Tumour necrosis factor inhibitor exposure and radiographic outcomes in Veterans with rheumatoid arthritis: a longitudinal cohort study

Abstract

Objectives

The aim was to estimate the impact of TNF inhibitor (TNFi) exposure on radiographic disease progression in US Veterans with RA during the first year after initiating therapy.

Methods

This historical longitudinal cohort design used clinical and claims data to evaluate radiographic progression after initiation of TNFi. US Veterans with RA initiating TNFi treatment (index date), ≥ 6 months pre-index and ≥ 12 months post-index VA enrolment/activity, and initial (6 months pre-index to 30 days post-index) and follow-up (10–18 months post-index) bilateral hand radiographs were eligible. The cumulative TNFi exposure and change in modified Sharp score (MSS) between initial and follow-up radiographs were calculated. The percentage of patients with clinically meaningful change in MSS (≥ 5) for each month of exposure was assessed using a longitudinal marginal structural model with inverse probability of treatment weights. Mean values and CIs were generated using 1000 bootstrapped samples.

Results

For 246 eligible patients, the mean (s.d.) age was 58 (11) years; 81% were male. The mean (s.d.) initial MSS was 19.6 (33.4) (range 0–214). The mean change (s.d.) in MSS was 0.3 (3.6) (median 0, range −19 to 22). Patients with the greatest exposure had the least radiographic progression for both crude and adjusted model analyses. Adjusted rates of MSS change ≥ 5 points (95% CI) were 10.6% (9.8%, 11.4%) for patients with 3 months of exposure compared with 5.4% (5.1%, 5.7%) for patients with 12 months of exposure.

Conclusion

One-year changes in radiographic progression were small. Patients with the greatest cumulative TNFi exposure experienced the least progression.

Key messages

• TNF inhibitors reduced radiographic progression in RA patients in clinical practice.

• These real-world observations are consistent with findings in randomized, controlled clinical trials and clinical registries.

• Rheumatologists can be confident that TNF inhibitor therapy has potential to reduce structural damage in practice.

Introduction

RA is an inflammatory disease characterized by a destructive arthritis with significant structural damage, morbidities and both personal and societal economic burdens [2]. An estimated 1.3 million adults in the US have RA [1], and the global prevalence of RA is estimated to be 0.24% [3]. The impact of RA on quality of life is significant, causing physical, social and mental disabilities [2].

Multiple effective therapies are available for the treatment of RA, including conventional synthetic DMARDs (e.g. MTX), biologic DMARDs [bDMARDs; e.g. TNF inhibitors (TNFi)] and targeted synthetic DMARDs (e.g. janus kinase inhibitors) [4]. Current guidelines emphasize the need for clinical disease assessment and the importance of individualized therapy [4]. Treatment of RA with DMARDs reduces the progression of structural damage of RA and has dramatically improved the management and quality of life for patients with RA. Among these DMARDs, TNFi therapies have reduced structural damage in patients with RA in randomized, controlled, clinical trials (RCTs).

In clinical trials, TNFi medications reduced radiographic progression compared with placebo in RA patients who were naïve to TNFi [5–8] or who had an inadequate response to the conventional synthetic DMARD MTX [5, 6, 9, 10]. A reduction in the rate of radiographic progression has also been reported in RA registries [11–14]. The gold standard used to measure radiographic damage and progression is the modified Sharp score (MSS), which assesses erosions and joint space narrowing. Greater degrees of radiographic damage and progression are associated with greater disability [15]. The impact of TNFi therapy on radiographic progression in real-world clinical practice outside of clinical trials and RA registries has not been analyzed extensively. The Veterans Health Administration (VHA) provides an ideal setting to study the impact of TNFi therapy in RA patients in a clinical practice setting across multiple sites of care. The objective of this study was to determine the percentage of patients with meaningful progression in structural damage on hand radiographs among US Veterans with RA associated with different levels of cumulative TNFi exposure during the first year of therapy.

Methods

Study design

This was an historical longitudinal cohort analysis using clinical and claims data produced during routine care. The index date was the date of first TNFi exposure between 1 January 2006 and 30 June 2015. The baseline period was the 6 months before first TNFi exposure index date (as early as 30 June 2005) and was used to evaluate baseline co-morbidities, medications and health-care utilization. The follow-up period for measuring TNFi exposure was 12 months after the index date (as late as 30 June 2016) (Fig. 1).

Fig. 1.

Study schema

TNFi exposure was measured after the index date until the time of follow-up hand X-ray. TNFi: tumor necrosis factor inhibitor.

The research protocol, patient privacy protocols and data protection plan were reviewed and approved by the institutional review board of the University of Utah and the Salt Lake City Veterans Affairs (VA) research committee to allow review of these patient records without individual informed consent by using secure records review.

Data sources

Data for this analysis were obtained from the VA Corporate Data Warehouse (CDW) and stored in the VA Informatics and Computing Infrastructure (VINCI) for research applications. Clinical and administrative data in VINCI include inpatient and outpatient encounters; pharmacy, laboratory, chemistry, radiology and electronic clinical notes; and time-stamped orders, consultations, dispensing and administration [16].

Identification and retrieval of hand radiographs

Hand radiographs were identified for incident TNFi users based on Current Procedural Terminology codes in the radiographic package in the CDW. The radiographs were collected through multiple methods, including the Veterans Information Systems and Technology Architecture (VistA) Web application, which is accessed through the National Compensation and Pension Record Interchange (CAPRI) application; VistA Imaging, which is accessed through the Advanced Web Image Viewer (AWIV) application; IMPAX Client System (for patients in the Denver, Colorado or Salt Lake City, Utah-affiliated regions); and collaboration with other VA facilities (i.e. radiographs were transferred through encrypted emails or mailed as de-identified images on compact discs and uploaded into the Salt Lake City, Utah VA system).

Patient eligibility

To be eligible, patients had to: have at least one dispensing or infusion of TNFi therapy; be enrolled and active in the VHA system for 6 months before the first administration/dispensing (index date); have an RA diagnosis based on International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 714.0 between 6 months before and 30 days after the index date; be ≥ 18 years of age on the index date; and have a bilateral hand X-ray between 6 months before and 30 days after the index date and again between 10 and 18 months after the index date. The selection of patients is outlined in Supplementary Fig. S1, available at Rheumatology Advances in Practice online.

RA diagnosis was confirmed by chart review, which required that a medical practitioner document the diagnosis of RA in the medical record and that the diagnosis was not determined only by administrative codes for RA in medical encounters. Exclusion criteria comprised: use of a bDMARD during the 6-month pre-index period; use of a non-TNFi bDMARD between the first and second bilateral hand X-rays; diagnosis of TNFi indication other than RA (including JIA, psoriasis, PsA, AS, Crohn’s disease or ulcerative colitis) at any time during their VA care; and uninterpretable radiographs.

TNFi exposure

All US Food and Drug Administration-approved TNFi medications were included in the analysis (adalimumab, certolizumab pegol, etanercept, golimumab and infliximab). Exposure was considered at the TNFi class level rather than at the individual agent level. Patients switching TNFi without gaps in their monthly exposure assessment were considered to have been exposed continually to TNFi, and data for the two different TNFi during the exposure period were combined and analysed as a single TNFi exposure.

TNFi exposure was handled as cumulative and a time-varying phenomenon influenced by effectiveness, side effects, tolerability and underlying changes in disease activity to avoid potential bias that could occur with traditional methods of evaluating the duration and intensity of medication exposure, such as the proportion of days covered [17, 18]. For example, traditional measures of medication adherence and persistence often condition on the future and overlap with the outcome measurement period [18]. Supplementary Fig. S1, available at Rheumatology Advances in Practice online illustrates how initial TNFi exposure during the index month is potentially confounded by baseline measures, whereas later TNFi exposure is confounded by intermediate variables between the index month and later TNFi exposure. As described in the section on potential confounders, we applied laboratory markers of inflammation and the use of CSs as proxy measures of time-varying RA disease activity because composite measures of RA disease activity were not readily available in our national RA cohort. Traditional statistical methods are biased when intermediate variables act as confounders for the treatment outcome process. This consideration was the justification for the use of marginal structural models (MSMs) with inverse probability of treatment weights to estimate the effect of various cumulative levels of TNFi exposure in our study population (described in detail in the Supplemental Materials, section Methods, available at Rheumatology Advances in Practice online).

Study outcomes

The primary outcome was the percentage of patients with a clinically meaningful deterioration in mean change in MSS for hand X-rays based on comparison of initial X-rays obtained at the beginning of TNFi therapy with follow-up X-rays after TNFi initiation (Fig. 1). A clinically meaningful deterioration was defined as a change of more than five in MSS [19], to account for reviewer error and inconsistencies in imaging positioning. The mean change in MSS in patients with a deterioration of MSS is also presented. This mean MSS change was calculated with the assumption of no progression in patients with a negative change in MSS and with the inclusion of negative MSS difference. For patients with multiple hand X-rays between 10 and 18 months post-index, the radiograph collected closest to 12 months after the index date was used (Fig. S1, available at Rheumatology Advances in Practice online). Hand X-rays were scored for erosions and joint space narrowing according to the MSS [20, 21] by an experienced evaluator (A.R.E.) in a fully blinded process, including patient identifiers, demographics, dates of radiographs, sequence of radiographs and site of radiographs. Erosions were scored on a scale of zero (absence of erosion) to five (extensive destruction of bone) for 34 joints, for a maximal total score of 170 [20]. Joint space narrowing was scored on a scale of zero (normal joint space) to four (ankyloses) for 32 joints, for a maximal total score of 128. An electronic data capture form was developed to collect structured and standardized MSS measurements reliably. To assess intra-observer reliability, reading of radiographs was repeated on 38 randomly selected blinded sets of hand radiographs; the Pearson correlation coefficient was 0.96 for these repeat readings.

Potential confounders

Baseline demographic and disease characteristics (age, smoking status), RF status, anti-CCP antibody status, co-morbidities (congestive heart failure, pneumonia, acute bronchitis, urinary tract infection, skin infection, septicaemia, shock, HIV infection, hepatitis, alcohol-related disorders and substance abuse disorders), concurrent medications (opioid and non-opioid analgesics, antirheumatic salicylates, non-salicylate NSAIDs, prednisone and proton pump inhibitors), counts of concurrent medications, rheumatic disease co-morbidity index score [22], health-care utilization (counts of rheumatology, urgent care, emergency department and inpatient visits)] and time-varying covariates (ESR, CRP and prednisone exposure) were assessed for clinical and statistical importance and included in the adjusted model in addition to the crude analysis. The last-value-carried-forward approach was used for monthly ESR and CRP, and the monthly updated proportion of days covered was used for prednisone exposure. If data were not available positively to confirm RF or anti-CCP status, co-morbidities or non-TNFi drug exposure, these variables were considered to be negative.

Results

Patients

A total of 246 patients across 83 VA sites were included in the analysis. The mean (s.d.) age was 57.7 (11.0) years; most patients were male (80.5%) and most were white (76.4%) (Table 1). The initial mean (s.d.) MSS was 19.6 (33.4) (range 0–214). The mean time (s.d.) from the index date to the follow-up radiograph was 14.6 (2.3) months (median 15 months; range 11–18 months). Baseline characteristics of the study population were similar to those reported in previous studies of Veterans treated with TNFi [23–26].

Table 1.

Baseline demographic and clinical characteristics

Characteristic	All patients (n = 246)
Age, mean (s.d.), years	57.7 (11.0)
Sex, n male (%)	198 (80.5)
Race, n (%)
White	188 (76.4)
Black	28 (11.4)
Other/unknown	30 (12.2)
Smoking status, n (%)
Current smoker	78 (31.7)
Former smoker	85 (34.6)
Never smoked	72 (29.3)
Unknown	11 (4.5)
RF seropositive, n (%)	168 (68.3)
Anti-CCP seropositive, n (%)	160 (65.0)
Select co-morbidities, n (%)
Congestive heart failure	1 (0.4)
Pneumonia	1 (0.4)
Acute bronchitis	5 (2.0)
Skin infection	5 (2.0)
Hepatitis	10 (4.1)
Alcohol-related disorders	20 (8.1)
Substance abuse disorders	8 (3.3)
RDCI score, mean (s.d.)	1.76 (1.47)
Concurrent medications, n (%)
nbDMARD	199 (80.9)
Opioid analgesics	130 (52.8)
Non-opioid analgesics	28 (11.4)
Non-salicylate NSAIDs	92 (37.4)
Prednisone	111 (45.1)
Proton pump inhibitors	106 (43.1)
MSS, mean score (s.d.)	19.6 (33.4)

MSS: modified Sharp score; nbDMARD: non-biologic DMARD; RDCI: rheumatic disease co-morbidity index.

TNFi exposure

Throughout the follow-up period, two patients had <1 month of cumulative exposure, 26 had ≤ 3 months, 33 had >3 to 6 months, 59 had >6 to 9 months and 126 had >9 months of cumulative exposure. Fig. 2 shows the number of patients with different durations of continuous TNFi exposure before a ≥ 90 day gap in therapy on initiating TNFi bDMARD therapy (blue bars). If patients restarted TNFi therapy during the observation period after a ≥ 90 day gap, the total or cumulative dose of TNFi therapy during the observation period was also calculated (Fig. 2, red bars).

Fig. 2.

Distribution of cumulative and continuous TNF inhibitor exposure in the first year

The numbers of patients with continuous (blue bars) and cumulative (red bars) exposure per month are shown.

Radiographic outcomes

The proportion of the study population with clinically meaningful progression from initial MSS decreased with increasing exposure to TNFi therapy and is reported for both crude and MSM models (Table 2). In the crude analysis, the percentage of patients with MSS change ≥ 5 (95% CI) in patients with 3 months of exposure was 10.1% (9.8%, 10.5%) and decreased to 7.3% (7.2%, 7.5%) for patients with 12 months of exposure. In the MSM, the percentage of patients with MSS change ≥ 5 (95% CI) with 3 months of exposure was associated with a 10.6% (9.8%, 11.4%) increase in MMS change ≥ 5, which decreased to 5.4% (5.1%, 5.7%) for patients with 12 months of exposure (Table 2).

Table 2.

Predicted risk for MSS change by TNFi exposure

Predicted mean changes and cumulative treatment	Crude model [% (95% CI)]	Adjusted model [% (95% CI)]
Patients with MSS changes (n = 246)
3 months	10.1 (9.8, 10.5)	10.6 (9.8, 11.4)
6 months	8.7 (8.6, 9.1)	7.6 (7.2, 8.0)
9 months	8.0 (7.8, 8.1)	6.1 (5.8, 6.4)
12 months	7.3 (7.2, 7.5)	5.4 (5.1, 5.7)

MSS: modified Sharp score.

The mean (s.d.) change from the initial MSS was 0.3 (3.6) (range −19–22) (Fig. 3). When negative MSS changes were considered as zero changes, the crude MSS mean change analysis showed an average difference in score (95% CI) for 3 months of exposure of 1.35 (1.33, 1.38) and for patients with 12 months of exposure of 1.25 (1.23, 1.27). The MSM-adjusted population average difference score for 3 months of exposure was 1.29 (1.32, 1.45), and for patients with 12 months of exposure it was 1.19 (1.14, 1.24). When MSS changes were included with their negative value, the MSS mean change analysis showed a crude average difference score for 3 months of exposure of 0.54 (0.51, 0.57) and for patients with 12 months of exposure of 0.54 (0.51, 0.56). The MSM-adjusted population average difference score for 3 months of exposure was 0.44 (0.36, 0.53), and for patients with 12 months of exposure it was 0.56 (0.50, 0.62) (Table 3).

Fig. 3.

Cumulative probability distribution of change in modified Sharp score from initial radiographs

MSS: modified Sharp score.

Table 3.

Predicted changes in modified Sharp score by TNF inhibitor exposure

Predicted mean changes and cumulative treatment	Crude model [mean change (95% CI)]	Adjusted model [mean percentage change (95% CI)]
Mean change in positive MSS (n = 246)a
3 months	1.35 (1.33, 1.38)	1.29 (1.32, 1.45)
6 months	1.32 (1.30, 1.34)	1.32 (1.27, 1.37)
9 months	1.29 (1.27, 1.30)	1.26 (1.20, 1.31)
12 months	1.25 (1.23, 1.27)	1.19 (1.14, 1.24)
Mean change in MSS (n = 246)
3 months	0.54 (0.51, 0.57)	0.44 (0.36, 0.53)
6 months	0.54 (0.51, 0.57)	0.48 (0.41, 0.55)
9 months	0.54 (0.51, 0.56)	0.52 (0.46, 0.58)
12 months	0.54 (0.51, 0.56)	0.56 (0.50, 0.62)

^aIncludes only MSS that were positive. MSS: modified Sharp score; TNFi: TNF inhibitor.

Discussion

To our knowledge, this is the first study to examine the impact of different levels of TNFi exposure on radiographic outcomes in real-world clinical practice outside RCTs. Although the important findings from RCTs and RA registries provide essential information to determine efficacy and safety profiles and to inform decisions in clinical practice, the populations included in RCTs are not fully representative of the general population because of restrictions required by enrolment criteria. Evaluations of clinical practice have suggested that only 14–34% of RA patients in clinical practice meet eligibility criteria to allow participation in RCTs [27]. Observational studies based in real-world clinical practice, although subject to bias as noted below, are needed to investigate whether the observations from RCTs can be confirmed in expanded populations. Thus, our study provides experience in a more diverse RA population than has been studied in RCTs.

In our study, we observed a reduction in the percentage of patients with meaningful progression of structural damage in patients with RA who had greater exposure to TNFi in comparison to patients with lower TNFi exposure. The observation of reduced progression of structural damage with TNFi therapy has also been seen in RCTs that have evaluated the clinical efficacy of the specific agents adalimumab [7], certolizumab pegol [28, 29], etanercept [5, 6, 8, 30], golimumab [31] and infliximab [10, 32] in RA patients. Although the population, concurrent medications and design of our study differ from those in RCTs, these prior RCTs have been consistent in demonstrating a significant benefit of TNFi therapy on reducing the progression of structural damage in RA patients in the clinical trial environment.

This work demonstrates the challenges of evaluating data from real-world cohorts and administrative claims to evaluate radiographic outcomes. These challenges are the definition of the radiographic outcome measured and the lack of specific protocols for routine collection of radiographic data. To address the first challenge, we adopted a strategy to use meaningful clinical change in structural damage as our primary outcome measure, which we propose to be the most relevant to real-world clinical practice and not subject to extreme outlier data, as is a mean score [19]. The primary outcome is presented as the percentage of patients with a clinically meaningful change in MSS (≥ 5), which can be interpreted as the population-level risk of disease progression associated with each level of cumulative TNFi exposure. This approach was presented as the primary outcome for three reasons. First, mean change in MSS does not take into account measurement error. Second, radiographic damage is generally thought of as progressive, and this was the justification for treating values less than zero (indicating reduction of damage) as zero for the mean change in MSS analysis. Third, the mean change score was not normally distributed; nevertheless, the mean change in MMS is a standard method for presenting these data in randomized trials [33]. Negative changes in MSS in our analysis could represent improvement or repair of structural damage in some patients during TNFi therapy. As noted in Table 3, this potential would reduce the impact of TNFi in our analysis. Although there were not large differences in the mean change in score, there were substantial and robust differences in the risk of experiencing clinically meaningful progression by the level of cumulative TNFi exposure. To address challenges inherent to real-world practice, we conducted our analysis using the longitudinal MSM with inverse probability of treatment weights to correct for these factors and confounders.

Our study had several strengths, including its basis in real-world practice, the involvement of a large nationwide cohort from 83 unique sites of care, and a chart review confirming the diagnosis of RA rather than merely a designation of RA by administrative data. Also, our study had the strength of a uniform reading system for each set of radiographs.

Our study also had several limitations. Our study population of US Veterans involves generally older male patients, which is different from most RA populations that are predominantly female. This demographic difference might make our findings difficult to generalize to non-VA settings, including non-US patients or those enrolled in Medicaid, Medicare or commercial health plans. The decisions to start TNFi were not defined by protocol and were not based on disease activity measures as in RCTs; however, it is reasonable to assume that our VA patients had active disease in the opinion of the prescribing providers. There might have been bias in selection of patients to receive TNFi therapy by the prescribing provider, thus resulting in an RA population with more severe disease than the general population of US Veterans with RA. The reasons for discontinuation of TNFi therapy were not available in our study as is reported in RCTs. The decision to perform hand radiographs was at the discretion of the practising provider and not part of a protocol. The decision to collect radiographs might have been biased by selecting for patients with lack of response to therapy or other reasons that could lead to a selection bias. The observation that ∼20% of our patients had initial and follow-up radiographs at different VA sites of care suggests that some patients might have received their radiographs as part of the process of establishing care at a new site.

Although a comparison of the rate of progression of structural damage in our study population, who received TNFi therapy, with the rate of progression in a control group of untreated RA patients would have been informative, this analysis raised potential concerns. The greatest concern was that we did not have the capacity to determine the baseline RA disease severity accurately, and the baseline disease activity of patients initiating TNFi would probably be significantly higher than that of patients not initiating TNFi, which would lead to confounding of the results because they could not be adjusted based on disease activity.

Concurrent medications, although recorded in our study, were not required to be stable during the observation period as is required in most RCTs. Randomized controlled trials have generally enrolled patients with active RA despite MTX therapy or patients naïve to MTX who were randomly assigned to receive a TNFi and/or MTX. In these RCTs, the control patients were maintained on MTX or initiated MTX. Although the use of concomitant medications was not controlled, these factors were considered in confounder adjustments. Exposure was measured using claims data and might not reflect real medication use. We had records only for medications dispensed by the VA, and the potential for outside TNFi therapy is a possibility, although a recent evaluation of US Veterans with RA showed that the vast majority of these patients received their care at VA facilities and, in particular, received biologic therapies exclusively at the VA [34]. As with any observational study, unobserved covariates might have influenced the study findings.

In summary, although our study has limitations, the observation of decreased progression in structural damage in patients with greater TNFi exposure is a significant and important finding that is consistent with the findings reported in RCTs and clinical registries. Although the progression of radiographic changes as measured by MSS was very small during the 1 year of observation in many patients, we identified a subset of patients with meaningful reductions in radiographic progression that was impacted by TNFi therapy. Larger, long-term studies to confirm our observation that longer cumulative TNFi exposure is associated with greater impact on radiographic progression are warranted. Our study design provides a new method for evaluating antirheumatic therapy in clinical practice and provides further evidence for the ability of these agents to provide significant benefit in patients with RA.