Original papers Eur J Health Econom 2004 · 5:64–69 DOI 10.1007/s10198-003-0203-4 Published online: 24.September 2003 © Springer-Verlag 2003
Jörg Ruof1 · Jan L. Hülsemann1 · Thomas Mittendorf1,2 · Silke Handelmann1 Rick Aultman2 · J. Matthias von der Schulenburg2 · Henning Zeidler1 · Sonja Merkesdal1 1 Division of Rheumatology,Hannover Medical School,Hannover,Germany 2 Center for Health Economics,University of Hannover,Hannover,Germany
Comparison of estimated medical costs among patients who are defined as having rheumatoid arthritis using three different standards
A
ccurate estimation of medical care costs raises a host of challenging issues, both practical and methodological [1, 2]. From a methodological point of view the level of aggregation of cost data is a major issue. While aggregate approaches (gross-costing) may be useful for some analyses,there is a clear trend to focus the construction of measures of costs from the components of resource units and their values directly (microcosting [2]). However,the availability and access to reliable and representative data sources to perform such microcosting poses a major challenge. A recent review of cost evaluations in rheumatic conditions cited the heterogeneity among the data sources as a major reason for the enormous variance in reported cost figures. Of 21 reviewed studies 7 relied on patient-derived data, 4 extracted costs from hospital or laboratory records, 8 used administrative sources or published data,and 2 did not specify their source of cost data. None of the studies reported evidence on validity, reliability, and comparability of their cost data source [3]. Additional difficulty arises when patient-derived measures are applied. An analysis of the features of frequently used cost questionnaires in rheumatic conditions showed a considerable difference with respect to major psychometric characteristics: length (from 3 to 113 items),recall period (between 1 week and 1 year), format (interview vs. self-administered),
64 |
Eur J Health Econom 1 · 2004
response categories,cost units (monetary vs. physical), and cost domains covered [4].Using administrative claims data may provide one option for improving the quality of cost estimates.However,despite the unquestionable validity of objective cost data sources major limitations must be considered such as: uncertainty regarding validity of diagnosis codes, poor accessibility of objective cost data sources, and lack of clinical information that match the economic data on a patient-per-patient level [5, 6, 7]. The aim of our study was to address some of the methodological uncertainties regarding the application of administrative claims data. In particular we examined whether pure administrative claims data without clinical validation of diagnosis allow for a reasonable estimate of disease-related costs in rheumatoid Arthritis (RA).
Patients and methods Perspective Following our approach to rely solely on healthcare payer’s cost data we decided to take the payer’s perspective.In particular we took the perspective of the major payers: the Allgemeine Ortskrankenkasse Niedersachsen (AOKN) and the Kassenärztliche Vereinigung Niedersachsen (KVN).While the AOKN is a sickness fund i.e.,a real payer,the physicians’asso-
ciation, KVN, acts primarily as a distributor i.e., they distribute the money that they receive as a lump sum from the AOKN among physicians.
Patient groups Three patient groups were identified and examined: Group A comprised patients who were enrolled in a randomized controlled clinical trial examining the introduction of disease management for RA in the region of Lower Saxony, Germany. Group B consisted of a sample of patients for whom administrative claims data reported the diagnosis of RA [International Classification of Diseases Version 10 codes [8]: M05, (seropositive RA) and/or M06 (seronegative RA)], and who had at least one prescription of a disease-modifying antirheumatic drug (DMARD) over the year under observation. Group C included patients for whom administrative claims data reported the diagnosis RA, but who had no DMARD prescription over the year under observation. Groups B and C were extracted as a single random sample (i.e.,we planned for a sample of about 1,000 randomly selected patients with the diagnosis M05 and/or M06) from the administrative claims data source.The subdivision of this random sample in a
J.Ruof and S.Merkesdal contributed equally to this manuscript
Table 1
Study design and hypothesis
Sociodemographic variables of the three patient groups
Females Age, mean±SD (years) Currently employed Yearly salary (€) per employed patient, mean±SD Retired
Group A (n=338)
Group B (n=303)
Group C (n=685)
258 (76.3%)
227 (74.9%)
515 (75.2%)
59.3±11.7
63.3±12.7
72.7±15.5
96 (28.4%)
53 (17.5%)
16,776±10,081
12,263±10,251
182 (53.8%)
179 (59.1%)
79 (11.5%) 11,670±8,958 445 (65.0%)
RA confirmed (group A), RA likely (group B), and RA possible (group C) Table 2
Annual health care utilization of the three patient groups
Visits to Generalist Rheumatologist Other specialists (RA related) Number of outpatient surgery procedures Number of non-physician service visits Patients receiving (%) DMARDsa Steroids NSAIDs Osteoporosis medication Analgetics Diagnostic procedures and tests Imaging (bones, chest) Laboratory tests Other procedures Devices and aids (no. of prescriptions) Acute and nonacute hospital facilities (without surgery; no. of inpatient days) Hospital facilities (surgery; no. of inpatient days) Transportation (number of trips)
Group A
Group B
Group C
1324 1052 248 7.39 52.9
1979 407 267 4.62 45.2
1608 135 180 5.98 57.1
92.0 75.4 68.6 55.9 31.4
100 71.3 68.6 32.7 34.3
0 26.9 52.7 10.8 32.6
124 6162 136 75 81
99 5695 117 95 83
42 2452 78 82 32
35 208
72 213
24 344
a This criterion was used to discriminate group B and group C RA confirmed (group A), RA likely (group B), and RA possible (group C).Only domains with incurred utilization are listed; all data are given as per 100 patients (NSAIDs nonsteroidal anti-inflammatory drugs,DMARDs disease-modifying antirheumatic drugs)
subset with and a subset without DMARD treatment was performed subsequently after the total study sample was drawn. Clinical data and administrative claims data were available for 338 patients with specialist-confirmed RA (group A).In addition administrative claims data for 988 randomly selected patients with the ICD10 diagnosis code M05 or M06 were received from the AOKN and KVN.Of these patients 303 received treatment with DMARDs (group B) and 685 patients did not (group C). The sociodemographic
data for all three patient groups are displayed in ⊡ Table 1. The clinical parameters at baseline for group A (including two patients in whom no administrative claims data were available) were (mean±SD): disease duration 8.4± 8.4 years,number of swollen joints 5.2±6.1, erythrocyte sedimentation rate 16.5± 15.4 mm/h),215 patients (63.8%) with positive rheumatoid factor, and 181 patients (60.3%) with erosive changes.
For all patients in group A both comprehensive clinical data and a full set of administrative claims data [9] were available. Diagnosis of RA (based on the ACR criteria [10] and clinically confirmed by a rheumatologist) was considered to be correct in this patient group.In contrast no clinical data were available for the patients in samples B and C and we had to rely solely on administrative claims data.Consequently there were considerable uncertainties regarding the diagnosis validity [5, 6]. Based on an extensive discussion among the clinically experienced authors (H.Z.,J.H.,J.R.,S.M.) we came to the decision that in the absence of clinical data the occurrence of DMARD treatment is the most valuable indicator for accuracy of diagnosis while lack of DMARD treatment indicates a more questionable level of diagnosis (other reasons for lack of DMARD treatment such as undertreatment were not in the scope of our analysis).Therefore we considered the diagnosis of RA to be likely in sample B and as possible in sample C.A similar approach has been suggested by von Ferber and Ihle [5]. The core element of our study design was the comparison of medical care costs between the three study samples.The payer’s (health care provider’s) perspective was taken for this analysis. Our hypothesis was that pure administrative claims data (extracted by combining nonvalidated ICD 10 M05 and/or M06 codes in conjunction with a treatment based algorithm i.e., DMARD therapy yes/no) allow for a reasonable estimate of medical care costs in RA. In particular we hypothesized (a) that medical care costs for groups A and B are closely related and (b) that costs in samples A and B are significantly higher than those in sample C.
Data and data sources Data were obtained from a major German statutory health insurance plan,the AOKN and the KVN. The AOKN [11] covers the medical care for 2.3 million members in the region of Lower Saxony, which is one of 16 regional states in Germany,the KVN represents the outpatient physicians (both Eur J Health Econom 1 · 2004
| 65
Abstract Eur J Health Econom 2004 · 5:64–69 DOI 10.1007/s10198-003-0203-4 © Springer-Verlag 2003
Jörg Ruof · Jan L. Hülsemann Thomas Mittendorf · Silke Handelmann Rick Aultman · J.Matthias von der Schulenburg Henning Zeidler · Sonja Merkesdal
Comparison of estimated medical costs among patients who are defined as having rheumatoid arthritis using three different standards Abstract Accurate estimation of medical care costs raises a host of challenging issues.We examined whether pure administrative claims data without clinical validation of diagnosis allow reasonable estimation of disease-related costs in rheumatoid arthritis (RA). Three patient groups were examined: patients with clinically confirmed RA (group A, n=338),patients with likely RA (administrative claims data reported the diagnosis of RA and patients were treated with disease modifying antirheumatic drugs,DMARDs; group B, n=303),and patients with possible RA (same as group B but patients had no DMARD treatment;group C, n=685).The payer’s perspective was taken for this analysis.Only direct costs were included in the analysis.Cost data and data for several covariates were obtained from a major German statutory health insurance plan,the AOK Niedersachsen.A patient-per-patient microcosting approach was performed.A repeated measures,fixed effects model was applied to examine differences between the three study groups.Mean±SEM annual RA-related direct costs were €2,017±183 per patient-year in group A,€1763±192 in group B,and €805±58 in group C.Outpatient (inpatient) costs were €1636 (328) in group A,€1344 (340) in group B,and €546 (136) in group C.DMARD costs were by far the single most important cost driver in groups A and B.The difference in total RA-related direct cost between groups A and B was not significant whereas the differences between groups A and C (group B and C respectively),were significant.Pure administrative claims data allowed for an accurate estimate of disease-related costs in RA patients that received DMARD therapy.However,the high number of patients for whom administrative claims data listed the diagnosis RA,but no DMARD treatment was given poses a challenge for further research. Keywords Rheumatoid arthritis · Cost of illness · Health economics
66 |
Eur J Health Econom 1 · 2004
generalists and specialists) in the region of Lower Saxony.Within the KVN diagnoses are recorded in the ICD-10 diagnostic codes.A random sample of approx. 1,000 KVN patients was targeted with the ICD10 diagnosis codes M05 (seropositive RA) and/or M06 (seronegative RA). A patient-per-patient micro-costing approach [2] was performed. Details of this approach are explained elsewhere [9]. In summary the majority of outpatient data (visits to physicians, nonphysician service utilization, diagnostic/therapeutic procedures and tests) were derived from the KVN database. All other data were obtained from the AOKN (medication, devices and aids, acute surgical and nonsurgical hospital facilities,transportation,home health care services).The data received from AOKN and KVN were matched in a single data base.The matching was performed on a patient-per-patient basis using the anonymous matching codes which were used for the transfer of the cost data from the two health care providers. The data transfer was performed on a quarterly basis and covered the time period of July 2000 to June 2001. The covariates gender, number of visits to medical providers (rheumatologist vs. general practitioner),insurance plan category, pension plan, and yearly income were as well obtained for the analysis.Patients were classified into six age groups (under 40 years,41–50 years,51–60 years, 61–70 years, 71–80 years, and over 80 years), and gross yearly income was classified to five levels. Different types of medical insurance plan coverage (e.g.,selfinsured member, family coverage) were identified and included in the analysis.
Data analysis The analysis was structured according to the recently published matrix of cost domains in RA [9, 12]. The current analysis was focused on RA-related costs only. Non-RA-related costs were not included. Based on a previously applied methodological approach [9,13] we created lists of medications (e.g., DMARDs, steroids), procedures (e.g.,radiography of hand and feet,specific laboratory investigations such as rheumatoid factor, blood sedimentation rate), and types of encounters that
were likely to have been indicated for RA. Physician visits were considered to be RArelated when the respective ICD-10 code included RA (M05 and M06).Transportation costs in conjunction with these visits were considered to be RA-related.All nonRA-related costs were excluded from the analysis. The mechanism to discriminate RA-related from non-RA-related costs was identical in the three study groups. Costs for gastrointestinal medication and hospital visits which might be due to RA (e.g., treatment related side effects,see [9]) were not separately accounted for and not included in the presented costing analysis. Data management was performed on Microsoft Access software (version 8.0). Based on personal preferences of the authors two different software packages were used (SPSS,SAS).While J.R.,T.M.,and S.H. did the initial calculations,the confirmatory analysis was conducted by R.A.Descriptive statistics were used to analyze health care utilization both in monetary and physical units. Patients who incurred no costs (group A n=0; group B n=0; group C n=24) were equally included in the analysis. Inferential statistics were used to test for differences in medical care costs between the three study samples A, B, and C. Therefore administrative claims data of all three study groups were pooled into one databank. Multiple time-points (i.e., data for the four quarters) were accounted separately for. The primary analysis consisted of a diagnosis based, repeated measures,fixed effects model to examine differences between the three study groups [14,15].In the analysis certainty of diagnosis (confirmed, likely, possible), time period (four quarters), age group, gender,income cohort,and type of retirement pension and medical insurance were included as fixed categorical effects (i.e., those factors with known discrete levels). In addition, the number of visits to rheumatologist (RH) and general practitioner (GP) were included as continuous covariates. A test for nonlinearity was assessed with the inclusion of a time-dependent quadratic term. The difference in mean RA direct cost by diagnosis category were determined by Bonferroni’s multiple comparison testing. All cost data including zero costs were included in the analysis. The final model is a nonintercept model
Table 3
RA-related direct costs (€) per patient-year by cost domains in the three patient groups Group A
Group B
Visits to physicians
323±9.3
Outpatient surgery
4±1.6
1±0.7
3±1.5
Nonphysician service visits
2±0.7
2±0.6
2±0.4
Medication DMARDs Steroids NSAIDs Osteoporosis medication Analgetics
949±140.9 723±138.6 47±3.7 84±12.1 73±7.8 22±5.3
772±114.3 568±113.1 48±5.0 78±7.6 45±6.7 34±8.1
90±7.1 0±0 15±1.7 43±3.9 14±2.5 18±3.9
Diagnostic procedures and tests Imaging (bones, chest) Laboratory tests Other procedures
185±5.7 27±1.3 140±4.4 18±1.9
72±4.5 9±1.0 49±2.7 14±2.5
29±1.8 3±0.7 18±0.7 8±1.2
173±35.2 163±54.4 100±52.7 65±27.9 53±10.9 2017±183
229±55.4 85±35.0 190±91.0 65±23.4 79±13.4 1763±192
233±33.7 24±12.8 34±18.8 78±18.8 123±15.4 805±58
Devices and aids Acute hospital facilities (without surgery) Acute hospital facilities (surgery) Nonacute hospital facilities (rehabilitation clinics) Transportation Total
268±10.6
Group C 189±6.4
RA confirmed (group A), RA likely (group B), and RA possible (group C).Only domains with costs >€0 are listed).Values are mean±SEM.Costs for gastrointestinal medication and hospital visits which might be due to RA (e.g., drug side effects) were not included (NSAIDs nonsteroidal anti-inflammatory drugs,DMARDs diseasemodifying antirheumatic drugs) Table 4
Least squares mean difference (per quarter during the period from July 2000 to June 2001) in RA-related direct costs between groups Groups
Mean cost difference (€) ±SEM
Adjusted 95% confidence interval (€)
Adjusted Pa
A vs. B
47.59±75.07
–132.29 to 227.46
1.0000
A vs. C
316.60±68.84
151.66 to 481.55
<0.0001
B vs. C
269.02±53.88
139.89 to 398.14
<0.0001
a Bonferroni’s adjusted α for multiple comparisons
RA confirmed (group A), A likely (group B), and RA possible (group C)
with fixed effects for certainty of diagnosis, patient age groups, and quarter year with number of GP and RH visits as continuous covariates. Only parameters that were significant in explaining variations in costs were included. The fixed effects model applied in the final analysis was: Yi,j,k=αi+βj+(αβ)i,j+γk+(αγ)i,k+η+αiη +λ+εi,j,k where: αi=certainty of diagnosis, βj=patient age group, γk=quarter-year, η=num-
ber of RH visits (continuous covariate), λ=number of RH visits (continuous covariate), εi,j,k=standard error,and i=1,2,3, j=1,... 6, k=1,... 4.
Results ⊡ Table 2 displays annual health care utilization of the three patient groups.To improve comparability all data are reported as per 100 patients. In terms of physician visits group A showed the highest frequency of visits to rheumatologists,while group
B had the highest frequency of visits to general practitioners and other specialists such as gastroenterologists, radiologists,gynecologists,ophthalmologists,and orthopedic surgeons. Groups A and B showed relatively similar patterns with regards to antirheumatic medication.However, the group with confirmed RA and higher frequency of rheumatologist visits had a higher percentage of osteoporosis treatment (55.9% vs. 32.7% in group B). Additional differences between group A and group B related to the annual number of inpatient surgery days with group B accounting for 72 days/100 patients and group A only accounting for 35 days/100 patients. Group C consistently showed lower utilization with respect to physician visits,antirheumatic medication,diagnostic procedures and tests, and acute and nonacute hospital facilities. RA-related direct costs are shown in ⊡ Table 3. The total annual (3rd quarter 2000–2nd quarter 2001) figures were €2017 per patient-year in group A,€1763 in group B, and €805 in group C. Outpatient (inpatient) costs were €1636 (328) in group A,€1344 (340) in group B,and €546 (136) in group C.Nonmedical direct costs were €53, €79, and €123 in groups A, B, and C, respectively. In groups A and B medication costs were by far the most important single cost driver, accounting for 47.1% of total RA-related direct costs in group A and 43.8% in group B. The least squares mean difference in total RA-related direct cost between groups A and B (€ 47 per quarter), was not found to be significant whereas the least squares mean difference in RA direct costs between groups A and C (€317 per quarter), and groups B and C (€269 per quarter), were highly significant (P<0.0001; ⊡ Fig. 1, ⊡ Table 4).
Discussion The two major outcomes of this study are (a) administrative claims data in conjunction with treatment based algorithms (i.e., DMARD therapy yes/no) provide an estimate of disease related medical care costs in RA which is closely related to that of a clinically defined cohort of RA patients, and (b) in 69% (685 of 988) of patients for whom administrative claims data listed Eur J Health Econom 1 · 2004
| 67
Original papers
Fig. 1 ▲ Total RA-related direct costs (€) of the three patient groups: RA confirmed (group A), RA likely (group B), and RA possible (group C) over four quarters (Q1–Q4). Differences between group A and B not significant, differences between groups A and C (B and C respectively) P<0.0001. Increases in costs in groups A and B in quarters 3 and 4 may be due to additional patients receiving treatment with antibiological treatments (etanercept and infliximab [9])
the diagnosis RA no DMARD treatment was given (group C). Medical care costs among those patients were considerably lower than in the other two patients groups. Similar to the majority of RA-related costing studies analyzed in a recent review [3] our group A reflected costs of patients that were included in a clinical trial. The high number of DMARD users within group A indicates that it might not be generalizable to all RA patients. However,as DMARDs are a very specific treatment for RA and the by far most relevant cost driver in RA, it was no surprise that the costs in groups A and B were very similar. No statistically significant difference was detected between these two samples after adjustment for a variety of sociodemographic variables. However, group A showed a higher frequency of rheumatology visits while group B was more likely to visit their generalist. This difference might be due to the selection process of group A which was recruited by consultant rheumatologists while group B reflected a random sample of patients with the respective diagnosis codes and occurrence of DMARD treatment.While differences in the clinical characteristics of the two study samples are possible (in absence of clinical data for group B we were not able to adjust for those) the fact that patients in group A were more likely to consult a rheumatologist and group B was more likely to visit the general practitioner did not result in a difference in medical care costs. This finding is confirmed by Gabriel et al.[16] who reported that RA
68 |
Eur J Health Econom 1 · 2004
care is not more costly when provided primarily by rheumatologists than by generalists. The other important finding of our study is the significant difference in RArelated medical care costs between group A and C (B and C,respectively).Although these results were expected they indicate a major problem when dealing with administrative claims data. Without clinical validation it cannot be clarified what number of these patients (a) are wrongly diagnosed,(b) did not receive appropriate treatment,(c) or are just mild to moderate stages of the disease. Nevertheless dealing with those patients poses a major challenge when estimating disease-related medical care costs. Into this category fell 69% of randomly selected patients with ICD-10 code M05 or M06.From both a payer’s and a societal perspective a better understanding of this cohort is disparately required to be able to target scarce resources in a most rational manner to those patients in whom long-term benefits may be achieved. However, none of the reviewed studies on costs in RA [17] took into account costs of the majority of patients with uncertain diagnosis. Instead they relied only on patients that were diagnosed by experts.From a health economic perspective this is a major shortcoming that might lead to a severe bias in cost estimates. Therefore naturalistic data as presented here taking into account uncertainty in diagnosis are much more likely to match the real costs incurred from both a payer’s and a societal perspective.
Another topic on the future research agenda addresses the identification and extraction of most relevant cost components. To ensure comprehensive collection and presentation of cost data we included in the presented study all domains in which RA-related direct costs were incurred. However, some specific domains such as outpatient surgery and nonphysician service utilization had a rather limited contribution to the total costs.Therefore it remains to be clarified whether a restriction to some distinct cost domains or cost components are sufficient to accurately predict total costs (i.e.,by means of a factor or principal component analysis), and which cost components have the highest power to predict changes in future costs. In conclusion, we found that pure administrative claims data allow accurate estimation of disease-related costs in RA patients who receive DMARD therapy. However,the high number of patients for whom administrative claims data listed the diagnosis RA, but no DMARD treatment was given poses a challenge for further research.
Corresponding author Jörg Ruof Division of Rheumatology, Hannover Medical School, Carl-Neuberg-Strasse 1, 30625 Hannover, Germany e-mail:
[email protected]
Acknowledgements We thank Brigitte Käser and Markus Dehning from the AOKN and Ernst Weinhold from the KVN for their encouraging support in realizing the costing study. The study team gratefully acknowledges Volker Kück from the AOKN for his continuous support with the transfer and management of the data.The study is funded by the research grant C5.1 of the Kompetenznetz Rheumatologie, a grant from the German Ministry of Education and Research.
References 1. Etzioni R, Riley GF, Ramsey SD, Brown, M (2002) Measuring costs.Administrative claims data, clinical trials, and beyond.Med Care 40 [Suppl III]:63–72 2. Luce BR, Manning WG, Siegel JE, Lipscomb J (1996) Estimating costs in cost-effectiveness analysis.In: Gold MR, Russell LB, Siegel JE,Weinstein MC (eds) Cost-effectiveness in health and medicine.Oxford University Press: Oxford 3. Ruof J, Huelsemann JL, Stucki L (1999) Evaluation of costs in rheumatic diseases: a literature review.Curr Opin Rheumatol 11:104–109
4. Ruof J, Merkesdal S, Huelsemann JL, Schoeffski O, Maetzel A, Mau W, Zeidler H (2001) Cost assessment instruments in rheumatology: evaluation of applied instrument characteristics.J Rheumatol 28:662–665 5. Ferber von L, Ihle P (1997) Analyse von Krankenkassendaten zur ambulanten Versorgung von Patienten mit chronischer Polyarthritis.Akt Rheumatol 22:S13–S18 6. Ferber von L, Koester I, Hauner H (1997) Eine Krankheitskostenstudie zum Diabetes mellitus.Gesundheitswesen 59:17–22 7. Newton KM,Wagner EH, Ramsey SD, McCulloch D, Evans R, Sandhu N, Davis C (1999) The use of automated data to identify complications and comorbidities of diabetes: a validation study.J Clin Epidemiol 52:199–207 8. World Health Organization (1992) International classification of diseases and related health problems, 10th revision, vol 1.WHO: Geneva 9. Ruof J, Huelsemann JL, Mittendorf T, Handelmann S, von der Schulenburg JM, Zeidler H, Merkesdal S (2003) Costs of rheumatoid arthritis in Germany.A micro-costing approach based on objective data sources.Ann Rheum Dis 62:544–550 10. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, et al (1988) The American Association 1987 revised criteria for the classification of rheumatoid arthritis.Arthritis Rheum 31:315–324 11. AOK–Die Gesundheitskasse für Niedersachsen (2001) Geschäftsbericht 2000.Grütter: Hanover 12. Merkesdal S, Ruof J, Huelsemann JL, Schoeffski O, Maetzel A, Mau W, Zeidler H (2001) Development of a matrix of cost domains in economic evaluation of rheumatoid arthritis.J Rheumatol 28:657–661 13. Lanes SF, Lanza LL, Radensky PW,Yood RA, Meenan RF, Walker AM, Dreyer NA (1997) Resource utilization and cost of care for rheumatoid arthritis and osteoarthritis in a managed care setting.Arthritis Rheum 40:1475–1481 14. Cook NR,Ware JH (1983) Design and analysis methods for longitudinal research.Annu Rev Public Health 4:1–23 15. Diggle PJ (1988) An approach to the analysis of repeated measurements.Biometrics 44:959–971 16. Gabriel SE,Wagner JL, Zinsmeister AR, Scott CG, Luthra HS (2001) Is rheumatoid arthritis care more costly when provided by rheumatologists compared with generalists.Arthritis Rheum 44:1504–1514 17. Merkesdal S, Ruof J, Mittendorf T, Mau W, Zeidler H (2002) Health economic research in rheumatoid arthritis.Z Rheumatol 61:21–29
ge i e z n A e t ein h e t s r e i nt H e m e s i t r adve n a s i s i Th
123 85 x 240 mm