Model decision trees. Risk-stratification tree represents risk stratification (standard testing with and without magnetic resonance imaging [MRI]) and “treat all” arms. Positive test result leads to baseline combination therapy. Treatment tree represents treatment regimens including optimized methotrexate monotherapy, escalated as needed; combination therapy of 2 or more traditional disease-modifying drugs (eg, triple therapy with hydroxychloroquine sulfate, sulfasalazine, and methotrexate); and biologic therapy (with methotrexate). Lack of treatment response (see text) leads to treatment escalation (from tier 1 to 2 and tier 2 to 3). ACR50 indicates American College of Rheumatology criteria for a 50% improvement in disease activity; RA, rheumatoid arthritis.
Threshold analysis for variables to which model output was sensitive in 1-way sensitivity analysis. Each horizontal bar represents the complete range of values for the variables listed on the vertical axis. Shaded areas represent those values that produced favorable (ie, <$100 000 per quality-adjusted life-year gained) incremental cost-effectiveness ratios (in dollars per quality-adjusted life-year gained) for a given strategy. For example, when the probability of remission after 6 months of methotrexate therapy is greater than 23.0%, standard risk stratification only is favored over the other strategies. Arrows indicate base case assumptions. MRI indicates magnetic resonance imaging; RA, rheumatoid arthritis.
Graphic depiction of test performance assumptions under which magnetic resonance imaging (MRI) plus standard risk stratification is the favored strategy (gray volume). Axes display plausible ranges for MRI sensitivity (horizontal) and specificity (depth) and standard risk stratification sensitivity (vertical, top) and specificity (vertical, bottom). Gray volume represents assumptions under which MRI is favored (ie, yields incremental cost-effectiveness ratios <$100 000 per quality-adjusted life-years gained [in dollars per quality-adjusted life-years gained]); the remaining space reflects assumptions under which standard risk stratification only is preferred.
Acceptability curve of the cost-effectiveness of adding magnetic resonance imaging (MRI) to standard risk stratification according to willingness to pay. The vertical axis represents the probability of cost-effectiveness, defined as producing an incremental cost-effectiveness ratio (ICER) below the willingness-to-pay threshold (in dollars per quality-adjusted life-years gained) listed on the horizontal axis for the MRI (solid line) and “treat all” (dotted line) strategies, respectively. For example, at an ICER threshold of $100 000 per quality-adjusted life-year gained, less than 10.0% of simulations yielded ICERs below $100 000 for the “treat all” strategy and less than 20.0% for the MRI strategy compared with standard risk stratification only.
Incremental cost-effectiveness ratio (ICER) scatterplot of adding magnetic resonance imaging (MRI) to standard risk stratification. Incremental effectiveness (in quality-adjusted life-years [QALYs]) and lifetime rheumatoid arthritis–related costs are plotted on the horizontal and vertical axes, respectively. Each dot represents the ICER for 1 simulation. The ellipse represents a 95% confidence ellipse around the ICERs (dots) for the MRI strategy compared with standard risk stratification only. The dashed line represents the commonly cited cost-effectiveness threshold of $100 000 per QALY gained, and every dot above this line exceeds this threshold. Therefore, only the dots located on the lower right of this diagonal dashed line represent cost-effective simulations for MRI compared with standard risk stratification at a threshold of $100 000 per QALY gained. These dots correlate with the less than 20.0% of simulations producing cost-effective ICERs for the MRI strategy in Figure 4. Data presented are for the lifetime analysis.
Suter LG, Fraenkel L, Braithwaite RS. Cost-effectiveness of Adding Magnetic Resonance Imaging to Rheumatoid Arthritis Management. Arch Intern Med. 2011;171(7):657-667. doi:10.1001/archinternmed.2011.115
Copyright 2011 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2011
Early, aggressive treatment of rheumatoid arthritis (RA) improves outcomes but confers increased risk. Risk stratification to target aggressive treatment of high-risk individuals with early RA is considered important to optimize outcomes while minimizing clinical and monetary costs. Some advocate the addition of magnetic resonance imaging (MRI) to standard RA risk stratification with clinical markers for patients early in the disease course. Our objective was to determine the incremental cost-effectiveness of adding MRI to standard risk stratification in early RA.
Using a decision analysis model of standard risk stratification with or without MRI, followed by escalated standard treatment protocols based on treatment response, we estimated 1-year and lifetime quality-adjusted life-years, RA-related costs, and incremental cost-effectiveness ratios (with MRI vs without MRI) for RA patients with fewer than 12 months of disease and no baseline radiographic erosions. Inputs were derived from the published literature. We assumed a societal perspective with 3.0% discounting.
One-year and lifetime incremental cost-effectiveness ratios for adding MRI to standard testing were $204 103 and $167 783 per quality-adjusted life-year gained, respectively. In 1-way sensitivity analyses, model results were insensitive to plausible ranges for every variable except MRI specificity, which published data suggest is below the threshold for MRI cost-effectiveness. In probabilistic sensitivity analyses, most simulations produced lifetime incremental cost-effectiveness ratios in excess of $100 000 per quality-adjusted life-year gained, a commonly cited threshold.
Under plausible clinical conditions, adding MRI is not cost-effective compared with standard risk stratification in early-RA patients.
Affecting 1.5 million Americans during their most productive years, rheumatoid arthritis (RA) has a mean age at onset of between 45 and 55 years, depending on sex and ethnicity, and results in an excess of $19.3 billion in direct and indirect costs each year.1- 5 Early aggressive treatment has been shown to improve outcomes and increase clinical remission6; however, early RA is also more likely to remit spontaneously,7 and aggressive treatment confers clinical and financial costs.8 Therefore, accurate risk stratification of early-RA patients is important to enable initiation of aggressive treatment in those at high risk for developing severe disease while sparing those at reduced risk from unnecessary treatment.
Standard risk stratification tools include clinical, laboratory, and radiographic evidence of disease activity and/or damage. Magnetic resonance imaging (MRI) provides supplementary information to standard risk stratification.9,10 Magnetic resonance imaging identifies bone erosions earlier than conventional radiography9 and can detect bone marrow edema and synovitis, possible erosion precursors.11,12 Therefore, MRI has been proposed as a more sensitive method of risk-stratifying RA patients to optimize treatment.13 However, few studies, to our knowledge, have directly compared it to standard risk-stratification tools. No randomized clinical trials, to our knowledge, have sought to determine the optimal role for MRI in the management of early RA. It would be preferential to use MRI only in scenarios that offer detectable clinical benefit at an acceptable cost.
It is unlikely that a trial to define the optimal risk-stratification approach could be undertaken because too many plausible MRI specifications and RA therapeutics exist to make exhaustive testing feasible, it would be difficult to enroll sufficient early-RA patients, and it would require many years of follow-up to capture clinically significant outcome differences. Decision analytic methods provide a rational, evidence-based, and updatable framework to inform clinical, research, and policy decisions. Decision analysis also enables assessment of giving aggressive treatment to all patients at baseline (ie, no risk stratification), given the benefits of early treatment and reduced sensitivity and specificity of risk stratification in early RA. Our objective was to determine the incremental benefits and costs of adding MRI to standard risk stratification in early RA and to compare each of these strategies to a no-risk stratification strategy (“treat all”).
We created a decision analysis model to examine differential outcomes achieved using standard risk stratification only vs with MRI vs a “treat all” strategy. Ours was a hypothetical population of individuals with recent-onset RA (≤12 months) and no baseline radiographic erosions (eAppendix).
The hypothetical patient population contained individuals at high risk for developing severe disease (ie, those who would develop plain radiographic erosions within 12 months, known as having a poor prognosis) and those at lower risk (ie, those who would not develop radiographic erosions, labeled as having a good prognosis). Risk stratification, using standard tests (ie, rheumatoid factor and/or anti–cyclic citrullinated peptide antibody positivity and disease activity assessment) only or with MRI, was used to discriminate between poor-prognosis and good-prognosis patients (Figure 1; “risk stratification”). On the basis of the results of testing, a treatment regimen was assigned. Treatment regimens in the model consisted of 3 tiers that represent accepted clinical practice in the United States and are consistent with recent practice guidelines14: optimized methotrexate monotherapy, escalated as needed; combination therapy of 2 or more traditional disease-modifying antirheumatic drugs (eg, triple therapy with hydroxychloroquine sulfate, sulfasalazine, and methotrexate)15; and biologic therapy (in combination with methotrexate). In response to initial risk stratification, patients could receive tier 1 (negative test result) or tier 2 (positive test result) therapy. Lack of treatment response led to treatment escalation (from tier 1 to 2 and tier 2 to 3). To explore the value of risk stratification in and of itself, we included the “treat all” strategy, in which we eliminated all risk stratification (ie, all patients received tier 2 therapy at baseline without testing having been performed).
Patients were assessed at 3-month intervals for survival, drug toxic effects, disease activity, and treatment response for the initial 12 months (Figure 1; “treatment”) based on data suggesting that clinical and radiographic progression is evident after 12 months.16 Drug-related adverse events (AEs) were divided into mild, which conferred a small decrement in quality of life but no associated costs or mortality effect, and moderate to severe (eg, infections requiring antibiotic therapy and/or hospitalization), which conferred more substantial but reversible decrements in quality of life, survival, and associated direct and indirect costs. To reflect the response of the physician to severe AEs, treatment was withheld for individuals experiencing moderate to severe AEs during the next 3-month interval, resulting in decreased treatment-related costs and clinical response at the next assessment.
Our model assessed disease activity at 3-month intervals to determine whether the patient had achieved remission. In the base case, remission was defined as a Disease Activity Score of 2.6 or less.17,18 Although it does not represent a complete lack of disease activity, this level of activity would likely prompt continuation of the existing treatment rather than escalation. Alternative definitions were explored in sensitivity analyses. For individuals whose conditions did not achieve remission, we assessed the American College of Rheumatology (ACR) criteria for a 50% improvement in disease activity (ACR50). Individuals whose conditions showed an ACR50 response were maintained with the same treatment, and those whose conditions did not show an ACR50 response were advanced to the next treatment tier.
Total (direct and indirect) RA-related costs and quality-adjusted life-years (QALYs) were tallied at the end of 12 months. These values were then extrapolated to estimate lifetime costs and quality of life using published life tables. Lifetime estimates considered the disease activity level (and associated productivity cost) and treatment assignment at the end of the first year. Poor-prognosis RA conferred an additional cost to allow for the inclusion of greater anticipated direct costs over time. Individuals with no treatment response at the end of 12 months of treatment were assumed to obtain an ACR50 response during the following cycle.
The hypothetical patient population consisted of individuals 45 years old with a new diagnosis of RA per the ACR 1987 consensus criteria of 12 months or less of disease and no evidence of plain radiographic erosions at baseline. We searched the published literature for the following input variables: RA-related costs, mortality, treatment response, and AE rates for each treatment tier, MRI and standard risk stratification sensitivity and specificity, and quality of life. Input assumptions for the base case analysis and ranges used in sensitivity analyses are listed in Table 1, described herein, and detailed in the eAppendix. Wherever plausible equivalent options existed for input assumptions, we selected those most favorable for MRI to evaluate a best-case scenario for MRI. To acknowledge limitations in available input data and ensure that our analysis captured the full spectrum of possible clinical values, we used estimate ranges in excess of published values for sensitivity analyses.
Treatment responses were derived from the published literature and stratified by poor-prognosis vs good-prognosis RA, treatment duration (3 vs 6 or more months of treatment duration), and the need to temporarily withhold treatment after a severe AE. To use the broadest plausible ranges for sensitivity analyses, AE rates were drawn from randomized clinical trials and prospective, observational cohort studies. Mild AE estimates included AEs described as “mild” and “any AE” estimates. Moderate to severe AE estimates included published data reported for “serious AEs” or “serious infections.”
Performance characteristics for standard risk stratification and MRI were derived from the published literature. We defined sensitivity and specificity as the ability of either testing approach to predict the likelihood of radiographic progression at 12 months.
The quality-of-life estimates in the model represented remitted RA, moderate to severe disease activity, and a partially treated state (ie, ACR50 response). To reflect our definition of remission (Disease Activity Score ≤2.6), we assumed a utility score of 0.95 for remitted RA and varied this assumption in sensitivity analyses.
All costs were converted to 2010 US dollars using the Bureau of Labor Statistics' Consumer Price Index for March 2010.81 Costs were assigned from the societal perspective and, where available, drawn from the published literature. Additional cost estimates were derived from Medicare reimbursement data and 2006 Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project data.58
Productivity costs related to RA disability and work losses were derived from the published literature68- 73 using a fractional estimate of the US national mean hourly wage for nonfarm workers and assuming a 40-hour workweek, 50 weeks per year. A broad range of productivity costs were tested in sensitivity analyses, including no RA-related and/or AE-related productivity costs.
We used standard life table methods derived from the Centers for Disease Control and Prevention's 2007 National Vital Statistics Report78 to estimate base mortality rates by age. The mortality increment due to moderate to severe AEs was derived from Healthcare Cost and Utilization Project data for in-hospital deaths related to infections and/or complications of medical treatment.58 Alternative assumptions were considered in sensitivity analyses.
Internal validation was achieved using quantitative and graphic representation of 1-way sensitivity analyses of all input variables across extreme ranges in which the expected outcome was obvious. External validity was demonstrated by comparing the model output to published data; our model estimated the mean unadjusted life expectancy for the populations to be 35.6 years compared with published estimates of 36.9 years.79
In the base analysis, we estimated 12-month and lifetime total RA-related costs and QALYs using the input assumptions in Table 1. The incremental cost-effectiveness ratios (ICERs) (in dollars per QALY gained) were calculated for all strategies compared with standard risk stratification only. The lifetime analysis used a 3.0% discount rate for costs and quality of life. Further details of the analysis are presented in the eAppendix.
One-way sensitivity analyses of all input variables were performed across the broadest range of clinically possible estimates for the 12-month and lifetime models. Additional multiway and threshold analyses (to determine when the favored strategy changed) were performed for key variables identified in 1-way analyses. Running the decision model as a Monte Carlo simulation, we performed a probabilistic sensitivity analysis, which traces the experiences (and thus costs and quality of life) of 10 000 hypothetical individuals through the model, and the values for each input variable were simultaneously varied using predefined distributions. Because the data for this model were limited and to provide the most conservative analysis, we report the results using uniform distributions for each variable (see the eAppendix for additional sensitivity analyses using alternative distributions). In addition, we adjusted our cost related to moderate to severe AEs using a cost-to-charge ratio (eAppendix).
In a hypothetical population of RA patients with no baseline radiographic erosions and disease duration of less than 12 months, estimated 1-year and lifetime ICERs for the MRI strategy compared with the standard risk stratification only were $204 103 and $167 783 per QALY gained, respectively.
One-year and lifetime RA-related costs, QALYs, and ICERs for standard risk stratification only and with MRI and the “treat all” strategy are listed in Table 2. Herein, we report the results of sensitivity analyses for the lifetime analysis; results of the 12-month analyses are provided in the eAppendix.
Model results were insensitive to wide variation in input variables. In 1-way sensitivity analyses (eAppendix), only 5 of 70 variables produced ICERs below the commonly accepted threshold of $100 000 per QALY gained82: the underlying prevalence of poor-prognosis RA in the population, standard testing sensitivity, standard testing specificity, MRI specificity, and the probability of remission after 6 months of methotrexate monotherapy. Threshold values in which the favored strategy (ie, the particular strategy producing ICERs <$100 000 per QALY gained) changed for each of these 5 variables are illustrated in Figure 2. Magnetic resonance imaging was favored over standard risk stratification only and the “treat all” scenario when the probability of remission after 6 months of methotrexate monotherapy was 21% to 23% (<21%, “treat all” favored; >23%, standard risk stratification favored); standard risk stratification sensitivity and specificity were less than 59% and less than 62%, respectively; the specificity of MRI plus standard risk stratification was 72% or higher; or the underlying prevalence of poor-prognosis RA in the population was 61% to 77%. The “treat all” strategy was favored compared with that of adding MRI when the probability of remission after 6 months of methotrexate monotherapy was less than 21% or the prevalence of poor-prognosis RA was greater than 77%. Data19- 25 suggest that the prevalence of poor-prognosis RA ranges from 1% to 34%. Therefore, none of these thresholds represents a clinically probable or consistently achievable value.
The lifetime model output was insensitive to simultaneously varying the definition of remission, occurrence of AEs, costs, and/or quality-of-life assumptions. The MRI strategy produced favorable ICERs when MRI sensitivity approached 100%, even if MRI specificity was lower than that of standard risk stratification only (Figure 3). Further sensitivity analyses are reported in the eAppendix.
Probabilistic sensitivity analysis of lifetime RA-related costs and QALYs yielded a mean ICER estimate of $302 251 (95% interpercentile range, $17 044-$697 006) per QALY gained for MRI compared with standard risk stratification only (Figure 4 and Figure 5), excluding the 11.1% of simulations in which MRI produced lower QALY estimates at a greater cost than standard risk stratification only. Most runs (83% and 79%) yielded ICERs (for adding MRI compared with standard risk stratification only) greater than commonly used willingness-to-pay thresholds (ie, $100 000 and $150 000 per QALY gained, respectively). We explored a wide range of alternative scenarios for our probabilistic sensitivity analyses (eAppendix). One scenario (assuming nonuniform distributions, using a cost-to-charge ratio for moderate to severe AE costs, and inversely correlating sensitivity and specificity) lowered the mean ICER for adding MRI to $22 868 per QALY gained, but 63.3% of runs still produced ICERs greater than $100 000 per QALY gained. For the “treat all” strategy, the lifetime probabilistic sensitivity analysis yielded a mean ICER estimate (compared with standard risk stratification only) of $268 263 (95% interpercentile range, $97 448-$563 513) per QALY gained, and 94.6% of runs yielded ICERs greater than $100 000 per QALY gained.
Using a decision analytic model of early-RA risk assessment, we found that the lifetime ICER for adding MRI to standard prognostic assessments was generally unfavorable, at $167 783 per QALY gained, and offered an incremental gain of fewer than 2 quality-adjusted days. Probabilistic sensitivity analysis suggested that this result was robust. Most simulations in probabilistic sensitivity analyses yielded ICERs for MRI above the commonly cited ICER threshold of $100 000 per QALY gained. Although decision analytic models are not designed to estimate the number needed to treat or harm as primary outcome measures, we can estimate that 37 patients would need to undergo MRI to identify 1 additional poor-prognosis RA patient. However, in performing these 37 MRIs, 5 additional good-prognosis RA patients would be inappropriately treated with baseline combination therapy and 4 additional patients would experience moderate to severe AEs from combination therapy during a 5-year period.
Although thresholds for acceptable health care value are controversial, most contemporary estimates of willingness to pay for health benefits in the United States range from $50 00082 to 3 times the per-person US gross domestic product (approximately $144 000) per QALY gained.83 Our data suggest MRI is unlikely to be a cost-effective addition to standard prognostic assessments in early RA, despite using highly conservative assumptions (that is, those biased in favor of MRI), including assumptions regarding the quality-of-life effect of RA or complications of its treatments. Although the cost-effectiveness of MRI was sensitive to the performance characteristics of MRI and standard risk stratification, our findings suggest that MRI must provide significantly greater sensitivity, and at least equal specificity, as standard testing only to deliver acceptable value in a population of early-RA patients. Although MRI provides greater sensitivity than standard testing, it is unclear how large an incremental increase in specificity it offers, if any. The specificity of MRI and standard risk stratification are reduced in very early RA84 and are likely to be further reduced with office-based extremity MRIs or in the hands of less-experienced readers.
It is noteworthy that our model included the option of not performing risk stratification (ie, the “treat all” strategy). The “treat all” strategy was preferred in comparison with adding MRI to standard risk stratification in the 12-month analysis. In the lifetime analysis, the “treat all” strategy was preferred when MRI sensitivity approached 100% and when the underlying prevalence of poor-prognosis RA in the population was sufficiently high (>77.0%) that most patients would benefit from aggressive treatment. Although such a high prevalence of poor-prognosis RA is unlikely, the incremental risk of serious AEs is small with combination or biologic treatment compared with methotrexate monotherapy, suggesting that overtreatment confers little additional health risk, particularly in the short term. Our finding that the “treat all” strategy was cost-effective in the first year of treatment supports alternative approaches to early-RA treatment, including induction and withdrawal or induction and maintenance strategies, which may offer economic in addition to clinical85 value and deserve further evaluation. In this setting, our work suggests it may be appropriate to shift focus away from risk stratification per se toward optimizing early diagnosis and early-RA treatment.
This analysis has several limitations. Because published data for our input variables were limited, we used some assumptions that relied on expert opinion. However, we chose estimates with strong clinical face validity, and we tested wide ranges of possible values for each input variable. We used data from patients with later-stage disease than our target population; as many as 62.0% of participants in studies included in the model had baseline plain radiographic erosions, likely representing populations with greater disease severity. However, this likely overestimated the favorability of MRI, because using MRI increases the proportion of the population receiving aggressive treatment at baseline due to the high sensitivity of MRI in identifying individuals at risk for progression. We did not consider the consequences of AEs after the first year of treatment and we did not include spontaneous remission in our model because data suggest it occurs in less than 8% of early-RA patients,7 both of which are reasons for possibly overestimating the favorability of MRI. Removing these biases would make the cost-effectiveness of MRI even less favorable. We did not use a cost-to-charge ratio in the calculation of costs related to moderate to severe AEs but found that adjusting for this resulting bias in sensitivity analyses did not qualitatively alter our results except in combination with other modifications. We used clinical rather than radiographic outcomes because we believe these correlate with quality-of-life measures and are in accordance with recommendations for RA clinical trials.86- 88
To our knowledge, ours is the first study to evaluate the cost-effectiveness of MRI for early-RA risk stratification. Other data examining the cost-effectiveness of MRI as a risk-stratification tool have found inconsistent results. A randomized clinical trial89 in early breast cancer demonstrated no reduction in subsequent operation rates in the MRI vs the no-MRI group, but a study90 of knee MRI after acute injury in patients with normal plain radiographs found reduced costs and improved clinical outcomes in the MRI group compared with the no-MRI group. Although the use of MRI in RA in the United States is unknown, an unpublished national survey of rheumatologists demonstrated that more than a third of respondents had used MRI in the treatment of their RA patients within the last year (Marissa Blum, MD, written communication, June 2010).
Our data suggest that adding MRI to standard risk stratification is unlikely to be a cost-effective alternative to standard testing only under commonly found clinical conditions and accepted willingness-to-pay thresholds. Given our findings in combination with the fact that nonradiologist MRI facility ownership is increasing,91 our data support a prudent approach to technology adoption in RA risk stratification. Data clearly defining the clinical benefit of MRI in early-RA treatment are urgently needed.
Correspondence: Lisa G. Suter, MD, Section of Rheumatology, Department of Internal Medicine, Yale University School of Medicine, 300 Cedar St, Room TAC S541, PO Box 208031, New Haven, CT 06520-8031 (email@example.com).
Accepted for Publication: October 11, 2010.
Author Contributions: Drs Suter and Braithwaite had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Suter, Fraenkel, and Braithwaite. Acquisition of data: Suter. Analysis and interpretation of data: Suter and Braithwaite. Drafting of the manuscript: Suter and Fraenkel. Critical revision of the manuscript for important intellectual content: Suter, Fraenkel, and Braithwaite. Statistical analysis: Suter and Braithwaite. Obtained funding: Suter. Administrative, technical, and material support: Suter. Study supervision: Fraenkel and Braithwaite.
Financial Disclosure: None reported.
Funding/Support: Dr Suter is funded by National Institutes of Health (NIH) K23 AR054095-01 Mentored Career Development and Arthritis Foundation Arthritis Investigator Awards and receives consultancy fees from the Yale–New Haven Health Services Corporation/Center for Outcomes Research and Evaluation. Dr Fraenkel receives support from the NIH, the Arthritis Foundation, the American College of Rheumatology Research and Education Foundation, the Donaghue Foundation, and the Department of Veterans Affairs. Dr Braithwaite receives support from the NIH, the Robert Wood Johnson Foundation, and the Agency for Healthcare Research and Quality.
Role of the Sponsors: The funding sources had no role in study design or data interpretation.
Online-OnlyMaterial: The eAppendix is available at http://www.archinternmed.com.