|Home | About | Journals | Submit | Contact Us | Français|
Although tumor necrosis factor alpha (TNF-α) antagonists are increasingly used in place of non-biologic comparator medications, their safety profile remains incomplete.
To determine whether initiation of TNF-α antagonists compared with non-biologic comparators is associated with an increased risk of serious infections.
Within a US multi-institutional collaboration, we assembled retrospective cohorts (1998–2007) of patients with rheumatoid arthritis (RA), inflammatory bowel disease (IBD), and psoriasis, psoriatic arthritis or ankylosing spondylitis (PsO-PsA-AS) combining data from Kaiser Permanente Northern California, New Jersey and Pennsylvania Pharmaceutical Assistance programs, Tennessee Medicaid and National Medicaid/Medicare. TNF-α antagonists and non-biologic regimens were compared in disease specific-propensity score (PS) matched cohorts using Cox regression models with non-biologics as reference. Baseline glucocorticoid use was evaluated as a separate covariate.
Infections requiring hospitalization (serious infections) during the first 12 months after initiation of TNF-α antagonists or non-biologic regimens.
Study cohorts included 10484 RA, 2323 IBD and 3213 PsO-PsA-AS PS-matched pairs using TNF-α antagonists and comparator medications. Overall, we identified 1171 serious infections, most of which (53%) were pneumonia and skin and soft tissue infections. Among RA patients, serious infection hospitalization rates were 8.16 (TNF-α antagonists) and 7.78 (comparator regimens) per 100 person-years (adjusted hazard ratio [aHR]: 1.07 (95% CI: 0.93–1.23)). Among IBD patients, rates were 10.91 and 9.60 per 100 person-years (aHR: 1.13, (0.85–1.50)). Among PsO-PsA-AS patients, rates were 5.41 and 5.19 per 100 person-years (aHR: 1.10, (0.80–1.53)). Among RA patients, infliximab was associated with a significant increase in serious infections compared with etanercept and adalimumab (aHRs: 1.27 (1.08–1.49) and 1.23 (1.02–1.48)). Baseline glucocorticoid use was associated with a dose-dependent increase in infections.
Among patients with autoimmune diseases, compared to treatment with non-biologic regimens, initiation of TNF-α antagonists was not associated with an increased risk of hospitalizations for serious infections.
Although the introduction of tumor necrosis factor (TNF)-α antagonists revolutionized the treatment of autoimmune diseases, concerns about the safety of these biologic drugs remain.1,2 Several studies reported serious infections in users of TNF-α antagonists.1–3 However, whether the risk of serious infections with TNF-α antagonists is greater than that with comparator non-biologic medications is unclear.
Available information from randomized clinical trials of TNF-α antagonists is limited because of insufficient power to assess safety outcomes conclusively.4 Moreover, the selected populations participating in the trials warrant caution in extrapolating results to the broader population of patients who receive these agents.4 Furthermore, many efficacy l trials of biologics were placebo-controlled, which limits inference about alternative treatment options for autoimmune diseases.5–10
Although observational studies have tried to fill this knowledge gap, several published studies had important limitations.2,3 Some studies aggregated TNF-α antagonists into a single category, precluding the assessment of individual medications. Similarly, serious infections were aggregated, but the role of TNF-α antagonists on specific infections, such as pneumonia, remains unknown. Furthermore, methodological differences may have contributed to dissimilar and sometimes conflicting results. Larger studies, addressing specific methodological concerns,11,12 are needed to quantify the risk of serious infections in users of specific TNF-α antagonists.
As part of a large US federally-funded multi-institutional initiative, the Safety Assessment of Biologic Therapy (SABER) project,13 we evaluated whether initiation of TNF-α antagonists was associated with an increased risk of serious infections among patients with autoimmune diseases, and whether risk varies by specific TNF-α antagonists.
This retrospective cohort study combined data from four large US automated databases.13 Exposure to TNF-α antagonists and other medications was determined using pharmacy and procedures data, and hospitalized serious infections were identified using discharge diagnoses and validated definitions. The incidence of serious infections between disease specific propensity score (PS)-matched exposure groups was compared using Cox proportional hazard regression models. Planned sensitivity and subgroup analyses evaluated the robustness of the main findings and key study assumptions.
Study databases encompassed: 1) National U.S. Medicaid and Medicare databases, excluding Tennessee (Medicaid Analytic eXtract, 2000–2005; Medicare, 2000–2006; and Medicare Part D, 2006); 2) Tennessee Medicaid (TennCare, 1998–2005); 3) The New Jersey’s Pharmaceutical Assistance to the Aged and Disabled, and the Pennsylvania’s Pharmaceutical Assistance Contract for the Elderly (PAAD/PACE, 1998–2006); and, 4) Kaiser Permanente Northern California (KPNC, 1998–2007). We used these data and a common programming algorithm to assemble retrospective cohorts of patients with autoimmune diseases who were initiating selected medications.
For each database, we identified patients with study-defined autoimmune diseases, using the earliest ICD9-coded healthcare encounter, who subsequently filled a prescription or received an infusion for a TNF-α antagonist or comparator medication (see below for medication details).13–15 We required a baseline period of 365 days with continuous enrollment in the respective database preceding the first infusion or prescription fill, to ascertain other selection criteria and study covariates. Using baseline information, patients were categorized in three mutually-exclusive groups: rheumatoid arthritis (RA), inflammatory bowel disease (IBD), and psoriasis (PsO), psoriatic arthritis (PsA) or ankylosing spondylitis (AS) (PsO-PsA-AS). Patients with diagnoses for >1 autoimmune disease were excluded (Figure 1).
Among potential cohort members, we identified new users of study medications,16 defined by a filled a prescription for a study medication after 365 baseline days without prescriptions filled for the specific study medication or others in the same group. This “first” filling date (t0) marked the beginning of follow-up. Follow-up continued through the earliest of the following dates: death, loss of enrollment, study outcome (see below), switch to another regimen or the discontinuation of the current regimen (30 days without medication), study end or 365th day of follow-up. We restricted the follow-up to 365 days as we were most interested in the period shortly after initiation of therapy,9,10 to limit the effect of time varying covariates such as glucocorticoids use, and because long term users likely differ from patients who discontinue therapy within the first year of treatment. A detailed description of the design of the SABER study has been reported elsewhere. This study was approved by the IRBs of all participating institutions and patient consent requirements were waived.13
We used pharmacy and procedures data to determine medication exposure. Study medications were classified in two groups: TNF-α antagonists (including infliximab, adalimumab and etanercept [not included for IBD]); and comparator medications. For RA, the comparator regimens were initiation of leflunomide, sulfasalazine or hydroxychloroquine after any use of methotrexate in the previous year (“Non-biologic regimens”). For IBD, the comparator group was initiation of azathioprine or 6-mercaptopurine (AZA or 6MP). For PsO-PsA-AS, the comparator group was initiation of methotrexate, hydroxychloroquine, sulfasalazine or leflunomide.
Exposed person-time encompassed all person-time covered by prescription fills and up to 30 person-days without subsequent medication available. This 30-day grace period was allowed because some residual effects of study medications could extend beyond the last day of use and to account for imperfect adherence. Both the TNF-α antagonist and non-biologic comparator regimens allowed concurrent use of methotrexate. Analyses of IBD allowed for continuation of or simultaneous initiation of AZA or 6MP in the TNF-α antagonist group.
Study outcomes were serious infections, defined as infections that required hospitalization.5,7,17 These infections were identified using definitions based on principal discharge diagnoses and included infections of the respiratory tract, skin and soft tissue, genito-urinary tract, gastrointestinal tract, central nervous system, septicemia/sepsis. Pneumonia, the most common infection, was also assessed separately. Considering medical chart reviews as reference, our definitions for serious infection hospitalizations have consistently shown positive predictive values ≥80%.18–21 Opportunistic infections and tuberculosis were not considered study outcomes.22
Baseline covariates included demographics: age, gender, race, residence (urban/rural), nursing home/community dwelling, area income, calendar year; generic markers of comorbidity and healthcare utilization: number of hospitalizations, outpatient and emergency room visits, number of different medication classes filled; surrogate markers of disease severity: extra-articular disease manifestations, number of intra-articular and orthopedic procedures, number of laboratory tests ordered for inflammatory markers, use of selected medications5,7,23–25; and, other known risk factors for infections: previous hospitalizations for infections, chronic obstructive pulmonary disease (COPD), diabetes and antibiotic use.7,13
Baseline use of oral glucocorticoids was categorized according to the average daily dose of prednisone equivalents in the 6-months preceding t0: 0, >0–<5 (low dose), 5–10 (medium dose) and >10 mg (high dose).26,27 This variable was not included in the matching strategy (see below) and its association with serious infections was ascertained separately in the final outcome models.
The effects of potential confounders were controlled for using a propensity score (PS)-matching strategy. For each database and disease group, logistic regression models estimated the predicted probabilities of exposure (to the reference regimen) for each episode of use. Thus, a PS value summarized covariate information for each episode, allowing confounding control through PS-matching and assuring shared data were not individually identifiable.13,28 Episodes of use were PS-matched using a greedy matching algorithm. Although subsequent analyses were restricted to matched episodes, crude hospitalization rates for infections estimated before and after PS matching were similar across exposures and diseases (data not shown).
Cox-proportional hazard regression models assessed the association between exposure groups and study outcomes, with stratification by database to allow the baseline hazard to vary. Since we compared PS-matched cohorts and since patients could contribute >1 episode of new use (with an updated set of covariates), we accounted for these additional correlations using the Huber-White “sandwich” variance estimator and calculated robust standard errors for all estimates.29 The proportional hazard assumption was verified for each study exposure.30 The final disease specific-outcome models for the PS-matched cohort analyses included only the exposure groups and the indicator for baseline glucocorticoid use. All statistical tests were two-sided, and a P value of less than 0.05 was considered to indicate statistical significance. Using available information, we estimated that for RA, our comparison of TNF-antagonists to non-biologic regimens would have a power of 80% to detect a hazard ratio ≥1.15 at a significance level of 0.05. Similarly, the detectable hazard ratios were ≥1.30 and ≥1.36 for IBD and Pso-PS-AS, respectively.31,32
Planned subgroup analyses included estimates by database, presence of baseline COPD, diabetes mellitus, hospitalization for a serious infection and glucocorticoid use. Each subgroup analysis required a separate PS-matching iteration. Sensitivity analyses evaluated the effects of restricting the TNF-α antagonists group to those who had used methotrexate during baseline among RA patients; restricting follow-up to 6 months; and, evaluating pneumonia hospitalizations as a separate outcome. All analyses were done in SAS.
We identified a total of 407,319 patients with autoimmune diseases who had filled prescriptions for study medications and had complete baseline data preceding that fill date. A total of 170,788 (42%) patients with other autoimmune diseases, more than one study diseases or initiating non-study regiments were excluded. We identified 35235, 7332 and 12905 patients who initiated study regimens for RA, IBD and Pso-PsA-As, respectively (Figure 1).
After application of selection criteria and PS-matching, the final RA cohorts included 10484 matched episodes of TNF-α antagonists and comparator medications use. The respective IBD and Pso-PsA-AS cohorts included 2323 and 3213 matched pairs. Overall, 20% of patients were aged ≥65 years. RA patients initiating TNF-α antagonists and comparator medications had similar baseline characteristics (Table 1). For RA patients, the median age was 58 years, 86% were women, 61% were white and 24% resided in rural areas. Of patients initiating TNF-α antagonists, 70% had used methotrexate during baseline. Similarly, after PS-matching, there were no substantial differences in the distribution of covariates between exposure groups for patients with IBD and PsO-PsA-AS (online-only material: eTables 1 & 2).
Overall, we identified 1171 serious infections, most of which (53%) were pneumonia and skin and soft tissue infections (online-only material: eTable 3).
The hospitalization rate for serious infections in the TNF-α antagonist group did not differ significantly from the rate for the non-biologic group (adjusted hazard ratio (aHR): 1.07, 95% CI: 0.92–1.23) (Figure 2). However, baseline glucocorticoid use was significantly associated with increased hospitalization risk compared with no baseline use (aHR: 1.29, 1.81 and 3.01 for low, medium and high doses of glucocorticoids, respectively) (Table 2).
Within TNF-α antagonists, the rate of serious infections among infliximab initiators was higher than that for non-biologic regimens (aHR: 1.23 (1.04–1.44)), whereas the respective rates for either etanercept or adalimumab were not. Rates were significantly higher for infliximab compared with etanercept and adalimumab (aHRs: 1.27 (1.08–1.49) and 1.23 (1.02–1.48, respectively). Rates did not differ significantly between adalimumab versus etanercept (aHR: 1.05 (0.88–1.26) (Figure 3). Concurrent use of methotrexate at t0/180 days of follow-up was 41%/41% for adalimumab vs. 44%/41% for etanercept. The respective proportions were 42%/42% vs. 53%/50% for adalimumab vs. infliximab, and 49%/48% vs. 41%/36% for infliximab vs. etanercept.
The rate of serious infections among initiators of TNF-α antagonists was not significantly higher than for initiators of AZA or 6MP (aHR: 1.13 (0.85–1.50)). Among IBD patients, there was not a significant increase in the risk of serious infections associated with baseline use of glucocorticoids (Table 2).
Among PsO-PsA-AS patients, the rate of serious infections for initiators of TNF-α antagonists was not significantly higher than for initiators of comparator medications (aHR: 1.10 (0.80–1.53)). Baseline use of glucocorticoids was associated with a significantly increased risk of serious infections, compared with no baseline use (aHR: 1.26, 1.70 and 3.68 for low, medium and high doses, respectively).
Subgroup analyses indicated that the presence of selected baseline characteristics (hospitalization for serious bacterial infections, diabetes mellitus, COPD and glucocorticoid use) increased the absolute risk of serious infections in both TNF-α antagonist initiators and in patients initiating comparator therapies to a similar degree. For example, patients with COPD had a 2–3 fold greater absolute risk of infection compared with non-COPD patients. However, within each subgroup, the relative risk of infection for TNF-α antagonists versus the comparator groups was similar, consistent with the main study findings. Similarly, estimates stratified by database generally yielded consistent results (online-only material: eTables 4 & 5), although for RA and Pso-PsA-As, crude infection rates were lower at KPNC compared to other databases.
A sensitivity analysis that compared initiation of TNF-α antagonists after baseline use of methotrexate (a subset of the main TNF-α antagonists group) with the non-biologic regimens among RA patients, yielded results similar to those from the main analyses: aHR: 1.11 (95% CI: 0.93–1.32). Similarly, when follow-up was truncated at 6 months, the aHR was 1.11 (95% CI: 0.94–1.31). Finally, restricting the study outcomes to pneumonia hospitalizations yielded results consistent with the main findings for TNF-α antagonists compared to comparator regimens, with aHRs: 1.07 (95% CI: 0.84–1.36), 1.08 (95% CI: 0.56–2.07) and 0.85 (95% CI: 0.47–1.53) for RA, IBD and PsO-PsA-AS.
In a large US multi-institution research initiative, we observed that initiation of TNF-α antagonists (as a group) was not associated with a significant increase in the risk of serious infections requiring hospitalization compared with initiation of comparator non-biologic medications. These findings were consistent across study diseases: RA, IBD and PsO-PsA-AS. Nevertheless, among RA patients, initiation of infliximab-based regimens was significantly associated with an increased risk of serious infections compared with other TNF-α antagonist regimens. We also observed a dose dependent increase in the risk of serious infections associated with baseline use of glucocorticoids among RA and PsO-PsA-AS patients.
A number of meta-analyses have summarized results from randomized clinical trials examining whether TNF-α antagonists increase the risk of infections, mainly in RA patients. These studies indicated that TNF-α antagonists increase the risk of infections (serious or non-serious) by 1.2–2.0-fold compared with placebo or other regimens, with the vast majority being placebo-controlled.4,33 A meta-analysis of randomized trials of TNF-α antagonists among IBD patients reported no difference in the frequency of serious infections between the anti-TNF and the placebo controlled groups.34 A recent systematic review of randomized trials that evaluated whether use of TNF-antagonists increased the risk of serious infections among patients with AS (mostly small and short-duration trials) reported no significant increase in the risk of serious infections compared with placebo.35 Thus, available clinical trial data are not consistent, and caution is warranted when interpreting data that combine studies with different comparators and selection criteria. Furthermore, underserved, vulnerable patients are typically excluded from clinical trials, and data on the safety of biologics for these populations are scarce.
Several observational studies examined the association of TNF-α antagonists use and the risk of serious infections (mostly in RA), but results again are inconsistent. For our analyses, we considered some of the methodological challenges that could explain differences in these studies.11,12 Some observational studies that used registry or administrative data observed an increased risk of serious infections associated with initiation or prevalent use of TNF-α antagonists compared with prevalent use of comparator drugs.5,9,10,36–38 Prevalent users have “survived” their initiation of therapy and their risk of serious outcomes may be lower than new users. Hence, to assure comparability of exposure groups, we applied a new-user design.16 Other observational studies that used a new-user design failed to identify significant increases in the risk of hospitalizations for serious infections among initiators of TNF-α antagonists compared with initiators of methotrexate.6–8,37,39 Similarly, a recent study in US veterans suggested that initiation of TNF-α antagonists was not associated with an increase in the risk of serious infections compared with initiation of other medications, including methotrexate, used to treat moderate disease.40
We reduced exposure misclassification by using pharmacy and procedure data to classify each day of follow-up during new episodes of medication use. Pharmacy records are an excellent source of exposure data since they are not subject to recall bias.41 We reduced outcome misclassification by using validated algorithms.18–22 These considerations are important for comparing our findings with other studies. For example, a registry study that reported an increased incidence of serious infections among users of TNF-α antagonists in RA used an inpatient database to identify a comparison group but information on their medications exposure was unavailable.42 Another registry-based study combined mild and severe infections but gave no specific estimate for serious infections requiring hospitalization.43
Since disease severity is an important predictor of treatment with TNF-α antagonists and could independently affect the risk of infections,44 it must be controlled during assessments of medication effects. Although direct measurements of disease severity are available from disease registries,36,38,42,43 studies that relied on administrative data, as ours, use available covariates as surrogates for disease severity. Nevertheless, these surrogates appear to be closely correlated with objective measurements of disease severity.5–7,11 Our study focused on initiation of study medications as a proxy for disease activity and minimized concerns about confounding by identifying an extensive list of relevant covariates and balancing their distribution between exposure groups using a PS-matching strategy.
Among patients with RA and PsO-PsA-AS, baseline glucocorticoid use was associated with an increased risk of serious infections in a dose-response manner irrespective of other medication regimens. Although glucocorticoid use (especially high doses) could also be a surrogate for disease severity, these associations persisted after adjustment for measured covariates, several of which are likely correlated with disease severity. Furthermore, this dose-dependent association has been consistently found in previous observational studies that addressed a similar research question.5–7,37,45–47
Unlike most prior studies, the SABER study was large enough to provide estimates for individual TNF-α antagonists among RA patients. We observed that initiation of an infliximab-based regimen was associated with a significant increase in the risk of serious infections, compared with other TNF-antagonists-regimens. This is consistent with recent studies conducted in US veterans40, in the German disease registry43, and two different large US Healthcare insurance databases.48,49 However, this differential effect was not observed in the British registry data.36 Differences in pharmacological properties, mechanisms of action and administration modes among TNF-antagonists have been postulated to explain these observations but a definitive explanation for this finding is lacking.2,43,48–51 This observation may have important implications for the interpretation of studies that report on TNF-α antagonists as a group, since the prevalence of infliximab use could influence the observed associations. Although our estimates evaluated the association between serious infections and infliximab-based regimens, as they are used in clinical practice, it must be noted that methotrexate was commonly used concurrently with infliximab to reduce the production of anti-infliximab antibodies. Disentangling the effects of individual drugs when used concurrently is difficult. Nevertheless, we noted that concurrent methotrexate use was similar for the other TNF-α antagonists.
Our findings must be interpreted in the light of several limitations. First, although pharmacy files provide excellent information on medications dispensed, the actual use of most medications is unknown. Secondly, we relied on coded information from administrative claims and other data not directly collected for clinical care to identify study outcomes. Misclassification of outcomes would make it more difficult to demonstrate true associations.5,18 However, we minimized outcome misclassification by using previously validated definitions. Third, despite the enormous effort to aggregate data from four major US data sources, our study had insufficient numbers to evaluate the role of specific TNF-α antagonists on serious infections for IBD and PsO-PsA-AS. Furthermore, our power to detect small risk increases in these groups was limited. Fourth, we were not able to study deaths due to infections as an outcome because information on cause of death was not available from all sources. Finally, the availability of clinical covariates for statistical adjustment was limited in our databases and we relied on surrogate measurements. However, although residual confounding could not be ruled out, our results were consistent in several subgroup and sensitivity analyses.
In conclusion, in this large retrospective cohort study of predominantly low income and vulnerable US patients with autoimmune diseases, we observed higher absolute rates of infection compared to previously-published cohort studies and randomized controlled trials. We found no increased risk of hospitalizations for serious infections among initiators of TNF-α antagonist (as a group) compared with initiators of comparator non-biologic therapies. However, our results also suggest that, among patients with RA, infliximab-based regimens were associated with an increase in the risk of serious infections compared with other TNF-α antagonists-based regimens. For RA and PsO-PsA-AS, this large study also demonstrated that glucocorticoid use was associated with a strong dose-dependent increase in the risk of serious infections requiring hospitalization.
eTable 1. Baseline characteristics of IBD cohorts after PS matching, SABER 1998–2007
eTable 2. Baseline characteristics of PsO-PsA-AS cohorts after PS matching, SABER 1998–2007
eTable 3. Distribution of serious infections leading to hospitalization by disease, SABER 1998–2007
*Due to Centers for Medicare & Medicaid Services (CMS) regulations, table cells with frequencies <11 cannot be reported
eTable 4. Rates of serious infections by exposure and baseline characteristics, SABER (1998–2007)
COPD, chronic obstructive pulmonary disease; DM, diabetes mellitus
eTable 5. Rates of serious infections by study database, SABER (1998–2007)
Medicaid/Medicare, National US Medicaid and Medicare databases; TennCare, Tennessee Medicaid; PAAD/PACE, New Jersey’s Pharmaceutical Assistance to the Aged and Disabled, and the Pennsylvania’s Pharmaceutical Assistance Contract for the Elderly; KPNC, Kaiser Permanente Northern California. Due to small counts, IBD estimates were not calculated for PAAD/PACE. National Medicare / Medicaid data do not include Tennessee.
Funding/Support: This work was supported by the Food and Drug Administration (FDA) US Department of Health and Human Services (DHHS) and the Agency for Healthcare Research and Quality grant U18 HS17919. Dr. Curtis receives support from the National Institutes of Health (AR053351) and AHRQ (R01HS018517). Dr. Beukelman was supported by NIH grant 5KL2 RR025776–03 via the University of Alabama at Birmingham Center for Clinical and Translational Science. Drs. Grijalva and Griffin receive support from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, grant 5P60AR56116.
Role of the Sponsors: Members of the funding organizations (Ouellet-Hellstrom [FDA] and Dr. Parivash Nourjah [AHRQ]) participated in the design and conduct of the study. The sponsors had the opportunity to review and comment on the final version of the report.
On behalf of the SABER collaboration: AHRQ, Parivash Nourjah; Brigham and Women’s Hospital, Robert Glynn, Mary Kowal, Joyce Lii, Jeremy Rassen, Sebastian Schneeweiss, Daniel Solomon; FDA, David Graham, Carolyn McCloskey, Rita Ouellet-Hellstrom, Kristin Phucas; Kaiser Permanente Northern California, Leslie Harrold, Lisa Herrinton, Liyan Liu, Marcia Raebel; University of Alabama at Birmingham, Lang Chen, Jeffrey Curtis, Elizabeth Delzell, Nivedita Patkar, Kenneth Saag, Fenglong Xie; University of Pennsylvania, Kevin Haynes, James Lewis, Vanderbilt University, Marie Griffin, Carlos Grijalva, Ed Mitchel.
Conflicts of interest Disclosures: ED received research support from Amgen. JWB reported consulting for Abbot. LJH received research support from Genentech, Centocor, and Procter and Gamble. DHS received research support from Amgen, Abbott, Lilly and Bristol-Myers Squibb. KLW reported consulting for Genentech, Abbott, CORRONA and Amgen. KGS received research support from Amgen, Genentech, Horizon and Merck. JRC received consultant fees and research grants from Roche/Genentech, UCB, Centocor, CORRONA, Amgen, Pfizer, BMS, Crescendo and Abbott. JDL has received research support from Centocor and consultant honoraria from Amgen and Pfizer. Other authors no conflicts.
Author Contributions: Dr. Chen had full access to all of the study data and takes responsibility for the integrity of the data and the accuracy of the data analysis.Study concept and design: Grijalva, Chen, Delzell, Griffin, Herrinton, Solomon, Lewis, Saag and Curtis.
Acquisition of data: Saag, Curtis, Griffin, Herrinton, Solomon
Analysis and interpretation of data: Grijalva, Chen, Delzell, Baddley, Beukelman, Winthrop, Griffin, Herrinton, Liu, Ouellet-Hellstrom, Patkar, Solomon, Lewis, Xie, Saag and Curtis.
Drafting of the manuscript: Grijalva.
Critical revision of the manuscript for important intellectual content: Grijalva, Chen, Delzell, Baddley, Beukelman, Winthrop, Griffin, Herrinton, Liu, Ouellet-Hellstrom, Patkar, Solomon, Lewis, Xie, Saag and Curtis.
Statistical analysis: Chen
Obtained funding: Saag, Curtis, Herrinton
Administrative, technical, or material support: Patkar
Study supervision: Patkar, Ouellet-Hellstrom, Saag and Curtis.
Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Food and Drug Administration, or the Agency for Healthcare Research and Quality.