To describe the patterns of care and direct medical costs of polymyalgia rheumatica (PMR) to test the hypothesis that the direct medical costs incurred by patients with PMR are higher than costs incurred by age- and sex-matched population-based controls from the same community.
The study population comprised 193 Olmsted County, Minnesota residents who were first diagnosed with PMR between January 1, 1987 and December 31, 1999. Inclusion criteria were as follows: age ≥50 years; bilateral aching and morning stiffness (lasting ≥30 minutes) persisting for at least 1 month and involving the neck, shoulders, or hip girdle regions; and an erythrocyte sedimentation rate (ESR) ≥40mm/hour. In patients who fulfilled the first 2 criteria, but had a normal ESR, a rapid response to low-dose corticosteroids served as the third criterion. A total of 695 age- and sex-matched subjects without PMR served as control subjects. Billing data from the Olmsted County Healthcare Expenditure and Utilization Database (OCHEUD) were used to provide estimates of nationally representative unit costs in the year 2002 inflation-adjusted dollars. All subjects were followed using the OCHEUD records until December 31, 2002 to assess the total direct medical costs. Generalized quantile regression modeling was used to estimate the effect of PMR on direct medical costs, after adjusting for age, sex, Charlson comorbidity score, number of hospital days, and number of radiographs.
During the first year following diagnosis, subjects with PMR used a substantially higher number of outpatient services and laboratory tests compared with controls, but during the subsequent 4 years, there were no differences between the 2 groups. In age- and sex-adjusted analysis, PMR was associated with a significant incremental cost of $2,233 at the 10th percentile of costs and $27,712 at the 90th percentile of costs. However, further adjustments for comorbidities, number of hospital days, radiographs, and imaging eliminated the incremental cost difference between the subjects with PMR and control subjects. PMR subjects were significantly more likely to have a history of myocardial infarction (odds ratio [OR] 1.78, 95% confidence interval [95% CI] 1.13, 2.82), peripheral vascular diseases (OR 2.21, 95% CI 1.37, 3.60), and cerebrovascular diseases (OR 1.60, 95% CI 1.08, 2.39) compared with the controls.
Incremental direct medical costs associated with the management of PMR can be substantial, especially early in the disease course. These incremental costs appear to originate mainly from comorbid cardiovascular conditions that were shown to be more prevalent among subjects with PMR.
Polymyalgia rheumatica (PMR) is a common, non-fatal illness occurring in middle-aged and older individuals, characterized by aching and morning stiffness in the proximal regions of the extremities and torso, and elevated markers of inflammation (1). The overall age- and sex-adjusted annual incidence per 100,000 population aged ≥50 years in Olmsted County, Minnesota was estimated at 58.7 (95% confidence interval [95% CI] 52.8, 64.7)(2). The management of PMR is challenging because it involves careful and repeated titration of the corticosteroid dose over a period of 1 to 2 years or longer. Therefore, patients with PMR may require a considerable amount of health care services due to the chronic nature of the disease and the difficulties with corticosteroid titration. To date, no studies have examined these factors in detail. Comprehensive cost of illness studies may, in fact, lead to more cost-effective disease management strategies, resulting in substantial savings in resource utilization and costs as well as providing new incentives to develop more standardized and efficient medical interventions. The objective of this study was to describe the patterns of care and direct medical costs of PMR to test whether the direct medical costs incurred by patients with PMR are higher than age- and sex-matched population-based controls from the same community.
PATIENTS AND METHODS
The study was conducted among the population of Olmsted County, Minnesota. This area is relatively isolated from other urban centers, and the majority of medical care is provided to the residents by a small number of providers. All health care providers use a unit medical record system whereby all medical information on each individual is accumulated within a single record. The potential of this data system for population-based studies has been described previously (3–5). This centralized system provides an ideal setting for an investigation of health services utilization and the direct medical costs of care associated with PMR because all aspects of health care services can be ascertained directly from the original medical records.
The study population comprised an inception cohort of all Olmsted County residents first diagnosed with PMR between January 1, 1987 and December 31, 1999 (2, 6, 7). A trained nurse abstractor screened the medical records of individuals who had received 1 or more diagnoses of PMR, giant cell arteritis (GCA), or temporal arteritis. Individuals were included as subjects with PMR if they fulfilled the following 3 criteria: age ≥ 50 years; bilateral aching and morning stiffness (lasting 30 minutes or more) persisting for at least 1 month involving 2 of the following areas: neck or torso, shoulders or proximal regions of the arms, and hips or proximal aspects of the thighs; and an erythrocyte sedimentation rate >40 mm/hour (Westergren method). Patients with suggestive clinical findings who fulfilled the first 2 of the 3 criteria, and who had a prompt response (definite improvement in symptoms within 24 hours) to low dose corticosteroid therapy (20 mg of prednisone per day or less) were also considered to have PMR. Except for GCA, the presence of another disease that could explain the symptoms, such as active rheumatoid arthritis, was considered an exclusion criterion. In cases where the diagnosis was questionable, 3 rheumatologists reviewed all the clinical information and reached consensus on the diagnosis.
As described previously, detailed data were collected on clinical characteristics, all physician encounters, specialty of care, and use of health services (2, 6, 7). Each PMR patient was individually matched with up to 7 age- and sex-matched community-based subjects without PMR (control subjects). Subjects were matched on the basis of age (± 5 years) and sex to ensure balance between the 2 groups. Each control subject was assigned an index date corresponding to the PMR incidence date of their matched subject with PMR. Comorbidities were assessed using the electronically available Medical Index (5). The Medical Index contains all clinical and pathologic diagnoses and surgical procedures recorded by all health care providers in Olmsted County since 1909. It captures and classifies diagnostic information from the patients' complete medical records (covering both inpatient and outpatient care from all local health care providers). The diagnoses assigned at each visit are coded and indexed continuously (5). Comorbidities were assessed for the period prior to PMR incidence date (or the index date for the control subjects) or at any time during the duration of followup (until December 31, 2002), using an electronic adaptation of the Comorbidity Index developed by Charlson et al (8), similar to that used by Deyo et al (9).
Cost data from the Olmsted County Healthcare Expenditure and Utilization Database were used to provide estimates of nationally representative unit costs. As a consequence of the geographic isolation and the small number of providers, >95% of all health care among Olmsted County residents occurs at the Mayo Clinic, Olmsted Medical Group, and their affiliated hospitals. Utilization and billing data systems for the 2 practice groups and the 3 affiliated hospitals have been linked, affording the capacity to uniquely identify individuals across institutions and capture inpatient and outpatient services delivered by these providers to residents of Olmsted County from January 1, 1987 to the present. The potential of this database for cost studies has been described previously (10–12). The database includes line item detail on date, type, frequency, and billed charge for all goods and services provided at each visit. Using a unit costing methodology, a standardized, inflation-adjusted estimate of the cost is assigned to each service and procedure, reflecting the national average cost. Specifically, using a bottom-up costing approach, the resource utilization was grouped according to the Medicare Part A and B classification system. Part A billed charges (i.e., hospital-billed items provided to inpatients, such as room and board, imaging, supplies, etc.) were adjusted by using hospital cost-to-charge ratios and wage indexes. Part B physician services (i.e., examinations and consultations, diagnostic and therapeutic procedures) were valued using Medicare reimbursement rates. The distinction between Medicare Part A and Part B was used solely to distinguish types of services according to the costing method used and does not imply that the database covers only Medicare patients. Costs for services with missing or unrecognized departmental codes in Part A and nonmatching line items in Part B were imputed. Followup lasted until December 31, 2002. Total direct medical costs included all inpatient and outpatient health care costs incurred by all local providers (excluding outpatient prescription drugs and nursing home care costs). Cost data for the control subjects were obtained over the same calendar period as their matched subjects with PMR. Hence, the categories of services included in this database include all hospitalization services (i.e., room charges, radiology, operating room, drugs, laboratory tests, supplies, and all other hospital-billed items), physician services (i.e., consultations, diagnostic and therapeutic procedures), and various other services provided to persons other than hospital inpatients (e.g., laboratory, radiology, physical therapy, etc.), but do not include outpatient prescription drugs, nursing home care, durable medical equipment, and ambulance and other transportation services.
Descriptive statistics were used to summarize the characteristics of the subjects with PMR and the controls. A conditional logistic regression analysis (with PMR status as the dependent variable) was used to assess the association of PMR status with these characteristics. Utilization of selected health care services (i.e., total visits, outpatient visits, laboratory tests) for subjects with at least 1 visit during followup was summarized using descriptive statistics. Outpatient visits included visits to any health care provider, and all of these outpatient visits were included under total visits.
The effect of PMR (versus non-PMR) status on direct medical costs was assessed using generalized quantile regression (13). One of the hallmark characteristics of medical cost data is their marked (positively) skewed distribution. Usual (least squares) regression models are sensitive to outliers present in skewed distributions. In contrast to the usual least-squares regression method for the (conditional) mean of Y as a function of specific covariates (X), which assumes equal variation about the expected value of Y, quantile regression examines specific percentiles of the distribution of Y in relation to the covariates. For example, quantile regression allows the covariates to affect the “tails” of the distribution of costs differently, and thus provides a more comprehensive assessment of the influence of the covariates on different portions of the cost distribution and is less sensitive to outliers. The response variable (Y) in our study was the 5-year total direct medical costs in the year 2002 dollars. The covariates examined in this analysis included demographics, Charlson comorbidity score, number of hospital days, and number of radiographs. The adequacy of the quantile regression models was examined using residual plots to assess the distribution of residuals in relation to the covariates. Although the subjects with PMR and the controls were matched on the basis of age and sex, these covariates were included in the models because they are associated with costs.
Because of the staggered entry into the study, the cost histories were administratively censored for some subjects (i.e., subjects who did not have complete 5-year costs). For example, a subject diagnosed with PMR on November 31, 1999 would only have 3 years (37 months) of cost data at study end (12/31/2002), and would therefore be labeled as administratively censored. A simple weighted quantile regression method was used to account for censoring, including both losses to followup and administrative censoring (14). This approach was implemented using the S-plus function, rq, for the 10th, 25th, 50th, 75th, and 90th percentiles of the cost distribution. Subjects with complete cost histories were weighted in this analysis, corresponding to the probability of not having a censored cost history for the total 5 years of followup. Subjects who died during the followup period were considered to have complete (uncensored) cost histories because subjects who died did not incur any additional medical costs. However, other losses to followup (e.g., those who moved out of Olmsted County) may have incurred medical costs, but these costs were no longer available for our study. The distinction between censoring of followup time and censoring of cost data (15) was the rationale for using the weighting method (14) applied to the quantile regression. The followup time for the control subjects was selected to match the followup time of the matched subjects with PMR.
The study population comprised 193 subjects with PMR and 695 age- and sex-matched controls. Approximately 37.3% of the subjects with PMR had more than 4 control subject matches. The mean age and the percentage of women were similar in both cohorts (mean age 73.6 years, 64.8% women and 72.1 years, 64.0% women, respectively) (Table 1). Of the 193 subjects with PMR, 148 (77%) either had complete followup (i.e., 5 years) or died during the study period, resulting in a median followup of 5.0 years. Among the 45 remaining subjects with PMR, 11 (6%) were lost to followup after a median followup of 3.26 years, and 34 (18%) were administratively censored after a median followup of 2.0 years. Because the followup time of the control subjects was selected to match their corresponding subject with PMR, followup for both groups was the same.
Table 1. Characteristics of the subjects with PMR and matched controls without PMR*
PMR subjects (n = 193)
Controls (n = 695)
OR (95% CI)
Unless otherwise indicated, values are the number (%). PMR = polymyalgia rheumatica; OR = odds ratio; 95% CI = 95% confidence interval; AIDS = acquired immunodeficiency syndrome.
Comorbidities diagnosed either prior to PMR incidence (or index date for the non-PMR subjects) or at any time during the duration of followup.
From a conditional logistic regression model. The final stepwise model found myocardial infarction, cerebrovascular disease, and peripheral vascular disease significant at the 0.05 level, but given these 3 comorbidities in the model, no others were significantly associated.
Subjects with PMR were significantly more likely to have a history of myocardial infarction (odds ratio [OR] 1.78; 95% CI 1.13, 2.82), peripheral vascular diseases (OR 2.21; 95% CI 1.37, 3.60), and cerebrovascular diseases (OR 1.60; 95% CI 1.08, 2.39) compared with the controls. All other comorbidities were similar in both cohorts (Table 1).
Health care utilization
The median (interquartile range [IQR]) number of outpatient and total visit days, and the median number of laboratory tests by years for the subjects with PMR and the controls are shown in Figure 1. During the first year, subjects with PMR had a substantially higher number of outpatient visit days (median 11; IQR 7, 16), higher total visit days (median 14; IQR 8, 20), and higher number of laboratory tests (median 28; IQR 4,12) compared with the control subjects. During the subsequent 4 years, the median number of visit days and laboratory tests remained relatively constant and similar to the controls (Figure 1). The median number of visit days and laboratory tests among the controls were relatively stable throughout the 5 years of followup.
Cost of PMR
The total median (IQR) direct medical costs over the 5 years of followup, for both the subjects with PMR and the controls are shown in Figure 2. For the subjects with PMR, the total yearly direct medical cost of care was highest during the first year of the disease (median $2,814; IQR $1,423, 6,820) and then stayed relatively constant at ∼$1,500/year in subsequent years. For the controls, the total yearly cost of medical care was relatively stable at ∼$1,000/year through the 5 years of followup.
The incremental cost difference between the 2 groups (i.e., the PMR effect) at different cost percentiles was examined further using quantile regression models (Table 2 and Figure 3A). Three models were examined. The first model showed that at the 10th percentile of costs, the incremental cost of care for PMR subjects over the 5-year period following diagnosis was $2,233 (95% CI 1,797, 3,017), after adjusting for age and sex. The PMR cost increment increased steadily towards the upper range of the cost distribution, reaching $14,067 (95% CI 8,589, 23,305) at the 75th percentile, and $27,712 (95% CI 14,355, 80,125) at the 90th percentile of costs. The incremental cost of care for PMR subjects was significantly higher than the controls in all cost percentiles. This model suggests that PMR subjects at the 10th percentile of total costs had incurred an additional $2,233 compared with their age and sex-matched peers, and those at the 90th percentile of costs had incurred an additional $27,712 (Figure 3A). The second model (Figure 3B) included Charlson comorbidity score and the number of hospital days, in addition to age and sex. After adjusting for the number of comorbidities and the number of hospital days during followup, the incremental cost of PMR was substantially reduced and was no longer significant, except at the 10th percentile of costs ($1,256; 95% CI 591, 1,659). Further adjustment for the number of radiographs and imaging in the third model essentially explained the incremental cost difference at all cost percentiles (Figure 3C).
Table 2. Total direct medical costs of care for subjects with polymyalgia rheumatica over a 5-year period according to cost percentiles (inflation-adjusted to 2002 dollars)*
Values are the total costs (95% confidence intervals).
Adjusted for age and sex.
Adjusted for age, sex, Charlson comorbidity score, and number of hospital days.
Adjusted for age, sex, Charlson comorbidity score, number of hospital days, number of radiographs, and imaging.
2,233 (1,797, 3,017)
1,256 (591, 1,659)
376 (−883, 765)
3,156 (1,911, 5,256)
956 (−42, 1,623)
315 (−387, 829)
7,691 (3,151, 11,041)
1,084 (−401, 2,817)
−58 (−1,121, 340)
14,067 (8,589, 23,305)
−468 (−1,562, 1,800)
−1,084 (−2,511, 527)
27,712 (14,355, 80,125)
527 (−2,561, 5,476)
−41 (−1,987, 2,670)
The quantile regression analyses were repeated by considering only the total costs during the first year. The incremental cost difference between the subjects with PMR and the controls persisted at all cost percentiles even after adjusting for age, sex, Charlson comorbidity score, and the number of hospital days.
The patterns of care and the direct medical costs of care for PMR over a 5-year period following diagnosis were examined in this community-based study. Our findings indicate that during the first year following diagnosis, subjects with PMR had a substantially higher number of outpatient services and laboratory tests compared with individuals of the same age and sex from the same community, but thereafter, there were no differences between the 2 groups. Similarly, PMR was associated with significantly higher incremental direct medical costs at all percentiles of total costs, but these incremental costs appeared to result mainly from several comorbid cardiovascular conditions (i.e., myocardial infarction, peripheral and cerebrovascular diseases) that were more prevalent in subjects with PMR. Once these comorbid conditions and their associated health care utilization characteristics (e.g., hospitalizations and imaging) were taken into account, management of subjects with PMR was not more costly than the management of any other person of the same age and sex in the same community. These results emphasize the importance of considering comorbid conditions when examining costs of medical care for PMR and likewise, any other disease of the elderly.
Although the patterns of health care utilization and the direct and indirect medical costs of care have been examined extensively in various other rheumatologic conditions, including rheumatoid arthritis (16, 17) and osteoporosis (18), only one study so far examined the costs of care in PMR (19). In that study of a cohort of 123 subjects with PMR who were referred to a Canadian tertiary referral clinic, the investigators evaluated the utilization and costs of diagnostic investigations ordered by family physicians to make the diagnosis. The accuracy of diagnosis of PMR by family physicians in that study was low, and the inappropriate and unnecessary diagnostic investigations resulted in a significant increase in costs (19).
Our findings indicate that the incremental direct medical costs associated with PMR could be explained by a higher prevalence of cardiovascular comorbidity among subjects with PMR. Subjects with PMR do not appear to have an increased risk of cardiovascular mortality because, as reported by several investigators, long-term survival in these subjects is not significantly different than that reported in the general population (2, 20–25). However, 3 studies reported an increased risk of cardiovascular mortality in GCA, and this was mainly due to ischemic heart disease (26–28). In the first 2 studies, it was concluded that the increased cardiovascular mortality was related to either insufficient control of inflammation or the effects of long-term corticosteroid therapy used in these patients (26, 27). In the third study, an increased mortality was found to be related to the development of dissection of the thoracic aorta in a small proportion of patients, a complication not seen in the present investigation (28). In fact, only 6 of the 193 subjects with PMR had experienced GCA in our study. Otherwise, there are no published studies on cardiovascular morbidity in PMR. Given the growing recognition of the inflammatory etiology of atherosclerosis and heart failure (29, 30), and the increased risk of cardiovascular disease in various systemic inflammatory conditions, such as systemic lupus erythematosus (31) and rheumatoid arthritis (32), there is an urgent need to examine the magnitude and the pathophysiology of cardiovascular diseases in subjects with PMR. Although the etiology of PMR and GCA are unknown, they are considered to be different phases of the same disease characterized by systemic inflammatory activation, and in the case of GCA, chronic vasculitis of large and medium-sized vessels (1, 33).
Our study has several strengths including the use of a large population-based incidence cohort of subjects with PMR, diagnostic accuracy using a standardized systematic approach for case ascertainment, the use of standardized unit costs, and the use of novel statistical analysis techniques to account for the skewed distribution of the cost data to distinguish the effect of PMR on costs. Specifically, the value of each unit of service was adjusted to national cost norms by use of widely accepted valuation techniques. The use of standardized unit costs is important because of the well-recognized discrepancies between billed charges, (which are also directly available in our utilization database), and true resource use, i.e., the “opportunity costs” in health care (34). Therefore, our cost data resources provide an estimated economic cost for each line item in the billing record and allow the aggregation of costs into categories (e.g., laboratory tests and outpatient visits) that are relevant to the management of subjects with PMR in the clinic.
Our results should be interpreted in light of some potential limitations. First, our findings may not be generalizable to nonwhite individuals because the Olmsted County population during the calendar years of the study was >95% white. With the exception of a higher proportion of the working population being employed in the health-care industry, and correspondingly higher education levels, the local population is socioeconomically similar to white Americans (4). Furthermore, according to the 2000 census, ∼25% of the Olmsted County population was of Scandinavian ancestry, where the incidence of PMR is particularly high (35). Second, the services and the utilization patterns in our study represent the clinical practice patterns in Olmsted County and may not be generalizable to other settings. However, the purpose of our study was to compare subjects with and those without PMR in an attempt to demonstrate whether health-care utilization and the direct medical costs are equal in both groups. If, for example, the health-care utilization patterns and the associated cost of care for subjects with PMR in Olmsted County were substantially lower (or higher) than that of subjects with PMR in other settings, then our results could be biased. This is unlikely because differences in patterns of utilization and associated costs would be present for both subjects with and those without PMR, preserving the validity of our analyses. Third, our cost database does not include several important categories of medical costs, including outpatient prescription drugs, nursing home care, durable medical equipment, and ambulance and other transportation services. However, most of these cost items are not expected to contribute substantially to the cost of care for subjects with PMR.
In conclusion, although subjects with PMR utilize a high number of outpatient services and laboratory tests during the first year of their disease, their health care utilization pattern is similar to their age- and sex-matched peers in subsequent years. PMR appears to be associated with significant incremental costs, regardless of the extent of the total costs; however, our results also indicate that the comorbid cardiovascular conditions appear to largely explain the incremental effect of PMR on costs. Future research addressing the cost of medical management of PMR, as with other chronic diseases of the elderly, should take into account the relationship of the comorbid conditions to PMR.
The authors wish to thank Margaret Donohue, RN for performing the data abstraction.