Quality of Care for Patients With Rheumatoid Arthritis FREE

Rheumatoid arthritis is a chronic disabling condition that affects 1% of US adults.¹ It causes substantial morbidity^2- 4 and is associated with a 5- to 15-year reduction in life expectancy.^5- 11 Early death stems not from the arthritis per se, but rather from comorbid diseases that commonly affect the general population.¹² Several factors may contribute to these premature deaths. First, pathophysiologic features of rheumatoid arthritis may accelerate the courses of other diseases.^13- 15 Next, medications used to treat rheumatoid arthritis may cause or exacerbate other diseases.^16- 17 Finally, patients with rheumatoid arthritis may not receive adequate treatment for comorbid diseases, perhaps because the attention of patients and/or physicians is focused on the arthritis. The few studies that have assessed comorbid disease care among patients with specific underlying chronic diseases suggest that the quality of health care for patients with comorbid diseases may be poor.^18- 21 However, no studies have examined how physician specialty affects quality of care across a spectrum of diseases in patients with a single chronic condition, such as rheumatoid arthritis.

We tested the following 3 hypotheses about the quality of care received by patients with rheumatoid arthritis: (1) patients with rheumatoid arthritis receive higher quality care for that disease than for prevention or treatment of comorbid diseases; (2) patients who see a specialist receive higher quality arthritis care, comorbid disease care, and health care maintenance; and (3) patients treated by a combination of specialists and primary care physicians receive the highest quality care.

Our primary data source was administrative data from a national health insurance company. During the study period (1991-1995), the company had an average annual enrollment of 250,000 individuals in either fee-for-service or discounted-fee-for-service plans. Enrollees lived throughout the United States, although the Southeast and Midwest were overrepresented. The administrative data consisted of 3 files. The patient eligibility file included information on patient demographics and health insurance characteristics. The claims file documented all submitted claims for inpatient and outpatient services, including diagnoses, procedures, and physician specialty associated with each claim. The pharmacy file documented all pharmacy claims, including the specific drug prescribed, the quantity, the number of days for which drugs were supplied, and the Drug Enforcement Agency number of the prescribing physician.

Other data sources for the study were the American Medical Association (AMA) Master File and the 1990 US Census file. The AMA Masterfile includes demographic and specialty data (that have been validated elsewhere^22- 23) about all practicing physicians in the United States. The US Census file includes demographic information about the US population including median household income.

The study population consisted of patients 18 years or older with rheumatoid arthritis. Patients were classified with rheumatoid arthritis and entered the study cohort if between 1991 and 1993 they had a minimum of 2 claims, at least 2 months apart, for a physician visit specifying a diagnosis of rheumatoid arthritis (International Classification of Diseases, Ninth Revision [ICD-9] codes 714, 714.0, 714.1, 714.2, 714.4, 714.8, 714.81, 714.89). We required 2 visits to increase the reliability of the diagnosis of rheumatoid arthritis. We required a period between claims to ensure that individuals who were initially evaluated, but subsequently ruled out for rheumatoid arthritis, were not classified as having rheumatoid arthritis. Patients remained in the cohort until the end of the study period or until they disenrolled from the insurance plan, whichever came first.

We obtained the age, sex, ZIP code, and type of health insurance for each subject from the eligibility claims file. Median household income was estimated by matching the patient's ZIP code with ZIP codes in the 1990 US Census file that were linked to median household income. For each patient, we constructed a chronological claims history that included claims for physician visits, procedures, and drugs. We reviewed a 5% sample of these constructed claims histories, and on implicit review they appeared to describe logical histories of health care. We considered health histories to be logical if the drugs prescribed for a patient were consistent with the diagnostic codes for that patient and if the chronological sequence of diagnostic, procedure, and drug claims was consistent with what occurs in clinical practice. For example, patients with claims for cardiac procedures and cardiac medications also had diagnostic codes for cardiac disease, and physician visits with diagnostic codes for acute infectious diseases were followed by prescriptions for antimicrobial agents.

We used process rather than outcome to measure quality because process measures are more sensitive to differences in quality.^24- 25 The process measures we used have links to outcomes that are either directly supported by randomized clinical trial evidence or by a combination of indirect evidence and expert consensus^26- 37 (Table 1). We selected measures to assess quality for arthritis care, comorbid disease care, and health care maintenance (Table 1, Table 2, and Table 3).

We selected diabetes mellitus, heart disease, and gastrointestinal bleeding as the comorbid diseases for 2 reasons. First, individuals with rheumatoid arthritis are at increased risk for these conditions as a consequence of the disease itself or of drugs used to treat it. Second, they were sufficiently common to give us statistical power to distinguish quality of care by physician specialty, and they have process of care that can be assessed using administrative data. Health care maintenance indicators included screening for breast, cervical, and colorectal cancer.

Overview of Our Approach. We developed a system for determining an overall performance score for each process measure, then aggregated these scores to obtain a score for each domain of care. This score represented the percentage of recommended procedures in the domain that an eligible patient received during a person-year. We also analyzed the scores for each process measure and the scores for each domain according to the patterns of specialty care for each patient.

Scores for Individual Measures. Each process measure had a specified period during which the process should be performed in eligible patients. For each individual measure, we determined the number of patients eligible to receive a process and the number of periods in which they were eligible. Next, we determined the number of eligibility periods during which the process was completed. Scores for individual measures were calculated by dividing the number of periods during which the measure was completed by the total number of eligible periods. We used eligibility periods as the unit of analysis because individuals were potentially eligible for each process measure more than once, and certain processes could occur more than once in a year. For example, an individual with diabetes who was enrolled in the study for 3 years would have 3 eligibility periods for the assessment of annual glycosylated hemoglobin. An individual taking methotrexate for 40 weeks during the study would have 5 eligibility periods that could occur within 1 year for the process measure of assessing white blood cell counts every 8 weeks. Patients were eligible for individual process measures only if they were enrolled in the plan for the entire eligibility period of that measure.

Scores for Each Domain. We used scores on individual process measures to develop aggregate scores for each of the 3 domains, using the person-year as the unit of analysis. We developed the aggregate score for each domain in a 2-step process. First, we determined for each person-year the number of eligibility periods during which processes in that domain were performed. We divided these by the total number of eligibility periods for those processes to calculate a performance score for that person-year. For example, if during 1 year a subject was eligible for and received 2 care processes pertaining to diabetes, and was eligible for but did not receive 1 care process pertaining to ischemic heart disease, the performance score for comorbid disease care for that person-year would be 0.67.

We averaged the domain-specific performance scores for each person-year to determine the overall performance score for each domain. These scores describe for each domain the average percentage of the process measures that were performed for eligible populations during each person-year. With our method, patients could contribute more than 1 eligibility period to each person-year and more than 1 person-year to each domain score, which could result in clustering or dependency of the scores within a patient. We adjusted for the effects of possible clustering in the analysis.

Analysis by Pattern of Specialty Care. We also analyzed scores for each individual process measure and for each domain of care by the following mutually exclusive patterns of care based on whether a patient received (1) specialty care without primary care in which a patient had 1 or more contacts with a relevant specialist, but no contacts with primary care physicians; (2) specialty and primary care in which a patient had 1 or more contacts with both relevant specialists and with a primary care physician; (3) primary care without specialty care in which a patient had contacts with a primary care physician, but not with relevant specialists; and (4) neither primary care nor specialty care in which a patient had contacts with neither relevant specialists nor a primary care physician.

We defined relevant specialists according to the specific disease and domain of care being assessed (Table 4). At the disease level, we required contact with any physician from the specialty considered relevant for that disease to classify a patient as having had contact with a relevant specialist. At the domain level, we classified patients in specialist practice patterns if they had contacts with any of the relevant specialists for that domain. For example, a patient with diabetes and myocardial infarction who had contacts with a rheumatologist and cardiologist, but neither a primary care physician nor endocrinologist, was classified as having had contact with a relevant specialist but not with a primary care physician for the arthritis and comorbid disease domains, and as having had contact with neither a relevant specialist nor primary care physician for the health care maintenance domain.

Physician contacts were defined both by physician visits and prescriptions written by physicians. For physician visits, the physician specialty was ascertained from the claims file. For prescriptions, we linked the Drug Enforcement Agency number associated with each prescription to the AMA Master File to determine the self-reported specialty of the prescribing physician.

We assigned each patient to 1 of the 4 patterns of care for each of the diseases studied and for each domain during each person-year of enrollment. Patterns of specialty care were assigned on the basis of person-years because we reasoned that being seen by a given physician could affect a patient's quality of care over some finite period, probably not longer than 1 year. We assigned person-years for each individual consecutively from the time he/she entered the study.

Process measures were assigned to person-years, and hence to pattern of specialty care, based on when in time the eligibility period for the process measure occurred relative to each patient's person-years. Process measures with eligibility periods that were 1 year or less were assigned to the person-year in which the eligibility period ended. For example, a patient with diabetes enrolled in the study for 3 years would be eligible for annual glycosylated hemoglobin testing during each of the 3 years of enrollment. Process measures that spanned multiple years were assigned to the person-year in which the process was completed, if it was completed. If the process was not completed, we assigned the process measure to the last year of the eligibility period. For example, a patient eligible for a Papanicolaou test every 3 years and enrolled in the study for 3 years would have 1 eligibility period for this process measure during the study period. This process measure would be assigned to person-year 1, 2, or 3 if the Papanicolaou test were performed in year 1, 2, or 3, respectively. If the Papanicolaou test were not performed, the process measure would be assigned to the practice pattern of year 3.

We expressed continuous variables as mean (SD). For individual process scores, differences between physician pattern groups were assessed by using logistic regression and a t test. At the domain level, adjusted performance scores that took into account the effects of age, sex, insurance characteristics (discounted vs nondiscounted fee-for-service reimbursement) and median household income were calculated using binomial logistic regression. We adjusted for these factors a priori based on potential correlates with health care quality. We used a t test to assess for differences between physician pattern groups for adjusted scores in each domain. To adjust for possible clustering of outcomes from repeated observations from the same individual, we used empirical SEs (the Huber correction).³⁸

We identified 1355 patients with 4878 person-years of observation that met our inclusion criteria; 1176 were eligible for at least 1 process measure during each of 4598 person-years. The 179 patients who were not eligible had claims for rheumatoid arthritis for less than 1 year of the study period.

As would be expected among individuals with rheumatoid arthritis, 76% of the population was female. Mean (SD) age was 52 (11) years; age ranged from 18 to 113 years. Median (SD) household income for this insured population was $30,000 ($11,000). On average, patients were followed up for 3.9 years.

Two hundred eighty-one patients (21%) had at least 1 comorbid condition; 65 patients (5%) had at least 2. One hundred twenty-seven (9%) had diabetes, 125 (9%) had ischemic heart disease, 40 (3%) had congestive heart failure, and 85 (6%) had serious gastrointestinal tract bleeding. There were 527 patients (38%) who were eligible for mammography, 671 (50%) eligible for Papanicolaou tests, and 222 (16%) eligible for colorectal screening.

With respect to the arthritis care, comorbid disease care, and health care maintenance, 329 patients were eligible to receive care in 1 domain of care, 540 patients were eligible in 2, and 307 patients were eligible in 3.

For the most part, performance scores for individual processes of care pertaining to arthritis care were higher than those for either comorbid diseases or health care maintenance. For example, patients were more likely to receive monitoring laboratory tests when treated with disease-modifying antirheumatic drugs than to have annual glycosylated hemoglobin levels checked if they had diabetes, or mammograms if they were eligible (Table 5; complete performance table available from C.H.M.).

For each of the 39 individual process measures we assessed and for which patients were eligible, and across all 3 domains of care, the average percentage of the measures performed was higher among patients who had contacts with a relevant specialist than among patients who did not. The difference was statistically significant for 19 process measures. For example, 83% of patients with diabetes whose care pattern included an endocrinologist received an annual test of glycosylated hemoglobin; the corresponding number was 20% for patients with diabetes who had no contact with an endocrinologist (P<.05). Likewise, among women eligible for breast cancer screening, 69% with care patterns that included a gynecologist received mammograms compared with 32% for women without contact with a gynecologist (P<.05).

The aggregate scores for each domain (Table 6) are consistent with the individual process scores discussed above. Recommended processes were performed, on average, 62% of the time for arthritis care during each person-year compared with 52% and 42% of the time for comorbid disease care and health care maintenance, respectively (P<.001). Overall, across all domains, 57% of recommended care was performed.

Across the arthritis care, comorbid disease care, and health care maintenance domains, 45% to 67% of the person-years were associated with physician patterns that did not include a relevant specialist. Within each of the domains (Table 6), patients that had relevant specialist contacts had performance scores that were 30% to 187% higher than those who did not (P<.001). There were no differences in the performance scores of individuals who saw both a primary care physician and a relevant specialist and those who saw a relevant specialist but not a primary care physician (P≥.23 for all domains). These results were essentially the same with and without adjustment for age, sex, insurance characteristics, and median household income (Table 6).

In the arthritis and comorbid disease domains, patients who had contacts with a primary care physician but no relevant specialist had performance scores that were slightly higher than those of patients who had contacts with neither, but these differences were not statistically significant (Table 6). In the health care maintenance domain, patients who had contacts with a primary care physician but no relevant specialist had performance scores that were 43% higher than those patients who did not have contact with either (P<.001; Table 6).

We found that quality of care for rheumatoid arthritis patients, as assessed by a number of process measures, fell far short of recommended guidelines and varied as a function of both health care domain and pattern of specialty care.

Although performance on arthritis-specific process measures has not been reported previously, our findings of suboptimal performance on individual process measures in the comorbid disease and health care maintenance domains are consistent with prior studies.^39- 44 For example, the overall performance rates for annual glycosylated hemoglobin and eye examinations among individuals with diabetes reported in this study are 27% and 30%, respectively. Weiner et al³⁹ have reported 16% and 46% performance among Medicare beneficiaries with diabetes on the glycosylated hemoglobin and eye examination measures, respectively. Similarly, the Health Plan Employer Data and Information Set (HEDIS) has reported 41% to 46% rates for annual eye examinations among individuals with diabetes.⁴⁰ Although the breast and cervical cancer screening rates (each 41%) reported in this study are lower than those reported in HEDIS (70% for each),⁴⁰ they are within the 35% to 47% rates reported by other investigators^41- 44 for breast cancer screening and the 15% to 76% rates reported for cervical cancer screening.^42- 44

Of note, prior studies that describe performance have been based on samples of patients with either the specific disease for which the process measures were relevant (eg, a sample of individuals with diabetes as opposed to a sample of individuals with rheumatoid arthritis and diabetes) or represented the general adult population. Our results demonstrate suboptimal quality of care for comorbid diseases and health maintenance among individuals with a specific underlying disease (ie, rheumatoid arthritis) that increases frequency of visits to physicians and thus the opportunity to provide higher quality care in these areas. Furthermore, the poorer overall scores for comorbid conditions and health care maintenance, compared with those for arthritis, suggest that patients with chronic rheumatoid arthritis may receive relatively better health care for their arthritis than for the treatment and prevention of comorbid diseases.

Our study demonstrates that patients with rheumatoid arthritis who had contact with a relevant specialist received substantially higher quality care along all measures of health care quality than did patients who did not. This finding is particularly noteworthy since nearly half of our study population never saw a specialist. Seeing a primary care physician made no significant difference in the quality scores for arthritis and comorbid diseases, but modestly improved performance in the area of health care maintenance.

Our finding about the effects of physician specialty on quality are consistent with numerous studies demonstrating that, in their areas of expertise, specialists have a greater bank of knowledge on which to draw and provide higher quality care, as measured by process and/or outcomes, than do generalists.⁴⁵ This study demonstrates the same differences in performance based on specialty patterns in care of patients with multiple diseases.

Our work has several limitations. First, the performance scores reported for each domain are a function of, and limited by, both the diseases we chose to assess and the specific performance measures assessed. Adding additional measures or conditions could significantly change our conclusions. Furthermore, this study does not assess health care quality as it pertains to individuals with undiagnosed diseases because the process measures we used were only applied to patients who had claims for, and hence had been diagnosed as having these diseases. However, the conditions we assessed are particularly important to individuals with rheumatoid arthritis and the measures we used cover a broad spectrum of medical care. Although we can never measure all the dimensions of quality, we did cover a range of dimensions and those that we did measure are believed to be important to producing good outcomes.

Second, not all of the process measures used can be linked to outcomes through randomized controlled trials. This is particularly true for the arthritis measures. However, the measures that are not directly linked to outcomes have indirect links to outcomes and are believed by experts to be important to producing good outcomes.

Finally, this research is based on administrative data, which raises 2 issues of validity: whether specific diagnoses or procedures that have occurred are coded, and whether recorded diagnostic and procedure codes represent truth.

The positive predictive value of using ICD codes to identify patients with rheumatoid arthritis (ie, the probability that a patient with a code for the disease actually has the disease) has been reported as 57%⁴⁶ and 95%⁴⁷ in 2 different studies. We expect that the positive predictive value of our selection method is much higher than that reported by the first study because that study was based on encounter data rather than claims submitted for reimbursement. Furthermore, that study reported the positive predictive value of a single ICD code for rheumatoid arthritis. We required at least 2 ICD codes for rheumatoid arthritis that were separated in time by at least 1 month. Our selection method probably has a lower positive predictive value than that reported in the second study because while that study was based on claims submitted by rheumatologists, claims in our data set were submitted by physicians of all specialties. Hence, our selection method allowed some patients who do not have rheumatoid arthritis into the study cohort.

The diagnostic codes used to define patients with comorbid diseases in this study all have high specificity (>90%), but only moderate sensitivity.^48- 51 Hence, we did not identify and assess all subjects with the comorbidities we were examining. It is possible that performance scores among patients we did not identify could differ from those of our study group. It is also possible that some processes actually occurred (eg, a specific laboratory test), but were not coded, falsely lowering performance scores. However, we are reassured that performance scores we report in this study are both accurate and generalizable because our results are consistent with those of other studies, many of which used data from medical records.^{39- 44,52- 56}

Our study has several strengths. First, because our sample was large, we were able to identify subsets of individuals with several different comorbid diseases and assess the quality of health care for each. Second, we used many well-established process measures that have randomized controlled trial data and expert opinion supporting process-outcome links. Third, as a result of our sampling method, our sample should be representative of the national fee-for-service population with rheumatoid arthritis.

Our findings provide support to those who have recently called for reevaluation of the optimal roles of generalists and specialists in the care of patients with complex conditions.^57- 60 Health care delivery models that use primary care physicians as overseers for overall care may not be the best models for patients with rheumatoid arthritis if primary care physicians restrict access to specialty care. Efforts are needed to improve the quality of care for rheumatoid arthritis patients and to increase physician awareness of comorbid diseases among patients with this chronic disease.