To our knowledge, the cohorts of patients described in this retrospective study, both in the full analysis set and propensity score-matched cohorts, represent the largest dataset reported for the real-world persistence and effectiveness of UPA in patients with RA. By 1:1 propensity score matching patients prescribed UPA to those prescribed other JAKis or TNFis, this study offers a descriptive comparison of the persistence and effectiveness of UPA with other advanced therapies to treat RA in clinical practice, which may inform treatment decisions. Furthermore, this study adds evidence supporting the growing body of literature on the effectiveness of switching between drugs with distinct mechanisms of action and provides a novel analysis of treatment persistence and effectiveness in patients who switched from other JAKis to UPA.

Consistent with patterns of use of newer b/tsDMARDs [33,34,35], UPA was more frequently initiated in later lines of advanced therapy across the 3-year study window, with more patients prescribed UPA in the third or greater line of advanced therapy than patients prescribed other JAKis or TNFis. Additionally, a larger proportion of patients prescribed UPA were over 65 years old, with a longer time since diagnosis and symptom onset for patients prescribed UPA versus patients prescribed TNFis. Given that UPA was the third JAKi available in Australia, a larger proportion of patients prescribed UPA also had prior exposure to other JAKis than patients prescribed TNFis or other JAKis. The higher rates of monotherapy use observed in patients prescribed UPA and other JAKis at index than patients prescribed TNFis are consistent with previous studies reporting on the use of tofacitinib and TNFis in the US [36, 37] and in Australia from the OPAL dataset [38]. This trend may reflect patient preference for monotherapy or alignment with EULAR guidelines suggesting that JAKis and IL-6 inhibitors may have some advantages versus other bDMARDs in patients who cannot use csDMARDs as co-medication [7, 39, 40].

This study found that patients prescribed UPA had a median time on treatment of over 2 years, with longer persistence in earlier lines of therapy. Treatment with UPA improved disease activity, with an absolute increase in remission rate of 38% from index at 3 months. Across all lines of advanced therapy, at least 69% of patients achieved DAS28CRP(3) remission by 3 months post-index, with the lowest remission rates at index reported in first-line patients. Treatment persistence and effectiveness were similar between patients prescribed UPA as monotherapy and patients prescribed UPA in combination with csDMARDs, both in a propensity score-matched cohort and in unmatched subgroups from the full analysis set. The real-world data presented here align with outcomes observed in phase 3 clinical trials of UPA, including consistent efficacy when UPA was assessed as monotherapy or in combination with csDMARDs and in patients with or without prior bDMARD treatment [11,12,13, 41]. These results are also consistent with studies of the effectiveness and/or persistence of UPA and other JAKis in clinical practice. Observational studies conducted in Germany and Canada showed similar improvements in disease activity in patients receiving UPA as monotherapy or in combination with csDMARDs [24, 42]. Additionally, a retrospective analysis of unmatched cohorts of patients with RA from the Services Australia 10% PBS dataset reported persistence rates for the different available JAKis (tofacitinib, baricitinib, and upadacitinib), TNFis, tocilizumab, and other bDMARDs in the total patient population and in patients prescribed monotherapy or combination therapy with csDMARDs [26]. Although shown in a smaller sample size for patients prescribed UPA (n = 574), the 12-month treatment persistence was similar between patients using UPA as monotherapy or in combination with csDMARDs and was consistent with the results in this study using the OPAL dataset.

In this study of large, propensity score-matched cohorts from the OPAL dataset, the median time on therapy for patients prescribed UPA was 1.6 times longer than patients prescribed other JAKis and approximately two times longer than patients prescribed TNFis. Patients prescribed UPA showed higher rates of DAS28CRP(3) remission than patients prescribed other JAKis at 3- and 9-months post-index and patients prescribed TNFis at 3-, 9-, and 15-months post-index. These results are consistent with the previously described study from the Services Australia 10% PBS dataset, which showed a greater 12-month persistence rate for UPA than tofacitinib or TNFis, although the persistence rate for UPA was not statistically different from baricitinib [26]. Furthermore, a recent study using data from an insurance claims database in the United States also showed that patients prescribed UPA were significantly more likely to adhere to treatment and less likely to discontinue treatment during the first 12 months than patients who initiated tofacitinib, baricitinib, or adalimumab, regardless of whether patients had previously used a TNFi [20].

In some real-world studies of persistence and/or effectiveness of JAKis versus TNFis or other bDMARDs, smaller or no differences have been observed between treatment groups. For example, an analysis of the real-world efficacy and safety in propensity score-matched cohorts of patients prescribed JAKis from The Kansai Consortium for Well-being of Rheumatic Disease Patients (ANSWER) cohort database in Japan found no difference for retention, discontinuation, or measures of LDA and remission at 6 months between JAKis (tofacitinib [n = 127], baricitinib [n = 153], peficitinib [n = 29], UPA [n = 52]) [23]. Additionally, a large retrospective analysis from the JAK-pot registries found no difference in drug discontinuation between cohorts of patients prescribed JAKi versus TNFi or IL-6i [43]. However, due to the timing of the study windows, no patients prescribed UPA were included within the JAKi cohort of the JAK-pot study. Other studies utilizing the OPAL dataset have shown that tofacitinib had similar persistence to bDMARDs in the overall population and when used as monotherapy or in combination with csDMARDs [35, 38]. Another study from the OPAL dataset in patients with RA receiving their first advanced therapy showed that the persistence of baricitinib was not clinically meaningfully different from that of TNFis [34].

The findings from the OPAL dataset reported in this study and those from existing literature suggest that in a real-world setting UPA may have different outcomes for the treatment of RA than other JAKis, in this case tofacitinib and baricitinib, and TNFis [20, 26, 34, 35, 38]. While there are no head-to-head randomized clinical trials directly comparing the JAKis (baricitinib, tofacitinib, and upadacitinib), these molecules, in combination with MTX, have been individually compared head-to-head with adalimumab plus MTX in patients with inadequate response to MTX in phase three studies. In the Oral Rheumatoid Arthritis triaL (ORAL)-Strategy trial, tofacitinib was non-inferior to adalimumab on ACR50 responses at 6 months, with comparable efficacy between the treatment strategies, including for remission measures [44]. The RA-BEAM trial demonstrated that baricitinib was superior to adalimumab on ACR20 responses and mean change in DAS28-CRP at week 12, while remission rates were not significantly different between treatments [45]. The SELECT-COMPARE trial showed that responses in ACR50, pain, and physical function at week 12 were superior with UPA compared to adalimumab, and statistically significantly higher remission rates with UPA compared to adalimumab were observed through 5 years [15, 19]. Notably, each JAKi has a different chemical structure, resulting in distinct modes of binding within the catalytic cleft of the target JAK and giving rise to distinct pharmacological and pharmacodynamic characteristics [46, 47]. The available agents also have differing selectivity for JAK isoforms, as well as off-target effects against non-JAKs in vitro. Together, these different features of the individual JAKis may translate into unique clinical profiles for these molecules.

With multiple TNFis, JAKis, and other bDMARDs available, when a patient is not achieving a desired treatment target, a decision must be made between switching within a treatment class or to a treatment with a different mechanism of action. To date, the evidence to guide this clinical decision is limited, particularly when assessing switches to newer molecules such as UPA. A previous analysis from the OPAL dataset, which evaluated the use and pattern of switching amongst JAKis in a study window of October 2015 to September 2021, demonstrated that switching from TNFis to JAKis and between different JAKis was frequent, but UPA was only available for 17 months within that study window [27]. With a larger dataset and a longer follow-up time for UPA-treated patients, this current OPAL study showed that patients switching from TNFi to TNFi had shorter median time on treatment and slower response to treatment (assessed by the proportion of patients in DAS28CRP(3) categories over time) than patients who switched from TNFis to UPA. These results support a recent study from the Adelphi RA Disease Specific Programme™ using data from Germany, France, Italy, Spain, the United Kingdom, Japan, Canada, and the United States captured between May 2021 and January 2022 which found that patients with RA who switched from first-line TNFi to second-line UPA demonstrated better clinical outcomes of remission, no pain, and complete adherence versus patients who switched between TNFis or patients who switched to a drug with another mechanism of action [28].

In this analysis, patients who switched to UPA from either a TNFi or other JAKi, displayed similar median time on treatment. Achievement of remission over time was also similar between these two subgroups, but different remission rates at index led to different absolute treatment effects. The higher rate of remission at index observed within the JAKi to UPA subgroup compared to the TNFi to UPA subgroup suggests that a proportion of patients may have changed treatment from JAKi as their first advanced therapy to UPA for reasons other than efficacy. Regardless, the response to UPA following the use of other JAKis further supports that individual JAKis may differ in their clinical profiles, potentially as a result of their differing pharmacological profiles, including selectivity [28].

In this current study, between 21 and 35% of patients prescribed UPA, other JAKis, and TNFis in the full analysis set were reported in DAS28CRP(3) remission at index. However, the remission rate was lowest when a given treatment was prescribed as a first-line advanced therapy (UPA: 9.7%; other JAKi: 13.0%; TNFi: 10.8%). In previous OPAL studies, 17% of patients prescribed baricitinib and 16% of patients prescribed tofacitinib were in DAS28-ESR remission at index [38], with 23% of patients prescribed tofacitinib as monotherapy in remission at index [35]. In Australia, patients may switch between therapies if clinically relevant improvements in disease activity are not achieved, or for other reasons such as toxicity or patient preference. Furthermore, patients in DAS28CRP(3) remission may still have active disease with greater than six swollen joints or increased patient global assessments, even if the other variables used to calculate the score are nearly normal [48].

One of the limitations of this study, and those of other real-world datasets, is the often high levels of missing data. Across the three treatment groups in the full analysis set, only half of patients had sufficient information available at index to allow for the calculation of a DAS28CRP(3) score. As a result, to achieve successful propensity score matching, a missing category was included in the DAS28CRP(3) covariate following the missing covariate indicator method approach, which outperforms complete case analysis but may still introduce bias, as there may be some discrepancy in the type of patient with a missing value between the two treatment groups [31]. However, in this study, this approach resulted in cohorts of well-matched patients, including those with a missing DAS28CRP(3) value, as detailed above. Importantly, in the propensity score-matched cohorts, the proportion of patients in DAS28CRP(3) remission at index was similar between the treatment groups. Furthermore, patient pain was reported infrequently and patient global assessment, although more routinely captured, had significant levels of missing data, thus limiting the ability to interpret the contribution of these patient-reported outcomes to initial disease activity. The extent of missing data suggests that patient-reported outcomes and formal disease activity measures may not be routinely recorded at clinical visits in Australia, perhaps because patient-reported outcomes and composite disease scores such as DAS28CRP are not required for access to PBS-funded advanced therapies. The COVID-19 pandemic occurring during the study window (May 2020 to June 2023), and subsequent shifts to telehealth by many clinicians could have also impacted the recording of these measures.

An additional limitation includes the use of concomitant corticosteroids and other medications such as NSAIDs were reported only at treatment index, without information about the dose or frequency of prescription of these and csDMARDs, including MTX, recorded in this study. Therefore, while the proportion of patients with concomitant use of csDMARDs was reported across time, it is not possible to interpret how the dose of these therapies, corticosteroids, and NSAIDs changed with time, nor how these additive therapies may have contributed to differences in the persistence or effectiveness of UPA, other JAKis and TNFis. Furthermore, the OPAL dataset captures data on the physician-prescribed medication; it does not capture data on whether the patient fills the prescription or takes the medication as prescribed. Finally, the OPAL dataset was not set up to formally collect robust safety outcomes and therefore no safety data are presented within this study. Additionally, it was not possible to accurately report on reasons for discontinuation because terms for capturing reasons for treatment discontinuation within the patient’s EMR were amended during the study window. These changes prevent a clear interpretation of the reasons for discontinuation during the study period.

A significant strength of this study is the large dataset with over 2600 patients prescribed UPA, which allowed for a robust analysis of persistence and effectiveness and a rigorous approach to propensity score matching when comparing treatments or treatment strategies. Notably, the inclusion of index year as a matching variable was done to minimize the impact of potential changes in prescribing behavior due to UPA being the newest therapy available, any influence of the COVID-19 pandemic, and the influence of the release of data from the tofacitinib ORAL surveillance study, which triggered concerns regarding risk of major cardiovascular events and malignancy with JAKi use in patients aged 65 years and older, and led to changes in labels in the United States in December 2021 [49] and Europe in January 2023 [50]. The labels for UPA, baricitinib, and tofacitinib in Australia were updated at the end of April 2023, which was after the end of the sampling window for this study [51, 52]. Notably, 41% of patients prescribed UPA and 40% of patients prescribed other JAKis in this study were aged 65 years or older. Other strengths of the OPAL dataset include that the data are derived from the EMRs of almost a third of practicing rheumatologists in Australia and that data collection is based on opt-out patient consent, thus avoiding issues in datasets that rely on patient enrollment in registries or insurance claims which may not always be representative of the overall rheumatology population [53, 54]. In addition, the findings in this study reflect prior observations from the 10% PBS dataset, which represents a random 10% of prescriptions across Australia [26]. Thus, the data contained in the OPAL dataset may be considered representative of Australian rheumatology practice. Furthermore, unlike in some clinical settings elsewhere in the world, the universal access to reimbursement for advanced therapies with the same patient cost regardless of treatment choice and the ability to readily switch treatments when required in Australia may reduce the potential for these factors to influence persistence outcomes, further strengthening the reliability of this robust dataset [53, 54].