Differential long-term retention of biological disease-modifying antirheumatic drugs in patients with rheumatoid arthritis by age group: an observational study

Background. The effectiveness and safety of biological disease-modifying antirheumatic drugs (bDMARDs) by age group (<65, 65–74, and ≥ 75 years) are uncertain. We examined retention rates reecting the effectiveness and safety of bDMARDs in actual clinical practice for clarifying optimal therapeutic strategies for rheumatoid arthritis (RA) by age groups. Methods. Data of patients who were treated with tumor necrosis factor inhibitors (TNFi), abatacept (ABA), and tocilizumab (TCZ) between February 2011 and April 2017 were retrospectively collected from an observational registry of RA patients. A total of 1,362 patients were enrolled, of which 695 were aged <65 years, 402 were aged 65–74 years, and 265 were aged ≥ 75. Primary outcome was the drug retention rate in adjusted data using inverse probability of treatment weighting based on generalized propensity scores. Results. In patients aged <65 years, three-year retention rates of TNFi, ABA, and TCZ were 43%, 47%, and 69%, respectively (ABA versus TCZ, p = 0.017; TNFi versus TCZ, p = 0.002). In patients aged 65–74 years, three-year retention rates of TNFi, ABA, and TCZ were 44%, 53%, and 60%, respectively (TCZ versus TNFi, p = 0.034). In patients aged ≥ 75 years, three-year retention rates for TNFi, ABA, and TCZ were 38%, 63%, and 58%, respectively (ABA versus TNFi, p = 0.017). Conclusions. We found that the effectiveness and safety of TCZ were maximized in patients aged <75 years and that patients aged ≥ 65 years were the most suited candidates for ABA therapy. The use of therapeutic strategies appropriate to each age group might improve the outcomes of bDMARD therapy for RA. respectively. The was retention rates of ABA, and TCZ 47%, 69%, retention TCZ


Background
Because the ratio of elderly people has markedly increased, therapeutic strategies in patients with elderly rheumatoid arthritis (RA) are emerging and important concerns to be addressed. Especially in Japan which is rapidly transitioning into a super-aged society, with approximately 30% in the country being at least 65 years of age [1]. Consequently, maintaining the psychosomatic health of people aged 65-74 years and encouraging them to participate in social activities are some important issues. A new de nition of true elderly people as those aged ≥ 75 years has been proposed [2]. In such a super-aged society, patients with RA are also increasingly becoming older. Thus, Japan may be a good model for developing therapeutic strategies against RA in a super-aged society that the world will face in the future.
For RA, methotrexate (MTX) and biological disease-modifying antirheumatic drugs (bDMARDs) are key options for achieving treat-to-target goals [3]. This stands true for treating elderly RA patients. However, because elderly patients exhibit age-related decreases in activities of daily living (ADL) and organ function, the incidence of adverse events (AEs) increases inevitably [4]. Accordingly, conventional synthetic DMARDs (csDMARDs) are typically restricted for use. Conversely, because bDMARDs are less likely to be affected by renal and hepatic disorders, appropriate use of bDMARDs for treating RA may be more important in elderly patients than in young patients. Thus, the optimal therapeutic strategy using bDMARDs for elderly and young RA patients must be reexamined. Thus, the effectiveness and safety of bDMARDs should be compared and analyzed across age groups.
Drug retention rate re ects its effectiveness and safety [5]. In elderly patients, retention rates of individual bDMARDs and of bDMARDs by age group have been investigated [6][7][8]. Because frailty-an important risk factor for impaired ADL, hospitalization, and death-rapidly increases in elderly people aged ≥ 75 years [9,10], the effectiveness and safety of bDMARDs may differ even among the elderly such as between patients aged 65-74 years and those aged ≥ 75 years. Therefore, the effectiveness and safety of bDMARDs by age group (< 65, 65-74, and ≥ 75 years) should be compared and analyzed. However, there has been no comparative analysis of retention rates of bDMARDs among these age groups. Moreover, observational studies on retention rates of individual drugs have predominantly been conducted in actual clinical practice; as such selection bias represents a problem for comparisons between drugs.
In this study, selection bias was reduced by applying inverse probability of treatment weighting (IPTW) based on generalized propensity score [11,12] and retention rates of bDMARDs were compared by age groups to clarify the optimal therapeutic strategies using bDMARDs.  Figure S1). Patients who received bDMARD therapy between February 2011 (MTX doses of up to 16 mg/week were approved for adult RA in Japan) and April 2017, who were aged ≥ 20 years, and who were followed up for at least 6 months were included in this study. Finally, 1,362 patients were enrolled, of which 695 were aged < 65 years, 402 were aged 65-74 years, and 265 were aged ≥ 75 years. Patient data were retrospectively collected from the FIRST registry. RA was diagnosed based on the 1987 American College of Rheumatology Classi cation Criteria or 2010 American College of Rheumatology/European League against Rheumatism classi cation criteria [13][14][15]. The study was approved by the ethics review board of the university, and written informed consent was waived because of the retrospective study design. bDMARD treatment. bDMARDs were prescribed to RA patients who were not adequately responsive to csDMARDs therapies. bDMARDs included both intravenous and subcutaneous agents and were administered based on the guidelines of the Japan College of Rheumatology.

Methods
Drug retention, discontinuation, and clinical effectiveness. The primary outcome was the drug retention rate in adjusted data analysis; secondary outcomes included reasons for discontinuation in non-adjusted data analysis and clinical effectiveness in non-adjusted and adjusted data analysis. Drug retention was analyzed according to the duration of treatment until drug discontinuation at the physician's judgment. Reasons for discontinuation were investigated using medical records and classi ed into discontinuation because of remission, effect insu ciency, AEs, or others (patient choice, economic, unspeci ed cause, and so on). Regarding discontinuation because of remission, physicians judged the state of the absence of disease activity based on composite measures such as the disease activity score using 28 joint counts (DAS28), the clinical disease activity index (CDAI), or the simple disease activity index [16,17]. Moreover, regarding effect insu ciency, physicians judged increase, return, or no change in disease activity based on composite measures. AEs were further classi ed according to Common Terminology Criteria for Adverse Events version 5.0 [18]. Because CDAI precisely re ects clinical response to bDMARDs, including TCZ [19], clinical effectiveness was evaluated as CDAI [20,21].
Statistical analysis. In analysis of baseline characteristics, summary statistics were presented using proportions for categorical data and medians and interquartile ranges for continuous data. In analysis of reasons for bDMARDs discontinuation, summary statistics were presented using of the number of incidents and incidence. Kruskal-Wallis and chi-square tests were used to assess differences among groups. In non-adjusted data analysis, the Kaplan-Meier method was used to assess drug retention rates.
In adjusted data analysis, IPTW was used based on generalized propensity scores. Generally, covariate adjustments using propensity scores are performed for comparisons between two groups. For comparisons among multiple groups, the adjustment cannot be performed without any changes. The expanded concept of generalized propensity scores is applied to match multiple groups [11,12]. In this study, for covariate adjustment in the three groups of patients treated with TNFi, ABA, or TCZ, age, disease duration, gender, history of bDMARD use, MTX dose, glucocorticoid (GC) dose, tender joint count, swollen joint count, patient global assessment, evaluator global assessment, Health Assessment Questionnaire-Disability Index (HAQ-DI), C-reactive protein (CRP) level, erythrocyte sedimentation rate (ESR), and rheumatoid factor (RF) level were used as baseline covariates, and generalized propensity scores were estimated using multinomial logistic regression. Area under the curve of generalized propensity scores was > 0.7 for each bDMARD. Balance following covariate adjustment was examined, and covariate adjustment was con rmed. bDMARD retention rates were analyzed using the Kaplan-Meier method and Cox proportional hazards model with adjustments using generalized propensity scores [12].

Results
Retention rates of bDMARDs in non-adjusted data. Table 1 summarizes characteristics of all patients and age groups. Patients aged ≥75 years exhibited higher incidence of advanced-stage RA, higher ABA and GC usage, higher preexisting lung disease prevalence, lower MTX usage, stronger in ammatory responses (e.g., CRP and ESR), higher RF levels, higher CDAI scores, and higher HAQ-DI scores than patients aged <65 and 65-74 years. Table 2 presents patient characteristics according to bDMARD usage. Additional le 1: Tables S1-3 present characteristics of each age group according to bDMARD usage. TNFi and TCZ usage decreased with age, whereas ABA usage remained comparable across age groups. TNFi-treated patients exhibited shorter disease duration, higher proportions of patients with Steinbrocker's stage I and II RA and biologics-naïve patients, higher MTX usage, and lower pre-existing lung disease prevalence than ABA-and TCZ-treated patients. ABA-treated patients showed longer disease duration, higher proportion of patients with Steinbrocker's stage III and IV RA, and higher pre-existing lung disease prevalence than those treated with other bDMARDs. TCZ-treated exhibited lower MTX usage, higher GC usage, higher CRP and ESR levels, and higher CDAI scores than those treated with other bDMARDs.
Risk factors for bDMARD discontinuation in non-adjusted data. In all patients, the risk factors for drug discontinuation included age (hazard ratio [ Table S7). bDMARD usage itself was a common risk factor for discontinuation in all age groups.
Retention rates of bDMARDs in adjusted data. Although non-adjusted data analyses suggested great differences in drug retention rates across bDMARD type, the retention rates were affected by patient selection bias. Thus, patient selection bias was minimized by applying IPTW based on generalized propensity score, and the retention rate of each bDMARD was compared among the age groups (Fig. 2). Differences in baseline characteristics were adjusted by IPTW (Additional le 1: Table S8). In all patients, three-year retention rates of TNFi, ABA, and TCZ were 43%, 51%, and 65%, respectively. The retention rate of TCZ was signi cantly higher than rates of ABA and TNFi (ABA versus TCZ, p = 0.021; TNFi versus TCZ, p < 0.001; Fig. 2A). In patients aged <65 years, three-year retention rates of TNFi, ABA, and TCZ were 43%, 47%, and 69%, respectively. The retention rate of TCZ was signi cantly higher than rates of ABA and TNFi (ABA versus TCZ, p = 0.017; TNFi versus TCZ, p = 0.002; Fig. 2B). In patients aged 65-74 years, three-year retention rates of TNFi, ABA, and TCZ were 44%, 53%, and 60%, respectively. The retention rate of TCZ was signi cantly higher than that for TNFi (p = 0.034; Fig. 2C). In patients aged ≥75 years, three-year retention rates for TNFi, ABA, and TCZ were 38%, 63%, and 58%, respectively. The retention rate of ABA was signi cantly higher than that of TNFi (p = 0.017; Fig. 2D). No statistical analysis was performed for TCZ usage data because of the small sample size. Overall, retention rates of TCZ were higher in patients aged <65 and 65-74 years, and retention rates of ABA increased with age.
Reasons for bDMARD discontinuation in non-adjusted data. Reasons for the discontinuation of each bDMARD were analyzed by age (Table 3 and Additional le 1: Table S9). In patients aged <65 years, discontinuation because of remission was more frequent following treatment with TNFi (16.9%) and less frequent following treatment with TCZ (3.4%). The prevalence of effect insu ciency was higher following treatment with ABA (22.0%) and comparable between treatments with TNFi (14.5%) and TCZ (15.5%). AE incidence was lower following treatment with ABA (4.9%). Particularly, there was no incidence of skin and subcutaneous tissue disorders. Similarly, in patients aged 65-74 years, discontinuation because of remission was more frequent following treatment with TNFi (10.5%) and the prevalence of effect insu ciency was lower following treatment with TCZ (8.5%). AE incidence was comparable among the three drugs. Likewise, in patients aged ≥75 years, discontinuation because of remission was more frequent following treatment with TNFi (9.9%). The prevalence of effect insu ciency was comparable among the three drugs, and AE incidence was lower following treatment with TCZ (6.9%) and ABA (4.3%). Among AEs, the incidence of infections was lower following treatment with ABA. Overall, the rate of discontinuation because of remission was higher following treatment with TNFi and AE incidence was lower following treatment with ABA in all age groups.
Retention rate of each bDMARD in adjusted data after excluding patients who discontinued treatment because of remission. Among the reasons for bDMARD discontinuation, the rate of discontinuation because of remission was higher following treatment with TNFi in each age group. Thus, patients who discontinued treatment because of remission were excluded from the analysis. The background characteristics of the remaining patients according to age and bDMARD usage are presented in Additional le 1: Tables S10-13. Retention rates were analyzed (Additional le 1: Figure S3) after adjusting for differences in baseline characteristics using IPTW (Additional le 1: Table S14). In all patients, three-year retention rates of TNFi, ABA, and TCZ were 48%, 55%, and 69%, respectively. The retention rate of TCZ was signi cantly higher than rates of ABA and TNFi (ABA versus TCZ, p = 0.029; TNFi versus TCZ, p = 0.001; Additional le 1: Figure S3A). In patients aged <65 years, three-year retention rates of TNFi, ABA, and TCZ were 52%, 52%, and 73%, respectively. The retention rate of TCZ was signi cantly higher than rates for ABA and TNFi (ABA versus TCZ, p = 0.025; TNFi versus TCZ, p = 0.011; Additional le 1: Figure S3B). In patients aged 65-74 years, threeyear retention rates of TNFi, ABA, and TCZ were 48%, 55%, and 63%, respectively. The retention rate of TCZ was signi cantly higher than that of TNFi (p = 0.048; Additional le 1: Figure S3C). In patients aged ≥75 years, three-year retention rates of TNFi, ABA, and TCZ were 42%, 65%, and 63%, respectively. The retention rate of ABA was signi cantly higher than that of TNFi (p = 0.019; Additional le 1: Figure S3D). No statistical analysis was performed for TCZ data because of the small sample size. Overall, similar results to those shown in Fig. 2 were obtained even after patients who discontinued treatment because of remission were excluded.

Discussion
In this study, after reducing selection bias by applying IPTW based on generalized propensity score, the long-term retention rates of bDMARDs were examined by age group in actual clinical practice. Consequently, the following two novel points were revealed. First, in all age groups, the highest retention rate was observed for TCZ. Second, in patients aged at least 65 years, the retention rates of ABA and TCZ were similar. Moreover, secondary assessment revealed the following two points. First, in all age groups, the lowest rate of discontinuation because of AEs was observed with ABA. Second, TNFi treatment improved disease activity more effectively and was associated with the highest rate of discontinuation because of clinical remission in all age groups.
In this study, the retention rate of TCZ was high in all age groups, particularly in patients aged < 65 years, and remained high in those aged ≥ 65 years. In a meta-analysis of studies on RA, TCZ and MTX combination was identi ed as the best intervention [22]. TCZ showed higher retention rates than TNFi and ABA in other studies [23,24], which supports the results of this study. The retention rates of ABA and TCZ were similar in patients aged ≥ 65 years, and this similarity more pronounced in those aged ≥ 75 years. As a possible explanation for these ndings, it is important that ABA showed the lowest discontinuation rate because of AEs in all age groups in this study and that the degree of improvement in disease activity was comparable among the drugs in patients aged ≥ 75 years. Regarding safety and age, an all-case postmarketing surveillance of bDMARDs in Japan revealed that age (≥ 65 years) contributed to severe infection onset with all biopharmaceuticals except for ABA [25][26][27][28]. Long-term safety of ABA [29,30] has been reported, and guidelines of the British Society for Rheumatology recommend ABA as the rst-line treatment for patients at a risk of infection [31]. The e cacy of ABA is similar to that of TNFi [32,33] and TCZ [34].
Elderly RA patients often present higher disease activity and greater functional impairment than young patients [35]. In recent years, immunosenescence has attracted much attention, and risk of in ammation and autoimmune diseases increases with age [36]. Immunosenescence is characterized by the lack of CD28 in T-cells [36], and CD28-negative Tcell count increases with age [37]. Lack of CD28 in CD4-positive T-cells is associated with chronic autoimmune diseases, including RA [36]. ABA therapy reduces CD28-negative CD4-positive T-cell counts, and this reduction in CD28-negative T-cell count is associated with the responsiveness of RA as assessed by DAS28 based on CRP level [38,39]. Thus, ABA may be suitable for improving immune dysregulation in elderly RA patients. However, previous studies at our institution as well as others reported that high CD28-negative CD4-positive T-cell count increases the risk of decreased responsiveness to ABA therapy [40], and the responsiveness to ABA therapy improves with decreasing pretreatment CD28-negative T-cell count [41]. That is, some elderly people may be at a risk of acquiring treatment resistance because of increasing CD28-negative T-cell count with increasing age. These previous studies suggest that heterogeneity increases in elderly RA patients. Thus, drug selection based on both safety and treatment responsiveness is preferable.
In terms of disease activity, non-adjusted data analysis demonstrated that the degree of improvement was comparable among the three bDMARDs, although adjusted data analysis identi ed differences in the degree of improvement among TNFi, ABA, and TCZ. These differences suggested that patient selection bias, as indicated by differences in patient characteristics in the non-adjusted data, maximized the effect of each bDMARD in actual clinical practice. In adjusted data analysis, the improvement achieved with TNFi was greater in all age groups, and this effect was especially pronounced in patients aged < 65 years. The Remission Induction by Remicade in RA (RRR) study using IFX at our institution [42], the Humira Discontinuation without Functional and Radiographic Damage Progression following Sustained Remission (HONOR) study using ADA at our institution [43,44], and other studies [45][46][47] reported that TNFi therapy may lead to discontinuation because of remission. Early-stage RA was also identi ed as highly reliable predictors of successful bDMARD tapering [3]. Although TNFi showed the lowest retention rate among the three bDMARDs groups, future studies are warranted to reveal characteristics of patient populations in which TNFi use is preferred and can achieve discontinuation because of remission.
This study has some limitations. First, the data were obtained from routine clinical practice, and no clear criteria were set for bDMARD discontinuation, with the decision being left at the attending physician's discretion. Second, because IPTW based on generalized propensity score cannot exclude the effects of unknown confounding factors, this study may not have su ciently eliminated biases compared with randomized controlled trials. Third, because IFX, ETN, ADA, GLM, and CZP were analyzed collectively as TNFi, the characteristics of each TNFi may not have been re ected. Fourth, the sample size of patients treated with TCZ was small; thus, the retention rate of TCZ among patients aged ≥ 75 years and changes in CDAI at 2 and 3 years with TCZ therapy in patients aged 65-74 and ≥ 75 years could only be used as a reference. Finally, this study did not include any Janus kinase inhibitors, the use of which will increase in the future.

Conclusions
Despite the aforementioned limitations, this study found that the effectiveness and safety of TCZ were maximized in patients aged < 75 years and that patients aged at least 65 years (particularly those aged ≥ 75 years) were the most Page 8/18 suited candidates for ABA therapy. Furthermore, in patients aged < 65 years, TNFi improved disease activity more effectively and was associated with increased frequently of discontinuation because of remission. Finally, the use of therapeutic strategies appropriate to each age group might improve the outcomes of bDMARD therapy for RA. Availability of data and materials. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Funding. This research received no speci c grant from any funding agency in the public, commercial, or not-for-pro t sectors.
Authors' contributions. AK, KN, and YT contributed to the study conception and design. AK, KN, SK, and YT contributed to the acquisition of data. AK, KN, SK, TA, and YT contributed to the analysis and interpretation of data. All authors drafted the article or revising it critically for important intellectual content and approved of the nal version of the article to be published.  Figure 1 Three-year retention rates of bDMARDs by age group in non-adjusted data Three-year retention rates of bDMARDs in all patients (A), in patients aged <65 years (B), in patients aged 65-74 years (C), and in patients aged ≥75 years (D). yr = years; No. at risk = number at risk; TNFi = tumor necrosis factor inhibitors; ABA = abatacept; TCZ = tocilizumab.

Figure 2
Three-year retention rates of bDMARDs in adjusted data using inverse probability of treatment weighting Three-year retention rates of bDMARDs in all patients (A), in patients aged <65 years (B), in patients aged 65-74 years (C), and in patients aged ≥75 years (D). yr = years; TNFi = tumor necrosis factor inhibitors; ABA = abatacept; TCZ = tocilizumab.