|Home | About | Journals | Submit | Contact Us | Français|
We tested the hypothesis that initiation of TNFα antagonists reduced the risk of fractures compared to nonbiologic comparator in patients with autoimmune diseases.
Using four large administrative databases, we assembled retrospective cohorts of patients with autoimmune diseases who initiated either a TNFα antagonist or a nonbiologic medication. We identified 3 mutually exclusive disease groups: rheumatoid arthritis (RA); inflammatory bowel disease (IBD); and psoriasis, psoriatic arthritis or ankylosing spondylitis (PsO-PsA-AS). We used baseline covariate data to calculate propensity scores (PS) for each disease group and used Cox regression to calculate hazard ratios (HR) and 95% confidence intervals (95%CI). We compared the risk of combined hip, radius/ulna, humerus, or pelvic fractures between PS-matched cohorts of new users of TNFα antagonists and nonbiologic comparator.
We identified 9,020, 2,014 and 2,663 new PS matched episodes of TNFα antagonist and nonbiologic comparator use in RA, IBD and PsO-PsA-AS cohorts, respectively. The risk of combined fractures was similar between new users of TNFα antagonists and nonbiologic comparators for each disease (HR: 1.17, 95%CI [0.91, 1.51]; HR: 1.49, 95%CI [0.72, 3.11]; and HR: 0.92, 95%CI [0.47, 1.82] for RA, IBD and PsO-PsA-AS, respectively). In RA, the risk of combined fractures was associated with an average daily dose of prednisone equivalents >10 mg/day at baseline compared with no glucocorticoid (HR: 1.54, 95%CI [1.03, 2.30]).
The risk of fractures did not differ between initiators of a biologic and a nonbiologic comparator for any disease studied. Among RA patients, use of >10mg/day of prednisone equivalents at baseline increased the fracture risk.
Patients with autoimmune disease, particularly rheumatoid arthritis (RA), have a higher risk of fractures compared to the general population (1–4) due to enhanced local and generalized bone loss (3, 5). In RA, osteoporosis and increased risk of fractures are associated not only with conventional risk factors such as aging, menopause and low body mass index (6), but also with disease-related factors such as prolonged use of glucocorticoids, ambulatory dysfunction, and inflammation (1, 6, 7).
Human and animal studies have shown that inflammatory cytokines have a major effect on bone metabolism (8, 9). Tumor necrosis factor-α (TNFα) induces osteoclast formation and activity (10, 11), and down-regulates osteoblast activity (12), thereby increasing bone loss in primary osteoporosis (13). Accordingly, inhibition of TNFα might be expected to reduce the risk of fractures by preventing bone loss in patients with inflammatory disease. Information about the effect of TNFα antagonists on bone physiology in patients with autoimmune diseases is restricted to measures of markers of bone turnover and bone mass (14, 15). Treatment with TNFα antagonists has been reported to decrease markers of bone resorption and to increase markers of bone formation (16–18). Furthermore, use of TNFα antagonists was associated with an increase in bone mass and/or attenuation of expected bone loss (17, 19–21). In addition, improved disease control with TNFα antagonists could decrease the use of glucocorticoids, a well known risk factor for osteoporotic fractures (22). Despite these favorable effects of TNFα antagonists on bone turnover and bone density, the effect on fracture risk is unclear. If indeed TNFα antagonists protect against fractures compared to nonbiologic comparators, they could be prescribed preferentially to patients at high risk for fractures with autoimmune diseases (if there is no contraindication) to avoid future fracture burden. Thus, as part of a U.S. multi-institutional initiative, SABER (Safety Assessment of Biologic thERapy), we assembled a large retrospective cohort of patients with autoimmune diseases to assess the hypothesis that TNFα antagonists would reduce the risk of fractures compared to nonbiologic therapy in patients with autoimmune diseases.
We studied a retrospective cohort of patients aged 18 years or older with study-defined autoimmune diseases using Tennessee’s Medicaid (TennCare, 1998–2005), Kaiser Permanente Northern California (KPNC, 1998–2007), New Jersey’s Pharmaceutical Assistance to the Age and Disabled and Pennsylvania Pharmaceutical Assistance Contract for the Elderly (PAAD/PACE, 1998–2006), and the National Medicare and Medicaid (MAX/MED, 2000–2006 - excluding Tennessee) databases. This cohort was categorized into three mutually-exclusive groups using the earliest ICD-9 coded health care encounters within the year before cohort entry (baseline): 1) rheumatoid arthritis (RA; ICD9-CM codes: 714.**, except 714.3*), 2) inflammatory bowel disease (IBD; ICD9-CM codes: 556.*, 555.*), and 3) psoriasis, psoriatic arthritis, or ankylosing spondylitis (PsO-PsA-AS; ICD9-CM 696.0, 696.1,720.0). Patients were required to have 365 days of baseline information to ascertain other study selection criteria and covariates. Records missing gender information and patients with other autoimmune diseases that might warrant biologic and non-biologic treatment (e.g. juvenile rheumatoid arthritis, systemic lupus erythematosus) or with other serious diseases (Paget’s disease, organ transplant, HIV/AIDS, dialysis, cancer, liver or lung failure) were excluded. Patients entered the cohort on the date (t0) they filled the first prescription for the specific study regimen if they met the following criteria: continuous enrollment for at least 1 year before entering the cohort (≤30 days gaps were allowed) and no prescription for the study regimen filled during this period.
To compare the risk of fractures between TNFα antagonists and nonbiologic comparator regimens, we first defined episodes of exposure. A new episode started on the date a first prescription for any of the study regimens was filled (t0), with no prescription filled for the same regimen during the year previous, and ended with the earliest of the following events: loss of enrollment, development of a serious disease (defined previously as exclusion criteria), death, end of the study, study outcome (first fracture), change in the medication regimen (e.g. filling a TNFα antagonist for nonbiologic users, or filling a different TNFα antagonist for biologic users - when comparing different TNFα antagonists), or 365 days without regimen medication available. We first identified episodes of TNFα antagonist use, and did not allow overlapping of person-time with other episodes. Within each episode of new use, we identified current use of each regimen, which extended from the start of the episode through the 30th day without medication available or the end of the episode, whichever occurred first. Only current use of study medication was included in the analyses.
We performed the following comparisons in each disease group:
The primary outcomes were: 1) a combined endpoint that included first hip, humerus, radius/ulna or pelvic fracture which are osteoporotic non-vertebral fractures that are accurately assessed in automated databases (23), 2) hip fracture alone. Due to a small number of events, the hip fracture endpoint was only evaluated among patients with RA. The secondary outcome for the RA cohort was the first clinical vertebral fracture. Fracture identification was performed using validated outcome algorithms for non-vertebral (23) and clinical vertebral fractures (24). We excluded fractures caused by major trauma and other conditions (see Supplement); and for clinical vertebral fractures, we excluded patients with a diagnosis/procedure compatible with vertebral fracture during baseline.
Baseline covariates included demographic factors: age, gender, race (white/black/other), residence (urban/rural), nursing home (yes/no), calendar year; generic markers of comorbidity: number of hospitalizations, outpatient and emergency room visits, use of selected medications, chronic obstructive pulmonary disease, diabetes, modified Charlson-Deyo score (an adapted comorbidity index that uses administrative data instead of clinical variables) (25); surrogate markers of disease severity: extra-articular disease manifestations, number of intra-articular or orthopedic procedures, number of tests for inflammatory markers, use of non biologic disease-modifying antirheumatic drugs (DMARDs); and other known risk factors for fractures: previous fractures, diagnosis of osteoporosis, use of oral glucocorticoids (average daily dose in prednisone equivalents), nonsteroidal anti-inflammatory drugs, narcotics, sedative hypnotics, muscle relaxants, antidepressants, antipsychotics, use of drugs that affect bone metabolism (e.g. bisphosphonates, estrogens, thiazides). Baseline glucocorticoid exposure was calculated as the average of daily dose of prednisone equivalents in the 6 months preceding t0and categorized in four dose levels: none, less than 5 mg/day, 5–10 mg/day and >10 mg/day.
In each dataset and disease group, a propensity score (PS) that included all baseline covariates above except oral glucocorticoids was calculated for each exposure episode using logistic regression models. This allowed control of potential confounders through PS greedy matching and assured that shared data were not individually identifiable. The variables and methodology for the creation of the PS have been described previously (26). The risk of fractures between disease specific PS-matched exposed groups was compared using Cox proportional hazard regression models and adjusted by baseline glucocorticoid use. Because patients could contribute ≥1 episodes of medication use, the Huber-White sandwich variance estimator was used to calculate robust standard errors for all hazard ratios (HR) and 95% confidence intervals [95%CI]. Planned subgroup analyses included stratification by age, previous fractures and baseline glucocorticoid use. The Tennessee’s Bureau of TennCare and the University’s Institutional Review Boards (IRB) of Vanderbilt, University of Alabama at Birmingham, Partners Healthcare, and Kaiser Foundation Research Institution approved the study protocol and waived consent requirements.
As previously reported (27), the SABER cohort consisted of a total of 407,319 patients with an autoimmune disease who filled a study DMARD prescription and had complete baseline information. After excluding 170,788 (42%) patients with disease that did not meet study criteria or those with more than one study disease, we identified 139,611 patients with RA, 45,188 with IBD and 51,732 with PsO-PsA-AS. After selection criteria were applied we had 9,020, 2,014 and 2,663 PS-matched episodes of TNFα antagonists and comparator nonbiologic medications use in the RA, IBD and PsO-PsA-AS cohorts respectively (Figure 1).
Baseline characteristics for both exposure groups were similar after PS-matching (Table 1). Patients with RA that started a new regimen were mainly white females with a median age of 58 years. The median follow-up period per episode was approximately 2 person-years. Baseline prednisone equivalents was not included in the PS matching, but was controlled for separately in our analyses. The use of >10 mg/day of prednisone equivalents at baseline was modestly higher in the TNFα antagonist users than the comparator group after PS matching. Of those using >10 mg/day of prednisone equivalents, 37% and 38% of nonbiologic and biologic users, respectively were using >15mg of prednisone equivalents daily, which is considered “high dose”. Although, by definition, all patients were prescribed methotrexate in the baseline year, approximately 51% of the patients in the HCQ/SSZ/LEF group and 38% in the TNFα antagonists group were a current user of methotrexate at t0. We found no differences in the incidence of hip fractures between exposure groups (Table 2, Figure 2). The survival curves for combined fractures, hip and clinical vertebral fractures were similar between both exposure groups (Figure 2A, 2B and 2C).
We found no differences in the incidence of combined fractures when we compared different TNFα antagonists: infliximab versus etanercept (HR: 1.13, 95%CI [0.85, 1.49]); adalimumab versus infliximab (HR: 0.92, 95%CI [0.63, 1.35]); and adalimumab versus etanercept (HR: 0.89, 95%CI [0.63, 1.26]).
We observed a trend for increased risk of fractures with baseline glucocorticoid use in all the fracture groups studied. The association reached statistical significance for the combined fractures at >10 mg/day of prednisone equivalents compared to no use of glucocorticoids at baseline (HR: 1.54, 95%CI [1.03, 2.30]) (Table 3).
In subgroup analyses, the incidence of combined fractures was similar between exposure groups when stratified by age, previous fractures and baseline glucocorticoid use: ≤65 years old (HR: 1.39, 95%CI [0.91, 2.11]), ≤ 65 years old (HR: 1.06, 95%CI [0.79, 1.42], no previous fracture in the baseline period (HR: 1.20, 95%CI [0.93, 1.54]), history of previous fracture in the baseline period (HR: 0.90, 95%CI [0.25, 3.23]), use of ≤ 5 mg/day of prednisone equivalents at baseline (HR: 1.16, 95%CI [0.86, 1.57]), and use of >5 mg/day of prednisone equivalents at baseline (HR: 1.09, 95%CI [0.71, 1.69]).
Additional stratification by baseline glucocorticoid use indicated that >10 mg/day prednisone equivalents was associated with an increased risk of combined fractures among patients ≤65 years old (HR: 1.92, 95%CI [1.08, 3.39]), in those without previous fractures during the baseline period (HR: 1.57, 95% CI [1.05, 2.34]), and when compared to 5–10 mg/day of prednisone equivalents at baseline (HR: 1.57, 95%CI [1.01–2.42]) (Table 4). The risk of fractures was higher with higher doses of prednisone equivalents for the others subgroups studied (>65 years old, history of previous fractures), and when compared to those using <5 mg/day of daily dose of prednisone equivalents at baseline, but these association did not reach significance likely due to small number of episodes in these subgroups (Table 4).
Baseline characteristics before and after matching by PS score are described in Supplement Table 2. Among patients with IBD, the incidence of combined fractures among new INF/ADA users was not significantly higher than new AZA/6-MP users (HR: 1.49, 95% CI [0.72, 3.11]) after adjusting for baseline glucocorticoid use (Table 2). In this disease group daily dose of prednisone equivalents at baseline was not associated with a statistically significant increased the risk of combined fractures when compared to no use (Table 3).
Baseline characteristics before and after matching by PS score are described in Supplement Table 3. Among patients with PsO-PsA-AS, there were no differences in the risk of combined fractures between new TNFα antagonists users and new users of a nonbiologic comparator (HR: 0.92, 95%CI [0.47, 1.82]) after adjusting for baseline glucocorticoid use (Table 2). Similar to IBD, >10 mg/day of prednisone equivalents at baseline did not significantly increase the risk of combined fractures when compared to no use (Table 3).
Our main finding is that fracture risk did not differ between initiators of a TNFα antagonist compared to a nonbiologic comparator regimen for any study disease (RA, IBD and PsO-PsA-AS). Notably, the risk of combined fractures was associated with higher daily doses of prednisone equivalents at baseline in RA patients.
Although some evidence indicates that treatment with TNFα antagonists may have positive effects on bone health in RA patients (16, 19, 21), we found no association between the initiation of TNFα antagonists and the occurrence of fractures. We also found that the risk of combined fractures was similar among the 3 different TNFα antagonists that we studied.
Most prior studies described the beneficial effects of TNFα antagonists on bone mineral density (BMD) and/or markers of bone turnover (16, 17, 21, 28, 29), but few tested this hypothesis in randomized placebo-controlled trials (20, 30, 31). Overall, the main finding of those studies was that treatment with TNFα antagonists resulted in arrest of bone loss at the hip and spine compared to baseline (19, 21, 28) or to other nonbiologic comparator regimens (16, 17), with less effects on bone loss in the metacarpals (20, 29, 30). However, in the Dutch BeSt trial, generalized BMD loss in the spine, hip and hands was similar among different treatment groups (including combination therapy with infliximab) after adjustment for baseline differences, and was associated with erosion progression (32). In the PREMIER trial, hand BMD loss was reduced in patients receiving the combination of adalimumab+methotrexate compared to methotrexate alone only in patients with high disease activity, but not in patients with low disease activity or in remission. Hand bone loss in the combination group was similar in patients with different disease activity and levels of loss were similar to those in patients in remission on methotrexate (31).
One study in female patients enrolled in the CORRONA (Consortium of Rheumatology Researchers of North America) registry, examined the fracture risk among different treatment regimens (33). In this study, monotherapy with a TNFα antagonist was associated with a decreased risk of overall fractures when compared with methotrexate monotherapy, but the expected differences in BMD were not observed. Instead, only the combination of methotrexate and other nonbiologic medication had a positive correlation with hip T score when compared with methotrexate monotherapy. Although this study included a large number of patients seen in clinical practice and used clinical covariates in the analysis, medication (exposure) groups were not clearly defined (33). Concordant with our findings, a recent study using data from a US commercially insured population and a Canadian Province found that the risk of fractures in RA patients did not differ among initiators of different regimens (including TNFα antagonists, methotrexate and other non biologic regimens) after adjusting by multiple clinical covariates(34).
Many studies have reported an increased risk of fractures at different anatomical sites with the use of glucocorticoids (22, 33, 35). We previously reported in this same cohort, that the initiation of both TNFα antagonists and other non biologic regimens were followed by a small reductions in the use of glucocorticoids (36). The percentage of patients taking glucocorticoids decreased less than 10% for new users of anti-TNF drugs and non biologic DMARDs 1 year after initiation of therapy (60.9 to 51.8% for anti-TNF users, 63.9 to 58.2% for LEF users, and 59.6 to 50.0% in HCQ/SSZ users). In the current study, we found an association between the use of >10 mg/d of prednisone equivalents at baseline and the risk of combined fractures, and that TNFα antagonists and other non biologic regimens had a similar risk of fractures at specific anatomic sites.
Our subgroup analyses suggest that >10 mg/day of prednisone equivalents at baseline was associated with an increased risk of combined fractures in patients ≤ 65 years old, and in those without previous fractures, compared to those with no use of baseline glucocorticoids. In the subgroup of patients >65 years old, and in those with history of previous fractures, the risk of fractures was also higher in those using >10 mg/day of prednisone equivalents at baseline, but did not reach statistical significance, probably due to the small sample size in each subgroup.
In the IBD and PsO-PsA-AS cohorts, we found that initiation of TNFα antagonists was not associated with a decrease in the risk of combined fractures. Few studies, with a small number of patients (37–39) and without a control group (38, 40), have assessed the effect of infliximab treatment on BMD in patients with spondyloarthropathies. In general, these studies reported that treatment with infliximab increased the spine and hip BMD (37–41). A post hoc and exploratory analysis of a randomized placebo-controlled trial in patients with AS reported that the median percent increase in spine and hip BMD score was higher in the infliximab treated group compared to placebo 24 weeks after treatment initiation (2.5% versus 0.5% in the spine, p<0.001; and 0.5% versus 0.2% at the hip, p<0.033) (18).
Despite the advantage of studying a large cohort of patients from real practice that allowed PS matching for multiple covariates, our study has several limitations: 1) we assumed that prescription fills represent medication use, but the actual use of study medications is unknown; 2) we relied on coded information to identify fractures which could have resulted in some misclassification, although we minimized outcome misclassification by using previously validated definitions; 3) clinical information is limited in these databases, so we relied on surrogate covariates to asses disease severity; 4) the study was not randomized and while we controlled for measured confounders through a PS matching method, major changes in time-dependent covariates could have unbalanced the study groups. Notably, this would be more likely to occur with outcomes that require long term exposures; 5) information related to lifetime cumulative doses of glucocorticoid, a risk factor for fractures (42), was not obtained. Instead, we used average daily dose of prednisone equivalents in the 6 months prior study entry in the analysis since some evidence suggested that average daily dose over a treatment period was more closely related to the risk of non-vertebral fractures than cumulative dose (35); 6) presence of unmeasured confounders (e.g.: BMD, physical activity, use of over the counter vitamin D and calcium supplements) that are balanced through randomization but not necessarily by PS matching could have biased our results; and 7) we had relatively short periods of follow-up that resulted in a relatively small number of events and limited our ability to detect small differences. Ideally, PsO, PsA, and AS would be analyzed separately since the clinical profile of these diseases are different. However, we aggregated data from these three seronegative spondyloarthropaties due to the small number of end points in each disease category. It is possible that TNFα antagonist users have less bone loss compared to nonbiologic users, but longer follow up time may be required to ascertain long term effects of specific therapies on fracture risk. For example, studies that examined the effects of antiresorptive drugs on the risk of osteoporotic fractures found evidence of protection in patients with pre-existing fractures and/or low BMD early in treatment (43–46). However, there was little effect among those without pre-existing fractures and preserved BMD even after long follow-up periods when compared to placebo (47, 48). Thus, the effects of anti-TNF therapy may be most evident in patient populations with low BMD, or after extended follow up in an unselected population.
On the other hand, inflammation generates an array of inflammatory mediators that can activate osteoclasts through pathways that are independent of TNFα leading to bone resorption (49, 50). Thus, it is possible that inhibition of TNFα alone may not prevent increased bone loss, if there is ongoing inflammation.
In conclusion, we observed that the risk of fractures in patients with autoimmune diseases did not differ between new users of TNFα antagonists and a nonbiologic comparator medication, but was associated with >10 mg/day of prednisone equivalents among RA patients. The role of TNFα antagonists in bone health is still poorly defined. Studies of larger cohorts of patients with longer follow up are needed to determine if changes in bone density translate into changes in fracture risk. Currently, there is no clear evidence that suggests that TNFα antagonists are strongly associated with the risk of fractures.
We are indebted to the Tennessee Bureau of TennCare of the Department of Finance and Administration, which provided the data on TennCare recipients On behalf of the SABER collaboration: Agency for Health-care Research and Quality (AHRQ), Parivash Nourjah; Brigham and Women’s Hospital, Robert Glynn, Mary Kowal, Joyce Lii, Jeremy Rassen, Sebastian Schneeweiss; Fallon Medical Center and University of Massachusetts, Leslie Harrold; Food and Drug Administration, David Graham, Carolyn McCloskey, Kristin Phucas; Kaiser Permanente Northern California; Kaiser Permanente Colorado, Marcia Raebel; University of Alabama at Birmingham, Nivedita Patkar, Kenneth Saag, Fenglong Xie; University of Pennsylvania, Kevin Haynes, James Lewis.
Funding statement: This work was supported by the Food and Drug Administration (FDA), the Agency for Healthcare Research and Quality (AHRQ) [U18 HSO17919-01, R01HS018517 to J.R.C.], the National Institutes of Health (NIH) [5T32GM007569-33 to V.K.K., AR053351 to J.R.C.,5P60AR56116 to C.G.G, C.M.S. and M.R.G], and Vanderbilt Physician Scientist Development Award (V.K.K).
Conflict of interest statement: JRC has received research support and consultant fees, from Abbott, Amgen, Centocor, Pfizer, BMS, Consortium of Rheumatology Researchers of North America (CORRONA), Roche/Genentech, and UCB. DHS has received research grants from Amgen, Lilly and Abbott and serves as a consultant to CORRONA. ED has received research support from Amgen. LH has received research support from Centocor, Genentech and Procter & Gamble. MG has received consultant fees from Rocky Mountain Poison and Drug Center funded by McNeil. The rest of authors have no conflict of interest to report.