Matching-adjusted indirect comparison of adalimumab vs etanercept and infliximab for the treatment of psoriatic arthritis

Abstract

Objectives:

No head-to-head trial has compared the efficacy of adalimumab vs etanercept and infliximab for psoriatic arthritis (PsA). This study implements a matching-adjusted indirect comparison technique to address that gap.

Methods:

Patient-level data from a placebo-controlled trial of adalimumab (ADEPT) were re-weighted to match average baseline characteristics from pivotal published trials of etanercept and infliximab. ADEPT patients were re-weighted by odds of enrollment in comparator trials, estimated using logistic regression. Matched-on characteristics included PsA duration, age, gender, severity, active psoriasis, and concomitant treatment. After matching, placebo-adjusted treatment arms were compared at weeks 12 (or 14) and 24. Outcomes included ACR20/50/70, PsARC, HAQ, and modified TSS. PASI50/75/90 were compared for patients with active psoriasis. Cost per responder (CPR) was assessed in the US and Germany using matching-adjusted end-points and drug list prices. Statistical significance was assessed using weighted t-tests.

Results:

After matching, adalimumab-treated patients had greater placebo-adjusted rates of ACR70 and PASI50/75/90 at week 24 compared with etanercept (all p < 0.05). Adalimumab patients had a higher placebo-adjusted rate of ACR70 than infliximab at week 14 (p = 0.034). Adalimumab treatment had lower CPR for ACR70 and PASI50/75/90 compared with etanercept at week 24, in both the US and Germany (all p < 0.02). Adalimumab had lower CPR than infliximab for all outcomes at week 24 (all p < 0.05).

Conclusion:

Adalimumab is associated with higher ACR70 and PASI50/75/90 response rates than etanercept at week 24 and a higher ACR70 response rate than infliximab at week 14. Adalimumab has significant advantages over etanercept and infliximab in CPR across multiple end-points.

Key limitations:

The matching-adjusted indirect comparison method cannot account for unobserved differences in patient characteristics across trials, and only a head-to-head randomized clinical trial can fully avoid the limitations of indirect comparisons. CPR findings are limited to the US and German markets, and may not be generalizable to other markets with different relative pricing.

Keywords::

• Psoriatic arthritis
• Tumor necrosis factor-alpha
• Indirect comparison
• Clinical trials
• Treatment efficacy
• Costs and cost analysis

Introduction

Psoriatic arthritis (PsA) is a chronic, inflammatory arthritis that commonly affects patients with psoriasis. It is characterized by painful, tender, and swollen joints1. Dactylitis (an inflammation of the entire digit) and enthesitis (inflammation at the site of the insertion of a tendon into the bone) are common features of PsA1. The majority of PsA patients experience a progression of clinical damage and increasing disability2,3. They also have reduced quality-of-life and function4–7. In a 2005 US survey, 39% of PsA patients rated their PsA as having a large impact on their daily life8.

Approximately 2.1% of US adults are affected with psoriasis9 and estimates of the prevalence of PsA among patients with psoriasis range from 6–39%10,11. Most recently, a 2005 study of the epidemiology of psoriatic arthritis based on a US survey found that 11% of individuals with psoriasis also had psoriatic arthritis8. The National Psoriasis Foundation estimated in 2009 that ∼30% of psoriasis patients will eventually develop PsA9. Psoriasis typically precedes PsA by an average of 8.5 years, but 11% of PsA patients surveyed developed PsA prior to psoriasis8. PsA generally develops between the ages of 30–55, and men and women are equally likely to be affected12.

Non-steroidal Anti-Inflammatory Drugs (NSAIDs) and traditional Disease-Modifying Anti-Rheumatic Drugs (DMARDs) are common first line therapies for PsA13. These treatments can be effective in some patients, having some effect on arthritis and skin disease, but little effect on radiological joint damage or axial disease13. Among the treatment options for PsA, tumor necrosis factor-alpha (TNF-α) inhibitors have demonstrated the greatest efficacy13,14. These include adalimumab, etanercept, and infliximab, and the more recently FDA-approved golimumab. These treatments are effective in treating all forms of PsA, including arthritis, skin and nail disease, spinal disease, enthesitis, and dactylitis13. They have also been shown to improve quality-of-life and prevent radiological joint damage13. TNF-α inhibitors are recommended for patients who have failed treatment with traditional DMARDs or as a first-line therapy for especially severe cases13.

Comparisons among TNF-α inhibitors for PsA have been limited by the absence of head-to-head randomized trials for these treatments. In the absence of head-to-head clinical trials, researchers rely on indirect comparisons of treatments. The most commonly used method for indirect comparisons is a meta-analysis15. There have been two prior studies the authors are aware of comparing the efficacy of adalimumab, etanercept, and infliximab for the treatment of PsA. One study relied on a meta-analysis of the TNF-α inhibitors and found no significant difference in the efficacy of adalimumab, etanercept, and infliximab16. The other employed a mixed treatment comparison (MTC) methodology to compare a single clinical end-point (20% improvement in the American College of Rheumatology criteria, ACR20) across adalimumab-, etanercept-, and infliximab-treated patients, and found no significant differences across treatment arms17. Despite this finding, the Bayesian analysis concluded that etanercpet was associated with a higher probability of efficacy when only one of several possible outcome measures was considered.

Indirect comparisons that rely on aggregate data are subject to limitations, including potential for bias due to differences in trial populations15,18–22. As noted by Gladman1, the placebo-controlled clinical trials of adalimumab, etanercept, and infliximab have different definitions for patient eligibility, including differences in the joints assessed. These differences, as well as any differences in disease severity across trials, could bias the results.

This study addresses the gap in comparative effectiveness literature for treatment of PsA by comparing the short-term response rates for adalimumab, etanercept, and infliximab using the matching-adjusted indirect comparison approach suggested by Signorovitch et al.23 The underlying rationale is to leverage individual patient data (IPD) for one treatment, adalimumab, and to re-weight those patients such that enrollment criteria and average baseline characteristics match those reported for the trials of the comparator treatments. Treatment outcomes with adalimumab can then be compared with those of etanercept and infliximab using observationally equivalent populations.

Once comparable clinical outcomes are available, cost per responder (CPR) can be assessed using the adjusted clinical outcomes and publically available drug list prices. The study assesses the cost per responder for adalimumab, etanercept, and infliximab across multiple end-points. As drug prices may vary across markets due to differing regulatory and competitive landscapes, the study includes CPR calculations for two major markets, one in North America (US), and one in Europe (Germany).

Patients and methods

The research approach consisted of five steps. First, a literature review was conducted and the relevant clinical trials identified. Second, after selecting the relevant clinical trials for etanercept and inflixmab, actual individual patient data from the adalimumab clinical trial described in Mease et al.24 (ADEPT) were used to adjust for differences in study design. Third, baseline characteristics available for matching were identified. Fourth, the average baseline characteristics were matched across trials by re-weighting patients in ADEPT. Fifth, after matching, outcomes were compared across the balanced trial populations.

Literature review

A detailed literature review identified all randomized, placebo-controlled, double-blind clinical trials of adalimumab, etanercept, and infliximab for the treatment of active PsA. The pivotal phase III placebo-controlled trials of adalimumab24, etanercept25, and infliximab26 were selected since they were most similar in patient selection criteria and length of follow-up. Differences across trials in patient eligibility, inclusion and exclusion criteria, and statistical analysis were noted. A Technology Assessment Report for the British National Institute of Clinical Excellence27 was also assessed and referenced for study outcomes not cited within comparator publications.

For etanercept treatment, the randomized, placebo-controlled trial described in Mease et al.25 was selected for analysis. This trial included patients aged 18–70 with at least three swollen and three tender joints. Joints assessed included the 66 or 68 joints that the American College of Rheumatology (ACR) criteria suggest for swelling or tenderness, respectively, as well as the 10 additional joints commonly involved in PsA. Outcomes included the proportion of patients who achieved at least a 20%, 50%, or 70% improvement according to ACR criteria (ACR20, ACR50, and ACR70) at weeks 12 and 24, the proportion of patients who achieved an improvement according to the PsA Response Criteria (PsARC) at weeks 12 and 24, the mean percentage change in Health Assessment Questionnaire (HAQ) disability index at weeks 12 and 24, and the annualized change in modified total sharp score (TSS) from baseline to 6 months. Among the sub-set of patients with active psoriasis, the proportion of patients achieving at least a 50%, 75%, or 90% improvement in Psoriasis Area and Severity Index (PASI50, PASI75, and PASI90) at week 24 was also observed.

The randomized, placebo-controlled trial described in Antoni et al.26 was selected for infliximab treatment. This trial included patients at least 18 years old with at least five swollen and five tender joints and a negative test for rheumatoid factor. Joints assessed included only the 66 or 68 joints that the ACR criteria suggest for swelling or tenderness, respectively. Outcomes included the proportion of patients who achieved ACR20, ACR50, ACR70, and PsARC at weeks 14 and 24, and the mean percentage change in HAQ at weeks 14 and 24. Among the sub-set of patients with active psoriasis, the proportion of patients achieving PASI50, PASI75, and PASI90 at weeks 14 and 24 was also observed.

The ADEPT trial24 was selected for adalimumab treatment. This trial enrolled patients at least 18 years old with at least three swollen and three tender joints. Joints assessed included the 66 or 68 joints that the ACR criteria suggest for swelling or tenderness, respectively, as well as the 10 additional joints commonly involved in PsA. Outcomes assessed included the proportion of patients who achieved ACR20, ACR50, ACR70, and PsARC at weeks 12 and 24, the mean percentage change in HAQ at weeks 12 and 24, and the change in modified TSS from baseline to week 24. Among the sub-set of patients with active psoriasis, the proportion of patients achieving PASI50, PASI75, and PASI90 at weeks 12 and 24 was also observed.

Another, smaller scale, phase III trial for adalimumab28 was not included in the analysis due to significant differences in study design which could not be accounted for using matching-adjusted indirect comparison methodology. Specifically, two main differences precluded use of this study. First, the double blind format was only maintained through week 12 of the trial, after which the patients entered an open-label extension. As such, comparison of outcomes at week 24 was not feasible. Second, the average weight of patients in Genovese et al.28 was considerably higher than that observed in the ADEPT trial (91.5 kg vs 86 kg for adalimumab-treated patients, respectively). As body weight may affect efficacy of treatment, one would need to account for this difference in the indirect comparison. However, average baseline weight was not reported for the infliximab and etanercept trials, and could not be matched in the analysis. Therefore, the Genovese et al.28 study was not included.

Adjusting for differences in study design

Individual patient data (IPD) were available for the adalimumab clinical trial ADEPT, while only published summary statistics were available for the etanercept and infliximab trials. Differences in study design were addressed using the ADEPT IPD.

For the comparison with etanercept, 13 patients over age 70 were excluded to match the study design of the etanercept trial. Likewise, mean change in modified TSS at 6 months was recalculated for ADEPT patients by linear extrapolation from week 24 change in modified TSS, in order to facilitate comparison with etanercept, for which radiographic progression was reported at 6 months.

To match the study design of the infliximab trial, patients in ADEPT were excluded from the comparison with infliximab if they tested positive for rheumatoid factor. Tenderness and swelling were only assessed for the joints that the ACR criteria suggest and patients were excluded if they had fewer than five tender or five swollen joints. A total of 76 ADEPT patients were excluded. Because outcomes in the infliximab trial were reported at week 14 rather than week 12, outcomes in ADEPT at week 14 were imputed as the average of a patient’s outcomes at weeks 12 and 16, or a weighted average of other available measurements surrounding week 14.

Identification of baseline characteristics available for matching

Aggregate baseline characteristics and outcomes were obtained from the publications of the etanercept and infliximab clinical trials. All baseline characteristics reported for comparators and available in ADEPT were used in the matching algorithm. The matched-on characteristics included PsA duration, age, gender, PsA severity, active psoriasis (%), and concomitant treatment (a complete listing of matched-on characteristics is reported in Tables 1 and 2). Patients in ADEPT with missing values for matched-on baseline characteristics were excluded (four for the etanercept comparison, and three for infliximab). Baseline characteristics before matching were compared using t-tests.

Table 1.  Baseline demographics and disease characteristics—adalimumab vs etanercept.

	Before matching			After matching
	ADEPT Pooled ADA and PBO arms	Mease et al.25 Pooled ETA and PBO arms	p-value	ADEPT Pooled ADA and PBO arms	Mease et al.25 Pooled ETA and PBO arms	p-value
Number of patients, n	296	205		296	205
Age (years), mean (SD)	47.8 (10.6)	47.4	–	47.4 (10.4)	47.4	1.000
Female, %	44.6	48.8	0.357	48.8	48.8	1.000
White, %	95.3	90.7	0.057	90.7	90.7	1.000
Duration of PsA (years), mean (SD)	9.5 (8.5)	9.1	–	9.1 (8.6)	9.1
Duration of Psoriasis (years), mean (SD)	17.1 (12.3)	19.0	–	19.0 (13.4)	19.0	1.000
Rheumatoid Factor Positive, %	9.8	6.5	0.172	6.5	6.5	1.000
Psoriasis Evaluation, %
Patients with ≥3% BSA affected  with Psoriasis	43.6	62.4	<0.001*	62.4	62.4	1.000
Radiographic Scores at Baseline, mean (SD)						1.000
Total Sharp Score	18.1 (33.4)	22.0	–	22.0 (41.6)	22.0	1.000
Erosion Score	9.4 (18.6)	10.7	–	10.7 (22.1)	10.7	1.000
Joint Space Narrowing Score	8.8 (16.1)	11.3	–	11.3 (20.7)	11.3	1.000
Concomitant Therapy During Study, %
Corticosteroids	12.5	17.1	0.161	17.1	17.1	1.000
Methotrexate	51.0	41.5	0.034*	41.5	41.5	1.000

ADA, adalimumab; ETA, etanercept; PBO, placebo; PsA, Psoriatic Arthritis; BSA, body surface area; SD, standard deviation.

p-values could not be calculated for continuous baseline characteristics because standard deviations were not reported in Mease et al.25

*p-value <0.05.

Table 2.  Baseline demographics and disease characteristics—adalimumab vs infliximab.

	Before matching			After matching
	ADEPT Pooled ADA and PBO arms	Antoni et al.26 Pooled INF and PBO arms	p-value	ADEPT Pooled ADA and PBO arms	Antoni et al.26 Pooled INF and PBO arms	p-value
Number of patients, n	234	200		234	200
Age (years), mean (SD)	48.8 (12.0)	46.8 (12.0)	0.084	46.8 (12.0)	46.8 (12.0)	1.000
Female, %	45.3	39.0	0.185	39.0	39.0	1.000
Duration of PsA (years), mean (SD)	9.0 (7.7)	8.0 (7.5)	0.137	8.0 (7.4)	8.0 (7.5)	1.000
ACR Component, mean (SD)
Number of Swollen Joints	14.9 (10.3)	14.2 (8.4)	0.424	14.2 (9.8)	14.2 (8.4)	1.000
Number of Tender Joints	24.7 (15.1)	24.9 (13.7)	0.900	24.9 (14.3)	24.9 (13.7)	1.000
CRP (mg/L)	14.0 (17.8)	21.0 (28.2)	0.003*	21.0 (29.6)	21.0 (28.2)	1.000
Physician’s Global Assessment of  Disease Activity	54.0 (15.0)	57.0 (17.5)	0.059	57.0 (14.5)	57.0 (17.5)	1.000
Patient’s Global Assessment of  Disease Activity	47.8 (21.9)	56.5 (21.5)	<0.001*	56.5 (21.1)	56.5 (21.5)	1.000
Patient’s Assessment of Pain	50.5 (21.3)	57.5 (22.0)	<0.001*	57.5 (21.5)	57.5 (22.0)	1.000
HAQ Disability Index	1.0 (0.6)	1.1 (0.6)	0.085	1.1 (0.6)	1.1 (0.6)	1.000
Patients with 1 or more Dactylitis Digits, %	39.3	40.5	0.802	40.5	40.5	1.000
Patients with Enthesopathy, %	37.2	38.5	0.778	38.5	38.5	1.000
Psoriasis Evaluation, %
Patients with ≥3% BSA affected  with Psoriasis	44.4	85.0	<0.001*	85.0	85.0	1.000
SF-36 Score (years), mean (SD)
Physical Component	32.9 (9.5)	32.0 (9.2)	0.314	32.0 (9.4)	32.0 (9.2)	1.000
Mental Component	47.7 (11.4)	46.3 (11.9)	0.213	46.3 (11.9)	46.3 (11.9)	1.000
Concomitant Therapy During Study, %
Corticosteroids	10.7	12.5	0.558	12.5	12.5	1.000
Methotrexate	51.7	46.0	0.236	46.0	46.0	1.000

ADA, adalimumab; INF, infliximab; PBO, placebo; PsA, Psoriatic Arthritis; BSA, body surface area; SD, standard deviation; ACR, American College of Rheumatology criteria; CRP, C-reactive protein; HAQ, Health Assessment Questionnaire.

*p-value <0.05.

Matching of baseline characteristics

For each comparison between adalimumab and etanercept or infliximab, all mean baseline characteristics reported from the comparator clinical trial and available in ADEPT were exactly matched to account for possible differences at baseline across trial populations. Mean baseline characteristics were matched by re-weighting individual patients from ADEPT by their odds of having been enrolled in the comparator trial, using the indirect comparison methodology developed by Signorovitch et al.23 Specifically, individual patients in the ADEPT trial were assigned weights such that two conditions were met: (a) weighted average baseline characteristics in the ADEPT trial precisely matched those reported in the comparator trials; and (b) a patient’s weight was equal to her odds of enrolling in ADEPT vs the comparator trials, estimated much like a standard propensity score model. In order to meet these conditions, weights were derived using a logistic regression model estimating the propensity to enroll in ADEPT vs comparator trials, with all available baseline characteristics included as predictors in the model. As only summary data were available for the comparator trials, the propensity model was estimated using the method of moments29.

The derived weights ensure that mean observable baseline characteristics exactly match across trials for the pooled trial population (i.e., placebo and treatment arms). Baseline characteristics after matching were compared using weighted t-tests.

Comparison of outcomes after matching

For the comparison with etanercept, outcomes compared were the proportion of patients who achieved ACR20, ACR50, ACR70, and PsARC at weeks 12 and 24, the mean percentage change in HAQ at weeks 12 and 24, and the annualized change in modified TSS from baseline to 6 months. Among the sub-set of patients with active psoriasis, the proportion of patients achieving PASI50, PASI75, and PASI90 at week 24 was also compared.

For the comparison with infliximab, outcomes compared were the proportion of patients who achieved ACR20, ACR50, ACR70, and PsARC at weeks 14 and 24 and the mean percentage change in HAQ at weeks 14 and 24. Among the sub-set of patients with active psoriasis, the proportion of patients achieving PASI50, PASI75, and PASI90 at weeks 14 and 24 was also compared.

After matching, outcomes for treatment with adalimumab were compared to those with etanercept and infliximab based on the difference between the effects of the biologic and the placebo arm (i.e., difference-in-differences). Outcomes were compared using weighted t-tests.

In binary outcome variables for which significant differences were found, a ‘numbers needed to treat’ (NNT) calculation was conducted. The NNT was calculated as the inverse of the absolute risk reduction, measured by difference in placebo-adjusted response rates.

Cost per responder analysis

Drug cost per responder (CPR) was calculated for all binary outcome variables. The comparison was calculated at weeks 12 and 24 for the adalimumab–etanercept comparison, and weeks 14 and 24 for the adalimumab–infliximab comparison. CPR was calculated based on US and German Wholesale Acquisition Costs for 201030,31. Adalimumab patients were assumed to receive the indicated dose of 40 mg every other week, and etanercept patients the indicated 50 mg per week. Total cost of treatment was calculated for each treatment over the relevant time period (e.g., over a 12-week period, an adalimumab patient would receive six 40 mg injections, and an etanercept patient would receive 12 50 mg injections). For infliximab, dosing is weight-based, but the list price is per 100 mg unit. The weight of patients in the infliximab study was not reported. Therefore, the average weight from the matched cohort from the ADEPT clinical trial (90 kg) was used to calculate the dose. Total cost of treatment over the relevant time frame was divided by the placebo-adjusted response rate for each outcome to arrive at the CPR. Weighted t-tests were used to compare CPR across trials.

Results

Baseline characteristics

Prior to matching, there were significant differences in baseline characteristics between ADEPT patients (pooled adalimumab and placebo arms) and patients in the etanercept and infliximab trials (pooled active and placebo arms). Compared with patients in the etanercept trial (Table 1), ADEPT patients were less likely to have active psoriasis (43.6% with ≥3% BSA vs 62.4%, p < 0.001) and more likely to have concomitant use of methotrexate (51.0% vs 41.5%, p = 0.034). Several other baseline differences were apparent (e.g., duration of psoriasis, baseline TSS), although their statistical significance could not be tested, as variance measures were not reported for the etanercept trial. Compared with patients in the infliximab trial (Table 2), patients in ADEPT had lower severity of PsA as measured by CRP (14.0 mg/L vs 21.0, p = 0.003), patient’s global assessment of disease activity (47.8 vs 56.5, p < 0.001), and patient’s assessment of pain (50.5 vs 57.5, p < 0.001), and they were less likely to have active psoriasis (44.4% with ≥3% BSA vs 85.0%, p < 0.001). After re-weighting of ADEPT IPD (separately for each comparison), all baseline characteristics were exactly matched to the etanercept and infliximab trials, respectively.

Comparison of clinical outcomes: adalimumab vs etanercept

Table 3 reports the results of the matching-adjusted indirect comparison of adalimumab vs etanercept at weeks 12 and 24. The comparison revealed no statistically significant differences in clinical outcomes at week 12, although the placebo-adjusted rate of ACR70 appeared higher among adalimumab-treated patients compared with etanercept-treated patients (20.6% vs 10.9%, p = 0.055). Note that PASI50/75/90 were not reported for etanercept at week 12.

Table 3.  Comparison of outcomes after matching—adalimumab vs etanercept.

	ADEPT			Mease et al.25			Difference in differences		Numbers needed to treat (NNT)
	ADA [A]	PBO [B]	Δ1 = [A] − [B]	ETA [C]	PBO [D]	Δ2 = [C] − [D]	Δ1 − Δ2	p-value
Number of patients, n	142	154		101	104
Week 12 outcomes
ACR20, %	62.0	13.8	48.2	59.4	15.4	44.0	4.2	0.609	–
ACR50, %	38.7	3.5	35.2	37.6	3.8	33.8	1.4	0.840	–
ACR70, %	21.1	0.5	20.6	10.9	0.0	10.9	9.7	0.055	10.3
PsARC, %	62.2	23.4	38.9	72.3	30.8	41.5	−2.7	0.761	–
Mean % change  in HAQ^a	49.0	−1.8	50.8	53.5	6.3	47.2	3.6	0.709	–
Week 24 outcomes
ACR20, %	59.3	12.0	47.3	49.5	13.5	36.0	11.3	0.161	–
ACR50, %	41.5	4.8	36.7	36.6	3.8	32.8	3.9	0.591	–
ACR70, %	24.6	0.6	23.9	8.9	1.0	7.9	16.0	0.002*	6.3
PASI50^b, %	71.2	10.7	60.5	47.0	17.7	29.2	31.3	0.001*	3.2
PASI75^b, %	55.6	0.5	55.0	22.7	3.2	19.5	35.5	<0.001*	2.8
PASI90^b, %	40.2	0.0	40.2	6.1	3.2	2.8	37.4	<0.001*	2.7
PsARC, %	60.2	19.2	41.0	70.3	23.1	47.2	−6.2	0.465	–
Mean change  in modified TSS^c	−0.20	1.57	−1.77	−0.03	0.53	−0.56	−1.21	0.080	–
Mean % change  in HAQ^a	49.1	−7.7	56.7	53.6	6.4	47.2	9.5	0.406	–

ADA, adalimumab; ETA, etanercept; PBO, placebo; Δ₁, ADA–PBO; Δ₂, ETA–PBO; ACR, American College of Rheumatology criteria; HAQ, Health Assessment Questionnaire; PASI, Psoriasis Area and Severity Index; PsARC, PsA Response Criteria; TSS, total sharp score.

*p-value <0.05.

^aPatients with HAQ disability index of zero at baseline were omitted.

^bOnly patients with psoriasis involving at least 3% of body surface area were evaluated for PASI responses.

^cModified TSS measured as mean change from baseline to 6 months. Mean change at 6 months was calculated for ADEPT patients by linear extrapolation from week 24 Modified TSS, to match the study design of Mease et al.25

At week 24, however, adalimumab-treated patients had greater placebo-adjusted rates of ACR70 (23.9% vs 7.9%, p = 0.002), PASI50 (60.5% vs 29.2%, p = 0.001), PASI75 (55.0% vs 19.5%, p < 0.001), and PASI90 (40.2% vs 2.8%, p < 0.001). Adalimumab patients also appeared to have a greater reduction from baseline relative to placebo in modified TSS at 6 months (−1.77 vs −0.56, p = 0.080). No statistically significant differences were found across the two treatment groups for rates of ACR20, ACR50, PsARC, and percent change in HAQ.

Comparison of clinical outcomes: adalimumab vs. infliximab

Results of the matching-adjusted indirect comparison of adalimumab vs infliximab are reported in Table 4. At week 14, adalimumab-treated patients had a higher placebo-adjusted rate of ACR70 compared with infliximab-treated patients (32.9% vs 14.0%, p = 0.034). No statistically significant differences were found for ACR20, ACR50, PASI50, PASI75, PASI90, PsARC, and percent change in HAQ. Similarly, no statistically significant differences were found between treatment groups at week 24.

Table 4.  Comparison of outcomes after matching—adalimumab vs infliximab.

	ADEPT			Antoni et al.26			Difference in differences		Numbers needed to treat (NNT)
	ADA [A]	PBO [B]	Δ1 = [A] − [B]	INF [C]	PBO [D]	Δ2 = [C] − [D]	Δ1 − Δ2	p-value
Number of patients, n	108	126		100	100
Week 14 outcomes
ACR20, %	56.1	14.6	41.5	58.0	11.0	47.0	−5.5	0.610	–
ACR50, %	38.7	1.2	37.5	36.0	3.0	33.0	4.5	0.635	–
ACR70, %	33.2	0.3	32.9	15.0	1.0	14.0	18.9	0.034*	5.3
PASI50^a, %	85.8	9.9	75.9	81.9	9.2	72.7	3.2	0.699	–
PASI75^a, %	70.0	0.6	69.4	63.9	2.3	61.6	7.8	0.368	–
PASI90^a, %	33.9	0.0	33.9	41.0	0.0	41.0	−7.1	0.481	–
PsARC, %	66.3	24.8	41.4	77.0	27.0	50.0	−8.6	0.447	–
Mean % change  in HAQ^b	50.5	−22.7	73.2	48.6	−18.4	67.0	6.2	0.751	–
Week 24 outcomes
ACR20, %	59.3	12.1	47.2	54.0	16.0	38.0	9.2	0.381	–
ACR50, %	44.4	3.7	40.7	41.0	4.0	37.0	3.7	0.704	–
ACR70, %	35.8	0.5	35.2	27.0	2.0	25.0	10.2	0.273	–
PASI50^a, %	83.2	11.7	71.6	74.7	8.0	66.7	4.9	0.576	–
PASI75^a, %	67.6	0.0	67.6	60.2	1.1	59.1	8.5	0.342	–
PASI90^a, %	44.3	0.0	44.3	38.6	0.0	38.6	5.8	0.578	–
PsARC, %	67.0	15.6	51.4	70.0	32.0	38.0	13.4	0.197	–
Mean % change  in HAQ^b	50.0	−15.6	65.7	46.0	−19.4	65.4	0.3	0.987	–

ADA, adalimumab; INF, infliximab; PBO, placebo; Δ₁, ADA – PBO; Δ₂, INF – PBO; ACR, American College of Rheumatology criteria; HAQ, Health Assessment Questionnaire; PASI, Psoriasis Area and Severity Index; PsARC, PsA Response Criteria.

*p-value <0.05.

^aOnly patients with psoriasis involving at least 3% of body surface area were evaluated for PASI responses.

^bPatients with HAQ disability index of zero at baseline were omitted.

Comparison of cost per responder: adalimumab vs etanercept

Using the post-match placebo-adjusted response rates, Table 5 reports the CPR calculations comparing adalimumab and infliximab. The table reports CPR in both the US and Germany, using list prices in each market. No significant differences were found in CPR at week 12, though the point estimates for cost per ACR70 responder were considerably lower for adalimumab in both the US and German markets (p = 0.060 and p = 0.068, respectively). The week 24 calculations, however, show a large difference in cost per ACR70 and PASI50/75/90 responders, suggesting that adalimumab is more cost-effective for these outcomes.

Table 5.  Comparison of cost per responder after matching—adalimumab vs etanercept.

	Placebo- adjusted response rate	Number of doses	US				Germany
			Drug cost per dose (2011 USD)^a	Drug cost per patient (2011 USD)	Drug cost per incremental responder (2011 USD)	p-value^b	Drug cost per dose (2011 EUR)^a	Drug cost per patient (2011 EUR)	Drug cost per incremental responder (2011 EUR)	p-value^b
Week 12
ACR20
Adalimumab	48.2%	6	838	5031	10,441		766	4594	9535
Etanercept	44.0%	12	427	5127	11,646	0.539	383	4594	10,436	0.612
ACR50
Adalimumab	35.2%	6	838	5031	14,288		766	4594	13,048
Etanercept	33.8%	12	427	5127	15,178	0.770	383	4594	13,601	0.841
ACR70
Adalimumab	20.6%	6	838	5031	24,414		766	4594	22,295
Etanercept	10.9%	12	427	5127	47,073	0.060	383	4594	42,183	0.068
PsARC
Adalimumab	38.9%	6	838	5031	12,949		766	4594	11,825
Etanercept	41.5%	12	427	5127	12,351	0.828	383	4594	11,068	0.761
Week 24
ACR20
Adalimumab	47.3%	12	838	10,062	21,261		766	9189	19,416
Etanercept	36.0%	24	427	10,254	28,448	0.148	383	9188	25,493	0.176
ACR50
Adalimumab	36.7%	12	838	10,062	27,453		766	9189	25,070
Etanercept	32.8%	24	427	10,254	31,273	0.532	383	9188	28,024	0.593
ACR70
Adalimumab	23.9%	12	838	10,062	42,037		766	9189	38,388
Etanercept	7.9%	24	427	10,254	128,986	0.007	383	9188	115,587	0.008
PsARC
Adalimumab	41.0%	12	838	10,062	24,527		766	9189	22,398
Etanercept	47.2%	24	427	10,254	21,714	0.527	383	9188	19,459	0.465
PASI50
Adalimumab	60.5%	12	838	10,062	16,629		766	9189	15,185
Etanercept	29.2%	24	427	10,254	35,081	0.002	383	9188	31,437	0.003
PASI75
Adalimumab	55.0%	12	838	10,062	18,280		766	9189	16,693
Etanercept	19.5%	24	427	10,254	52,578	<0.001	383	9188	47,116	<0.001
PASI90
Adalimumab	40.2%	12	838	10,062	25,024		766	9189	22,851
Etanercept	2.8%	24	427	10,254	361,702	0.011	383	9188	324,129	0.011

ACR, American College of Rheumatology criteria; PASI, Psoriasis Area and Severity Index; PsARC, PsA Response Criteria.

^aWholesale Acquisition Costs were used for drug prices.

^bCost per responder was compared using weighted t-tests.

Comparison of cost per responder: adalimumab vs infliximab

Similar calculations for the adalimumab-infliximab comparison are reported in Table 6. Using drug costs for Germany, treatment with adalimumab is associated with significantly lower cost per responder for all outcomes evaluated at weeks 14 and 24, with the exception of PASI90 at week 14 (p = 0.066). Using US drug costs, adalimumab-treated patients had lower cost per ACR50/ACR70/PASI50/PASI75 responder at week 14, and lower cost per responder for all end-points at week 24 (p < 0.05). No significant differences in CPR were found for other end-points at week 14.

Table 6.  Comparison of cost per responder after matching—adalimumab vs infliximab.

	Placebo- adjusted response rate	Number of doses	US				Germany
			Drug cost per dose (2011 USD)^a	Drug cost per patient (2011 USD)^b	Drug cost per incremental responder (2011 USD)	p-value^c	Drug cost per dose (2011 EUR)^a	Drug cost per patient (2011 EUR)^b	Drug cost per incremental responder (2011 EUR)	p-value^c
Week 14
ACR20
Adalimumab	41.5%	7	838	5869	14,154		766	5360	12,925
Infliximab	47.0%	3	2971	8913	18,963	0.247	3659	10,977	23,355	0.019
ACR50
Adalimumab	37.5%	7	838	5869	15,647		766	5360	14,289
Infliximab	33.0%	3	2971	8913	27,008	0.038	3659	10,977	33,263	0.001
ACR70
Adalimumab	32.9%	7	838	5869	17,835		766	5360	16,287
Infliximab	14.0%	3	2971	8913	63,661	<0.001	3659	10,977	78,405	<0.001
PsARC
Adalimumab	41.4%	7	838	5869	14,172		766	5360	12,941
Infliximab	50.0%	3	2971	8913	17,825	0.377	3659	10,977	21,953	0.042
PASI50
Adalimumab	75.9%	7	838	5869	7731		766	5360	7060
Infliximab	72.7%	3	2971	8913	12,254	<0.001	3659	10,977	15,092	<0.001
PASI75
Adalimumab	69.4%	7	838	5869	8460		766	5360	7725
Infliximab	61.6%	3	2971	8913	14,479	<0.001	3659	10,977	17,832	<0.001
PASI90
Adalimumab	33.9%	7	838	5869	17,333		766	5360	15,828
Infliximab	41.0%	3	2971	8913	21,757	0.426	3659	10,977	26,796	0.066
Week 24
ACR20
Adalimumab	47.2%	12	838	10,062	21,306		766	9189	19,457
Infliximab	38.0%	5	2971	14,854	39,090	0.012	3659	18,295	48,144	<0.001
ACR50
Adalimumab	40.7%	12	838	10,062	24,726		766	9189	22,580
Infliximab	37.0%	5	2971	14,854	40,147	0.049	3659	18,295	49,445	0.001
ACR70
Adalimumab	35.2%	12	838	10,062	28,556		766	9189	26,077
Infliximab	25.0%	5	2971	14,854	59,417	0.013	3659	18,295	73,178	<0.001
PsARC
Adalimumab	51.4%	12	838	10,062	19,578		766	9189	17,879
Infliximab	38.0%	5	2971	14,854	39,090	0.003	3659	18,295	48,144	<0.001
PASI50
Adalimumab	71.6%	12	838	10,062	14,062		766	9189	12,842
Infliximab	66.7%	5	2971	14,854	22,286	<0.001	3659	18,295	27,448	<0.001
PASI75
Adalimumab	67.6%	12	838	10,062	14,888		766	9189	13,596
Infliximab	59.1%	5	2971	14,854	25,138	<0.001	3659	18,295	30,960	<0.001
PASI90
Adalimumab	44.3%	12	838	10,062	22,700		766	9189	20,730
Infliximab	38.6%	5	2971	14,854	38,528	0.029	3659	18,295	47,451	<0.001

ACR, American College of Rheumatology criteria; PASI, Psoriasis Area and Severity Index; PsARC, PsA Response Criteria.

^aWholesale Acquisition Costs were used for drug prices.

^bThe mean weight of infliximab patients was assumed to be identical to ADEPT patients mean weight after matching.

^cCost per responder was compared using weighted t-tests.

Discussion

This study utilized a matching-adjusted indirect comparison methodology to assess the comparative effectiveness of adalimumab vs etanercept and adalimumab vs infliximab for treatment of PsA. In the absence of head-to-head clinical trials comparing these treatments, researchers must rely on such indirect comparisons. Common methods for indirect comparison, such as meta-analyses, use aggregate clinical trial data. These methods are subject to important limitations, as differences in study design and baseline patient characteristics can bias the results. To overcome these limitations, this study compared the three treatments while controlling for differences in study design and baseline patient characteristics using the matching-adjusted indirect comparison methodology described in Signorovitch et al.23 With access to individual patient data (IPD), ADEPT patients not meeting the inclusion criteria from the etanercept or infliximab trials could be excluded for that comparison. Patients could additionally be assigned weights such that their re-weighted baseline characteristics exactly matched the mean characteristics reported from the etanercept and infliximab clinical trials.

The methodology used in this study has several advantages over other indirect comparison methods and observational studies. Using individual patient data, a greater number of characteristics can be balanced across trials compared with studies relying on aggregate data alone. Moreover, use of placebo-adjusted measures better addresses unobserved confounders, which could bias observational studies. Finally, the method is considerably less sensitive to an arbitrary effect measure32. However, certain important limitations remain. The indirect comparison methodology cannot replace a well-controlled head-to-head trial, and is still observational in nature. Related to this, despite matching on observables and considering placebo-adjusted outcomes, unobserved differences between trials which potentially affect treatment and placebo arms differently could still bias the results. Finally, feasible adjustments may be limited: (a) patients can be removed from individual patient data to match selection criteria of published studies, but if the criteria are more restrictive in the trial for which IPD are available (ADEPT in this case), the reverse is not possible; and (b) only characteristics reported for each pair-wise comparison can be matched on.

Of the treatment options available for PsA, TNF-α inhibitors have been shown to be the most effective13,14, but only two prior studies have attempted to compare the efficacy of adalimumab, etanercept, and infliximab in the treatment of PsA16,17. The studies found no significant differences in efficacy between the three treatments. While one study purports to have demonstrated greater likelihood of attaining ACR20 when treated with etanercept17, no attempt was made to address differences in study design and baseline characteristics across trials. As noted above, such differences could bias the results. Moreover, the comparison was limited to a single outcome measure (ACR 20), and did not address efficacy in terms of radiographic progression or treatment of skin symptoms.

The results reported above support the general finding that all three TNF-α inhibitors are effective in treating PsA. However, once study design and baseline differences are accounted for across trials, this matching-adjusted indirect comparison has found that adalimumab, etanercept, and infliximab do exhibit differences in efficacy.

Compared with etanercept-treated PsA patients with active psoriasis, adalimumab-treated patients were more likely to experience improvement in their psoriasis skin symptom severity as measured by PASI50, PASI75, and PASI90 at week 24. Patients treated with adalimumab were also more likely to achieve a large improvement (≥70%) in their joint symptoms (ACR70) at weeks 12 (p = 0.055) and 24 (p = 0.002), even though the two patient groups did not have significant differences in terms of lower levels of improvement in joint symptoms (≥20% or ≥50%).

Compared with infliximab-treated patients, adalimumab patients had a greater likelihood of experiencing a large improvement (≥70%) in joint symptoms (ACR70) at week 14. However, that difference was no longer significant at week 24, and no other statistically significant differences in outcomes between adalimumab and infliximab patients were observed. Note that matching the study design of the infliximab trial resulted in the exclusion of 76 ADEPT patients (compared with only 13 in the etanercept comparison), which may have reduced precision of the estimated differences between adalimumab and infliximab.

Patients treated with all three biologics appear equally likely to have at least a moderate improvement in psoriatic arthritis symptoms (ACR20, ACR50). Even though the likelihood of experiencing some improvement is similarly high on any of the three treatments, the greater ACR70 response rate for adalimumab-treated patients compared with etanercept (week 24) and infliximab (week 14) patients suggests that the improvement is more likely to be a large one when treated with adalimumab.

In terms of cost per responder, the comparison of adalimumab and etanercept mirrors the comparative effectiveness findings, as the two treatments are nearly identically priced in both the US and Germany. Treatment with adalimumab has a lower cost per PASI50/75/90 and ACR70 responder at week 24, and suggests a possible advantage in cost per ACR70 responder at week 12. The comparison between adalimumab and infliximab demonstrates that, despite generally comparable matching-adjusted comparative effectiveness, the higher cost of infliximab treatment greatly increases its cost per responder across multiple end-points.

As the indicated dose of infliximab is weight-based, and IPD (including weight) were not available for that trial, the infliximab CPR analysis relies on the assumption that the mean weight of matching-adjusted adalimumab and inflixmab patients is the same. Given that the trials were matched on all the other patient characteristics for which data were available in ADEPT, the authors contend that this is the most appropriate approach. While for lighter weight patients the CPR calculation would be less advantageous to adalimumab, the difference in weight across trial would need to be very large to substantively alter the findings. Moreover, the cost calculations for infliximab assumed administration of fractional units with no waste of remaining medication (e.g., for patients with a mean weight of 90 kg, 4.5 units of 100 mg each were assumed per each infusion). If the remainder of partially infused units cannot be used, this calculation possibly under-states true infliximab costs.

Note that when comparing the adalimumab and infliximab CPR findings, special attention must be given to the specific time frame in question, as it can differentially affect the total cost of treatment used in the calculation. The infliximab regimen indicated for PsA calls for infusions at weeks 0, 2, 6, and every 8 weeks thereafter. This implies that the comparison at week 14 is based on only three infusions, possibly favoring infliximab. On the other hand, the comparison at week 24 is based on five infusions, the last of which was administered at week 22, possibly putting infliximab at a cost disadvantage.

This study was able to overcome some of the challenges of indirect comparisons by incorporating individual patient data, but the study is still subject to important limitations. While all observable baseline characteristics were matched across trial populations, it is possible that there were unobserved differences between the patients in each of the trials that could confound the results. A randomized, head-to-head clinical trial would be the only way to avoid all limitations of an indirect treatment comparison. In addition, the analysis was restricted to short-term follow-up (24 weeks), based on the reported results from the pivotal phase III trials for all treatments. Further research is necessary to establish whether the differences in efficacy observed in this study continue to hold over longer follow-up periods.

As noted above, the study includes CPR calculations for two major markets, one in North America (US), and one in Europe (Germany). The pricing information in these two markets may not be generalizable to other markets. If the relative pricing of the treatments varies across countries, the CPR findings would be affected.

It is important to note that the CPR analysis in not a substitute for a full cost-effectiveness analysis. The CPR focuses on the total expenditure necessary to attain an expected response for a given outcome measure. A full cost-effectiveness model would need to consider many additional factors, including progression, quality-of-life, and rates of adverse events.

Conclusion

After accounting for observable differences in trial populations and in study design, this indirect comparison has found that, compared with etanercept treatment, adalimuab-treated PsA patients have a greater likelihood of achieving a number of clinical end-points, primarily skin-related, and that this advantage is also manifested in a lower cost per responder for these end-points. While the efficacy of adalimumab and infliximab were found to be generally comparable, infliximab’s higher costs result in significantly higher costs per responder across multiple end-points.

Transparency

Declaration of funding

Research support was provided to Analysis Group, Inc. by Abbott Laboratories (now AbbVie Inc.).

Declaration of financial/other relationships

Dr Kirson and Dr Birnbaum are employees of Analysis Group, Inc. Mr Kantor and Mr Wei were employees of Analysis Group, Inc. at the time the analysis was conducted. Dr Rao and Dr Cifaldi are employees of Abbvie Inc. JME Peer Reviewers on this manuscript have no relevant financial relationships to disclose.

Acknowledgments

No assistance in the preparation of this article is to be declared.