An Observational Study of Cardiovascular Risks Associated with Rheumatoid Arthritis Therapies: A Comparison of Two Analytical ApproachesLisa J Herrinton, PhD; G Thomas Ray, MBA; Jeffrey R Curtis, MD, MS, MPH; Perm J 2018;22:17-101 [Full Citation] https://doi.org/10.7812/TPP/17-101E-pub: 07/19/2018ABSTRACTBackground and Objectives: Comparative safety studies typically use hierarchical treatment categories that lump monotherapy and combination therapy. The consequence of this approach on study results is not clear. For example, studies of tumor necrosis factor inhibitors usually lump users regardless of whether they are using the drug alone or in combination with other agents. This study explored the importance of lumping vs splitting users of monotherapy and combination therapy. We also explored whether the timing of disenrollment from Health Plan membership was informative as an outcome variable when interpreting unmeasured, time-varying confounding. INTRODUCTIONA methodologic challenge associated with observational drug safety research is treatment classification. Research has shown that to minimize bias and provide direct contrasts of therapeutic options, users of drug A should be compared with users of a hypothetical drug B that is in therapeutic equipoise with drug A with respect to known effectiveness, safety, and access. In contrast, comparing users of drug A with nonusers of drug A is regarded as poor methodology because of the potential for selection bias.1 Many drugs are used for many diseases, however, and exposure to these drugs changes dynamically during the course of a patient’s treatment trajectory. Several approaches can be used to classify combination therapy. The most conventional option is to lump treatments for the most severe and intractable disease without regard for whether the treatment is used alone or in combination. A second option is to split monotherapy from combination therapy to create more categories. The relative advantages and disadvantages of these approaches have not been described. We had the opportunity to study the importance of treatment classification in an observational study of cardiovascular disease in patients with rheumatoid arthritis (RA). Patients with RA are at increased risk for cardiovascular disease, with traditional cardiovascular risk factors playing a limited role.2-4 Observational studies indicate that treatment using biologic and nonbiologic disease-modifying antirheumatic drugs (DMARDs) may reduce cardiovascular risk.5-9 Studies summarized in meta-analyses indicate that use of both methotrexate (MTX) and tumor necrosis factor (TNF) inhibitors is associated with reduced risk for cardiovascular events in RA.10-22 However, few studies have distinguished monotherapy (anti-TNF alone) from combination therapy (eg, anti-TNF plus MTX).20-24 The objective of this report was to provide an empiric example that contrasted lumping and splitting of treatment classification and to assess whether this approach leads to different results. We also explored the association of treatment with disenrollment from the Kaiser Foundation Health Plan. Disenrollment data are available in many data sets and often reflect a patient’s employment stability, income, insurance quality, and baseline health status.25-27 We used disenrollment as a measure of selection factors that may be associated with treatment. This report has implications for improving the transparency of observational drug safety research. METHODSSettingKaiser Permanente Northern California is a nonprofit, integrated health care delivery system using an electronic medical record that provides data on diagnostic, laboratory, and pharmacy information. Identification of Patients with Rheumatoid Arthritis Receiving MethotrexateWe conducted a new-user observational cohort study.28 We extracted all visits to rheumatologists for RA (International Classification of Diseases, Ninth Revision [ICD-9] 714.xx) between 2002 and 2013, retaining those whose first diagnosis occurred during 2003 or later. We then restricted the cohort to patients treated with MTX subsequent to their initial RA visit. The index date was the date of first MTX use. We excluded patients who were younger than age 18 years on their index date; those who did not have continuous Health Plan membership during the year before their initial RA diagnosis or their index date; and those who used the following drugs during the year before their RA diagnosis or index date: MTX, TNF inhibitors, and other biologic DMARDs (abatacept, anakinra, rituximab, or tocilizumab). We further excluded those who, during the year before their index date, had a serious cardiovascular event; had a diagnosis of psoriasis (ICD-9 696.1x); or had transplantation, advanced kidney or liver disease, or cancer other than nonmelanoma skin cancer (Supplemental Table 1, available online at: www.thepermanentejournal.org/files/2018/17-101-Suppl.pdf). Cardiovascular eventsCardiovascular events were defined based on hospital or Emergency Department encounters for acute myocardial infarction, stroke, or coronary revascularization (Supplemental Table 2, available online at: www.thepermanentejournal.org/files/2018/17-101-Suppl.pdf). Rheumatoid Arthritis Treatment ClassificationsFor each patient, we extracted all prescription fills and infusions of MTX, TNF inhibitors, DMARDs, prescription nonsteroidal anti-inflammatory drugs (NSAIDs), and glucocorticoids (Supplemental Table 3, available online at: www.thepermanentejournal.org/files/2018/17-101-Suppl.pdf). The US Food & Drug Administration approved the first TNF inhibitors, infliximab and etanercept, in 1998; adalimumab in 2002; and certolizumab and golimumab in 2013. The latter two drugs were not used in this cohort during the study period. We classified treatment in 2 ways: 9 split categories vs 5 lumped categories. The 9 split categories classified each patient’s daily use into 1 of these categories: 1) TNF inhibitor monotherapy; 2) concurrent use of TNF inhibitor and only MTX; 3) concurrent use of TNF inhibitors with MTX plus 1 or more other nonbiologic DMARDs; 4) concurrent use of TNF inhibitor and non-MTX nonbiologic DMARDs; 5) concurrent use of MTX and non-MTX nonbiologic DMARDs; 6) non-MTX nonbiologic DMARDs only; 7) RA biologics other than a TNF inhibitor (abatacept, anakinra, rituximab, and tocilizumab), which were rarely used by this cohort; 8) no systemic RA treatments; and 9) MTX monotherapy. The 5 lumped categories combined groups 1 to 4 into a single group (TNF inhibitor mono or combination therapy) and groups 5 and 9 into a single group (MTX mono or combination therapy without TNF inhibitor). The lumping approach is similar to that used by most other studies.11,12,23 Daily use of glucocorticoids was separately classified into none, less than 10 mg/d of prednisone equivalent, and 10 or more mg/d of prednisone equivalent. For each drug category, we constructed medication episodes with start and end dates allowing for stockpiling. To allow for less than 100% adherence with instructed use and potentially continuing cardiovascular effects after discontinuation, we overlooked gaps in use of up to 90 days and added 90 days to the end of each episode. In a sensitivity analysis, a 180-day gap was used. Cardiovascular risk factors and other baseline measuresWe derived cardiovascular risk factors and other baseline measures recorded during the year before the index date from pharmacy and laboratory data, vital signs, and diagnoses (Supplemental Tables 4 and 5, available online at: www.thepermanentejournal.org/files/2018/17-101-Suppl.pdf). Risk factors included use of cardiovascular medications, hypertension, body mass index, dyslipidemia, tobacco use, and the Charlson comorbidity index.29 Statistical analysisWe used extended Cox regression analysis to estimate hazard ratios (HR) and 95% confidence intervals (CIs) for the association of RA treatments with risk for a major cardiovascular event.30 The comparator was use of MTX monotherapy. We ran separate models for RA treatments coded as 9 split categories vs 5 lumped categories. The extended Cox models were fit with a separate record per person per day, with the key variables specifying the RA treatments used on that day. The outcome was the date of the first cardiovascular event. Patients were censored on the date of disenrollment, diagnosis of psoriasis, transplantation, diagnosis of advanced kidney or liver disease, diagnosis of cancer (other than nonmelanoma skin cancer), death, or end of the study on October 31, 2014. For each analysis, we ran five statistical models with increasing levels of adjustment. In the most fully adjusted models, glucocorticoid use was included as a time-varying covariate. In a second set of analyses, we assessed disenrollment as the outcome variable. Analyses were performed using SAS, Version 9.3 (SAS Institute Inc, Cary, NC). The study was approved by the institutional review board at the Kaiser Foundation Research Institute. RESULTSThe final study cohort included 5885 adults who met eligibility criteria during the study period (2002-2013). This was after excluding 1896 patients who lacked continuous Health Plan membership and 2803 patients who used an RA-related drug during the year before their first RA diagnosis or index date. We also excluded 86 patients who had experienced a serious cardiovascular event, 268 who had psoriasis, and 296 who underwent transplantation or had advanced kidney or liver disease or cancer during the year before their index date. Women comprised three-quarters of the final study cohort; their average age was 57 years, and 50% were white. Among these patients, 13% had diabetes, 33% had dyslipidemia, and 34% had hypertension (Table 1). The study population contributed 24,344 person-years of follow-up from the index date to the censoring date, or an average of 4.24 years per patient. The average time from RA diagnosis to initiation of MTX therapy was 205 days, and for the 1557 patients who subsequently started TNF inhibitors, the mean time from MTX initiation to TNF inhibitor initiation was 575 days. TNF-inhibitor use was more common in younger patients, whereas other treatments were more common in older patients (TNF-inhibitor average age 55 years; other treatments, average age 60 years; no treatment, average age 60 years; Table 2). Also, TNF-inhibitor use was more common in patients who did not have diabetes, hypertension, or dyslipidemia at baseline. There were 238 first cardiovascular events, yielding a crude incidence rate of 9.78 per 1000 person-years. The crude annual incidence rate of cardiovascular events per 1000 person-years was 3.63 (CI = 0.07 - 7.20) among current users of TNF inhibitor monotherapy, 4.79 (CI = 1.24 - 8.34) for current users of TNF inhibitors plus MTX only, 10.10 (CI = 7.81 - 12.38) for current users of MTX monotherapy, and 10.90 (CI = 8.14 - 13.66) for those who were not current users of any systemic RA treatments (Table 2). The association of RA treatment with major cardiovascular events is shown in Table 3 and Figure 1. Table 3 shows the confounding effects of observed covariates. With the exception of confounding by age and sex, these results did not change to an important degree with the inclusion of larger numbers of covariates, suggesting that the cohorts were comparable in the balance of observed covariates among the treatment groups. Table 3 and Figure 1 illustrate the effects of using different approaches to coding RA treatment after adjusting for all available covariates. Using 9 split-treatment categories, the adjusted HR and CI for the association of current use of various regimens involving TNF inhibitor therapy compared with current use of MTX monotherapy ranged between 0.70 (CI = 0.25 - 1.96) for TNF inhibitor monotherapy and 1.82 (CI = 0.86 - 3.85) for TNF inhibitor therapy combined with one or more other non-MTX nonbiologic agents. About one-quarter of anti-TNF use was combined with a non-MTX nonbiologic agent. The HR for MTX combination therapy compared with MTX monotherapy was 1.28 (CI = 0.88 - 1.87). Using 5 lumped categories, the HR for the association of TNF inhibitor therapy (mono or combination therapy) compared with MTX (mono or combination therapy) was 0.86 (CI = 0.54 - 1.37). We then evaluated the HR for the association of current drug treatment with disenrollment from the Health Plan membership after adjusting for covariates (Table 4, Figure 2). We found no disenrollment association with current use of TNF inhibitor monotherapy compared with MTX monotherapy (HR = 0.97; CI = 0.72 - 1.30). However, there was a nonsignificant disenrollment association for current users of TNF inhibitor combination therapies (compared with MTX monotherapy, HRs = 0.67 - 0.80 for categories A2-A4). Risk for disenrollment was highest when patients were not using any RA-related therapy (compared with MTX monotherapy; HR = 1.55; CI = 1.33 - 1.82); it is important to bear in mind that every patient started on MTX therapy. In sensitivity analyses, we found no substantive changes in the associations using a 180-day gap vs a 90-day gap. DISCUSSIONMost new-user observational studies comparing the safety of RA treatments in a head-to-head fashion lumped TNF inhibitor monotherapy with TNF inhibitor combination therapy.11,12 To assess whether lumping treatment classification might influence study interpretation, we conducted an exploratory analysis using a dataset of RA and major cardiovascular outcomes. We compared two approaches for drug classification, one in which we used five lumped categories and the other in which we used nine split categories to separate monotherapies and specific drug combinations. In the example we present, current users of combinations of at least three drugs involving TNF inhibitor plus MTX plus other nonbiologic drugs appeared to be at higher risk for a major cardiovascular event than patients using a TNF inhibitor (with or without MTX) without a third nonbiologic agent. We also observed that people not currently using any RA-related drug or using multiple nonbiologic drug combinations were at relatively high risk for a major cardiovascular event. These results stress the value of assessing the heterogeneity of treatment effects among users of monotherapies vs various drug combinations. Treatment heterogeneity can result from patient selection factors or genuine differences in the effects of treatments on cardiovascular risk. For example, both TNF inhibitor and MTX are associated with lower cardiovascular risk for patients with RA, and it is reasonable to consider that users of TNF inhibitor and MTX combination therapy are at lower risk than users of either agent taken alone. We further observed a relatively low risk for disenrollment among current users of most combination therapies. This association is not statistically significant, but, in the analysis of bias, statistical significance is not essential. We further noted a relatively high risk for disenrollment among those not currently using an RA-related drug. Every patient started as an MTX user; however, during the course of follow-up, some experienced periods without using any RA drug during which they were more likely to disenroll. Patients currently using various drug combinations differed from those using monotherapy with respect to factors that influence disenrollment. Similarly, persons without current exposure to any RA-related drug differed from those with current use of any drug. Factors that influence disenrollment include employment, disease severity, patient frailty, and health-seeking behavior, among others. These same unobserved differences among patients may influence risk for a major cardiovascular event. Thus, lumping TNF inhibitor users into a single category without regard for the number or nature of cotherapies can reduce study transparency because the category combines different treatment effects and different types of patients using different combinations of therapies in unknown proportions. The HR for the association of disenrollment with current nonuse of RA treatment was 1.55 (CI = 1.33-1.82). This is not a notable difference, which suggests that the magnitude of confounding may not be large in this controlled study. However, this association is more substantial than the cardiovascular disease benefit observed when using TNF inhibitor plus MTX compared with MTX monotherapy (HR = 0.80; CI = 0.61-1.05). Because cardiovascular disease is relatively common in RA, small effects are clinically significant, and this question has generated a large number of published reports from observational studies.10-23 Like many observational comparative safety studies that use clinical data, we did not have access to information on the severity or activity of RA because these data were not recorded in the medical record. Among measured covariates, we found that only age and sex operated as confounders of the relationship between treatments and cardiovascular risk factors. The results did not change when we controlled for cardiovascular risk factors, presumably because these factors were relatively well balanced among patients in the treatment categories. Because traditional cardiovascular risk factors play only a limited role in RA-related cardiovascular risk, however, this balance does not rule out bias related to differences in RA severity and RA-related cardiovascular risk.31-37 Indeed, these study findings suggest that bias may exist. A limitation of the study was its sample size even though this cohort represented about 1% of patients with RA in the US. In the analysis of 9 split vs 5 lumped categories, differences among HRs were modest, whereas several CIs were quite wide. Typically, study validity has primacy over study precision; nonetheless, the differences in HR across exposure categories may be attributable to chance and should be confirmed in other populations. Yet the results of the disenrollment analysis (excluding disenrollment related to death) were more precise in showing that current use of more complex drug regimens was associated with decreased risk for disenrollment, and nonuse of treatment drugs was associated with increased risk for disenrollment. Also, this study involved a community-based population in which RA severity may be lower than in academic settings or rheumatology practices. Regarding study external validity, these results are similar to those reported by Greenberg and colleagues,11 who analyzed data from approximately 100 US rheumatology practices and examined a similar composite cardiovascular end point. A strength of their study was the ability to adjust for RA severity. However, they used a hierarchy to classify treatment into 3 groups after excluding use of other biologic DMARDs and periods with no treatment. When comparing with nonbiologic DMARDs excluding MTX, they observed the following HRs: TNF inhibitor combination therapy (including MTX): HR = 0.39 (CI = 0.19-0.82) and MTX alone or in combination with any other nonbiologic DMARD: HR = 0.94 (CI = 0.49-1.80). When we used 5 lumped categories (but treated other nonbiologic DMARDs as the reference group to allow direct comparison), we observed the following HRs: TNF inhibitor combination therapy HR = 0.73 (CI = 0.42-1.25) and MTX HR = 0.85 (CI = 0.56-1.27). Although trends in the present study were similar to those reported by Greenberg and colleagues,11 we observed more modest differences between TNF inhibitor combination therapies and other therapies. However, considering that the TNF inhibitor combination therapy group includes an indeterminate amount of MTX use and MTX use includes some combination of MTX and other nonbiologic DMARDs, these lumped classifications can be difficult to fully interpret or compare, particularly considering the associations of more complex drug regimens with continued enrollment. The analyses using 9 split categories of treatment groups indicate possible heterogeneity among patients using TNF inhibitors, depending upon whether they use TNF inhibitors only or TNF inhibitors in combination with MTX and/or other DMARDs. In our population, the proportion of follow-up time with any RA-related therapy was 77%. This statistic has not been reported in other studies. However, the percentage of patients who receive at least one dispensing (or infusion) of an RA-related drug during 1 year is a quality metric and has been reported. In 245 Medicare managed care plans (2005-2008), the number of patients with at least 1 dispensing in each year ranged between 16% and 87%.38 In the TennCare population, this proportion was 71% in 2004.39 ConclusionDrug discoveries and drug costs are accelerating, increasing the need for research on associated benefits and risks. In the case of TNF inhibitor therapy, numerous observational studies have reported benefits associated with anti-TNF and MTX therapies and major cardiovascular outcomes. These outcomes were not evaluated in clinical trials, and selection bias could result in the benefits of these drugs being overstated or understated in some studies. A randomized controlled trial on this issue has not been published, and prescribing decisions are based on observational studies. The methods we recommend for classifying monotherapy and combination therapy will increase the transparency of studies and meta-analyses that assess treatment and comparison groups across settings and studies. Disclosure StatementThe author(s) have no conflicts of interest to disclose. AcknowledgmentsWe are grateful to an anonymous reviewer who recommended reconsideration of the scope of this work. Brenda Moss Feinberg, ELS, provided editorial assistance. How to Cite this ArticleHerrinton LJ, Ray GT, Curtis JR, et al. An observational study of cardiovascular risks associated with rheumatoid arthritis therapies: A comparison of two analytical approaches. Perm J 2018;22:17-101. DOI: https://doi.org/10.7812/TPP/17-101 References1. Brookhart MA, Stürmer T, Glynn RJ, Rassen J, Schneeweiss S. Confounding control in healthcare database research: Challenges and potential approaches. Med Care 2010 Jun;48(6 Suppl):S114-20. DOI: https://doi.org/10.1097/mlr.0b013e3181dbebe3.
|
ETOC
Click here to join the eTOC list or text ETOC to 22828. You will receive an email notice with the Table of Contents of The Permanente Journal.
CIRCULATION
2 million page views of TPJ articles in PubMed from a broad international readership.
Indexing
Indexed in MEDLINE, PubMed Central, EMBASE, EBSCO Academic Search Complete, and CrossRef.