Background
Arthritis and other rheumatic conditions are a significant public health issue, and are estimated to affect more than 21% of adults [
1]. Osteoarthritis (OA), the most common form of arthritis, is a major cause of disability [
2,
3] and presents a significant burden to health care providers [
4]. In approximately 10% of the world's population aged 60 years or more, OA-related joint pathology causes significant clinical problems [
5]. Rheumatoid arthritis (RA) is less prevalent but is associated with high medical expenses because of the requirement for continuous treatment to slow disease progression, and a high incidence of joint replacements [
6]. Hospitalization represents a significant component of the costs associated with arthritis [
7,
8].
Anemia, defined by the World Health Organization [
9] as a hemoglobin concentration below 12 g/dl in women and 13 g/dl in men, is common in people with arthritis. Anemia is associated with increased morbidity, length of hospitalization, and cost of care delivery [
10]. In RA, it is estimated that 30-60% of patients are anemic [
11‐
14]. One of the most frequent causes of anemia in RA patients is "iron deficiency anemia," which can result from gastrointestinal (GI) bleeding related to nonsteroidal anti-inflammatory drug (NSAID) use [
15,
16]. "Anemia of chronic disease," which does not usually respond to iron supplementation, is another major cause of anemia in patients with RA [
17,
18]. In a study of 225 patients with RA, anemia of chronic disease accounted for 77% and iron deficiency anemia for 23% of observed anemia [
14]. There are few data on the prevalence of anemia in patients with OA, although the prevalence of both conditions is known to increase with age [
5,
19‐
21]. Sex also appears to influence disease prevalence; OA appears to affect more women than men, while current estimates suggest that women are up to three times more likely to develop RA than men [
22]. Women are also at a greater risk of becoming anemic than men, particularly during menstruation or pregnancy, when iron requirements are increased [
23].
Information on the impact of anemia in arthritic populations is also limited, although there is evidence that anemic RA patients have more severe arthritic disease than nonanemic patients [
14,
24,
25]. Studies in other populations have demonstrated that the clinical impact of anemia is substantial: for example in chronically ill patients, anemia has been associated with an increased risk of mortality and morbidity while also having a negative impact on quality of life [
18]; Anand and colleagues also demonstrated that anemic patients with chronic heart failure have greater disease severity and have a higher risk of hospitalization or death [
26]. Adverse outcomes related to anemia may be of particular importance in the elderly, in whom anemia (and arthritic disease) is common [
19]. Penninx et al. have associated anemia in elderly populations with an increased risk of hospitalization (adjusted relative risk 1.27 [95% confidence interval [CI] 1.12-1.45]) and mortality (relative risk 1.61 [95% CI 1.34-1.93]) [
27], as well as disability, poor physical performance, and decreased muscle strength [
28‐
30]. In community-dwelling older women, even mild anemia and low-normal hemoglobin levels have been identified as independent risk factors for frailty: compared with a hemoglobin concentration of 13.5 g/dl, adjusted odds for frailty of 11.5 g/dl and 12 g/dl were 1.9 (95% CI 1.1-3.4) and 1.5 (95% CI 1.0-2.1) [
31].
Anemia has also been linked to increased health care costs and resource utilization [
32], with direct medical costs for anemic patients with comorbid conditions up to twice those for nonanemic patients with the same comorbid condition [
33,
34]. However, data on anemia-associated resource use and cost in people with arthritis remain very sparse.
The objective of this study was to assess differences in health resource utilization patterns among arthritis (OA and RA) patients with concomitant anemia compared with those without anemia. We analyzed data from French hospital admissions to test the hypothesis that arthritis patients with concomitant anemia are associated with more health care resource use than nonanemic arthritis patients.
Methods
Study design
This retrospective cohort study utilized data on secondary care in France from the Programme de Médicalisation des Systèmes d'Information (PMSI) database. For public and private hospitals, two cohorts were identified from the hospitalizations that occurred during the 2001 calendar year: hospitalizations where there was a primary/secondary diagnosis of arthritis without anemia; and hospitalizations where there was both a primary/secondary diagnosis of arthritis and a primary/secondary diagnosis of anemia. Thus, both cohorts were comprised of arthritis patients, but differed in the presence or absence of anemia diagnosis, respectively. The cohorts were compared for the following measures of hospital resource utilization: length of stay, number of procedures, and mean total cost.
The PMSI database covers more than 90% of private and public hospital activity in France and is used by government and regional health authorities as a tool to provide hospital activity indicators for allocating annual budget and forecasting medical needs and resources. Even though the PMSI database is not exhaustive of French hospitals, it guarantees a standardized collection of data that allows the unbiased identification of cases and controls for epidemiological studies. For each hospital stay, the PMSI database includes information on the patient's age, sex, and postal region; their primary, secondary, and related diagnoses; procedures undertaken; and length of stay. Use of PMSI to assess the epidemiological and economic burden of illness is recommended by the French guidelines for health economic evaluation [
35].
Patients admitted to a hospital were classified by primary diagnosis and then allocated a randomized code to maintain anonymity. Access to patient information thereafter was available only at the hospital admission or hospital stay level. The two study cohorts were identified from the PMSI database using
International Statistical Classification of Diseases and Related Health Problems, 10th Revision (
ICD-10) codes [
36]. Arthritis, defined as RA and/or OA, included seropositive RA (
ICD-10 code M05), other RA (M06), polyarthrosis (M15), coxarthrosis/hip arthrosis (M16), gonarthrosis/knee arthrosis (M17), arthrosis of first carpometacarpal joint (M18), and other arthrosis (M19). A diagnosis of anemia included iron deficiency anemia (D50.0, D50.1, D50.8, D50.9), vitamin B
12 deficiency anemia (D51.0-D51.3, D51.8, D51.9), folate deficiency anemia (D52.0, D52.1, D52.8, D52.9), other nutritional anemias (D53.0-D53.2, D53.8, D53.9), and acquired hemolytic anemias (D59.0-D59.6, D59.8, D59.9). Patients with alpha and beta thalassemia were not included.
Costs
Health care resource utilization for length of stay, number of procedures, and total cost of stay in public and private hospitals was analyzed for both study populations. Costs were assessed from the health care system perspective, and were calculated according to the French Diagnosis-Related Group (DRG) system. This system classifies every hospital patient into one of several hundred DRG groups that are intended to be clinically meaningful and homogenous with respect to resource use. The DRG assignment was recorded on the PMSI dataset for each hospitalization.
Due to differences in the reference costs and financing systems of public and private hospitals in France, reference cost information was not equivalent in the two sectors and, therefore, was not directly comparable. For individual public hospitals in each region, the mean cost per DRG is calculated and expressed in a synthetic index, called the ISA (Index Synthétique d'Activité); the number of ISA points represents an index of hospital productivity. For public hospitals, the average ISA value is, in part, calculated by dividing the "short-term stays" activity budget by the number of ISA points for this activity. However; for private hospitals or clinics, the ISA point value is calculated from the expenses reimbursed by the public health insurance fund (Social Security Sick Funds) to these hospitals. Furthermore, because the financing systems differ, costing for public and private hospitals has to be performed and interpreted separately.
For public hospitals (which represent three quarters of all hospitalized patients), two published lists of reference costs were used to calculate costs for hospital stays for the study populations [
37]. The reference costs, calculated from accountability data from a subgroup of 40 public hospitals, were given per DRG for an average length of stay for each DRG, allowing calculation of unit costs per day for each DRG. Total costs of admissions in each of the two study populations were calculated by summing the product of the length of stay for each admission and the appropriate DRG-related unit cost per day for each admission.
The reference unit costs per hospital stay per DRG comprised the following: medical and paramedical care (salaries of clinicians, nurses, and other medical staff); pharmaceuticals and drugs; anesthesia (including operation suite); laboratory tests and procedures; intensive care (medical staff salaries, pharmaceuticals, amortization, maintenance, and medical logistics); logistics (amortization and medical material maintenance, medical logistics, food service, central laundry, hospital management); structure (amortization of building and facilities and committed fixed costs); and reanimation costs (costs associated with acute care; similar but not equivalent to intensive care in the US). The average total cost per stay in public hospitals was broken down into these categories.
For private hospitals we used the reference cost lists, calculated on all medical fees reimbursed to patients by the Social Security Sick Funds [
38], to calculate total costs per stay. Because of the method used for its development, the list of reference costs for private hospitals did not include assessment of detailed items of costs, such as in the public sector. Only total costs of stay per DRG, DRG-related procedures, and medical care corresponding to an invoice sent to the Social Security Sick Fund for reimbursement, were available. Thus, part of the fixed costs, maintenance, logistics, and salaries are not included in the private total costs per stay.
Statistical analysis
Our primary analysis was to describe the resource utilization of hospitalized patients with arthritis in the two study groups (arthritis with anemia cohort and arthritis-only cohort). The description of resource utilization data was done by their means and 95% CIs. Five confounders of the impact of anemia on resource utilization were identified: sex (male, female), age (0-59, 60-69, 70-79, 80+ years), type of primary diagnosis of arthritis (presence or absence of each of the following diagnoses: M05, M06, M15, M16, M17, M19), type of secondary/associated diagnosis of arthritis (presence or absence of each of the following diagnoses: M05, M06, M15, M16, M17, M19), and number of associated diagnoses/comorbidities (none, 1-2, 3-4, 5-6, 6+).
In order to eliminate the possible effect of these confounders on resource utilization when comparing arthritis with anemia and arthritis-only cohorts, we randomly generated "matched-control" samples from the arthritis without anemia population based on the five confounders. Simple random sampling was performed in the arthritis-only cohort stratified by confounders. The stratum sampling rates were specified as being equal to the actual percentages observed in the arthritis with anemia cohort. Five separate replications of this random process were performed leading to five independent matched-samples. This approach allowed obtaining five control samples (no anemia) comparable to the population with anemia in terms of sociodemographics, diagnostics, and comorbidities. Then the resource utilization data were described and compared between "matched-control" samples and the arthritis with anemia population by their means and 95% CIs.
To further confirm these results, we also conducted multivariable analyses on the entire study sample, including the confounding variables and the variable "anemia yes/no." We described and compared the adjusted resource utilization for the two study groups (arthritis with anemia cohort and arthritis-only cohort) using analysis of covariance. All data processing and analyses were performed using SAS software (Statistical Analysis System, version 8. 2, SAS Institute, Inc., Cary, NC).
Discussion
This retrospective cross-sectional study assessed the 2001 hospital health care utilization and costs related to anemia in arthritis patients in France. The study showed that, in public hospitals, arthritis patients with concomitant anemia use more health care resources than those without anemia. After adjustment for confounders, the length of stay and number of procedures attributable to anemia were greater in public than in private hospitals. Although it was not possible to directly compare the two populations due to the way in which these costs were calculated, the cost of stay per hospitalization in public hospitals was greater in patients with concomitant anemia than in those without anemia, while in private hospitals anemia was associated with a modest decrease in the total cost of stay.
To our knowledge, there are no previous studies looking specifically at the impact of anemia on health care resource use in hospitalized populations in Europe. However, studies assessing overall annual health care utilization have shown that anemia is associated with increased resource use and costs [
32‐
34]. In a US study based on administrative claims data for anemic adults with RA or one of five other comorbid conditions, utilization of key medical services was significantly higher (
P < 0.001) for anemic than nonanemic patients [
34]. In addition, anemic patients had higher care (including inpatient) costs than nonanemic patients with the same comorbid condition, and inpatient and outpatient costs more than double those for nonanemic patients after adjustment for confounders. Another US study identified a difference in direct costs between anemic and nonanemic RA patients of more than $7,000 per year, and showed that anemia also impacts indirect health care costs [
33]. Penninx et al. [
27] examined the relationship of anemia with death and hospitalization outcomes in a community-based sample of older people. They identified a significant association between anemia and subsequent mortality and hospitalization (relative risk 1.61 [95% CI 1.34-1.93] and relative risk 1.27 [95% CI 1.12-1.45], respectively) and, consistent with the results from the French public hospital sector in the current study, found that people with anemia who were hospitalized had a significantly longer length of stay than nonanemic patients (25.0 vs 13.7, respectively,
P < 0.001) [
27].
The use of a national database representative of hospitalizations across France, in conjunction with the large size of the two study populations involved (303,648 hospitalizations, representing 1.6% of all admissions to French hospitals in 2001) adds weight to the findings regarding the impact of anemia in an arthritic population. In addition, the adjustment for key factors that could influence health care resource use, particularly the primary/secondary diagnosis and number of associated diagnoses, helps guard against the possibility that anemia is simply a marker for greater morbidity and severity of underlying disease. Indeed, analysis of the number and types of procedures performed in anemic and nonanemic patients suggests that, in this study, the anemic population has less severe arthritic disease than the nonanemic population.
However, this study also has several limitations. First, as the analyzed data is derived from retrospective analysis of medical claims, the findings do not denote causality, but rather focus on identifying temporal association among patient outcomes. We are thus identifying association, not causality, between anemia and increased resource utilization. Second, care must be taken when using DRGs as these are sometimes re-coded for billing purposes and may, therefore, no longer accurately reflect the actual final diagnosis [
39]. Indeed, even though DRGs are intended to be clinically consistent with respect to resource use, the calculated costs remain an average estimate that does not reflect the heterogeneity among severity of diseases included in the DRG. Third, the prevalence of anemia in the hospitalized arthritis patients (0.9%) seems low compared with previous estimates of 30-60% prevalence of anemia in people with RA [
11‐
14], which may have biased the outcome. However, most patients in this analysis were OA patients, in whom the prevalence of anemia may not be as high. Moreover, our study did not include patients with "anemia of chronic disease," a common type of anemia in people with RA [
18,
17]. Another possible reason for the low percentage of anemic patients in this study is that anemia is often presented and treated in the primary care setting, and thus may have been successfully treated in a proportion of the study population before they were hospitalized. A fourth limitation is that the database identifies "admission" and not "patient," so some patients may have been counted twice (if they had two admissions for their arthritis or anemia during the year). Additionally, the database identifies only admissions for which the anemia or arthritis required a specific treatment of the patient during his or her admission. Arthritic/anemic patients who were admitted to the hospital for a reason other than their arthritis/anemia were not identified. Last, anemia was measured as the hemoglobin threshold and did not take into account its severity and clinical impact.
It is interesting that the results of this study show an anemia-attributable increase in resource utilization and cost in public but not in private hospitals. A small increase was observed in the length of stay of anemic versus nonanemic patients in the private sector, but this did not translate into increased costs, probably because of the less detailed way in which costs are reported for this sector. In private hospitals, because only total costs of stay per DRG (comprising procedures 0 and medical care corresponding to an invoice sent to the Social Security Sick Fund for reimbursement) were available, part of the fixed costs such as, maintenance, logistics, and salaries, were not included in the private total costs per stay 0.
In private hospitals, the anemia-related increase in length of stay was less than half that observed for public hospitals, and there was no difference in the number of procedures undergone by anemic and nonanemic patients. It is possible that the presence of anemia has a greater impact in patients who are generally more unwell than in those who are not as sick. In this study, diagnoses of patients in private hospitals were generally less severe (for example, there were fewer RA patients and the mean number of associated diagnoses was significantly lower) than those in public hospitals, reflecting the different characteristics of the two sectors: public hospitals, which are mostly teaching hospitals, offer more technical and innovative procedures and allow treatment of more severe cases than private centers, which are more sensitive to activity profitability. Furthermore, differences in financial incentives, corresponding to an increased stay in either public or private hospitals, may have also influenced these findings, though this is beyond the current scope of the paper and further studies are warranted.
These study findings are important given that iron deficiency, which is an important cause of anemia in arthritis patients, is partially preventable. Although not proven, the upper GI complications associated with nonselective NSAID use, including ulceration, perforation, and bleeding, could contribute to iron deficiency anemia in arthritis patients using these drugs [
15,
16,
40]. Lower GI events are also an important contributor to safety throughout the entire GI tract [
41‐
43] and may contribute to anemia, although there is no literature in anemic patients per se. Our data showing that anemia increases health care utilization in patients with arthritis suggest that treatment options should be examined carefully with consideration of the complete patient profile.
Acknowledgements
Ruth Diazaraque, Gergana Zlateva, and Liviu Niculescu are all full-time employees of Pfizer Inc. Murial Viala-Danten is a full-time employee of Mapi Values, France and was a paid consultant to Pfizer in connection with this research.
We would also like to thank Florence Baron-Papillon, formerly of Mapi Values, France, for her valuable input in this study.
We would like to thank Leigh Prevost, BSc, of PAREXEL, who provided medical writing services and was funded by Pfizer Inc.
Competing interests
R. Diazaraque, G. Zlateva, and L. Niculescu are full-time employees of Pfizer Inc. M. Viala-Danten is a full-time employee of Mapi Values, France, who were paid consultants to Pfizer in connection with this research. This study was funded by Pfizer Inc.
Authors' contributions
GZ contributed to the data analysis, interpretation and writing of the manuscript; RD contributed to the data interpretation and writing of the manuscript; MVD contributed to the statistical analysis, interpretation and writing of the manuscript; LN contributed to the study design, analysis, interpretation and writing of the manuscript. The manuscript has not been submitted or is not simultaneously being submitted elsewhere, and all authors have read and approved the final version of the manuscript.