Abstract
Background:
Low-dose glucocorticoids are frequently used for the management of rheumatoid arthritis (RA) and other chronic conditions, but the safety of long-term use remains uncertain.
Objective:
To quantify the risk for hospitalized infection with long-term use of low-dose glucocorticoids in patients with RA receiving stable disease-modifying antirheumatic drug (DMARD) therapy.
Design:
Retrospective cohort study.
Setting:
Medicare claims data and Optum’s deidentified Clinformatics Data Mart database from 2006 to 2015.
Patients:
Adults with RA receiving a stable DMARD regimen for more than 6 months.
Measurements:
Associations between glucocorticoid dose (none, ≤5 mg/d, >5 to 10 mg/d, and >10 mg/d) and hospitalized infection were evaluated using inverse probability–weighted analyses, with 1-year cumulative incidence predicted from weighted models.
Results:
We identified 247 297 observations among 172 041 patients in Medicare and 58 279 observations among 44 118 patients in Optum. After 6 months of stable DMARD use, 47.1% of Medicare patients and 39.5% of Optum patients were receiving glucocorticoids. The 1-year cumulative incidence of hospitalized infection in Medicare patients not receiving glucocorticoids was 8.6% versus 11.0% (95% CI, 10.6% to 11.5%) for glucocorticoid dose of 5 mg or less per day, 14.4% (CI, 13.8% to 15.1%) for greater than 5 to 10 mg/d, and 17.7% (CI, 16.5% to 19.1%) for greater than 10 mg/d (all P < 0.001 vs. no glucocorticoids). The 1-year cumulative incidence of hospitalized infection in Optum patients not receiving glucocorticoids was 4.0% versus 5.2% (CI, 4.7% to 5.8%) for glucocorticoid dose of 5 mg or less per day, 8.1% (CI, 7.0% to 9.3%) for >5 to 10 mg/d, and 10.6% (CI, 8.5% to 13.2%) for greater than 10 mg/d (all P < 0.001 vs. no glucocorticoids).
Limitation:
Potential for residual confounding and misclassification of glucocorticoid dose.
Conclusion:
In patients with RA receiving stable DMARD therapy, glucocorticoids were associated with a dose-dependent increase in the risk for serious infection, with small but significant risks even at doses of 5 mg or less per day. Clinicians should balance the benefits of low-dose glucocorticoids with this potential risk.
Primary Funding Source:
National Institute of Arthritis and Musculoskeletal and Skin Diseases.
Toc blurb
Patients with rheumatoid arthritis often receive low-dose glucocorticoids as bridging therapy when starting or changing disease-modifying antirheumatic drug (DMARD) treatment. This observational study examined the risk for serious infection in patients on stable DMARD therapy who received low-dose glucocorticoids over the long term.
Glucocorticoids are effective for the treatment of rheumatoid arthritis (RA) when added to disease-modifying antirheumatic drugs (DMARDs) (1-3). They are widely accepted for short-term use as bridging therapy in patients starting or changing treatment with DMARDs (4, 5), with the goal of their eventual withdrawal. Yet, 30% to 60% of patients with RA remain on long-term glucocorticoids, especially at low doses (6-9). Low-dose glucocorticoid therapy is also a common treatment of other rheumatic and nonrheumatic conditions, such as chronic obstructive pulmonary disease (COPD).
Controversy on the appropriate use of long-term glucocorticoids stems from continued uncertainty about the safety of low-dose therapy (10). Although the risk for infection and other toxicity with higher-dose therapy is well established (11-15), data about low-dose therapy are less clear. Existing randomized trials are short term and are not adequately powered to assess safety. Some observational studies have suggested a greater risk for serious infection even at prednisone doses of 5 mg or less per day (11, 12, 14-18), but concerns about residual confounding remain. Prior studies have not separated long-term glucocorticoid use from short-term bridging therapy, which may be more prone to confounding and misclassification introduced by greater fluctuations in disease activity, concomitant DMARD and biologic use, and short-term glucocorticoid dose changes.
In this study, we aimed to quantify the risk for infection with long-term, stable, low-dose glucocorticoids in a population of patients with RA who were also receiving stable DMARD therapy, including biologics. We hypothesized that low-dose glucocorticoids (≤5 mg/d) would carry a modest risk for infection in this context.
Methods
Data Sources
This retrospective cohort study evaluated patients with RA using national Medicare claims data from 2006 to 2015 and Optum’s deidentified Clinformatics Data Mart database from 2001 to 2015, including patients with an index date between 1 January 2007 and 31 August 2015. Medicare is a public health plan covering more than 90% of U.S. adults aged 65 years or older (19). Younger persons with disabilities, such as RA, may also be covered. Clinformatics is a deidentified U.S. administrative claims database from a nationally representative commercial health care insurer with sociodemographic, inpatient, outpatient, and prescription claims available for more than 60 million unique members. Laboratory data are available for approximately 10% of the database.
Cohort Identification
Patients were included if they were aged 18 years or older, had 2 physician International Classification of Diseases, Ninth Revision (ICD-9) codes for RA (714.xx) at least 7 days apart, and were receiving a DMARD (20). We identified patients receiving a stable DMARD course for 6 months or more (180 days), defined as either continuous use of methotrexate with no biologic or targeted synthetic (ts) DMARD (tofacitinib) or continuous use of a biologic or tsDMARD, with or without methotrexate. Continuous use was defined as no gaps in treatment greater than 90 days based on the days' supply from prescription fills or expected infusion intervals for infused therapies (21). Rituximab courses were required to be 210 days or longer to exclude patients receiving 2 initial infusions with no subsequent infusions at 6 months.
The index date anchored the start of follow-up and was defined as 6 months after the start date of the stable DMARD course (Supplement Figure 1, available at Annals.org). All patients were required to have at least 6 months of data available before the DMARD course start date; this 6-month period and the first 6 months of the DMARD course made up the 1-year baseline period. We excluded patients with a diagnosis of psoriatic arthritis, ankylosing spondylitis, inflammatory bowel disease, systemic lupus erythematosus, cancer, or HIV during the baseline period (codes are shown in Supplement Table 1, available at Annals.org). Patients could contribute multiple observations if they had several distinct stable DMARD courses.
Exposure
The exposure of interest was glucocorticoid use, categorized as none, 5 mg or less per day, >5 to 10 mg/d, and greater than 10 mg/d. We assessed average daily glucocorticoid dose in the 90 days before the index date on the basis of oral prescriptions for prednisone, prednisolone, and methylprednisolone, using prescribed dose in prednisone equivalents and days' supply to determine each daily dose and truncating prescriptions if a new prescription was filled before the prescription end date. We also measured cumulative glucocorticoid dose in the first 9 months of the baseline period but did not include this in primary analyses because of concerns about collinearity with baseline glucocorticoid dose.
Outcome and Follow-up
The outcome of interest was time to first infection occurring during an acute care hospitalization (hereafter referred to as "hospitalized infection"). The primary outcome definition was the presence of an ICD-9 diagnosis of infection (Supplement Table 1) in any position of the discharge diagnoses from an acute care hospitalization (positive predictive value >80%) (13, 22). In secondary analyses, we evaluated an alternative definition requiring an ICD-9 diagnosis of infection in the primary position of the discharge diagnoses.
Outcomes were assessed after the index date (after 6 months of stable DMARD treatment). Patients with a hospitalized infection before the index date could still be included. Censoring occurred at the end of enrollment in the health plan (30 September 2015), at the end of the stable DMARD course (gap >90 days or initiation of a new biologic or tsDMARD), at death (available in Medicare only), or when there was a change in glucocorticoid dose (to directly assess the risk associated with long-term use of a stable dose), whichever came first. To evaluate changes in glucocorticoid dose, we recalculated the average daily dose in each fixed 90-day interval during follow-up, beginning on the index date (Supplement Figure 1). In the primary analysis, patients were censored at the end of the 90-day interval in which a dose change was identified to ensure that outcomes that led to dose changes were captured. In a sensitivity analysis, patients were censored at the beginning of the 90-day interval in which a glucocorticoid dose change was identified.
Covariates
Covariates measured during the baseline period included age, sex, race, geographic region, calendar year, comorbidities, an adaption of the Charlson Comorbidity Index (23), health care use (outpatient visits, rheumatologist outpatient visits, emergency department visits, and hospitalizations), previous hospitalized infections, influenza vaccination, skilled-nursing facility stay, and use of durable medical equipment. In addition to including the DMARD defining the stable DMARD treatment course, we assessed included prescription fills for nonsteroidal anti-inflammatory drugs, opioids, methotrexate, sulfasalazine, leflunomide, hydroxychloroquine, other conventional synthetic DMARDs, proton-pump inhibitors, and antibiotics in the 90 days before the index date. We also used all available data in the data set before the index date to assess pneumococcal and zoster vaccination, cancer screening, and number of previous biologics used. Disability status, dual Medicare/Medicaid eligibility, and quintiles of median household income based on ZIP code from the 2009–2013 American Community Survey (24) were available in Medicare, whereas education and household income were available in Optum.
Statistical Analysis
Primary Analyses
Analyses were done separately in the Medicare and Optum cohorts. Associations between glucocorticoid dose and time to first hospitalized infection were assessed using cause-specific proportional hazards models with cluster-robust SEs to account for patients contributing multiple observations (clustering by the treating rheumatologist led to similar results). We used propensity score–based inverse probability weights to balance confounders across treatment groups. Because the clinical importance of hazard ratios can be difficult to interpret, we used our cause-specific proportional hazards models to calculate the predicted 1-year cumulative incidence of infection for patients in each glucocorticoid dose category. Proportional hazards assumptions were assessed using log–log plots and were not violated.
Propensity Scores
To account for potential confounders, generalized propensity scores based on the probability of being in each exposure group were generated separately in each cohort using multinomial logistic regression models (25, 26) (included variables are shown in Supplement Tables 2 and 3, available at Annals.org). Propensity scores were used to create stabilized inverse probability treatment weights (26-28) truncated at the first and 99th percentiles (29). The results of the weighted cause-specific proportional hazards models estimate the average treatment effect—the average effect, at the population level, of patients receiving stable DMARD therapy moving from one glucocorticoid dose category to another—assuming the Stable Unit Treatment Value Assumption, exchangeability, and positivity. The balance of covariates across treatment categories was assessed compared with the reference group (no glucocorticoids), with a standardized mean difference of 0.1 or less indicating good balance. Unbalanced covariates were added as covariates to weighted models, as described in the tables and figures.
Subgroup and Sensitivity Analyses
Prespecified subgroup analyses included evaluating associations between glucocorticoid dose and hospitalized infection among patients using methotrexate versus a biologic or tsDMARD and among older versus younger patients. Because of differences in age distributions in the 2 cohorts, we evaluated patients younger than 75 years and those aged 75 years or older in Medicare and patients younger than 65 years and those aged 65 years or older in Optum. Propensity score–based inverse probability weights were recalculated for each subgroup analysis.
Sensitivity analyses (all done using inverse probability weights) included 1) treating glucocorticoid dose as time-varying and not censoring with glucocorticoid dose changes (to evaluate the potential for informative censoring in patients changing dose), 2) censoring at the beginning of 90-day intervals indicating a glucocorticoid dose change, 3) excluding patients with a hospitalized infection in the 180 days before the index date, 4) excluding patients with COPD, asthma, or extra-articular RA (who may receive glucocorticoids for these indications), 5) including cumulative glucocorticoid dose in the first 9 months of the baseline period as a covariate in weighted models, and 6) including only patients who had received glucocorticoids in the 3 months before the DMARD course initiation (comparing patients who stopped vs. continued glucocorticoid treatment).
To assess associations between other risk factors and the outcome of interest, we created a traditional multivariable model (without inverse probability weights) that included all variables from the propensity score model as covariates.
To assess the effect of an unmeasured confounder (disease activity) on results, we did a residual bias assessment (30) using C-reactive protein (CRP) level in a subset of patients in Optum. C-reactive protein data were used to estimate plausible associations between disease activity and both the exposure and outcome as well as prevalence of this confounder. Results were used to calculate the effect on study results if this unmeasured confounder had been included in the study (see the Methods section of the Supplement, available at Annals.org).
The data set was created with SAS, version 9.4 (SAS Institute), and the analysis was done using Stata, version 15.1 (StataCorp) (commands are shown in Supplement Table 4, available at Annals.org). The protocol was approved by the University of Pennsylvania and University of Alabama at Birmingham institutional review boards.
Role of the Funding Source
This study was funded by the National Institute of Arthritis and Musculoskeletal and Skin Diseases. The funder had no role in the design, conduct, or reporting of the study.
Results
We identified 247 297 qualifying medication courses among 172 041 patients in Medicare and 58 279 medication courses among 44 118 patients in Optum (Figure 1). Courses of biologics or tsDMARDs made up 47% of Medicare medication courses and 50% of Optum courses. At the index date (after 6 months of stable DMARD use), 47.1% of Medicare patients and 39.5% of Optum patients were receiving glucocorticoids, most commonly at doses of 5 mg or less per day.
Figure 1.
Cohort identification. Observations are based on DMARD courses with patients able to contribute multiple observations if they had multiple distinct DMARD courses. Missing data were region of residence or urban versus rural status. AS = ankylosing spondylitis; DMARD = disease-modifying antirheumatic drug; IBD = inflammatory bowel disease; PsA = psoriatic arthritis; RA = rheumatoid arthritis; SLE = systemic lupus erythematosus; ts = targeted synthetic.
Patient characteristics are shown in Table 1 and Supplement Tables 2 and 3, with standardized mean differences in characteristics before and after weighting shown in Supplement Figure 2 (available at Annals.org). Compared with patients not receiving glucocorticoids, those receiving 5 mg or less per day were more likely to have received opioids or antibiotics; had a higher frequency of COPD; had more frequent outpatient visits; and were more likely to have had an emergency department visit, hospitalization, or hospitalized infection in the past year (standardized differences >0.1). These differences were more pronounced among patients receiving >5 to 10 mg or more than 10 mg of glucocorticoids per day, and these patients also had different rates of prior biologic use, disability, certain comorbidities, and use of durable medical equipment. Older patients were more likely to receive glucocorticoids but less likely to receive more than 10 mg of glucocorticoids per day, especially in Medicare. After inverse probability weights were applied, measured patient characteristics were well balanced in patients receiving 5 mg or less or >5 to 10 mg of glucocorticoids per day versus no glucocorticoids in both data sets. For patients receiving more than 10 mg/d, small residual imbalance remained for opioid use, outpatient visits, and hospitalizations in both data sets and emergency department visits in the Medicare data set (Supplement Figure 2).
Table 1.
Select Baseline Cohort Characteristics Before Inverse Probability Weighting*
| Characteristic | Average Glucocorticoid Dose in the Past 3 Months | Total | |||
|---|---|---|---|---|---|
| None | ≤5 mg/d | >5 to 10 mg/d | >10 mg/d | ||
| Medicare cohort | |||||
| Observations, n | 130 822 | 76 491 | 31 621 | 8363 | 247 297 |
| Unique patients, n | 90 976 | 53 159 | 22 050 | 5856 | 172 041 |
| Mean age (SD), y | 68.6 (11.8) | 69.4 (11.6) | 68.0 (11.8) | 65.1 (12.7) | 68.7 (11.8) |
| Female, n (%) | 106 854 (81.7) | 62 129 (81.2) | 24 223 (76.6) | 6061 (72.5) | 199 267 (80.6) |
| White, n (%) | 92 724 (70.9) | 55 153 (72.1) | 22 948 (72.6) | 6162 (73.7) | 176 987 (71.6) |
| Disability, n (%)† | 56 280 (43.0) | 32 468 (42.4) | 15 249 (48.2) | 4742 (56.7) | 108 739 (44.0) |
| Stable DMARD course, n (%) | |||||
| Methotrexate without biologic | 70 689 (54.0) | 41 761 (54.6) | 16 576 (52.4) | 4035 (48.2) | 133 061 (53.8) |
| TNF inhibitor | 40 757 (31.2) | 21 644 (28.3) | 8946 (28.3) | 2563 (30.6) | 73 910 (29.9) |
| Non–TNF inhibitor biologic/tsDMARD | 19 376 (14.8) | 13 086 (17.1) | 6099 (19.3) | 1765 (21.1) | 40 326 (16.3) |
| Prior biologic use (all available data), n (%) | 41 512 (31.7) | 26 958 (35.2) | 12 443 (39.4) | 3531 (42.2) | 84 444 (34.1) |
| Opioid use in the past 90 d, n (%) | 55 649 (42.5) | 38 184 (49.9) | 18 574 (58.7) | 5678 (67.9) | 118 085 (47.8) |
| Antibiotic use in the past 90 d, n (%) | 39 263 (30.0) | 28 033 (36.6) | 12 424 (39.3) | 3882 (46.4) | 83 602 (33.8) |
| Median Charlson Comorbidity Index score (IQR) | 2 (0–3) | 2 (0–4) | 2 (0–4) | 2 (1–5) | 2 (0–4) |
| Diabetes, n (%) | 29 548 (22.6) | 16 200 (21.2) | 7294 (23.1) | 2224 (26.6) | 55 266 (22.3) |
| COPD, n (%) | 13 923 (10.6) | 10 996 (14.4) | 5460 (17.3) | 1911 (22.9) | 32 290 (13.1) |
| Extra-articular RA, n (%) | 3009 (2.3) | 2266 (3.0) | 1197 (3.8) | 516 (6.2) | 6988 (2.8) |
| Hospitalizations in the past year, n (%) | 28 136 (21.5) | 20 408 (26.6) | 10 188 (32.2) | 3428 (41.0) | 62 160 (25.1) |
| Emergency department visits in the past year, n (%) | 46 175 (35.3) | 32 528 (42.5) | 15 370 (48.6) | 4753 (56.8) | 98 826 (40.0) |
| Hospitalized infections in the past year, n (%) | 10 821 (8.3) | 8834 (11.5) | 4882 (15.4) | 1781 (21.3) | 26 318 (10.6) |
| Median outpatient visits in the past year (IQR), n | 12 (8–18) | 14 (9–20) | 14 (10–21) | 16 (10–23) | 13 (8–19) |
| Optum cohort | |||||
| Observations, n | 35 251 | 15 504 | 5889 | 1635 | 58 279 |
| Unique patients, n | 26 449 | 11 774 | 4632 | 1263 | 44 118 |
| Mean age (SD), y | 57.4 (13.7) | 56.5 (13.8) | 58.8 (13.5) | 59.1 (13.2) | 57.6 (13.1) |
| Female, n (%) | 27 469 (77.9) | 11 962 (77.2) | 4213 (71.5) | 1056 (64.6) | 44 700 (76.7) |
| White, n (%) | 25 505 (72.4) | 11 076 (71.4) | 4229 (71.8) | 1206 (73.8) | 42 016 (72.1) |
| Stable DMARD course, n (%) | |||||
| Methotrexate without biologic | 16 390 (46.5) | 8448 (54.5) | 3231 (54.9) | 878 (53.7) | 28 947 (49.7) |
| TNF inhibitor | 15 173 (43.0) | 5303 (34.2) | 1881 (31.9) | 476 (29.1) | 22 833 (39.2) |
| Non–TNF inhibitor biologic/tsDMARD | 3688 (10.5) | 1753 (11.3) | 777 (13.2) | 281 (17.2) | 6499 (11.2) |
| Prior biologic use (all available data), n (%) | 10 002 (28.4) | 4727 (30.5) | 1973 (33.5) | 586 (35.8) | 17 288 (29.7) |
| Opioid use in the past 90 d, n (%) | 10 279 (29.2) | 6106 (39.4) | 2714 (46.1) | 884 (54.1) | 19 983 (34.3) |
| Antibiotic use in the past 90 d, n (%) | 8494 (24.1) | 4832 (31.2) | 1882 (32.0) | 582 (35.6) | 15 790 (27.1) |
| Median Charlson Comorbidity Index score (IQR) | 0 (0–1) | 0 (0–2) | 0 (0–2) | 1 (0–2) | 0 (0–2) |
| Diabetes, n (%) | 4954 (14.1) | 2193 (14.1) | 921 (15.6) | 312 (19.1) | 8380 (14.4) |
| COPD, n (%) | 1747 (5.0) | 1152 (7.4) | 585 (9.9) | 212 (13.0) | 3696 (6.3) |
| Extra-articular RA, n (%) | 809 (2.3) | 395 (2.5) | 200 (3.4) | 83 (5.1) | 1487 (2.6) |
| Hospitalizations in the past year, n (%) | 4919 (14.0) | 2829 (18.2) | 1375 (23.3) | 496 (30.3) | 9619 (16.5) |
| Emergency department visits in the past year, n (%) | 9293 (26.4) | 4923 (31.8) | 1904 (32.3) | 598 (36.6) | 16 718 (28.7) |
| Hospitalized infections in the past year, n (%) | 1419 (4.0) | 932 (6.0) | 544 (9.2) | 208 (12.7) | 3103 (5.3) |
| Median outpatient visits in the past year (IQR), n | 15 (9–23) | 17 (11–27) | 19 (12–29) | 21 (13–33) | 16 (10–25) |
| Mean C-reactive protein level (SD), mg/L | 6.1 (11.8) (n = 6724) | 8.0 (13.7) (n = 3142) | 10.7 (19.5) (n = 1111) | 13.6 (23.3) (n = 308) | 7.3 (13.8) (n = 11 285) |
COPD = chronic obstructive pulmonary disease; DMARD = disease-modifying antirheumatic drug; IQR = interquartile range; RA = rheumatoid arthritis; TNF = tumor necrosis factor; tsDMARD = targeted synthetic DMARD.
Values shown are based on observations (distinct stable DMARD courses) rather than distinct patients.
Patients who qualified for Medicare on the basis of receiving Social Security disability benefits.
Median follow-up time after the index date with stable DMARD use and stable glucocorticoid dose was 180 days (interquartile range, 90 to 433 days) in Medicare and 148 days (interquartile range, 90 to 347 days) in Optum. Follow-up time was shorter in patients receiving higher glucocorticoid doses, driven primarily by glucocorticoid dose changes (31%, 51%, 62%, and 61% of censoring events in Medicare and 23%, 48%, 55%, and 50% of censoring events in Optum for patients receiving no glucocorticoids, ≤5 mg/d, >5 to 10 mg/d, or >10 mg/d, respectively). Censoring due to end of enrollment in the health plan occurred in 3.7% of patients in Medicare and 14.0% of patients in Optum.
Associations Between Glucocorticoids and Hospitalized Infection
There were 20 963 (10.9 per 100 person-years) and 2177 (5.4 per 100 person-years) hospitalized infections in Medicare and Optum, respectively, when any discharge diagnosis was used. Rates were lower (12 568 [6.4 per 100 person-years] and 1586 [3.9 per 100 person-years]) when a primary discharge diagnosis was used. The most common infections were urinary infection, pneumonia, bacteremia or septicemia, and skin or soft tissue infections (Supplement Table 5, available at Annals.org). Higher rates of infection were seen in patients receiving higher glucocorticoid doses (Table 2).
Table 2.
Association of Glucocorticoid Dose With Hospitalized Infection
| Glucocorticoid Dose Group |
Observations, n* |
Any Discharge Diagnosis | Primary Position Discharge Diagnosis | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Person- Years |
Infections, n (inc/100py) |
Unadjusted HR (95% CI) |
Adjusted HR (95% CI)† |
Person- Years |
Infections, n (inc/100py) |
Unadjusted HR (95% CI) |
Adjusted HR (95% CI)† |
||
| Medicare | |||||||||
| None | 130 822 | 128 317 | 10 719 (8.4) | Reference | Reference | 131 201 | 6281 (4.8) | Reference | Reference |
| ≤5 mg/d | 76 491 | 48 274 | 6514 (13.5) | 1.54 (1.49–1.59) | 1.29 (1.25–1.34) | 49 638 | 3952 (8.0) | 1.61 (1.55–1.68) | 1.34 (1.29–1.40) |
| >5 to 10 mg/d | 31 621 | 13 120 | 2816 (21.5) | 2.31 (2.22–2.42) | 1.72 (1.65–1.81) | 13 442 | 1746 (13.0) | 2.53 (2.39–2.67) | 1.86 (1.75–1.97) |
| >10 mg/d | 8363 | 2897 | 914 (31.6) | 3.34 (3.11–3.58) | 2.16 (1.99–2.34) | 2992 | 589 (19.7) | 3.78 (3.47–4.13) | 2.36 (2.14–2.61) |
| Optum | |||||||||
| None | 35 251 | 30 139 | 1201 (4.0) | Reference | Reference | 30 356 | 865 (2.8) | Reference | Reference |
| ≤5 mg/d | 15 504 | 7875 | 584 (7.4) | 1.76 (1.59–1.95) | 1.32 (1.18–1.47) | 7947 | 424 (5.3) | 1.80 (1.60–2.03) | 1.35 (1.19–1.53) |
| >5 to 10 mg/d | 5889 | 2113 | 282 (13.3) | 3.03 (2.64–3.47) | 2.07 (1.78–2.41) | 2133 | 211 (9.9) | 3.24 (2.77–3.79) | 2.17 (1.83–2.58) |
| >10 mg/d | 1635 | 494 | 110 (22.3) | 4.93 (4.04–6.01) | 2.76 (2.19–3.47) | 500 | 86 (17.2) | 5.55 (4.43–6.95) | 3.27 (2.51–4.26) |
HR = hazard ratio; inc/100py = crude incidence per 100 person-years.
Refers to the number of total observations, with some patients contributing multiple observations.
Adjusted hazard ratios are from inverse probability–weighted cause-specific hazards models. Variables that were imbalanced across glucocorticoid categories after inverse probability weighting were added as covariates to weighted models (opioid use, outpatient visits, and hospitalizations in both data sets and emergency department visits in Medicare).
In inverse probability–weighted analyses, glucocorticoids were associated with a dose-dependent increase in the risk for hospitalized infection by both definitions, with some attenuation compared with unadjusted results (Table 2). Predicted 1-year cumulative incidence of hospitalized infection from these models is shown in Figure 2 and Supplement Figure 3 (available at Annals.org). Results were similar when a primary discharge diagnosis was used (Supplement Figure 3).
Figure 2.
Association between glucocorticoid dose and hospitalized infection (any discharge diagnosis). Predicted 1-year incidence of hospitalized infection calculated from inverse probability–weighted cause-specific hazards models. Confidence intervals are not available for the reference group, which represents the baseline incidence at 1 year. Variables that were imbalanced across glucocorticoid categories after inverse probability weighting were added as covariates to weighted models (opioid use, outpatient visits, and hospitalizations in both data sets and emergency department visits in Medicare).
A traditional multivariable model including all covariates of interest showed similar results and demonstrated that older age and prior hospitalized infection were strongly associated with the risk for hospitalized infection, with other factors, such as COPD and opioid use, also associated with risk (Supplement Table 6, available at Annals.org).
Subgroup and Sensitivity Analyses
In both data sets, associations were similar in patients receiving a biologic or tsDMARD and in patients receiving methotrexate without a biologic or tsDMARD. Also, although absolute rates of infection differed substantially in older versus younger patients, the relative associations between glucocorticoids and hospitalized infections were similar (Table 3; Supplement Table 7, available at Annals.org).
Table 3.
Subgroup Analyses Evaluating Association Between Glucocorticoid Dose and Hospitalized Infection (Any Discharge Diagnosis) in Nonbiologic Users, Biologic/tsDMARD Users, Younger Patients, and Older Patients*
| Glucocorticoid Dose Group |
Methotrexate Without Biologic/tsDMARD |
Biologic/tsDMARD | Younger Patients† | Older Patients† | ||||
|---|---|---|---|---|---|---|---|---|
| Infections, n (inc/100py) |
Adjusted HR (95% CI)* |
Infections, n (inc/100py) |
Adjusted HR (95% CI)* |
Infections, n (inc/100py) |
Adjusted HR (95% CI)* |
Infections, n (inc/100py) |
Adjusted HR (95% CI)* |
|
| Medicare | ||||||||
| None | 6178 (8.9) | Reference | 4541 (7.7) | Reference | 5769 (6.4) | Reference | 4950 (12.8) | Reference |
| ≤5 mg/d | 3929 (14.7) | 1.32 (1.26–1.37) | 2585 (12.0) | 1.26 (1.20–1.33) | 3173 (10.1) | 1.26 (1.20–1.32) | 3341 (20.0) | 1.31 (1.25–1.37) |
| >5 to 10 mg/d | 1566 (23.2) | 1.71 (1.61–1.82) | 1250 (19.6) | 1.75 (1.63–1.88) | 1677 (17.5) | 1.73 (1.63–1.84) | 1139 (32.0) | 1.74 (1.62–1.86) |
| >10 mg/d | 498 (37.2) | 2.26 (2.02–2.52) | 416 (26.7) | 2.08 (1.84–2.35) | 656 (28.0) | 2.22 (2.01–2.44) | 258 (46.8) | 2.23 (1.92–2.58) |
| Optum | ||||||||
| None | 609 (4.4) | Reference | 592 (3.6) | Reference | 500 (2.4) | Reference | 701 (7.4) | Reference |
| ≤5 mg/d | 360 (8.2) | 1.34 (1.16–1.54) | 224 (6.5) | 1.31 (1.11–1.55) | 191 (4.1) | 1.23 (1.03–1.46) | 393 (12.4) | 1.34 (1.17–1.54) |
| >5 to 10 mg/d | 169 (14.5) | 2.07 (1.71–2.51) | 113 (12.0) | 2.18 (1.71–2.78) | 118 (8.6) | 2.14 (1.69–2.72) | 164 (21.9) | 2.01 (1.65–2.45) |
| >10 mg/d | 54 (21.0) | 2.03 (1.48–2.78) | 56 (23.7) | 3.22 (2.26–4.59) | 56 (16.0) | 2.65 (1.82–3.85) | 54 (37.4) | 2.55 (1.84–3.54) |
HR = hazard ratio; inc/100py = crude incidence per 100 person-years; tsDMARD = targeted synthetic disease-modifying antirheumatic drug.
Adjusted hazard ratios are from inverse probability–weighted cause-specific hazards models. Variables that were imbalanced across glucocorticoid categories after inverse probability weighting were added as covariates to weighted models (Medicare: opioids, prior hospitalizations, emergency department visits, and outpatient visits for all analyses; chronic pain for methotrexate analyses; and antibiotics for analyses of older patients; Optum: opioids, prior hospitalizations, and outpatient visits for all analyses; chronic pain for methotrexate analyses; proton-pump inhibitors, prior biologics, Charlson Comorbidity Index score, hypertension, emergency department visits, and outpatient rheumatology visits for biologic analyses; DMARD type and Charlson Comorbidity Index score for analyses of younger patients; and prior hospitalized infections for analyses of older patients).
Because of different age distributions in the data sets, in Medicare, younger patients were aged <75 y and older patients were aged ≥75 y. In Optum, younger patients were aged <65 y and older patients were aged ≥65 y.
Results were similar in sensitivity analyses that censored before glucocorticoid dose changes; treated glucocorticoid dose as time varying; excluded patients with hospitalized infection for 180 days or fewer before the index date; or excluded patients with COPD, asthma, or extra-articular RA. Neither adding cumulative glucocorticoid dose in the first 9 months of the baseline period nor restricting to patients receiving glucocorticoids at the time of DMARD course initiation changed results substantially (Supplement Tables 8 to 13, available at Annals.org).
Residual Bias Analysis
The potential effect of residual bias due to differences in disease activity in exposure groups was assessed using data from 11 285 patients with CRP values in Optum. The amount of residual bias required to explain the observed results (for example, odds ratio >2 for association between ≤5 mg of glucocorticoids per day and moderate or high disease activity; relative risk >4 for association between disease activity and serious infection) was substantially greater than the residual bias estimated, even with conservative assumptions (Supplement Tables 14 to 17, available at Annals.org).
Discussion
In this retrospective cohort study, we found that long-term use of glucocorticoids was common in patients with RA (>40% of patients), and that even low-dose glucocorticoids (≤5 mg/d) were associated with a small but statistically significant and clinically meaningful increase in the risk for hospitalized infection. Associations were consistent in the 2 data sets and were similar in older versus younger patients and in biologic versus non–biologic-treated patients. This study advances the existing literature by directly evaluating the risks of a stable dose of low-dose glucocorticoids in patients receiving a stable DMARD regimen, a situation frequently faced by clinicians. We have also focused on a period in which the risk for confounding due to elevated disease activity and RA treatment changes would be expected to be lower. In addition to informing the treatment of RA, these results may inform the management of other patient populations receiving low-dose glucocorticoids.
The approximately 2-fold greater risk for infection with more than 10 mg of glucocorticoids per day is similar to results from prior studies (11-15). Long-term use of glucocorticoids at this dose, however, was uncommon. More commonly, patients with RA were receiving low-dose glucocorticoids. Even glucocorticoid doses of 5 mg or less per day were associated with a small but potentially clinically meaningful risk for hospitalized infection. Although relative risks were similar in the subgroups studied, the absolute risk associated with glucocorticoids depends on the baseline risk in the population—for this reason, we found a larger absolute risk in Medicare patients than in the healthier and younger Optum patients. The association between low-dose glucocorticoids and infection was remarkably consistent with several other observational studies (11,12, 14-18), despite our unique design characteristics.
To put the observed risk into perspective, the risk for hospitalized infection associated with 5 mg or less of glucocorticoids per day was similar to the risk associated with biologic therapies in prior studies, with an absolute difference in the 1-year incidence of 1.2% (95% CI, 0.7% to 1.8%) in Optum and 2.4% (CI, 2.0% to 2.9%) in Medicare with 5 mg or less per day versus no use (31). With the highly publicized risks of biologics, physicians and patients may assume that these medications carry much higher risks than low-dose glucocorticoids. Indeed, older patients and patients with comorbidities are less likely to receive biologics but are equally likely to receive glucocorticoids for their RA (32).
Several randomized studies have demonstrated that even low-dose glucocorticoids can provide a substantial benefit for control of disease activity, although these studies were not powered to assess safety and typically did not enroll higher-risk patients with advanced age or high comorbidity burdens, as are cared for in real-world settings (1-3). The safety results from this study can assist clinicians making treatment decisions. Our results support guideline recommendations to minimize the long-term use of glucocorticoids and attempt tapering, but the risks of low-dose glucocorticoids may be acceptable in patients who are obtaining substantial benefit and who have increased disease activity when attempts are made to taper. In addition, explaining that infection risks with low-dose glucocorticoids and biologics are similar may be helpful when counseling patients who require higher doses of glucocorticoids but who are hesitant to start a biologic treatment because of infection concerns. We did not evaluate the risks for other adverse outcomes (for example, fracture, cataracts, diabetes, and cardiovascular events), which have been associated with low-dose glucocorticoids in some studies (18). The potential for confounding is a key concern in any observational study, particularly observational studies involving glucocorticoids. In our study, differences in patient characteristics at baseline were especially pronounced for those receiving higher doses of glucocorticoids. Characteristics were much more similar for patients receiving low-dose glucocorticoids and were well balanced after inverse probability weights were applied. We included health care use and other surrogates of disease activity and severity. Prior hospitalized infections and certain comorbidities, such as COPD, were associated with glucocorticoid use and outcomes, but sensitivity analyses excluding patients with recent infections or with COPD, asthma, or extra-articular RA showed similar results. In addition, CRP values obtained in a subset of Optum patients showed relatively small differences in this measure of disease activity in patients receiving 5 mg or less of glucocorticoids per day versus none. Although residual confounding from disease activity remains possible, our residual bias assessment using CRP data suggests that substantially more confounding than suggested by our data would be required to fully explain our results. Although CRP level is an imperfect measure of disease activity, the associations between disease activity and infection (relative risk >4) would also need to be substantially greater than those shown in previous work (incidence rate ratio, 1.56 [CI, 0.94 to 2.59]) (11).
Several other limitations should be noted. Glucocorticoid dose was based on filled prescriptions, but prescriptions may not match physician instructions or patient behavior (for example, physicians prescribing extra tablets in case of a disease flare). For this reason, we averaged glucocorticoid dose over 3 months. In general, this misclassification of dose may be expected to overestimate glucocorticoid exposure and bias toward the null. We could not compare the safety of alternate-day versus daily glucocorticoids and did not include intravenous or intramuscular glucocorticoid use, although this is expected to be infrequent compared with oral use. Given that patients tend to reduce glucocorticoids over time, censoring was more frequent in patients receiving higher doses of glucocorticoids, but we found similar results in analyses allowing glucocorticoid dose to vary over time. We only evaluated infections occurring during an inpatient hospitalization because these have been well validated; we did not evaluate outpatient infections. We were specifically interested in patients receiving stable DMARD therapy to evaluate a clinical situation in which glucocorticoid use is particularly controversial and to minimize confounding due to disease activity and RA treatment changes (expected to be higher immediately after DMARD initiation); results may not be generalizable to patients initiating DMARDs or patients who discontinue DMARD treatment before 6 months.
In conclusion, for patients with RA receiving stable DMARD therapy, continued use of glucocorticoids is associated with an increased risk for hospitalized infection. Even low-dose therapy of 5 mg or less per day is associated with a small but significant risk for infection, which is similar in magnitude to risks seen with biologic therapies in prior studies. Clinicians should avoid long-term use of higher-dose glucocorticoids and should weigh the benefits of low-dose therapy in individual patients with these potential risks.
Supplementary Material
Acknowledgments
Grant Support: Dr. George is supported by grant 1K23AR073931-01 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases.
Footnotes
Disclosures: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M20-1594.
Reproducible Research Statement: Study protocol: See the Supplement. Statistical code: Available from Dr. George (, michael.george@pennmedicine.upenn.edu). Data set: Data use is governed by data use agreements. Medicare data are available through the Centers for Medicare & Medicaid Services. OptumInsight data are available to license through Optum.
Contributor Information
Michael D. George, University of Pennsylvania, Perelman School of Medicine, Philadelphia, Pennsylvania.
Joshua F. Baker, University of Pennsylvania and Philadelphia Veterans Affairs Medical Center, Philadelphia, Pennsylvania.
Kevin Winthrop, Oregon Health & Science University, Portland, Oregon.
Jesse Y. Hsu, University of Pennsylvania, Perelman School of Medicine, Philadelphia, Pennsylvania.
Qufei Wu, University of Pennsylvania, Perelman School of Medicine, Philadelphia, Pennsylvania.
Lang Chen, University of Alabama at Birmingham, Birmingham, Alabama.
Fenglong Xie, University of Alabama at Birmingham, Birmingham, Alabama.
Huifeng Yun, University of Alabama at Birmingham, Birmingham, Alabama.
Jeffrey R. Curtis, University of Alabama at Birmingham, Birmingham, Alabama.
References
- 1.Bakker MF, Jacobs JW, Welsing PM, et al. ; Utrecht Rheumatoid Arthritis Cohort Study Group. Low-dose prednisone inclusion in a methotrexate-based, tight control strategy for early rheumatoid arthritis: a randomized trial. Ann Intern Med. 2012;156:329–39. [PMID: 22393128] doi: 10.7326/0003-4819-156-5-201203060-00004 [DOI] [PubMed] [Google Scholar]
- 2.Svensson B, Boonen A, Albertsson K, et al. Low-dose prednisolone in addition to the initial disease-modifying antirheumatic drug in patients with early active rheumatoid arthritis reduces joint destruction and increases the remission rate: a two-year randomized trial. Arthritis Rheum. 2005;52:3360–70. [PMID: 16255010] [DOI] [PubMed] [Google Scholar]
- 3.Buttgereit F, Mehta D, Kirwan J, et al. Low-dose prednisone chronotherapy for rheumatoid arthritis: a randomised clinical trial (CAPRA-2). Ann Rheum Dis. 2013;72:204–10. [PMID: 22562974] doi: 10.1136/annrheumdis-2011-201067 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Singh JA, Saag KG, Bridges SL Jr, et al. 2015 American College of Rheumatology guideline for the treatment of rheumatoid arthritis. Arthritis Rheumatol. 2016;68:1–26. [PMID: 26545940] doi: 10.1002/art.39480 [DOI] [PubMed] [Google Scholar]
- 5.Smolen JS, Landewé R, Bijlsma J, et al. EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs: 2016 update. Ann Rheum Dis. 2017;76:960–977. [PMID: 28264816] doi: 10.1136/annrheumdis-2016-210715 [DOI] [PubMed] [Google Scholar]
- 6.Caplan L, Wolfe F, Russell AS, et al. Corticosteroid use in rheumatoid arthritis: prevalence, predictors, correlates, and outcomes. J Rheumatol. 2007;34:696–705. [PMID: 17299838] [PubMed] [Google Scholar]
- 7.Makol A, Davis JM 3rd, Crowson CS, et al. Time trends in glucocorticoid use in rheumatoid arthritis: results from a population-based inception cohort, 1980–1994 versus 1995–2007. Arthritis Care Res (Hoboken). 2014;66:1482–8. [PMID: 24821680] doi: 10.1002/acr.22365 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Thiele K, Buttgereit F, Huscher D, et al. ; German Collaborative Arthritis Centres. Current use of glucocorticoids in patients with rheumatoid arthritis in Germany. Arthritis Rheum. 2005;53:740–7. [PMID: 16208641] [DOI] [PubMed] [Google Scholar]
- 9.Wallace BI, Lin P, Kamdar N, et al. Patterns of glucocorticoid prescribing and provider-level variation in a commercially insured incident rheumatoid arthritis population: a retrospective cohort study. Semin Arthritis Rheum. 2020;50:228–236. [PMID: 31522762] doi: 10.1016/j.semarthrit.2019.09.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Palmowski Y, Buttgereit T, Dejaco C, et al. "Official view" on glucocorticoids in rheumatoid arthritis: a systematic review of international guidelines and consensus statements. Arthritis Care Res (Hoboken). 2017;69:1134–41. [PMID: 28029750] doi: 10.1002/acr.23185 [DOI] [PubMed] [Google Scholar]
- 11.Au K, Reed G, Curtis JR, et al. ; CORRONA Investigators. High disease activity is associated with an increased risk of infection in patients with rheumatoid arthritis. Ann Rheum Dis. 2011;70:785–91. [PMID: 21288960] doi: 10.1136/ard.2010.128637 [DOI] [PubMed] [Google Scholar]
- 12.Grijalva CG, Chen L, Delzell E, et al. Initiation of tumor necrosis factor-α antagonists and the risk of hospitalization for infection in patients with autoimmune diseases. JAMA. 2011;306:2331–9. [PMID: 22056398] doi: 10.1001/jama.2011.1692 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Curtis JR, Patkar N, Xie A, et al. Risk of serious bacterial infections among rheumatoid arthritis patients exposed to tumor necrosis factor alpha antagonists. Arthritis Rheum. 2007;56:1125–33. [PMID: 17393394] [DOI] [PubMed] [Google Scholar]
- 14.Wolfe F, Caplan L, Michaud K. Treatment for rheumatoid arthritis and the risk of hospitalization for pneumonia: associations with prednisone, disease-modifying antirheumatic drugs, and anti-tumor necrosis factor therapy. Arthritis Rheum. 2006;54:628–34. [PMID: 16447241] [DOI] [PubMed] [Google Scholar]
- 15.Smitten AL, Choi HK, Hochberg MC, et al. The risk of hospitalized infection in patients with rheumatoid arthritis. J Rheumatol. 2008;35:387–93. [PMID: 18260176] [PubMed] [Google Scholar]
- 16.Dixon WG, Abrahamowicz M, Beauchamp ME, et al. Immediate and delayed impact of oral glucocorticoid therapy on risk of serious infection in older patients with rheumatoid arthritis: a nested case-control analysis. Ann Rheum Dis. 2012;71:1128–33. [PMID: 22241902] doi: 10.1136/annrheumdis-2011-200702 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Weaver A, Troum O, Hooper M, et al. Rheumatoid arthritis disease activity and disability affect the risk of serious infection events in RADIUS 1. J Rheumatol. 2013;40:1275–81. [PMID: 23772079] doi: 10.3899/jrheum.121288 [DOI] [PubMed] [Google Scholar]
- 18.Wilson JC, Sarsour K, Gale S, et al. Incidence and risk of glucocorticoid-associated adverse effects in patients with rheumatoid arthritis. Arthritis Care Res (Hoboken). 2019;71:498–511. [PMID: 29856128] doi: 10.1002/acr.23611 [DOI] [PubMed] [Google Scholar]
- 19.Centers for Medicare & Medicaid Services. Brief summaries of Medicare & Medicaid. Accessed at www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/MedicareProgramRatesStats/SummaryMedicareMedicaid.html on 20 June 2020.
- 20.Kim SY, Servi A, Polinski JM, et al. Validation of rheumatoid arthritis diagnoses in health care utilization data. Arthritis Res Ther. 2011;13:R32. [PMID: 21345216] doi: 10.1186/ar3260 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Cannon GW, Mikuls TR, Hayden CL, et al. Merging Veterans Affairs rheumatoid arthritis registry and pharmacy data to assess methotrexate adherence and disease activity in clinical practice. Arthritis Care Res (Hoboken). 2011;63:1680–90. [PMID: 21905260] doi: 10.1002/acr.20629 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Schneeweiss S, Setoguchi S, Weinblatt ME, et al. Anti-tumor necrosis factor α therapy and the risk of serious bacterial infections in elderly patients with rheumatoid arthritis. Arthritis Rheum. 2007;56:1754–64. [PMID: 17530704] [DOI] [PubMed] [Google Scholar]
- 23.Gagne JJ, Glynn RJ, Avorn J, et al. A combined comorbidity score predicted mortality in elderly patients better than existing scores. J Clin Epidemiol. 2011;64:749–59. [PMID: 21208778] doi: 10.1016/j.jclinepi.2010.10.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.U.S. Census Bureau. American Community Survey (ACS). Accessed at www.census.gov/programs-surveys/acs on 20 June 2020.
- 25.Spreeuwenberg MD, Bartak A, Croon MA, et al. The multiple propensity score as control for bias in the comparison of more than two treatment arms: an introduction from a case study in mental health. Med Care. 2010;48:166–74. [PMID: 20068488] doi: 10.1097/MLR.0b013e3181c1328f [DOI] [PubMed] [Google Scholar]
- 26.Feng P, Zhou XH, Zou QM, et al. Generalized propensity score for estimating the average treatment effect of multiple treatments. Stat Med. 2012;31:681–97. [PMID: 21351291] doi: 10.1002/sim.4168 [DOI] [PubMed] [Google Scholar]
- 27.McCaffrey DF, Griffin BA, Almirall D, et al. A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med. 2013;32:3388–414. [PMID: 23508673] doi: 10.1002/sim.5753 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Austin PC, Schuster T. The performance of different propensity score methods for estimating absolute effects of treatments on survival outcomes: a simulation study. Stat Methods Med Res. 2016;25:2214–2237. [PMID: 24463885] [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Cole SR, Hernán MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168:656–64. [PMID: 18682488] doi: 10.1093/aje/kwn164 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Schneeweiss S Sensitivity analysis and external adjustment for unmeasured confounders in epidemiologic database studies of therapeutics. Pharmacoepidemiol Drug Saf. 2006;15:291–303. [PMID: 16447304] [DOI] [PubMed] [Google Scholar]
- 31.Singh JA, Cameron C, Noorbaloochi S, et al. Risk of serious infection in biological treatment of patients with rheumatoid arthritis: a systematic review and meta-analysis. Lancet. 2015;386:258–65. [PMID: 25975452] doi: 10.1016/S0140-6736(14)61704-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.George MD, Sauer BC, Teng CC, et al. Biologic and glucocorticoid use after methotrexate initiation in patients with rheumatoid arthritis. J Rheumatol. 2019;46:343–350. [PMID: 30275262] doi: 10.3899/jrheum.180178 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.