Merging Veterans Affairs rheumatoid arthritis registry and pharmacy data to assess methotrexate adherence and disease activity in clinical practice

Authors


Abstract

Objective

The Veterans Affairs (VA) Rheumatoid Arthritis (VARA) registry and the VA Pharmacy Benefits Management database were linked to determine the association of methotrexate (MTX) adherence with rheumatoid arthritis (RA) disease activity.

Methods

For each patient, the medication possession ratio (MPR) was calculated for the first episode of MTX exposure of a duration of ≥12 weeks for both new and established MTX users. High MTX adherence was defined as an MPR ≥0.80 and low MTX adherence was defined as an MPR <0.80. For each patient, the mean Disease Activity Score with 28 joints (DAS28) score, erythrocyte sedimentation rate (ESR), and C-reaction protein (CRP) level observed during registry followup were compared in high- versus low-adherence groups.

Results

In 455 RA patients, the prescribed doses of MTX (mean ± SD 16 ± 4 mg versus 16 ± 4 mg; P = 0.6) were similar in high-adherence patients (n = 370) in comparison to low-adherence patients (n = 85). However, the actual observed MTX doses taken by patients were significantly higher in the high-adherence group (mean ± SD 16 ± 5 mg versus 11 ± 3 mg; P < 0.001). DAS28 (mean ± SD 3.6 ± 1.2 versus 3.9 ± 1.5; P < 0.02), ESR (mean ± SD 24 ± 18 versus 29 ± 24 mm/hour; P = 0.05), and CRP level (mean ± SD 1.2 ± 1.3 versus 1.6 ± 1.5 mg/dl; P < 0.03) were lower in the high-adherence group compared to those with low MTX adherence. These variances were not explained by differences in baseline demographic features, concurrent treatments, or whether MTX was initiated before or after VARA enrollment.

Conclusion

High MTX adherence was associated with improved clinical outcomes in RA patients treated with MTX. Adjustment for potential confounders did not alter the estimated effect of adherence. These results demonstrate the advantages of being able to merge clinical observations with pharmacy databases to evaluate antirheumatic drugs in clinical practice.

INTRODUCTION

Rheumatoid arthritis (RA) is associated with significant morbidity and increased mortality (1, 2). Disease-modifying antirheumatic drugs (DMARDs) have dramatically improved the management of RA patients (3–6). While numerous controlled clinical trials have demonstrated that methotrexate (MTX) and other DMARDs are effective in RA, controlled clinical trials are generally limited to highly selected patients often with moderate to severe disease activity and limited comorbidities. Surveys of rheumatology clinic populations estimate that only 5–16% of patients with RA would qualify for participation in these clinical trials (7). To understand how controlled clinical trials' estimates of DMARD effectiveness and safety translate into the clinical setting, it is critical to evaluate DMARD performance in observational studies of routine clinical practice. This work employed the detailed information provided on the medication history of US veterans collected through the Veterans Affairs (VA) Pharmacy Benefits Management (PBM) (8, 9). PBM not only lists medications dispensed, but also captures information on the specifics of each prescription, including the number of tablets dispensed and the expected duration of each prescription and prescribing information (“sig”). The potential to combine these detailed administrative pharmacy data sets with comprehensive clinical outcome measures provides an opportunity to evaluate DMARD therapy of RA in a “day-to-day” clinical practice setting.

The Veterans Affairs Rheumatoid Arthritis (VARA) registry prospectively records data from clinical encounters (1, 10) and has recently been merged with VA administrative pharmacy data collected in a centralized PBM program that allows for the evaluation of the use of and adherence to DMARDs in clinical practice. This study was undertaken to examine the following: first, whether adherence to MTX is inversely proportional to RA disease activity measures, and second, whether other factors such as demographic and clinical diseases parameters confound this relationship. Furthermore, the study demonstrates the potential for combining data from disease registries and administrative databases such as PBM to evaluate the clinical issues in a “day-to-day” community clinical practice setting.

Significance & Innovations

  • Clinical data in the Veterans Affairs Rheumatoid Arthritis (VARA) registry can be merged with the VA Pharmacy Benefits Management (PBM) database.

  • PBM data can be used to measure methotrexate (MTX) adherence.

  • Rheumatoid arthritis (RA) disease activity may be correlated with MTX adherence and the impact of potential confounder on this association evaluated.

  • Higher MTX adherence is associated with improvement in clinical outcomes for RA patients in clinical practice.

PATIENTS AND METHODS

Patients.

The VARA multicenter observational cohort began enrollment in January 2003 and is described elsewhere (1, 10). This analysis evaluated patients enrolled through August 31, 2009, and included observation up to this date. At the time of this analysis, PBM data were only available from the Dallas, Denver, Jackson, Omaha, Salt Lake City, and Washington, DC, VA medical centers, limiting the analysis to these 6 VARA sites. Using the PBM pharmacy database, patients were required to have recorded treatment with MTX (i.e., 1 or more dispensed MTX prescriptions for oral or subcutaneous MTX) prior to August 31, 2009. This analysis was limited to the first course of MTX therapy within the VA, and patients could have been either new or established (i.e., prevalent) MTX users. In addition, the analysis required at least one Disease Activity Score with 28 joints (DAS28) assessment (11) within the VARA registry occurring on or after the 90th day of MTX therapy during this first MTX course, an interval after which MTX benefit has been observed in clinical trials (4, 6). Patients contributed data to analyses specific to 2 of the 3 following groups: 1) all patients receiving MTX during VARA observation (full cohort), 2) patients initiating MTX more than 30 days prior to VARA enrollment but continuing therapy beyond enrollment (established MTX cohort), and 3) patients initiating MTX within 30 days of VARA enrollment or after VARA enrollment (incident MTX cohort).

The registry has received institutional review board approval at each site. This study was approved by the VARA Scientific and Ethics Advisory Committee for the VARA registry.

Determination of MTX exposure.

The details of MTX therapy were determined from PBM outpatient dispensing records for each patient. PBM includes the following information on each prescription: dispense date, number of days' supply, units dispensed (total number of tablets or liquid vials), unit dose, and the directions for use (“sigs”) (8). Each single prescription of MTX was defined as a dispensing episode. A drug course was defined as a period of continuous MTX treatment consisting of 1 or more dispensing episodes as described below and in Supplementary Appendix A (available in the online version of this article at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)2151-4658).

Calculation of the duration and dose of MTX for each dispensing episode.

The dispense date and expected duration of each MTX prescription were needed to estimate the duration of each dispensing episode. An algorithm was used to calculate the expected duration of each dispensed prescription (Figure 1). The average weekly dose (AWD) was calculated from the days' supply, unit strength, and number of units dispensed. If the AWD was within the range of 5 to 30 mg per week, then the number of days' supply information was assumed to be accurate and reported as the expected duration of that prescription. In order to correct for rare outlier prescriptions with improbable weekly dosages (e.g., greater than 50 mg/week) or durations (e.g., of more than 6 months of prescribed medication at a single dispensing of MTX), we employed the dispensing information in this algorithm. The application of this algorithm affected 9 (2%) of the 445 MTX courses, in which we incorporated dosing directions (i.e., “sigs”) to estimate expected dosages/durations. The corresponding author (GWC) reviewed all of the unique doses and “sig” combinations and from these, calculated an estimated “sig weekly dose.” If the AWD was below 5 mg or above 30 mg, then the total amount of the drug dispensed (number of units dispensed multiplied by the unit strength) was divided by the “sig weekly dose” and multiplied by 7 days (to convert weeks into days) to obtain the sig estimated days' supply (SEDS). The SEDS was then used to populate the expected duration. Examples and further illustration of the application of this algorithm are shown in Supplementary Appendix A (available in the online version of this article at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)2151-4658).

Figure 1.

Algorithm for determining the expected duration of each medication dispensing. The term “sig” refers to the physician prescribing instructions with the individual prescription. * = total dose in mg dispensed = number of units (tablets, vials) × unit strength; ** = average weekly dose (AWD) = total dose dispensed/days' supply; MTX = methotrexate; *** = “sig” estimated days' supply (SEDS) = total dose dispensed/“sig” weekly dose × 7 days.

Calculation of gaps in MTX medication dispensing and drug course.

The dispensing date and expected duration were used to calculate the expected end date for each prescription dispensed. Differences between the expected end of a prescription and the dispensing date of the subsequent medication refill were considered gaps in therapy. A patient was defined as receiving continuous MTX therapy if there was less than a 90-day gap between refills of MTX, which has been used in other analyses of adherence (12–14). A course ended when a 90-day gap occurred as defined above. The duration of the course was defined from the date of the first MTX dispensed prescription until the expected end date for the last prescription before a 90-day gap or end of the observation period. This study was limited to the first course of MTX identified from PBM data for each patient, whether established (beginning before VARA enrollment) or incident (occurring on or after VARA enrollment).

Calculation of MTX adherence.

For the first course of MTX, we calculated a prescribed cumulative dose, course duration, average prescribed dose, prescribed duration average observed dose, and medication possession ratio (MPR). The prescribed duration of drug exposure during a course was defined as the sum of the expected durations for each prescription dispensing within a course. The average dose prescribed was calculated as the total dose dispensed divided by the prescribed duration for a course. The average dose observed was the total dose dispensed divided by the total course durations, which thereby takes into account the potential for less than full adherence. The MPR was calculated as the number of prescribed days of MTX during a course divided by the total duration of days of a course (12, 14).

VARA outcome variables.

The primary outcome variable was the DAS28 (11). An average DAS28 score was calculated for each subject by including all DAS28 scores from 90 days after the MTX dispensing date (i.e., impact date) until the end date of the first MTX course. Secondary outcome variables evaluated were tender joint count, swollen joint count, patient global disease assessment (100-mm scale), patient pain (10-point scale), physician global disease assessment (100-mm scale), Multidimensional Health Assessment Questionnaire (15, 16), Westergren erythrocyte sedimentation rate (ESR), and C-reactive protein (CRP) level. Radiographic outcomes were not measured.

Covariates.

At enrollment into the VARA registry, we collected presence of RA classification criteria (e.g., nodules and radiographic erosions, based on medical documentation), medical history, smoking status (never, former, or current), sociodemographics (education, race/ethnicity, age, sex), duration of RA since initial diagnosis, and prior and current use of DMARDs (both biologic and nonbiologic). A serum sample collected on enrollment was evaluated for rheumatoid factor (RF) and anti–cyclic citrullinated peptide antibody (anti-CCP) (17). RF was determined by nephelometry (Siemens Healthcare Diagnostics) and was considered positive at levels ≥15 IU/ml. Standardized anti-CCP (IgG) was measured using a commercially available second-generation enzyme-linked immunosorbent assay (Diastat, Axis-Shield Diagnostics) and was considered positive at levels ≥5 units/ml. Charlson-Deyo comorbidity scores were calculated for all of the subjects based on aggregated VA national administrative data (18). For the incident MTX user cohort, we also adjusted for baseline DAS28, although this was not possible for the full and established cohorts, as MTX treatment for these patients was initiated prior to VARA enrollment.

Statistical analyses.

All statistical analyses were carried out using Stata, version 11 (StataCorp). Descriptive statistics for continuous variables are summarized as means and SDs; categorical variables are shown as percentages. Differences between high- and low-adherence groups were analyzed using t-tests and large sample binomial tests of proportions. To determine whether covariates were responsible for any differences in mean DAS28 scores between high- and low-adherence groups, multivariable linear regression was used. For all 3 cohorts (full, incident MTX, established MTX), we constructed multivariable models that included demographic characteristics and then demographic characteristics plus DMARD/biologic exposures (i.e., full model). Because overspecification was possible, we generated propensity scores by modeling the probability of having high adherence, and used the predicted probabilities as continuous and categorical adjustment variables. Since the findings between the propensity score and standard regression models were nearly identical, only results of standard regression are shown.

RESULTS

Patient characteristics.

There were 1,412 enrollees in VARA prior to August 31, 2009, of which 1,354 had available PBM data (Figure 2). Of these 1,354 eligible subjects, 1,029 (76%) were dispensed at least 1 MTX prescription. MTX courses for 499 veterans ended before VARA enrollment, making these veterans ineligible for the study. Of the 530 VARA enrollees receiving MTX, 455 veterans had at least one DAS28 score more than 90 days following treatment initiation (32.2% of all VARA enrollees). In these 455 subjects, 71 (16%) had their first observed MTX course on or after their VARA enrollment date (i.e., incident MTX course) and 384 (84%) initiated their first MTX course prior to VARA enrollment (i.e., established MTX course). Of the 455 subjects, 85 (19%) had an MPR of <0.80 and 370 (81%) had an MPR ≥0.80 calculated after a mean ± SD duration of followup of 42.7 ± 31.2 months.

Figure 2.

Veterans Affairs Rheumatoid Arthritis (VARA) registry population with a full breakdown of subjects fulfilling inclusion criteria and the fraction of subjects with high adherence (medication possession ratio [MPR] ≥0.80) and low adherence (MPR <0.80) to methotrexate (MTX). DAS28 = Disease Activity Score with 28 joints; * = percentage of veterans of the 455 who met inclusion criteria.

Patient characteristics are shown in Table 1. Patients were predominantly men (92%; mean ± SD age 64 ± 11 years) and had moderate to high levels of disease activity at the time of VARA enrollment reflected by a mean ± SD DAS28 score of 3.9 ± 1.6. There were no differences in age, disease duration at VARA enrollment, smoking status, autoantibody status, or the use of concurrent therapies based on high versus low MTX adherence overall or in groups defined by incident or established MTX use (Table 1). Among incident MTX users, men and white race were more likely to be in the high-adherence group versus the low-adherence group. African American patients were more likely to be in the low-adherence group versus the high-adherence group for all cohorts. Higher adherence to MTX was also associated with lower disease activity at enrollment based on DAS28, tender and swollen joint counts, and ESR overall and in the cohort defined by established MTX use. These baseline differences were not seen in the incident user cohort.

Table 1. Clinical features of the cohorts*
 Full cohortIncident user cohortEstablished user cohort
MPR <80% (n = 71)MPR ≥80% (n = 384)PMPR <80% (n = 17)MPR ≥80% (n = 68)PMPR <80% (n = 54)MPR ≥80% (n = 316)P
  • *

    MPR = medication possession ratio; VARA = Veterans Affairs Rheumatoid Arthritis; DAS28 = Disease Activity Score with 28 joints; MDHAQ = Multidimensional Health Assessment Questionnaire; ESR = erythrocyte sedimentation rate; anti-CCP = anti–cyclic citrullinated peptide antibody; MTX = methotrexate; DMARDs = disease-modifying antirheumatic drugs; TNFα = tumor necrosis factor α.

  • P < 0.05.

  • Traditional DMARDs included auranofin, azathioprine, cyclosporine, injectable gold, hydroxychloroquine, leflunomide, minocycline, and sulfasalazine.

  • §

    TNFα blockers included adalimumab, etanercept, and infliximab.

  • Other biologic agents included abatacept and rituximab.

Demographics at time of VARA enrollment         
 Age at enrollment, mean ± SD years62 ± 1264 ± 110.1763 ± 1262 ± 120.6462 ± 1265 ± 110.08
 Disease duration, mean ± SD years9.4 ± 10.39.4 ± 11.70.946.5 ± 7.59.4 ± 120.510.2 ± 119.5 ± 12.80.66
 Men, no. (%)63 (89)355 (92)0.3714 (83)66 (97)0.0249 (91)289 (91)0.86
 Race/ethnicity, no. (%)         
  White51 (72)316 (82)0.0511 (65)58 (85)0.0540 (74)258 (82)0.19
  African American18 (25)49 (13)0.015 (29)9 (13)0.1113 (24)40 (13)0.03
  Hispanic1 (1)11 (3)0.441 (6)1 (1)0.500 (0)10 (3)0.18
  Other/not specified1 (1)8 (2)0.440 (0)0 (0)1 (2)8 (3)0.35
 Ever tobacco user, no. (%)50 (70)300 (78)0.0712 (71)60 (88)0.0740 (74)236 (75)0.38
Enrollment clinical measures         
 DAS28, mean ± SD4.4 ± 1.83.8 ± 1.60.024.9 ± 1.64.6 ± 1.70.454.2 ± 1.83.7 ± 1.50.03
 Tender joints (range 0–28), mean ± SD7.1 ± 8.05.0 ± 6.70.027.4 ± 7.67.6 ± 8.00.907.0 ± 8.24.7 ± 6.3< 0.01
 Swollen joints (range 0–28), mean ± SD6.7 ± 7.44.2 ± 5.4< 0.019.0 ± 7.66.2 ± 6.30.126.0 ± 7.83.8 ± 5.2< 0.01
 MDHAQ (range 0–3.0), mean ± SD0.9 ± 0.70.9 ± 0.60.981.0 ± 0.71.0 ± 0.50.790.9 ± 0.70.9 ± 0.60.99
 Pain (range 0–10), mean ± SD4.7 ± 3.24.4 ± 2.80.424.6 ± 3.34.5 ± 2.80.904.7 ± 3.24.3 ± 2.80.41
 Patient global (range 0–100), mean ± SD41 ± 2940 ± 260.8846 ± 2046 ± 260.9934 ± 2435 ± 220.76
 ESR, mean ± SD mm/hour32 ± 2724 ± 210.0236 ± 2427.1 ± 210.1530.3 ± 28.124.5 ± 21.50.04
 Rheumatoid factor positive, no. (%)61 (86)298 (78)0.1312 (71)55 (81)0.3549 (91)243 (77)0.02
 Anti-CCP positive, no. (%)53 (75)290 (76)0.8513 (76)54 (79)0.9740 (74)236 (75)0.90
 Deyo index, mean ± SD1.6 ± 2.11.5 ± 1.80.671.8 ± 1.71.2 ± 1.60.201.6 ± 2.21.6 ± 1.80.94
Concurrent therapy with MTX, no. (%)         
 Traditional DMARDs39 (55)226 (59)0.547 (41)33 (49)0.5932 (59)193 (61)0.80
 TNFα blockers§29 (41)143 (37)0.567 (41)12 (18)0.0422 (41)131 (41)0.92
 Other biologic DMARDs1 (1)12 (3)0.431 (6)0 (0)0.000 (0)12 (4)0.16

MTX therapy and adherence.

Prescribed weekly MTX doses were similar in high- and low-adherence groups across cohorts (Table 2). However, observed weekly doses and duration of use were both greater in the high-adherence group compared to the low-adherence group. These differences were statistically significant in the overall group and in the group defined by established MTX use.

Table 2. Summary of drug course information*
CourseFull cohortIncident user cohortEstablished user cohort
MPR <80% (n = 71)MPR ≥80% (n = 384)PMPR <80% (n = 17)MPR ≥80% (n = 68)PMPR <80% (n = 54)MPR ≥80% (n = 316)P
  • *

    Values are the mean ± SD. MPR = medication possession ratio.

  • The course total dose (mg) during the course divided by the total prescribed duration (weeks).

  • The course total dose (mg) during the course divided by the total course duration (weeks).

  • §

    P < 0.05.

  • The sum of all of the expected prescription durations for each prescription over the course.

  • #

    The time from the date of the first prescription to the end of the final prescription in a course.

  • **

    The total number of milligrams dispensed with each prescription over the course.

Average prescribed weekly dosage, mg/week16 ± 416 ± 40.615 ± 316 ± 40.4416 ± 416 ± 40.95
Average observed weekly dosage, mg/week11 ± 3§16 ± 5§< 0.001§10 ± 2§16 ± 6§< 0.001§12 ± 3§16 ± 5§< 0.001§
Total prescribed duration, days654 ± 599§1,345 ± 975§< 0.001§350 ± 320§605 ± 382§0.02§749 ± 634§1,504 ± 1,395§< 0.01§
Observed course duration, days#902 ± 783§1,370 ± 980§< 0.001§497 ± 410592 ± 3740.361,029 ± 803§1,536 ± 14,727§< 0.01§
Course total dose, mg**1,492 ± 1,425§3,154 ± 2,490§< 0.001§768 ± 7701,421 ± 1,0790.021,720 ± 1,307§3,527 ± 3,245§< 0.01§

Clinical outcomes during MTX therapy.

Mean ± SD disease activity values during VARA followup are shown in Table 3. In these unadjusted analyses, high MTX adherence was associated with a significantly lower mean DAS28 score compared to low MTX adherence across the study groups. Higher adherence to MTX was also associated with improvements in secondary outcomes, including ESR (borderline in the established use cohort), CRP level, and tender and swollen joints in the overall group and in the established use group.

Table 3. Clinical outcome measures in the full, incident methotrexate user, and established methotrexate user cohorts*
CourseFull cohortIncident user cohortEstablished user cohort
MPR <80% (n = 71)MPR ≥80% (n = 384)PMPR <80% (n = 17)MPR ≥80% (n = 68)PMPR <80% (n = 54)MPR ≥80% (n = 316)P
  • *

    Values are the mean ± SD. MPR = medication possession ratio; DAS28 = Disease Activity Score with 28 joints.

  • P < 0.05.

Primary outcome measure         
 DAS283.9 ± 1.53.6 ± 1.20.024.2 ± 1.43.6 ± 1.10.103.8 ± 1.53.5 ± 1.20.09
Secondary outcome measures         
 Erythrocyte sedimentation rate, mm/hour29 ± 2424 ± 180.0535.2 ± 2625 ± 170.0527.4 ± 23.024.2 ± 18.40.26
 C-reactive protein level, mg/dl1.6 ± 1.51.2 ± 1.30.031.8 ± 1.61.4 ± 1.90.511.6 ± 1.51.2 ± 1.20.03
 Tender joints (range 0–28)5.4 ± 6.23.6 ± 4.5< 0.014.9 ± 4.73.6 ± 4.40.285.6 ± 6.713.7 ± 4.60.02
 Swollen joints (range 0–28)4.3 ± 4.83.2 ± 3.30.024.6 ± 4.63.2 ± 3.50.194.3 ± 4.93.2 ± 3.20.03
 Health Assessment Questionnaire score (range 0–3.0)0.9 ± 0.560.9 ± 0.520.931.0 ± 0.450.9 ± 0.540.660.9 ± 0.60.9 ± 0.90.75
 Visual analog scale38 ± 2439 ± 200.6337 ± 2132 ± 190.2735 ± 2839 ± 200.22
 Patient global37 ± 2439 ± 200.3041 ± 2341 ± 210.97534 ± 2433 ± 170.40
 Pain (range 0–10)4.4 ± 2.74.2 ± 2.20.713.8 ± 2.94.4 ± 2.10.334.5 ± 2.64.21 ± 2.20.34

Multivariable models for DAS28.

Multivariable linear regression was then used to examine whether the association of higher MTX adherence with improved DAS28 scores was independent of potentially confounding variables (Table 4). Differences in mean DAS28 scores based on MTX adherence remained statistically significant across cohorts after multivariable adjustments, although this difference was greater in magnitude among incident MTX users compared to established users. An analysis limited to male subjects produced very similar results to those shown. The number of female subjects was too small for a separate analysis.

Table 4. Mean difference in DAS28 based on adherence to MTX and other covariates, calculated using univariate (crude) and multivariable linear regression*
 Full user cohortIncident user cohortEstablished user cohort
Crude (95% CI)Demographic adjusted (95% CI)Full model (95% CI)Crude (95% CI)Demographic adjusted (95% CI)Full model (95% CI)Crude (95% CI)Demographic adjusted (95% CI)Full model (95% CI)
  • *

    Quartile disease duration cut points: full cohort, 0.05, 2.3, 9.9, and 11.1; incident cohort, 0.06, 1.1, 5.4, and 15.7; established cohort, 0.02, 0.47, 4.1, and 13.4, respectively. DAS28 = Disease Activity Score with 28 joints; MTX = methotrexate; 95% CI = 95% confidence interval; MPR = medication possession ratio; Q = quartile; anti-CCP = anti–cyclic citrullinated peptide antibody; DMARDs = disease-modifying antirheumatic drugs; TNFα = tumor necrosis factor α.

  • P < 0.05.

MPR ≥80%−0.34 (−0.68, −0.06)−0.35 (−0.66, −0.04)−0.37 (−0.67, −0.07)−0.54 (−1.18, 0.11)−0.56 (−1.2, 0.08)−0.40 (−1.11, 0.30)−0.38 (−0.67, −0.05)−0.29 (−0.65, −0.07)−0.37 (−0.72, −0.02)
Age at enrollment, years 0 (−0.01, 0.01)0.01 (0.00, 0.02) −0.01 (−0.04, 0.01)0.01 (−0.03, 0.01) 0.01 (−0.01, 0.02)0.01 (0.00, 0.02)
Disease duration Q2 −0.12 (−0.43, 0.20)−0.36 (−0.68, −0.04) −0.13 (−0.04, 0.01)−0.15 (−0.88, 0.59) −0.05 (−0.41, 0.31)−0.17 (−0.52, 0.17)
Disease duration Q3 0.08 (−0.24, 0.40)−0.13 (−0.45, 0.20) 0.13 (−0.59, 0.85)−0.01 (−0.75,0.74) 0.03 (−0.38, 0.33)−0.17 (−0.52, 0.17)
Disease duration Q4 −0.11 (−0.43, 0.20)−0.24 (−0.56, 0.07) 0.33 (−0.4, 1.08)0.26 (−0.50, 1.03) −0.07 (−0.42, 0.29)−0.11 (−0.46, 0.22)
Male sex −0.16 (−0.61, 0.28)−0.18 (−0.61, 0.25) 0.88 (−0.31, 2.07)0.68 (−0.55, 1.91) −0.40 (−0.88, 0.09)−0.41 (−0.87, 0.06)
African American race 0.08 (−0.24, 0.40)0.09 (−0.22, 0.41) 0.10 (−0.59, 0.79)−0.17 (−0.53, 0.87) 0.02 (−0.34, 0.39)0.1 (−0.34, 0.36)
Ever tobacco user 0.31 (0.03, 0.59)0.30 (0.03, 0.57) 0.08 (−0.66, 0.82)0.08 (−0.68, 0.85) 0.32 (0.02, 0.62)0.32 (0.03, 0.61)
Rheumatoid factor positive 0.27 (−0.04, 0.59)0.26 (−0.05, 0.50) −0.43 (−1.19, 0.34)−0.45 (−1.21, 0.32) 0.42 (0.07, 0.77)0.40 (0.02, 0.61)
Anti-CCP positive 0.22 (−0.08, 0.52)0.21 (−0.08, 0.50) 0.36 (−0.39, 1.11)−0.34 (−0.41, 1.10) 0.16 (−0.16, 0.49)0.14 (−0.17, 0.46)
Baseline DAS28  0.28 (0.52, 5.1)0.26 (0.21, 4.89) 
Concurrent therapy with MTX         
 Traditional DMARDs  0.28 (0.06, 0.50)  0.07 (−0.57, 0.72)  0.43 (0.17, 0.69)
 TNFα blockers  0.39 (0.15, 0.63)  −1.76 (−0.75, 4.2)  1.18 (0.45, 1.91)
 Other biologic DMARDs  1.37 (0.62, 2.02)  1.73 (−0.74, 4.21)  2.38 (1.43, 3.33)

DISCUSSION

In this investigation, high adherence to prescribed MTX was associated with improved clinical outcomes in both new and established MTX users. Our analysis was limited to the initial course of MTX during VARA observation to provide a more homogenous cohort. We found that the majority of US veterans showed high levels of medication adherence during their first course of therapy, with 84% having an MPR ≥0.8. This is similar to prior reports of MTX adherence in other RA populations receiving MTX (13, 19), but higher than that seen with other medications commonly used in RA, including nonsteroidal antiinflammatory drugs (20). This relatively favorable rate of treatment adherence is also higher than commonly reported in patients receiving medications for chronic conditions such as hypertension (21), hyperlipidemia (21, 22), and diabetes mellitus (23). Our observation of relatively high adherence may have been in part a result of our large established cohort, although the percentage of subjects with high adherence was similar in the incident and established cohorts. While the observations were very similar in the incident and established groups, the power to confirm that the differences observed in the incident group was limited by the small sample size in this subpopulation.

In all groups, the mean MTX dosage prescribed, 16 mg per week, was within the established standard of care and consistent with doses employed in recent RA clinical trials (6). In the high MTX adherence groups, the observed dose (the estimated dose that the patient actually received) approximated the prescribed dosage of 16 mg per week. The low-adherence group, however, had an estimated dosage of 11 mg per week, which is substantially lower than the prescribed dose and generally not considered to be as effective as the higher MTX dose (24). The lower-adherence group also exhibited longer delays in refilling their MTX during their first MTX course, which may have additionally affected clinical outcomes. These combined observations may explain the association between MTX adherence and disease outcome measures, with lower adherence associated with higher DAS28, ESR, and CRP values in patients with RA.

A similar association of MTX adherence with CRP level was recently reported in a Danish population of RA patients (25). Using administrative databases, this study did not examine clinical outcome measures other than laboratory values. In that study, the reported MPR (87.3%) was very similar to the adherence observed in our study. These investigators also noted that in patients initiating MTX, strong perceptions of “a personal need for the treatment” were associated with higher treatment adherence (19). Consistent with findings from our study, other investigators have measured persistence on MTX and demonstrated that patients with higher persistence had improved DAS28, CRP levels, and patient-reported outcomes (26).

Several factors could serve to confound the relationship of treatment adherence with outcomes. Results from our multivariable analyses showed that some factors such as smoking were significantly associated with disease activity but did not significantly confound the relationship between MTX adherence and DAS28 score. Furthermore, the effect estimates in each cohort were not highly influenced after accounting for these potential confounders. It is important to note that enrollment clinical measures are a mix of baseline and intermediate variables for the full cohort, necessitating the evaluation of adherence in the established and incident MTX user subsets. For the established cohort (and therefore, the full cohort), we were unable to adjust for baseline DAS28, as MTX treatment was initiated prior to VARA registry enrollment. At enrollment, clinical disease activity measures were lower in the high-adherence groups in the full and established groups, which likely reflect the beneficial effects of higher adherence and more optimal MTX use even prior to VARA enrollment. Of note, there were no differences in baseline DAS28 at the time of enrollment for the incident cohort, suggesting a similar clinical disease state at the time of enrollment for patients that ultimately demonstrated either high or low adherence. This observation also suggests that baseline disease activity does not predict subsequent adherence; therefore, it is unlikely to confound the relationship between adherence and posttreatment average DAS28. However, it is recognized that unmeasured confounding could impact our findings.

Estimation of days' supply was a key component to inferring the duration of drug courses, observed weekly dose, and MPR. While VA administrative pharmacy databases have been used for evaluation of medication adherence in other clinical settings (22, 27), medications for patients with RA present special challenges related to complex dosing schedules, varying formulations, and different routes of administration. Because of this complexity, errors in data input for days' supply and units dispensed can occur and result in improbable estimates of exposure duration and weekly dosing. These problems are particularly relevant to MTX, the most commonly used DMARD in RA, because of its varying routes of administration and sometimes complex dosing schedules. To address these problems, we employed an algorithm that added information from the dispensing instructions (“sig”) to the calculation of prescription duration, thus improving the accuracy of the estimated course duration and measure of adherence (28).

Our study has the additional strengths of the prospective data collection in VARA and the documented “gold standard” diagnosis of RA by a rheumatologist in all of the study subjects. In contrast, other studies of this nature have frequently relied on administrative data to establish the diagnosis of RA (13, 19, 25), a method prone to misclassification (29). Other strengths include the large number of comprehensive clinical outcome measures and other potential confounders available for analysis, in addition to the collection of these data by clinicians who were not specifically informed about the patient medication adherence history.

In contrast to other RA populations, VARA is composed of predominantly older men with substantial comorbidity (1), potentially limiting the generalizability of our results. While most enrolled veterans received their comprehensive care within the VA, the potential for concurrent medication from non-VA sources cannot be excluded. Additionally, this study included many patients for whom disease activity measures were not available for much of their MTX course before entry into the VA system, preventing adjustment for baseline disease activity among these established MTX users. An analysis of incident and established users, however, showed that higher adherence was associated with lower disease activity in both groups. The effect estimates were lower in the established cohort, a result that likely reflects early unmeasured changes in DAS28, since VARA enrollment occurred after initiation of MTX in this group. The limitation of small sample size in the incident cohort is also noted, as well as the impact on statistical analysis in this subgroup because of limited power.

Our study does not specifically address the reasons for high or low adherence in our patients. Future work will be directed at this issue specifically. There are reports that high medication adherence is associated with “healthy behaviors” that could more globally impact clinical outcomes. For example, high adherence to placebo in osteoporosis intervention trials has been associated with a decrease in fracture rate (30). Treatment adherence has been associated with several other factors, including cognitive (31, 32) and psychological impairment (31, 33, 34), complex dosing schedules (33), treatment of an asymptomatic disease (35), medication side effects (34), substance abuse (27, 36), barriers to obtaining medication (including costs) (34, 37–39), consistency of followup (35, 40), ethnicity (41), and provider/patient communication difficulties. Further investigations will be needed to identify the reasons for poor adherence to medications in this population and potentially develop strategies to improve patient compliance with prescribed therapies.

In summary, this investigation demonstrates that high adherence to prescribed MTX is associated with improved clinical outcome measures in RA, an effect that is independent of other important demographic and treatment-related clinical factors. These data support the clinical benefit of MTX therapy in “day-to-day” clinical practice and imply that strategies focused on optimizing MTX adherence may have the important potential to improve patient outcomes. Moreover, these findings demonstrate that VA administrative pharmacy databases can be combined with our observational cohort data to evaluate clinical outcomes during DMARD therapy. This methodology has the potential for expanded applications for the continued evaluation of community practice DMARD experience.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Cannon had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Cannon, Mikuls, Hayden, Curtis, Reimold, Caplan, Kerr, Richards, Johnson, Sauer.

Acquisition of data. Cannon, Mikuls, Hayden, Reimold, Caplan, Kerr, Richards, Johnson, Sauer.

Analysis and interpretation of data. Cannon, Mikuls, Hayden, Ying, Curtis, Reimold, Caplan, Kerr, Richards, Johnson, Sauer.

Ancillary