Modeling the progression of rheumatoid arthritisA two-country model to estimate costs and consequences of rheumatoid arthritis.код для вставкиСкачать
ARTHRITIS & RHEUMATISM Vol. 46, No. 9, September 2002, pp 2310–2319 DOI 10.1002/art.10471 © 2002, American College of Rheumatology Modeling the Progression of Rheumatoid Arthritis A Two-Country Model to Estimate Costs and Consequences of Rheumatoid Arthritis Gisela Kobelt,1 Linus Jönsson,2 Peter Lindgren,2 Adam Young,3 and Kerstin Eberhardt4 Objective. Two simulation models were developed to analyze the cost-effectiveness of new treatments that affect the progression of rheumatoid arthritis (RA). Methods. We used data from 2 cohorts of patients with early RA who had been followed up since disease onset (up to 15 years). In the Swedish study, 183 patients were followed up for a mean of 11.3 years. In the UK study, 916 patients were followed up for a mean of 7.8 years. Disease progression over 10 years was modeled as annual transitions between disease states, defined by Health Assessment Questionnaire (HAQ) scores. A regression model was used to estimate transition probabilities conditional on age, sex, and time since onset of disease, in order to allow simulation of different patient cohorts. Costs and utilities associated with different HAQ levels were based on data from the cohort studies and cross-sectional surveys. Results. Costs increase and quality of life decreases as RA progresses. In Sweden, total annual costs range from $4,900 to $33,000 per patient, compared with $4,900 to $14,600 in the UK. Cumulative costs over 10 years for patients starting in disease state 1 (HAQ < 0.6) are $54,600 in Sweden and $26,600 in the UK. The cumulative numbers of quality-adjusted life-years (QALYs) are 5.5 and 5.6, respectively. Both costs and QALYs were discounted at 3%. Conclusion. The 2 models, which were based on different patient cohorts, reach a similar conclusion in terms of the effect of RA over 10 years. They appear to accurately capture disease progression and its effects and can therefore be useful in estimating the costeffectiveness of new treatments in RA. The prevalence of rheumatoid arthritis (RA) is estimated at 0.5–1% worldwide (1,2), but the progressive nature of the disease and its onset relatively early in life lead to a considerable social and economic impact (3–5). The costs to society associated with RA are substantial, because the disease may cause not only restricted joint mobility, chronic pain, fatigue, and functional disability, but also psychological distress (6,7). Half of patients will be work-disabled within 10 years after disease onset (8–10), making productivity losses the predominant economic burden of the disease (5,11–13). Direct health care consumption represents about one-fourth of all costs and is dominated by inpatient care. For example, in Sweden, the total annual cost associated with RA in 1999 was estimated at 3 billion Swedish kronor (SEK) ($285 million; $1.00 ⫽ 10.5 SEK), and the average estimated cost per patient was 49,650 SEK ($4,730). Direct costs represent 28%, and, of these, inpatient care represents 45% (14). In the UK, the total annual cost associated with RA in 1992 was estimated at £605 million ($900 million; $1.00 ⫽ 0.67£), and the cost per patient was ⬃£2,600 ($3,880). Direct costs represent ⬃20%, and, of these, inpatient and community day care represent 28% and 22%, respectively (15). In both Sweden and the UK, drugs currently represent a minor cost: 3–4% of total costs and 13–15% of direct costs. However, the introduction of several new treatments of RA will change the distribution of costs associated with this disease. Both the widely used generic 1 Gisela Kobelt, MBA: Health Dynamics International, Speracedes, France; 2Linus Jönsson, MD, Peter Lindgren, MSc: Stockholm Health Economics Consulting AB, Stockholm, Sweden; 3Adam Young, MD: City Hospital, St. Albans, UK (for the Early Rheumatoid Arthritis Study Group); 4Kerstin Eberhardt, MD: Lund University Hospital, Lund, Sweden. Address correspondence and reprint requests to Gisela Kobelt, Health Dynamics International, 492 Chemin des Laurens, 06530 Speracedes, France. E-mail: firstname.lastname@example.org. Submitted for publication December 12, 2001; accepted in revised form May 3, 2002. 2310 MODELING DISEASE PROGRESSION IN RA disease-modifying antirheumatic drugs (DMARDs) and nonsteroidal antiinflammatory drugs (NSAIDs) are being replaced with treatments that are more potent and/or more tolerable but also are more expensive. Therefore, drug budgets will increase. The interesting economic question is whether savings in other resources will offset the added cost of drugs, or whether overall costs in RA will increase. If overall costs increase, the relevant question is whether there is an associated gain in health and whether, from a societal perspective, this gain justifies the additional expenditure. This question immediately poses the problem of how to make predictions of both costs and outcome when no or only limited data on the use of such treatments in clinical practice are available, and any assessment must be based on rather short-term clinical trials in selected patient groups. The health benefits, as well as any potential economic benefit associated with these new treatments, that affect disease progression will, however, be most evident in the longer term, because delaying the development of functional disability will lead to lower levels of resource consumption and maintain the ability of patients to work longer. Several studies have also shown that patients’ quality of life decreases as RA progresses (13,16). As a consequence, slowing disease progression can be expected to maintain quality of life at a higher level for a longer period of time. Thus, a baseline is required that represents disease progression, resource consumption, and the quality of life in patients using current treatments, against which the new treatments can be evaluated within a time frame that exceeds that of the clinical trials. Epidemiologic, clinical, and economic data must be combined, and modeling becomes unavoidable. Economic evaluation implies, in most cases, the use of some type of model, and models are increasingly accepted as a good—and often the only—way to illustrate disease processes and their economic impact and to estimate the impact of changes in treatment strategies. Economic evaluations compare treatment strategies in terms of their costs (resources used) and their effectiveness (health benefits), and results are expressed as the extra cost for each additional unit of health benefit gained with one treatment strategy compared with another (17–19). A disease model that serves as a baseline for analysis of the cost-effectiveness of new treatments of RA used in short-term clinical trials must incorporate good epidemiologic data over a relevant period of time, detailed information about resource consumption by patients at all levels of disease severity, 2311 and an effectiveness measure that is generally available in both epidemiologic studies and clinical trials. The model must also be adaptable to different patient populations, because different clinical trials select patients at different levels of disease severity or at different times since disease onset. We now propose a framework for costeffectiveness analysis of new treatments of RA based on 2 cohort studies in Sweden and the UK. The approach is based partly on our earlier work in this field of study (13) but uses a novel technique for modeling disease progression and covers a longer period of time. PATIENTS AND METHODS Disease progression in chronic diseases is most often illustrated using Markov models (20). Markov models classify patients into a finite number of distinct and mutually exclusive disease states at baseline, usually defined by disease severity. Changes in disease are represented as transitions from one state to another over time. Spending a given amount of time (cycle) in a particular state is associated with a cost and a certain level of quality of life, and the models thus calculate expected costs and outcomes in a given cohort of patients over a defined number of cycles. To make models credible and useful for decisions about allocation of resources, large amounts of different types of data are required. We constructed 2 models that had an identical structure but were based on 2 different inception cohorts, one in Sweden and the other in the UK (21–24). The studies were similar in that in the late 1980s, they enrolled consecutive patients with early RA (according to criteria of the American College of Rheumatology; formerly, the American Rheumatism Association) (25) and conducted followup of these patients for up to 15 years. Epidemiologic data. Table 1 shows the demographic characteristics of the 2 cohorts included in the models, and Figure 1 shows the development of average Health Assessment Questionnaire (HAQ) (26) scores of the 2 patient populations. The data cover 10 years for the Swedish cohort but only 9 years for the UK cohort, because of the small number of patients in the latter study for whom 10-year followup was complete at the time of our analysis. The Lund Study (Sweden). This prospective study of the course and outcome of RA was conducted at the Department of Rheumatology at Lund University Hospital. Criteria for inclusion in the study were definite RA, with duration of symptoms less than 24 months and age 18 years or older. Patients were enrolled irrespective of disease activity and severity. The enrollment period lasted from 1985 until 1989, and most of the patients were referred from primary care units in the surrounding health districts. In total, 183 patients, constituting ⬃50% of all new cases of RA during that period of time, were enrolled. Table 1 shows the demographics and some disease characteristics of these patients at the time of enrollment. Erosive disease was defined as a Larsen score of 2 in at least 1 joint in the hands and/or feet. The patients were treated according to routine clinical practice but had at least 1 special 2312 KOBELT ET AL Table 1. Characteristics of cohorts at the time of inclusion* Female, % Age, years, mean ⫾ SD Duration of symptoms, months, mean ⫾ SD Rheumatoid factor positive, % Functional class, no. of patients I II III IV Erosive disease, % HAQ score, mean ⫾ SD Years of followup, mean ⫾ SD HAQ score at 9/10-year followup, mean ⫾ SD Lund cohort, Sweden (n ⫽ 183) ERAS cohort, UK (n ⫽ 916) 63.4 52.4 ⫾ 12.5 11.1 ⫾ 6.1 75 66.6 54.8 ⫾ 13.6 8.2 ⫾ 6.1 73 16 166 1 0 49 0.93 ⫾ 0.6 11.3 ⫾ 2.7 1.10 ⫾ 0.7 299 540 74 3 24 1.11 ⫾ 0.7 7.8 ⫾ 2.1 1.26 ⫾ 0.9 * Ten-year followup data were used for the Sweden cohort, and 9-year followup data were used for the UK cohort. ERAS ⫽ Early Rheumatoid Arthritis Study; HAQ ⫽ Health Assessment Questionnaire. annual followup assessment using standardized measures at a team care unit. Two- and 5-year followup data have been published previously (21,22). Recently, a 10-year followup was performed, and this data set was used to build the disease model (23). The Early Rheumatoid Arthritis Study (ERAS), UK. ERAS started in 1987 with the primary aims of recruiting patients with early RA prior to their use of second-line drugs and following up these patients for at least 10 years, in order to characterize the disease on the basis of age, sex, disease pattern and severity, and treatment response (9,24). Currently, ERAS includes 1,473 patients; of these, 916 with minimum 5-year followup data have been used for the model. From the late 1980s until the early 1990s, it was normal practice in the Figure 1. Development of mean Health Assessment Questionnaire (HAQ) scores in the epidemiologic cohorts in Sweden (n ⫽ 183) and the UK (n ⫽ 916). HAQ scores improved or remained stable in the first years after diagnosis and entry into the studies, likely illustrating the effect of early treatment, particularly in the UK. The average increase in HAQ score over 10 years is limited, because rheumatoid arthritis (RA) progresses slowly in a majority of patients and possibly because of the use of early intervention, which is now standard practice in RA management. UK to initiate treatment with DMARDs early in the course of RA, using sulfasalazine first, followed by methotrexate for nonresponders or patients who experienced adverse effects. The Markov model. Figure 2 illustrates the structure of the Markov model. Markov states are defined according to functional disability measured with the HAQ (26). Since its inception, the HAQ has been widely used in cohort studies and is included routinely in all clinical studies. It has also been Figure 2. Structure of the Markov model of disease progression. Patients can start in any of the Health Assessment Questionnaire (HAQ)–derived states except “dead.” After each cycle (1 year), the model redistributes patients in the different states, depending on whether their HAQ scores improved, remained stable, or worsened during that cycle, or whether the patient died during the cycle. [⫹] indicates that this redistribution, as illustrated for state 1, takes place at each state every year. MODELING DISEASE PROGRESSION IN RA shown to correlate better than radiologic scores with quality of life (utility) and resource utilization (13), making it the ideal measure to use in economic evaluation. Six disease states were defined previously by a panel of rheumatologists (13). Each state represents 0.5 on the HAQ scale, with cutoffs at 0.6, 1.1, 1.6, 2.1, and 2.6; a state for death was included. The model was developed using DATA Pro (TreeAge Software, Williamstown, MA). Transition probabilities. Disease progression is modeled as transitions between states over time. Transition probabilities between states can vary over time and depend on patient characteristics but not on previous events (transitions). The cycle length in this model is 1 year. Ordered probit regressions were used to estimate probabilities of transition between HAQ-derived states, controlling for the age and sex of the patient and the time since onset of disease. Using this technique, it is possible to adjust the Markov model to match a clinical trial population in terms of age, sex, time since disease onset, and disease severity. Thus, although the model was originally designed for use in patients with early RA, it is possible to use the transitions between HAQ states that occurred in the later years of cohort studies to match a clinical trial in which patients are, for instance, in their seventh or eighth year of disease. Simulations will then use this hypothetical cohort over a defined time frame (e.g., the duration of the clinical trial or longer). In the ordered probit regression, a latent variable, y, determines the current HAQderived state of a patient. This latent variable is explained by the following regression equation: y ⫽ ␤0 ⫹ ␤1 䡠 age ⫹ ␤2 䡠 sex ⫹ ␤3 䡠 distime ⫹ ␥2 䡠 s2 ⫹ ␥3 䡠 s3 ⫹ ␥4 䡠 s4 ⫹ ␥5 䡠 s5 ⫹ ␥6 䡠 s6 ⫹ where ␤0 is a constant, age is the age of the patient at the time of disease onset, distime is the time since onset of disease, s2–s6 are dummy variables for the HAQ state in the preceding period (1 year earlier), and is the (normally distributed) error term. Thus, the model predicts the probability of the patient being in each HAQ state during the current period, conditional upon the HAQ state during the previous period (among other factors). This is equal to the transition probability between different states over 1 year. Mortality. The evidence regarding the effect of RA on life expectancy is conflicting. Several epidemiologic studies have shown increased mortality in patients with severe RA, and the mean standardized mortality ratio has been estimated at 1.87 (27). A link between functional status (HAQ score) and mortality has also been reported (28). However, more recent studies have not demonstrated any effect of RA on mortality (29,30). Because data for estimating an age-specific and sexspecific mortality effect at different HAQ levels are not readily available, the model includes only normal age- and sexadjusted mortality in the population. Outcome measures. Economic evaluation requires that effectiveness be expressed with one measure, in order to estimate the cost per additional unit of measure gained by using an intervention. In chronic progressive diseases such as RA, there is no obvious single outcome measure, because there is a continuous effect on several functions over a long period of time. The quality-adjusted life-year (QALY), which combines quality of life with time, is therefore the measure of 2313 effectiveness that is used most often. In addition, use of the QALY allows comparison of outcomes in different diseases, which is a key requirement if such economic evaluations are to be used to support decisions about resource allocation. QALYs are calculated by adjusting life-years with a quality weight (utility). Utility is defined as the preference patients and/or the general population have for given states of health. It is expressed as a value on a scale between 0 (death) and 1 (full health). Therefore, a year spent at a utility of 0.5 is equal to spending half a year at full health. Utilities are measured using interview-based techniques from decision analysis or preference-based quality of life instruments, such as the EuroQol-5D (EQ-5D) (31–33). The EQ-5D is selfadministered and easy to use, particularly in large populations, and has been validated in a large number of studies, including studies in RA (13,16). Utilities measured with the EQ-5D have also been shown to correlate with HAQ scores (13). We therefore use country-specific average utilities measured with the EQ-5D for each disease state in the models. In a cross-sectional study in Sweden, individuals in a subsample of the Lund cohort (n ⫽ 103) were asked to complete the EQ-5D and the HAQ (13). These utilities have been updated with additional data from the cohort study, and the mean utility and standard deviations for the Markov states were estimated. Similarly, in the UK, EQ-5D and HAQ scores were collected cross-sectionally from a subsample (n ⫽ 107) of the ERAS cohort. In addition, to decrease variability of the data, scores obtained in a large study investigating healthrelated quality of life in patients with RA (n ⫽ 240) were available and were included in the model (16). For the purpose of modeling, the utility scores are estimated for a given HAQ-derived state (controlled for age and sex) rather than for a defined patient population; therefore, combining the 2 data sets does not present any problem. Resource utilization and unit costs. The use of health care resources and the loss of work capacity were estimated based on information from the cohort studies, and unit costs were obtained from hospital accounting and official price lists. In Lund, all utilization data had been collected throughout the study, and complete information on hospitalizations, surgical interventions, outpatient visits, medication, and drug monitoring were available in the database. The ERAS database contained similar data but was less complete in terms of outpatient visits, because in the UK patients were not followed up exclusively at the rheumatology clinic, as was the case in Sweden. Information on medical visits and the use of community services was therefore collected in the same crosssectional survey as that used to collect utility data, based on recall of utilization in the past 1–6 months. All observations for patients in a given state, at any year in the followup period, are used to calculate the average annual cost and standard deviations for each state. One limitation is that in both cohorts, the number of patients with very severe disability (HAQ score ⬎2.6) was rather limited; therefore, the direct costs associated with state 6 must be considered with caution. However, it is interesting to note that in neither country did patients in state 6 undergo surgical interventions, which may partially explain the lower direct costs. Direct costs included were hospitalizations, surgical interventions, ambulatory and community care, and RA med- 2314 KOBELT ET AL ication. NSAIDs were not included, because almost all patients used them, and our earlier work in Sweden showed that use of NSAIDs did not differ significantly between patients in the various states. Also, NSAIDs represented a very low cost, because the more expensive cyclooxygenase-2 (COX-2) inhibitors were not yet available during these followup studies. Also excluded were nonmedical direct costs and costs of informal care, because this information was not included in the data sets. Hospitalization was based on the number of inpatient days in different wards. The cost of surgical interventions (when not included in the hospital day) was based on the type of intervention and its duration multiplied by the cost per minute for use of the operating room. Outpatient care was based on the number of visits to different health care professionals. The cost of RA drugs was calculated from the number of months of usage of each drug, associated with the cost of standard drug-monitoring protocols in place in the rheumatology departments participating in the cohort studies. Indirect costs were calculated as the loss of work capacity of patients in the more advanced disease states compared with that of patients in state 1 (HAQ score ⬍0.6). The rationale behind this approach is that only productivity losses attributable to RA should be included. Hence, instead of absolute work capacity, the relevant measure is the difference in work capacity as disability worsens. In the Swedish cohort, patients in state 1 had the same average work capacity as that of healthy persons with the same age and sex distribution (13). We therefore considered that for patients with HAQ scores ⬍0.6 (state 1), only short-term sick leave represented a loss of work capacity. In economic evaluations, the loss of work capacity is generally calculated using the human capital approach, by which an individual’s productivity is valued at the market price (i.e., at the gross income including employers’ contribution) until retirement. Indirect costs in the models are thus calculated as the number of productive years lost at each Markov state compared with state 1, multiplied by the average gross annual income including employers’ contributions (SEK 327,000 [US $31,145] in Sweden and £17,658 [US $26,355] in the UK). Table 2. Comparison of modeled and actual distribution (%) of patients in the different disease states* Sweden cohort HAQ score ⬍0.6 0.6 ⬍ 1.1 ⬍ 1.6 ⬍ 2.1 ⬍ ⱖ2.6 UK cohort 10-year followup Baseline Actual 1.1 1.6 2.1 2.6 27 33 26 10 4 0 18 30 30 16 4 2 9-year followup Modeled Baseline Actual Modeled 16 30 31 18 4 1 27 25 21 15 8 4 35 16 15 14 11 8 35 22 16 13 9 5 * Only patients who were alive at followup were included. HAQ ⫽ Health Assessment Questionnaire. Table 2, where the cohort distributions in the epidemiologic studies are compared with the distribution resulting from the models, at 10 and 9 years for Sweden and the UK, respectively. Utilities. The average utility scores in the 2 countries (according to disease state) confirmed earlier findings that the EQ-5D is able to discriminate between relatively small differences in disability. As shown in Table 3, utility decreases as functional disability increases. Costs. Table 4 shows mean annual costs for the different disease levels. Costs in all categories of resources increased as patients’ functional disability increased, but these costs were dominated by productivity losses. Indirect costs were much higher in Sweden than in the UK, and the difference was predominantly because far fewer patients remained in the work force in Sweden, as shown in Figure 3. Also, the variability in all cost categories was very high. This is generally the case in data sets of this size, and it is therefore important to use the full range of all costs for the simulations and RESULTS Disease progression. The results of the probit regressions were similar for both cohorts. All coefficient estimates were significant at the 1% level. Longer time since onset of disease and older age at onset were associated with higher rates of disease progression. The progression rates among men were lower than those among women. As expected, the coefficient was higher for patients starting in more severe disease states; i.e., being in a severe state at the beginning of a cycle increased the risk of being in a severe state at the end of a cycle (1 year later). The models accurately reproduced the disease progression in the epidemiologic cohorts, as shown in Table 3. Utility scores according to Health Assessment Questionnaire (HAQ) level* Utility score HAQ score Sweden cohort UK cohort ⬍0.6 0.6 ⬍ 1.1 ⬍ 1.6 ⬍ 2.1 ⬍ ⱖ2.6 0.7274 ⫾ 0.1609 0.6358 ⫾ 0.2298 0.6111 ⫾ 0.2317 0.4223 ⫾ 0.3034 0.2370 ⫾ 0.3639 0.2245 ⫾ 0.0502 0.7459 ⫾ 0.1402 0.6491 ⫾ 0.2053 0.4692 ⫾ 0.2678 0.4419 ⫾ 0.2873 0.2556 ⫾ 0.2908 0.2538 ⫾ 0.3514 1.1 1.6 2.1 2.6 * Utilities are measured on a scale between 0 (death) and 1 (full health) and represent preferences that patients and the general population have for certain health states. Values are the mean ⫾ SD. MODELING DISEASE PROGRESSION IN RA 2315 Table 4. Mean annual direct, indirect, and total costs according to disease severity level, in US dollars* HAQ score ⬍0.6 0.6 ⬍ 1.1 ⬍ 1.6 ⬍ 2.1 ⬍ ⱖ2.6 1.1 1.6 2.1 2.6 Sweden cohort UK cohort Direct Indirect Total Direct Indirect Total 723 1,293 1,924 3,672 3,363 1,782 0 5,997 8,524 15,588 24,838 27,067 723 7,290 10,448 19,260 28,201 28,849 1,228 3,152 2,091 3,087 3,401 2,697 148 2,524 3,474 5,300 8,070 8,407 1,376 5,676 5,565 8,387 11,471 11,104 * $1.00 ⫽ 10.50 Swedish kronor ⫽ £0.67. HAQ ⫽ Health Assessment Questionnaire. calculations of total costs over a given time frame, rather than the mean costs. Modeling results. To illustrate costs and QALYs in the models, we present results for patients in the different disease states at baseline as well as for the 2 epidemiologic cohorts, using the actual sex distribution and mean values for age and disease onset. These results were estimated using Monte Carlo simulations (10,000 individual simulations). For costs and utilities, the actual values observed in the data sets were used. To include uncertainty about transition probabilities, the parameters estimated with the probit regression were varied according to their standard errors, assuming normality. We used the estimated standard errors of the parameters in the regression models, which permits calculation of standard deviations for our estimates. In the UK, Figure 3. Proportion of patients (ages 65 years and younger) at different Health Assessment Questionnaire (HAQ)–derived states who are able to work. Participation in the work force is calculated from the total number of years worked by patients in the different disease states. To estimate the loss of participation attributable to rheumatoid arthritis, state 1 is used as the reference (no impact of rheumatoid arthritis on work, other than short-term sick absence), and the loss of work capacity of patients in the more severe states is compared with that of patients in state 1. costs are generally discounted at 6%, and effects are discounted at 1.5%; however, for comparison purposes, we used the more general discount rate of 3% in both countries. Table 5 shows expected average 10-year costs and utilities, including standard deviations, for patients starting in different disease states at baseline, for both cohorts (costs and QALYs discounted at 3%). The total expected costs for patients starting with an HAQ score ⬍0.6 was SEK 573,448 ($54,614) in Sweden and £17,824 ($26,603) in the UK. The average number of QALYs for the same patients over 10 years was 5.5 in Sweden and 5.6 in the UK. The corresponding expected values for patients starting in state 3 (HAQ score 1.1 ⬍ 1.6) were SEK 962,982 ($91,713) and £26,914 ($40,170) and 4.9 QALYs in both countries. In comparison, a group of healthy people with a similar sex distribution and age would have about 6.1 QALYs per person during the same time period. DISCUSSION Several new treatments for RA have been introduced recently, and more are in late stages of development. These products are effective in reducing symptoms and have been shown in short-term clinical trials to have an effect on disease progression. The cost of these new treatments is higher than that of standard DMARDs, and it is therefore important to assess the cost of treatments in relation to their potential long-term benefits. Such an assessment requires disease models that relate disease progression to costs and extend over a longer period of time than that of the clinical trials, and against which the new treatments can be tested. We previously proposed a structure for such a disease model, using 5-year followup data of a subgroup from the Lund study. The 2 new models presented in the current report are based on the same structure, but the underlying data sets are much larger and cover up to 15 years of followup. Use of these data sets allows simulation of patient cohorts that are more advanced in their disease compared with the patients at baseline in the 2 cohort studies. Using a new approach to calculate the probabilities of transition between different states of functional disability, the models also allow controlling for the effect of sex, age, age at the time of disease onset, and disease severity. Thus, the models can be used to create a hypothetical patient population that precisely matches, for instance, the populations of a clinical trial. Transitions between HAQ states that occurred in the trial can then be directly compared with transitions in 2316 KOBELT ET AL Table 5. Expected costs and QALYs for patients starting at different Health Assessment Questionnaire (HAQ) levels* Sweden UK Expected total cost Expected total cost HAQ score Swedish kronor, mean ⫾ SD US dollars, mean Expected QALYs mean ⫾ SD British pounds, mean ⫾ SD US dollars, mean Expected QALYs, mean ⫾ SD ⬍0.6 0.6 ⬍ 1.1 1.1 ⬍ 1.6 1.6 ⬍ 2.1 2.1 ⬍ 2.6 ⱖ2.6 All scores 573,448 ⫾ 621,458 770,100 ⫾ 802,639 962,982 ⫾ 844,297 1,179,148 ⫾ 833,602 1,335,294 ⫾ 783,365 1,267,971 ⫾ 837,054 828,570 ⫾ 745,683 54,614 73,343 91,713 112,300 127,171 120,759 78,911 5.479 ⫾ 0.666 5.212 ⫾ 0.850 4.932 ⫾ 0.745 4.541 ⫾ 0.931 4.243 ⫾ 0.842 4.369 ⫾ 0.827 5.102 ⫾ 0.749 17,824 ⫾ 16,495 23,921 ⫾ 26,374 26,914 ⫾ 8,470 31,090 ⫾ 28,992 36,011 ⫾ 30,667 39,072 ⫾ 30,598 25,682 ⫾ 23,999 26,603 35,703 40,170 46,403 53,748 58,316 38,331 5.595 ⫾ 0.784 5.24 ⫾ 0.617 4.908 ⫾ 0.493 4.677 ⫾ 0.585 4.256 ⫾ 0.669 4.045 ⫾ 0.590 5.05 ⫾ 0.636 * Values were discounted at 3%. QALY ⫽ quality-adjusted life-year. the cohort studies for the duration of the trial and extrapolated beyond the trial using the epidemiologic data. Two examples of this method, using data from clinical trials with infliximab and leflunomide, have been previously reported (34,35). Simulation models in chronic disease require a number of different types of data that are not usually found in one single data set. Therefore, we included cross-sectional surveys for quality of life as well as epidemiologic data. Combining data from different sources is currently a well-accepted methodology in economic evaluation, provided that the data sets are comparable. In our study, the majority of the crosssectional data came from a subgroup of the cohort study, and including them therefore presents no difficulty. The number of patients included in the crosssectional study of resource utilization in the UK is very small, and direct costs other than DMARD usage must be interpreted with care, considering the variability in resource consumption that is generally observed. However, direct costs in the UK study were quite similar to those in the Swedish study, in which information for the entire group of 183 patients was available at each annual followup assessment. In both cohort studies, patients were enrolled within 2 years of the onset of RA and were followed up annually in a similar manner. It could be argued that the 2 data sets should be combined to increase the number of patients in each disease state. This method would be similar to that of international clinical trials, in which it is generally accepted that clinical effects are similar across countries. However, there are 2 arguments against combining the data sets. First, because 2 similar epidemiologic studies in the same time period are rarely available, it is more interesting to compare the 2 disease models. Also, economic evaluations include clinical practice as one of the drivers of cost, and these evaluations are, therefore, country specific. It is generally difficult to compare costs between countries, because health care systems and both absolute and relative prices differ. However, both countries in our study have a national health service, and the cost of the different resources is based on accounting data rather than on tariffs or billings. It thus appears acceptable to draw some conclusions from the comparison. Second, development of the HAQ states in the 2 groups differed somewhat during the first 2 years of followup, and there may be several reasons for this. The Lund study is population-based and recruited patients from primary care units, irrespective of disease activity and severity. Approximately 50% of all patients in the region who had new RA were thus enrolled. Patients received treatment that was conventional at that time (1985–1989), and DMARDs were prescribed only to patients with active disease. During the first 10 years of the study, the majority of patients received DMARDs, predominantly chloroquine or D-penicillamine. In the later years, methotrexate was used. ERAS is an inception cohort that recruited patients who were referred to 9 rheumatology clinics in 9 different regions in England during the period 1987–1995. Thus, it is likely that a higher proportion of ERAS patients had active disease at the time of enrollment, and a higher proportion than in Sweden received DMARDs, both because of increased disease activity in the ERAS patients and the study objective of early intervention using sequential monotherapy. In both studies, patients switched DMARDs frequently or received no treatment for periods of time. Therefore, integrating the effectiveness or toxicity of individual treatments into the economic analysis is dif- MODELING DISEASE PROGRESSION IN RA ficult, but also was not the objective. Rather, effectiveness and toxicity were captured as a consequence on costs through increased use of health care resources, because management of toxicities and switching to a different drug would lead to additional patient monitoring. It is interesting to note that direct spending in the 2 countries is very similar across the disease states. Differences in total costs are entirely attributable to differences in the loss of work capacity. The number of patients with severe disease who are working is far fewer in Sweden than in the UK, which may be attributable to the very small number of patients in these disease states. An additional reason may be that Sweden is more generous with disability pensions compared with the UK. Last, the average salary in Sweden was found to be higher than that in the UK. Use of NSAIDs was not included in the current analysis, because previous studies showed that such use did not differ significantly between different levels of disease severity, and costs were minimal. Clearly, this will change in coming years as use of COX-2 inhibitors increases in this patient population. However, for economic models designed to capture the effect of disease progression, such as the model presented here, the relevant parameter is the difference in costs between the disease states, not the absolute costs within a disease state. It is through this difference that the effect of delaying progression to more severe disease (and therefore higher costs) is captured. A similar argument is made for indirect costs, for which patients at the lowest level of disability serve as reference. This approach allows incorporation of indirect costs due only to the worsening of disease rather than to other factors. We used the QALY as the effectiveness measure in the model. The concept of the QALY has long been controversial, because it is based on several assumptions. First, it assumes that the QALY is able to capture the concept of aggregate health improvement, thus implying that more QALYs are always better. This assumption appears to be reasonable for the vast majority of health care programs, in which the objective is to increase the quantity and/or quality of individuals’ lives. The assumption underlying economic evaluation is, therefore, that QALYs should be maximized within the available resources, and reimbursement agencies in several countries, including the UK and Canada, are using the QALY when making decisions about resource allocation. The appeal of the QALY is its ability to capture health improvements in terms of both life expectancy and quality of life, which also implies that the 2 sources of health benefits are considered equal. This may not be 2317 the case for all health care programs, and society may prefer to give priority to certain groups of patients. However, the use of QALYs in economic evaluations does not preclude the ability to assign priority, because cost-effectiveness analyses only provide data; decisionmakers can then set their own priorities. A second assumption behind the QALY is that utility weights (i.e., scores on a scale between 0 and 1) express overall quality of life and can be measured accurately. Clearly, the difficulty is less with the concept than with the methods used to assess utility. The EQ-5D has been shown in a very large number of studies in all types of diseases to be a valid instrument, and its use is therefore well accepted. In the Swedish study (13), we previously showed that in RA in particular, the EQ-5D can discriminate between patients within the same functional class who have different HAQ scores, and data from the UK in the current study provide further proof of the sensitivity of this instrument. The utility values in the 2 countries are very similar, confirming the correlation between utilities and HAQ-derived states, irrespective of the cohort of patients. For the purpose of the models, it is this relationship that is important, because utility values are set for each Markov state, and a patient’s utility will thus depend solely on his or her HAQ state. In Sweden, the scores for the severe HAQ states are based on a very small number of patients and should be considered with caution. However, in view of the fact that they are virtually identical to the UK scores, we believe that they also accurately reflect utility values in advanced disease states. An additional question is whether the HAQ is an appropriate measure for assessing disease progression. Several studies have shown that disability begins to develop early in RA (36), increases over time (37,38), but is variable among individuals (39). Initial disability has been shown to be a very strong predictor of final disability (40–43), but it has also been demonstrated that using the HAQ to predict the prognosis in individual patients is difficult (39). However, although this may be an issue in clinical management, it is not important in economic evaluation, in which analysis always involves groups of patients rather than individuals. It has also been suggested that functional disability is influenced by different factors early and late in disease (44): in early disease, disability is mainly caused by disease activity, and in late disease, joint damage becomes more important. This explains the finding in our earlier modeling work that radiologic scores had a poor correlation with 2318 KOBELT ET AL quality of life and argues for use of the HAQ as the relevant disease measure in disease progression models. In economic evaluation of chronic diseases with several different symptoms, the important criteria for the choice of a disease measure are whether it correlates well with an overall measure of quality of life and whether it is able to distinguish between patients at different levels of disease severity. Our data show that the HAQ correlates well with utilities measured with the EQ-5D, and that the difference in utility between the different HAQ groupings in the model is significant. Also, the HAQ has been used widely for several years and is therefore available in both longitudinal followup studies and clinical trials. The ordered probit regressions estimate the probabilities of transition to a different disease state depending on sex, age at the time of disease onset, time since disease onset, and HAQ score for the previous period (1 year earlier). All of these variables were shown to have a significant (at the 1% level) effect on progression. This is consistent with clinical findings that women have a higher risk of disability, disability increases over time, and HAQ scores from year to year are correlated, particularly in the later stages of RA (39). Recently, the use of mean values in costeffectiveness analyses has been questioned in view of the variability of outcomes, particularly costs. Because the data for all parameters are stochastic, it is possible to estimate the uncertainty by using Monte Carlo analysis. This method also allows the estimation of confidence intervals for cost-effectiveness calculations when clinical trial data are incorporated into the model. The accuracy of the cohort distributions at the end of the simulations confirms that the models reflect disease progression as seen in these 2 cohorts. Also, the similarity between the groups in the utility estimates and the overall outcome (QALYs) after 10 years suggests that the models reflect the disease and quality of life in a general way. In contrast, costs are clearly influenced by the health care systems and therefore are very different. Therefore, we conclude that these disease models can be used for cost-effectiveness analyses in countries other than Sweden and the UK, provided that resource utilization is adjusted to reflect local conditions. ACKNOWLEDGMENTS We are grateful to Nigel Hurst (Edinburgh) for making the UK EQ-5D data available, to Simon Dixon (Sheffield) for helping with costing, and to Elisabet Lindquist (Lund) and Cathy Mayes (St. Albans) for support with the cohort data. REFERENCES 1. Silman A, Hochberg MC. Epidemiology of rheumatic diseases. Oxford: Oxford University Press; 1993. 2. Symmons D, Silman A. The epidemiology of rheumatoid arthritis. In: Rheumatoid arthritis: pathogenesis, assessment, outcome and treatment. Pincus T, editor. New York: Marcel Dekker; 1994. p. 131–50. 3. Meenan RF, Yelin EH, Nevitt M, Epstein WV. The impact of chronic disease: a sociomedical profile of rheumatoid arthritis. Arthritis Rheum 1981;24:544–9. 4. Pincus T. The underestimated long term medical and economic consequences of rheumatoid arthritis [review]. Drugs 1995;50 Suppl 1:1–14. 5. Allaire SH, Prashker MJ, Meenan RF. The cost of rheumatoid arthritis [review]. Pharmacoeconomics 1994;6:513–22. 6. Katz PP. The impact of rheumatoid arthritis on life activities. Arthritis Care Res 1995;8:272–8. 7. Pincus T, Callahan LF, Sale WG, Brooks AL, Payne LE, Vaughn WK. Severe functional declines, work disability, and increased mortality in seventy-five rheumatoid arthritis patients studied over nine years. Arthritis Rheum 1984;27:864–72. 8. Fex E, Larsson BM, Nived K, Eberhardt K. Impact of rheumatoid arthritis on work status and social and leisure time activities in patients followed 8 years from onset. J Rheumatol 1998;25:44–50. 9. Young A, Dixey J, Kulinskaya E, Cox N, Davies P, Devlin J, et al. Which patients stop working because of rheumatoid arthritis: results of 5 years’ follow up in 732 patients from the Early RA Study (ERAS). Ann Rheum Dis 2002;61:335–40. 10. Barrett EM, Scott DJG, Wiles NJ, Symmons DP. The impact of rheumatoid arthritis on employment status in the early years of disease: a UK community-based study. Rheumatology (Oxford) 2000;39:1403–9. 11. Stone CE. The lifetime economic cost of rheumatoid arthritis. J Rheumatol 1984;11:819–27. 12. Joensson B, Kaarela K, Kobelt G. Economic consequences of disease progression in rheumatoid arthritis in Finland: a Markov model. Stockholm School of Economics EFI research report no. 6574; 1997. 13. Kobelt G, Eberhardt K, Jönsson L, Jönsson B. Economic consequences of the progression of rheumatoid arthritis in Sweden. Arthritis Rheum 1999;42:347–56. 14. Jonsson D, Husberg M. Socioeconomic costs of rheumatic diseases: implications for technology assessment. Int J Technol Assess Health Care 2000;16:1193–200. 15. McIntosh E. The cost of rheumatoid arthritis. Br J Rheumatol 1996;35:781–90. 16. Hurst NP, Kind P, Ruta D, Hunter M, Stubbings A. Measuring health-related quality of life in rheumatoid arthritis: validity, responsiveness and reliability of EuroQol (EQ-5D). Br J Rheumatol 1997;36:551–9. 17. Johannesson M. Theory and methods for economic evaluation of health care. Boston: Kluwer Academic Publishers; 1996. 18. Gold M, Siegel JE, Russell LB, Weinstein MC. Cost-effectiveness in health and medicine. New York: Oxford University Press; 1996. 19. Drummond M, O’Brien B, Stoddard GL, Torrance GW. Methods for the economic evaluation of health care. Boston: Kluwer Academic Publishers; 1997. 20. Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Med Decis Making 1993;13:322–38. 21. Eberhardt KB, Rydgren LC, Pettersson H, Wollheim FA. Early rheumatoid arthritis: onset, course, and outcome over 2 years. Rheumatol Int 1990;10:135–42. 22. Eberhardt K, Fex E. Clinical course and remission rate in patients with early rheumatoid arthritis: relationship to outcome after 5 years. Br J Rheumatol 1998;37:1324–9. 23. Lindqvist E, Saxne T, Geborek P, Eberhardt K. Ten-year outcome MODELING DISEASE PROGRESSION IN RA 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. in a cohort of early rheumatoid arthritis patients: health status, disease process and damage. Ann Rheum Dis. In press. Young A, Dixey J, Cox N, Davies P, Devlin J, Emery P, et al. How does functional disability in early rheumatoid arthritis (RA) affect patients and their lives: results of 5 years of follow-up in 732 patients from the Early RA Study (ERAS). Rheumatology (Oxford) 2000;39:603–11. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al. The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 1988;31:315–24. Fries JF, Spitz P, Kraines RG, Holman HR. Measurement of patient outcome in arthritis. Arthritis Rheum 1980;23:137–45. Guedes C, Dumont-Fischer D, Leichter-Nakache S, Boissier MC. Mortality in rheumatoid arthritis. Rev Rhum Engl Ed 1999;66: 492–8. Soderlin MK, Nieminen P, Hakala M. Functional status predicts mortality in a community based rheumatoid arthritis population. J Rheumatol 1998;25:1895–9. Lindqvist E, Eberhardt K. Mortality in rheumatoid arthritis patients with disease onset in the 1980s. Ann Rheum Dis 1999;58: 11–4. Kroot EJ, van Leeuwen MA, van Rijswijk MH, Prevoo ML, Van ’t Hof MA, van De Putte LB. No increased mortality in patients with rheumatoid arthritis: up to 10 years of follow up from disease onset. Ann Rheum Dis 2000;59:954–8. The EuroQol Group. EuroQol: a new facility for the measurement of health-related quality of life. Health Policy 1990;16:199–208. Dolan P, Gudex C, Kind P, Williams A. A social tariff for EuroQol: results from a UK general population survey. York: Centre for Health Economics, University of York; 1995. Brooks R. EuroQol: the current state of play [review]. Health Policy 1996;37:53–72. 2319 34. Kobelt G, Lindgren P, Young A. Costs and effects of leflunomide in the UK [abstract]. Arthritis Rheum 2001;44 Suppl 9:S311. 35. Kobelt G, Jonsson L, Lindgren P, Eberhardt K, Young A. Economic consequences of the progression of rheumatoid arthritis in Sweden and the United Kingdom. Arthritis Rheum 2001;44 Suppl 9:S312. 36. Wolfe F, Hawley DJ, Cathey MA. Clinical and health status measures over time: prognosis and outcome assessment in rheumatoid arthritis. J Rheumatol 1991;18:1290–7. 37. Leigh JP, Fries JF, Parikh N. Severity of disability and duration of disease in rheumatoid arthritis. J Rheumatol 1992;19:1906–11. 38. Sherrer YS, Bloch DA, Mitchell DM, Young DY, Fries JF. The development of disability in rheumatoid arthritis. Arthritis Rheum 1986;29:494–500. 39. Wiles N, Barrett J, Barrett E, Silman A, Symmons D. Disability in patients with early inflammatory polyarthritis cannot be “tracked” from year to year: an examination of the hypothesis underlying percentile reference charts. J Rheumatol 1999;26:800–4. 40. Wolfe F, Cathey MA. The assessment and prediction of functional disability in rheumatoid arthritis. J Rheumatol 1991;18:1298–306. 41. Harrison BJ, Symmons DP, Brennan P, Bankhead CR, Barrett EM, Scott DG, et al. Inflammatory polyarthritis in the community is not a benign disease: predicting functional disability one year after presentation. J Rheumatol 1996;23:1326–31. 42. Van der Heide A, Jacobs JW, Haanen HC, Bijlsma JW. Is it possible to predict the first year extent of pain and disability for patients with rheumatoid arthritis? J Rheumatol 1995;22:1466–70. 43. Eberhardt K, Fex E. Functional impairment and disability in early rheumatoid arthritis: development over 5 years. J Rheumatol 1995;22:1037–42. 44. Guillemin F, Briancon S, Pourel J. Functional disability in rheumatoid arthritis: two different models in early and established disease. J Rheumatol 1992;19:366–9.