. In rheumatoid arthritis, the higher rates of infection, herpes zoster and gastrointestinal adverse events warrant careful risk stratification, especially in older patients and those receiving glucocorticoids. For psoriatic arthritis and axial spondyloarthritis, the persistence data suggest comparable durability to TNF inhibitors, though the study design cannot remove all confounding from treatment selection.
This is where real-world evidence adds texture that trial data often cannot provide. Randomized studies are ideal for efficacy under controlled conditions, but registries capture the clinical sorting that occurs every day: which patient gets a JAK inhibitor after several prior biologics, which patient remains on a TNF inhibitor because it has already proved tolerable, and which patient drops off therapy because comorbidity shifts the risk-benefit balance. The BIOBADASER data reflect that messiness rather than hiding it.
A direct clinical takeaway emerges from the pattern of discontinuation: comorbidity, steroid use and greater disease activity are not background noise, they are the forces most likely to derail long-term treatment success.
Limits of Observational Comparisons Remain Built In
Registry studies can measure what happens in practice, but they cannot fully equalize the patients who receive different drugs. Prescribing of JAK inhibitors to older, more treatment-experienced and more comorbid patients raises the possibility of channeling bias, even with multivariate adjustment. Event adjudication, residual confounding and differences in monitoring intensity can also shape apparent safety patterns.
Even so, the overall picture is clinically coherent. TNF inhibitors and JAK inhibitors showed similar persistence in this large cohort, while rheumatoid arthritis patients on JAK inhibitors experienced more infections and herpes zoster, along with more gastrointestinal adverse events. The differences in discontinuation risk by diagnostic group and background treatment reinforce a familiar point in inflammatory arthritis care: the right drug choice depends as much on the patient in front of the clinician as on the disease label attached to the chart.