Estimation of optimal adherence threshold for tumor necrosis factor inhibitors in rheumatoid arthritis

Treatment with disease-modifying antirheumatic drugs (DMARDs) is the only method for reducing inflammation and slowing joint damage caused by rheumatoid arthritis (RA) [1]. A tumor necrosis factor (TNF) inhibitor is recommended as second-line therapy in patients with moderate or high disease activity, according to the American College of Rheumatology’s 2015 treatment guidelines after the failure of monotherapy with methotrexate [2]. Medication adherence, or taking the medication at the prescriber-recommended dose, frequency, and timing [3], has been found to be associated with patient outcomes in various chronic disease states, including RA [4‐7]. These studies support the positive effect of medication adherence, especially when it is dichotomized at the 80% threshold, on patient outcomes. For example, patients who self-reported taking their TNF inhibitor as directed had better responses on the 28-joint disease activity scale at 6 months [6]. In patients with various autoimmune disorders, a medication possession ratio (MPR) ≥ 80% for self-administered TNF inhibitors was associated with a 12% decreased odds of hospitalization and 18% decreased odds of an emergency room visit [7]. However, it is unknown if there are lower adherence thresholds that can yield favorable outcomes.

Currently, the 80% adherence threshold is used to dichotomize patients into adherent and non-adherent groups. This threshold was initially supported by Haynes et al., who found that patients who used at least 80% of their antihypertensive medications for 6 months had significantly lower diastolic blood pressure [8, 9]. However, the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) recommends that adherence be kept as a continuous variable unless a threshold has been previously validated [10]. The 80% adherence threshold has been validated for other chronic disease states but not in RA. For example, in hypercholesterolemia, an MPR ≥ 80% to statins is associated with achieving low-density lipoprotein cholesterol goals [11]. The Pharmacy Quality Alliance (PQA) set their measure for adherence to non-infused biologics for RA to a proportion of days covered (PDC) of 80% [12]. The 80% threshold for adherence has become the norm in RA studies, cited as the value typically used [5, 13, 14]. Although the 80% threshold is normally used to categorize patients as adherent and non-adherent, the determination of this threshold in RA patients using RA-related outcomes has not been published [15].

An evidence-based threshold for adherence is needed for a couple of reasons. First, healthcare providers need a threshold for determining if non-response to medication is due to ineffectiveness of the drug or non-adherence. Patients may have their medications unnecessarily switched when increasing adherence would improve their response to the medication. The second reason is for research and quality improvement. Much current research on adherence to DMARDs for RA classifies patients as adherent or non-adherent based on the 80% threshold and relates this classification to outcomes, costs, and predictors [5, 7, 13, 16‐19]. If adherence to self-injected TNF inhibitors for RA becomes a quality measure for health plans, it is important that the threshold is set using an evidence-driven approach so that outcomes are maximized if a higher threshold is needed, and providers are not unfairly penalized if a lower threshold is needed. The objective of this study was to determine the optimal threshold for adherence for TNF inhibitors with or without methotrexate among Medicare-enrolled adults with RA.

Of 13,020 beneficiaries who had a claim for a TNF inhibitor between 2012 and 2018, 1190 met the inclusion criteria (Supplement Fig. 2). Table 1 contains the baseline characteristics of the final sample. Female beneficiaries (80.5%) comprised most of the sample, and 164 (13.8%) were DMARD naïve. Most beneficiaries were enrolled due to disability benefits. Only 420 (35.3%) were enrolled in Medicare by meeting the age requirement. A total of 865 (72.7%) patients experienced at least one outcome, with hospitalizations (39.1%) being the most frequent, followed by oral GC use (36.6%), ER use (25.0%), injectable GC use (20.8%), and serious infections (16.1%). The median follow-up time, which was calculated from the index date to the first occurrence of either the outcome or a censoring event, ranged from 270 to 707 days, depending on the outcome under examination (Table 2).

Table 3 shows time-varying characteristics at days 1, 90, 180, and 365 of follow-up. The mean 90-day PDC was 0.92 (SD: 0.16) on day 1. On days 90, 180, and 365 of follow-up these numbers slightly decreased, ranging from 0.77 (SD: 0.34) at day 90 to 0.62 (SD:0.43) at day 365. Figure 1 shows histograms of the percent of patients in each PDC decile at follow-up days 1, 90, 180, and 365. Other time-varying characteristics also show a similar declining trend for average weekly methotrexate dose, NSAID use, patient out-of-pocket prescription costs, and the daily average number of concomitant medications used. The percentage of patients using only mail-order pharmacies to fill their TNF inhibitor prescriptions increased from day 1 (7.1%) to day 365 (10.6%).

Survival models using Cox regression were run for each outcome variable. Every 10% increase in PDC was statistically significantly associated with a decreased risk of the composite outcome (HR 0.98, 95% CI 0.96–1.00), oral GC use (HR 0.93, 95% CI 0.91–0.96), and hospitalizations (HR 0.96, 95% CI 0.94–0.99). The hazard ratio for PDC was significantly above 1 for ER visits, indicating that higher PDCs to TNF inhibitors are associated with a higher risk of ER visits (HR 1.04, 95%, CI 1.01–1.07). The hazard ratios for injected GCs (HR 0.99, 95% CI 0.96–1.03) and serious infections (HR 0.97, 95% CI 0.93–1.00) were not statistically significant. Table 4 shows the adjusted HRs and their confidence intervals for each variable in each survival model.

Optimal PDC thresholds were calculated for the three outcomes, composite, oral glucocorticoid use, and hospitalizations, that demonstrated favorable outcomes at higher PDCs. The AUCs for these curves were estimated in the range of 0.5 to 0.6 and were all significantly greater than 0.5 (Table 5). The optimal thresholds for the composite outcome, oral GC use, and hospitalization were 0.64, 0.59, and 0.56, respectively (Table 5). Dichotomizing adherence at these thresholds showed a decrease in the hazard of all three outcomes in both adjusted and unadjusted models. However, the association of the dichotomized PDC was not found to be statistically significant for the composite outcome.

This is the first study to determine an optimal PDC threshold for TNF inhibitor therapy among adults with RA, which was found to be approximately 60%. Adherence to TNF inhibitors decreased as time on therapy increased. In the survival analysis, an increase in PDC was significantly associated with a decreased risk of experiencing oral GC use, hospitalization, and the composite outcome.

The PDCs found in this study for TNF inhibitors in a Medicare sample are about on par with previous studies. Calip et al. found that the mean MPR for TNF inhibitors among patients with RA was 62% over the first year [7]. Chu et al. found 1-year PDC in patients with RA who were on adalimumab or etanercept was 77% or 66% when including patients who discontinued treatment [16]. Patients with RA in a study by Grijalva et al. on adalimumab and etanercept had an MPR of 83%–85%, but those on concomitant methotrexate had lower MPRs of 72% and 64%, respectively [36]. In a study involving data from a specialty pharmacy, the mean PDC for patients on bDMARDs and tsDMARDs from index date to last fill was 86% [37]. Similarly, 2-year PDC for Optum patients with RA ranged from 81 to 87% [18]. This study found a decreasing trend in PDCs over the follow-up. While this trend may be an artifact of patients discontinuing due to being in remission or the measurement method, as there are more potential gaps in therapy over a longer duration, the low PDC values are consistent with the literature and suggest the need for continued emphasis on medication adherence among patients with RA.

Higher PDC was associated with a lower risk of the composite outcome, hospitalizations, and oral GC use, but a higher risk of ER visits. Previous studies have shown poor adherence to DMARDs usually leads to worse disease activity in RA. For example, MPR ≤ 80% to bDMARDs in the previous 180 days was associated with higher odds of a higher Disease Activity Score-28 (DAS28) score, indicating worse RA disease activity [17]. A meta-analysis found that patients adherent to DMARD therapy had lower DAS28 scores, lower erythrocyte sedimentation rates, and smaller tender joint counts, showing better adherence, which leads to improvement in these disease activity outcomes [38]. Calip et al. found having an MPR ≥ 80% was associated with an 18% lower odds of hospitalizations [7]. The increased risk of ER visits with higher PDC is surprising as it is contrary to a previous study that found an MPR ≥ 80% over 365 days was associated with a 12% lower odds of an ER visit [7]. This finding may result from confounding by indication, where patients with more severe disease are more likely to be adherent and need ER visits for pain management. While this study controlled for a claims-based disease severity score, it is possible this proxy did not eliminate this bias entirely. Interestingly, the relationship between PDC and the outcomes of interest was inconsistent in the sensitivity analysis using a time-invariant PDC measurement window. While the discrepancy in results may also be due to the low discriminatory ability of PDC, this likely demonstrates the robustness of the time-varying approach for evaluating the impact of PDC. Therefore, it should be considered the standard for future research assessing the relationships between PDC and outcomes.

The optimal PDC thresholds for oral GC use, hospitalization, and composite outcome were similar at 365 days of follow-up. This time point of interest (365 days) is in accordance with existing quality measures and other extant literature. The AUCs for these are considerably poor, although consistent with previous literature [39]. Clinical outcomes for RA—as with many other chronic diseases—are multifactorial, and adherence to medication is only one component that influences the risk of these outcomes. However, when determining a threshold for PDC, these other factors were not considered, leading to low expected AUCs. Additionally, changes in prescription fill patterns, including an increase in the use of 90-day fills and automatic refills, may lead to a departure of true medication adherence behavior from what can be captured using administrative claims data, further contributing to lower AUCs [40, 41]. The 60% adherence threshold for TNF inhibitors yielded significant adjusted HRs of less than 1 for oral GC use and hospitalization. This finding has significant implications for payers, patients, and clinicians. While this threshold is lower than what is used in existing quality measures, there may still be ample room for improvement in adherence based on the findings in this study. Although the threshold is relatively low, it is important to note that higher PDCs were associated with better outcomes. Future studies should examine the change in threshold at other times during follow-up in order to optimize potential interventions to improve medication adherence. Further research is also needed prior to changing benchmarks for quality measures—but using less stringent benchmarks provides the potential to allocate scant resources elsewhere in RA disease management. Not many studies have researched the optimal medication adherence threshold, especially in autoimmune diseases. Govani et al. determined the optimal cumulative MPR threshold for adalimumab in patients with inflammatory bowel disease was 86% and 87% for certolizumab based on a combined outcome of hospitalizations and GC prescriptions [24].

There are several limitations to the present study. Because PDC was measured based on the previous 90 days, this design potentially ignores outcomes occurring within the first 90 days after the start of TNF inhibitor therapy. Next, claims data are not collected for research purposes and, therefore, do not contain clinical information, such as RA disease activity and quality of life assessments, nor do such data capture prescriptions or medical services paid for or used outside of the Medicare program. Adherence calculations using claims data also may not reflect patients’ true medication use patterns since receipt of the medication does not indicate that the patient is actually taking the medication. Third, other patient factors and comorbidities may affect the outcomes of this study. The gold standard for measuring the clinical efficacy of TNF inhibitors is a 20% improvement in the American College of Rheumatology core set measures (ACR20) [42], which is not available in claims data. Future studies should assess the relationship between time-varying adherence to TNF inhibitors and RA disease activity. Finally, this sample of patients with RA may not represent all patients with the disease. Generalizations to other populations should be made with caution.

The average 90-day PDC of self-injectable TNF inhibitors in Medicare patients with RA initially starts above 80% but then decreases to about 60% after 1 year. Higher PDC was associated with a reduced risk of hospitalization, oral GC use, and the composite outcome. The optimal PDC threshold for TNF inhibitors in RA was found to be around 60% at 365 days. More research is needed to determine the optimal adherence threshold in DMARD therapy for RA over different time intervals and at various time points after the index date than the 90-day intervals and 365 days of follow-up used in this study.