Costs, Outcomes, and Patient Satisfaction by Provider Type for Patients with Rheumatic and Musculoskeletal Conditions: A Critical Review of the Literature and Proposed Methodologic Standards

Methods

Literature Search

We searched the MEDLINE database for studies published between 1966 and 1 April 1996. Paired medical subject (MeSH) headings used for keyword and text word searching included musculoskeletal disease, rheumatic disease, arthritis, low back pain, sprains or strains, osteoarthritis, gout, crystal arthritis, septic arthritis, rheumatoid arthritis, chiropractor, physiatrist, osteopath, rheumatologist, orthopedic surgeon, specialist, specialty care, physician practice patterns, appropriateness, cost of health care, and delivery of health care. We did not examine such conditions as lupus nephritis or rheumatoid vasculitis, which are clearly in the domain of specialists and have not been covered in studies comparing generalists with specialists. In addition, we consulted with experts in health services research, rheumatology, orthopedic surgery, and epidemiology of musculoskeletal disease to identify proprietary and other unpublished references. Finally, we reviewed the reference lists of all articles obtained to identify additional articles of interest.

Selection Criteria

We reviewed only English-language articles in which 1) patients had rheumatic or musculoskeletal conditions or both and 2) providers of generalist and specialist care were compared with respect to clinical outcomes, resource utilization, patient satisfaction, or appropriateness of care. Although abstracts were not included in the review, authors of published abstracts [6-10] were asked to provide manuscripts submitted for publication or in press.

Study Evaluation

The studies were evaluated by using the methodologic criteria summarized in Table 1. These criteria pertain to several aspects of each study, including the organization of the physician and patient cohorts, measurement of the specified outcomes, and data analysis.

We examined three aspects of the studies with respect to providers: physician training, the physical environment and financial organization of the practice setting, and the assignment of patients to providers. With regard to patient cohorts, we determined whether diagnoses were similar and whether differences in severity of disease were adjusted for in analyses.

Several types of outcomes were measured in the studies that we included: patient-centered outcomes (that is, clinical outcomes or patient satisfaction with care), resource utilization (duration of hospitalization or cost of care), and appropriateness of the process of care. We examined the potential for bias in the personnel and instruments used to assess outcomes. Specifically, we determined whether outcome data were obtained by self-administered questionnaires (obviating the potential for observer bias) or by interviewers. If an interviewer or chart review was the source of the outcome data, we determined whether assessors were involved directly in patient care (creating the potential for observer bias) and whether they were blinded to provider assignment. Finally, we determined whether outcome instruments were standardized and validated in the relevant populations. In addition, if appropriateness measures were applied, we examined whether such measures were based on evidence-based criteria or on a consensus approach. We also determined whether appropriate sample size estimates were calculated.

Results

The MEDLINE search identified 70 studies, 17 of which fulfilled the inclusion criteria. The excluded studies did not clearly identify provider comparison groups or patient conditions. Investigations that were included fell into five categories of disease: low back pain, work-related injuries, osteoarthritis, acute arthritis (crystal-induced and septic arthritis), and rheumatoid arthritis. Results are reported by disease category (Table 2 and Table 3).

Low Back Pain

Differences between generalists and specialists have been studied more frequently with regard to low back pain than to any other regional musculoskeletal condition. Meade and colleagues [11, 12] performed the only randomized, controlled trial to date comparing chiropractic with conventional hospital outpatient care of patients with mechanical low back pain. The authors studied 608 patients 18 to 65 years of age who presented to 1 of 11 hospitals or chiropractic clinics in the United Kingdom. Patients who had been seen in the previous month for low back pain or in the previous 2 years at the same clinic were excluded. Demographic characteristics and functional disability, assessed by using the Oswestry low back pain disability questionnaire [28], were similar in the two study groups at baseline. The authors found that patients who reported less severe functional disability at study entry had similar outcomes with either type of provider, whereas those with severe disability at study entry had greater improvement with chiropractic care at 6 months. These differences persisted at 3 years.

This study has numerous strengths but also several limitations (Table 3). The randomized trial design minimized selection bias, but approximately 85% of patients screened for the study were not eligible or were excluded, limiting the generalizability of the study results. In addition, chiropractic care was delivered in private offices outside of the National Health Service's hospital environment, where conventional care was administered. This raises the question of whether differences in organizational and physical environment between the private sector and the National Health Service may have overshadowed differences between providers.

In an observational study, Carey and coworkers [13] prospectively compared the outcomes and costs for nonreferred patients with acute low back pain who were cared for by primary care practitioners, chiropractors, and orthopedic surgeons practicing in various organizational settings in North Carolina. The authors contacted a random cohort of clinicians, and 74% agreed to enroll consecutive patients who presented with a first episode of low back pain of less than 10 weeks' duration. Small demographic differences between patient cohorts were noted at baseline, but mean pain scores did not differ among the groups. The study results showed that although time to self-reported recovery did not differ between provider types, chiropractors saw their patients approximately five times more often than did other providers, and chiropractors and orthopedic surgeons ordered twice as many radiographs as primary care practitioners did. Patients of orthopedic surgeons had the highest rates of hospitalization. Chiropractors and orthopedic surgeons were twice as expensive as primary care physicians, but patients reported significantly greater satisfaction with chiropractic care than with care delivered by other providers. This study was large and well-designed and included predetermined standardized outcomes and adequate description of provider groups (Table 3). However, although the authors attempted to adjust for baseline differences, the nonrandom allocation of patients may have resulted in the referral of more challenging or severe cases to specific providers, particularly surgeons.

Shekelle and associates [14] compared the cost of an episode of back pain treated by different types of providers in the RAND Health Insurance Experiment. This large randomized trial was conducted in the 1970s to examine the effects of different payment plans (such as fee-for-service or prepayment) on various end points; each type of insurance included coverage for chiropractic care. Shekelle and associates analyzed the resources used by each type of provider for 1020 episodes of back pain. Chiropractors served as the primary care providers for 40% of the back pain episodes and had the highest mean outpatient costs. Although chiropractors charged less for each episode than other practitioners did, they saw patients an average of 10.4 times per episode, approximately twice as often as any other type of provider. General practitioners saw their patients only 2.3 times per episode and had the lowest mean outpatient and overall costs.

Several important limitations of this study must be noted. Most important is that the care was delivered approximately 20 years ago; thus, the generalizability of the study results to current practice is unclear. In addition, the study randomly assigned patients to payment plans but not to types of providers, thereby creating the possibility for patient self-selection bias.

Cherkin and MacCornack [15] surveyed members of a health maintenance organization who were 18 to 64 years of age and chose to see either a family physician or a private chiropractor for low back pain. The health maintenance organization paid the first $200 per year for chiropractic care. Compared with patients who saw family physicians, patients visiting chiropractors reported greater satisfaction and fewer days with limited ability to carry out usual activities.

In this study, patients were not randomly assigned to providers; thus, selection bias cannot be excluded (Table 3). The authors adjusted for baseline differences in clinical severity in an attempt to overcome this limitation, but they did not account for such nonclinical factors as smoking, obesity [29], and education [30], which can influence outcomes for low back pain.

Another observational study [16] compared data from medical records on chiropractic and medical care for low back pain in a random sample of workers' compensation claimants. Claimants who saw medical physicians lost more time from work and incurred greater costs than did those who saw chiropractors. Because functional status was not measured at baseline or follow-up, the chiropractors' patients might have had better outcomes simply because they had less severe disability at baseline (Table 3). Little information was given about the actual process of care.

Two studies used large administrative databases to assess resource utilization for low back pain. Stano and Smith [17] used MEDSTAT, an insurance claims dataset that includes chiropractic care and covers approximately 2 million insurance beneficiaries. These authors compared the duration and cost of care for patients with low back pain who presented to medical physicians and chiropractors. Patients who saw chiropractors had lower total payments after adjustment for demographic factors, insurance status, and severity of disease. On the basis of their own unpublished analyses, the authors claim that the diagnostic codes can be used to adjust severity. However, problems with claims-based severity adjustments are well documented [31]. The authors did not describe the method used to calculate payments, but because they used an administrative database it is unlikely that out-of-pocket payments for over-the-counter medications were included in their totals.

In another analysis of administrative data, Hart and colleagues [18] used the National Ambulatory Medical Care Surveys to examine resource utilization for low back pain across provider types. Generalists (family physicians, general physicians, and internists) wrote the most prescriptions for low back pain; orthopedic surgeons and neurologists ordered the most radiographs of areas other than the chest.

Both of these studies [17, 18] used large administrative databases and lacked clinical data; this limits the precision of case-mix adjustment (Table 3).

Work-Related Musculoskeletal Injuries

Johnson and associates [19] compared chiropractic, medical, and osteopathic care for work-related sprains and strains in a workers' compensation population. Approximately one third of the claimants who were contacted responded to a mailed questionnaire; 90% of respondents represented claims that had been settled. Claimants who saw chiropractors lost 1 less day of work, on average, than claimants who saw other practitioners (8 compared with 9 days of work lost).

Several limitations of this study must be pointed out (Table 3). Because only one third of claimants responded to the survey, the external validity of the findings is uncertain. No adjustments were made for baseline demographic characteristics or clinical severity. In addition, claimants selected providers themselves. These limitations may have favored the providers who treated patients with less severe and more self-limited disease.

Osteoarthritis

In a study comparing the self-reported practices of rheumatologists with those of primary care physicians, Mazzuca and coworkers [20] surveyed family physicians and internists about their therapeutic approaches to hypothetical cases of osteoarthritis of the hip. They mailed the same questionnaire to a random sample of practicing rheumatologists in the United States, whose responses served as the standard of care [21]. Approximately two thirds of each type of provider responded. Compared with rheumatologists, primary care physicians more often used nonsteroidal anti-inflammatory drugs (NSAIDs) at lower, analgesic doses (that is, non-anti-inflammatory doses) for uncomplicated osteoarthritis. In addition, primary care physicians recommended misoprostol to accompany an NSAID less often than rheumatologists did for patients with a history of ulcer disease, and they did not use sulindac as often in patients with a history of renal insufficiency. In addition, rheumatologists prescribed nonpharmacologic therapies (for example, splinting and physical therapy) more often than primary care physicians. The calculated cost of care recommended by rheumatologists did not differ from that recommended by non-rheumatologists for complicated patients but was greater for uncomplicated osteoarthritis of the hip.

This study has several limitations (Table 3). Responses to hypothetical patient scenarios do not necessarily reflect actual practice. Further, only scant data support the prescription of particular NSAIDs, use of specific dosages of NSAIDs, or use of rehabilitation therapies for osteoarthritis of the hip [32]. Hence, although primary care physicians and rheumatologists report different beliefs about optimal care for osteoarthritis, it is not possible from current data to determine which view is more appropriate.

Acute Arthritis

Five studies compared the care provided by different practitioners for acute mono- or oligoarthritis. Panush and colleagues [22] retrospectively reviewed the charts of 55 patients who were hospitalized with acute arthritis at one university-affiliated community hospital. Four rheumatologists managed the care of 20 of the patients; general internists, subspecialists, and orthopedic surgeons cared for the other 35. Demographic characteristics, comorbidity, and severity of illness at baseline did not differ between the patients of rheumatologists and those of nonrheumatologists. Rheumatologists performed radiography (65% compared with 31%) and synovial fluid analysis (75% compared with 34%) significantly more often than nonrheumatologists did. Rheumatologists established a definite diagnosis more frequently. All aspects of the quality of care delivered by the rheumatologists were considered to be superior on the basis of a quality rating scale developed by the authors. In addition, patients of rheumatologists had a significantly shorter mean duration of symptoms (3.5 days compared with 6.6 days), a shorter mean duration of hospitalization (7.4 days compared with 14.7 days), and a lower mean cost of hospitalization ($8755 compared with $14 750). The small sample size, the nonrandom allocation of patients to providers, and the use of a post hoc scale to measure quality of care limit the study findings (Table 3).

Using different methods, Walker and associates [23] also studied the management of acute arthritis. They identified consecutive patients who presented to a general practitioner, presented to an emergency department, or were discharged from an inpatient rheumatology unit with acute crystal-induced or septic arthritis. Only 1 of 30 patients with acute monoarthritis of the first metatarsophalangeal joint who presented to a general practitioner or emergency department had aspiration. On the other hand, 82% of the patients with crystal-induced arthritis who were discharged from the rheumatology unit had aspiration. Generally, the rheumatology patients did not have podagra; half had gout in other joints, and the others were given a diagnosis of pseudogout. Rheumatologists gave 13 patients a diagnosis of septic arthritis, whereas only 2 patients seen in the emergency department received such a diagnosis. No cases of septic arthritis were missed, and all but 1 patient with septic arthritis received antibiotics within 24 hours. No patients in either group died. The lack of formal statistical analysis and adjustment for disease severity limit the external validity of these findings (Table 3). Also, because there is debate about whether acute monoarthritis in the first metatarsophalangeal joint requires aspiration, the failure of nonrheumatologists to aspirate these joints cannot be considered inappropriate.

Two studies, one in Canada and one in New Zealand [24, 25], surveyed general physicians and rheumatologists to assess the pharmacologic management of acute and chronic gout and asymptomatic hyperuricemia. For acute gout, all physicians used NSAIDs most frequently. The frequency of use of colchicine was similar in the two groups. On average, Canadian rheumatologists allowed a greater number of acute attacks of gout before starting urate-lowering therapy. In addition, rheumatologists in both countries prescribed colchicine significantly more frequently to cover the introduction of urate-lowering therapy. It is interesting that 59% of nonrheumatologists and 70% of rheumatologists said that they would prescribe urate-lowering therapy for asymptomatic patients, a practice that has no support in the literature [33]. Because these studies used hypothetical cases, it is unclear how their results reflect actual practice.

Evans and coworkers [26] compared actual patterns of usage of intravenous colchicine at two academic hospitals with usage as recommended in published guidelines. A retrospective review of all inpatient pharmacy records over an 18-month period yielded orders for intravenous colchicine for 19 patients who had crystal-induced arthritis and complete medical records. Eight of the 19 patients did and 11 did not receive rheumatology consultations. The authors applied literature-based guidelines [34] to evaluate the appropriateness of all orders for colchicine and found that 5 of the 11 patients who did not have rheumatology consultations received intravenous colchicine inappropriately, whereas none of the 8 patients who had consultations received colchicine inappropriately (P = 0.045). No adverse events were noted in either group.

The authors asserted that, on the basis of their findings, more teaching with regard to the use of intravenous colchicine is needed, but several limitations of their study must be pointed out (Table 3). First, like most appropriateness guidelines, those used in this study were based not on trial data but on expert review of case reports. Second, the authors are rheumatologists who were among the developers of the published guidelines; rheumatologists at other institutions might behave differently. Third, leukocyte count, which is an intermediate end point for colchicine toxicity, decreased overall in both patient groups. Hence, guideline compliance did not clearly change outcomes.

Rheumatoid Arthritis

Two studies examined whether rheumatologic care for patients with rheumatoid arthritis is associated with better outcomes. In a 10-year prospective cohort study, Ward and coworkers [27] evaluated the progression of disability in a cohort of 282 patients with rheumatoid arthritis. Patients receiving continuous rheumatologic care had less decline in Health Assessment Questionnaire scores than did those receiving intermittent care from a rheumatologist, even after controlling for clinical and demographic variables. Patients who received continuous care from a rheumatologist were given disease-modifying antirheumatic medications and underwent joint replacements more frequently than patients who received care from other providers. The prospective study design, with careful adjustment for severity and use of standardized functional status outcomes, strengthened these findings (Table 3). However, the study was not randomized and thus may have been confounded by factors not accounted for in the multivariate analysis.

Criswell and associates found similar trends in a different cohort of patients with rheumatoid arthritis who were followed in the University of California, San Francisco, longitudinal rheumatoid arthritis panel (Criswell LA. Personal communication). Patients who saw a rheumatologist primarily or in consultation were more likely to receive disease-modifying antirheumatic medications, and their functional status was significantly better than that of patients who were not seen or were seen intermittently by a rheumatologist in the year after the initiation of therapy. These findings persisted after adjustment for baseline demographic variables and Health Assessment Questionnaire scores before treatment. These results, together with the similar findings of Ward and coworkers [27], suggest that patients with rheumatoid arthritis may have less disability if they are seen on an ongoing basis by a rheumatologist.

Discussion

We have critically reviewed published studies comparing specialist with generalist care for rheumatic and musculoskeletal conditions. Although relatively few such studies are available, several findings are noteworthy. Clinical outcomes for low back pain, which is often an acute and self-limited condition, seem to be similar across types of providers. However, resource utilization was higher in patients seen by chiropractors and orthopedists and satisfaction was highest in patients seen by chiropractors. Hence, generalist care seems to be as effective and less expensive-but also less satisfying-to patients with low back pain. Whether these generalizations extend to other regional disorders, such as shoulder or knee pain, remains an important topic for further study. For rheumatoid arthritis (a chronic and complex condition), however, specialists seem to produce better outcomes. It is not clear whether this is true for other systemic rheumatic disorders; this also merits further investigation.

Johnson and associates [19] claimed that patients with work-related musculoskeletal injuries who saw chiropractors lost less time from work than did those who saw practitioners other than chiropractors. The studies of osteoarthritis [20, 21] compared self-reported processes of care and found differences between rheumatologists and primary care physicians in the use of NSAIDs and physical therapies. The studies of acute arthritis [22-26] suggested that rheumatologists give more appropriate care and use less resources but found no clear variation in clinical outcomes.

In all of these studies, methodologic limitations hinder interpretation. Provider comparisons constitute an emerging area of research with few relevant methodologic standards [35]; hence, readers must be aware of potential pitfalls in research design, cohort assembly, outcome definition, and sample size calculations. We propose criteria with which to evaluate the design, outcomes, and analysis of such studies and have applied these criteria to the studies included in this review (Table 1 and Table 3).

Studies that provide relevant comparisons between generalists and specialists should report the qualifications of the providers and their practice experience. Because practice setting, method of payment, and various financial incentives may influence clinical outcomes and patient satisfaction, these factors should be described in detail and controlled for in analyses. In addition, the role of the different providers needs to be defined in a uniform manner because subspecialists, such as rheumatologists, function not only as specialists but also as primary care physicians for certain patients.

Assignment to a type of provider should ideally be random to balance patient characteristics (known and unknown) that might affect outcomes. Imbalances in disability and comorbid medical conditions at baseline can introduce selection bias. Because randomization is seldom possible, clinical severity should be measured carefully and controlled for in analyses. In addition, such nonclinical risk factors as health habits, education, and socioeconomic status should be carefully accounted for because they may influence the outcomes being studied.

Outcomes and resource utilization should be measured or collected, or both, by observers using standardized instruments who are uninvolved in the care of patients. If possible, observers should be blinded to hypothesis and type of provider. Self-administered questionnaires largely obviate the potential for observer bias. To be useful in the formulation of policy, a broad range of predetermined outcomes, including cost of care, patient satisfaction, symptom relief, functional status, and appropriateness, should be studied. When appropriateness is used as an outcome, the source of the practice standard should be explicitly described and the evidence supporting the practice guideline should be noted and assessed critically.

Finally, to avoid false-negative results, adequate sample sizes should be used as determined by a priori power calculations. Sample size calculations may need to be corrected by using an intraclass correlation coefficient because individual physicians may see multiple patients.

One methodologic issue worthy of special comment is whether provider opinion surveys are valuable, particularly in the absence of evidence-based guidelines establishing the appropriateness of management strategies. For example, studies by Mazzuca and coworkers [20, 21] of prescribing patterns for osteoarthritis are difficult to interpret because the literature does not clearly support one treatment over others. In contrast, provider opinions about the management of asymptomatic hyperuricemia are more useful because allopurinol is not indicated for this condition [33]. Surveys about process of care are probably most useful when a standard management algorithm has been established on the basis of well-accepted data. For instance, Ayanian and associates [2] surveyed cardiologists, internists, and family physicians about care after myocardial infarction, an area for which evidence-based recommendations exist. These authors found that noncardiologists were less likely to report using aspirin, β-blockers, and thrombolytic agents, all of which have been shown to improve outcomes.

Disease-specific comparisons of generalists and specialists may help us understand the optimal role of each provider in specific clinical circumstances. However, these comparisons do not provide guidance on 1) when, how often, and for how long patients should be referred to different types of providers over the course of a chronic illness or 2) which aspects of specialization explain observed differences in outcomes. Do specialists use distinctive processes of care, or do they provide similar care more skillfully? Which aspects of specialist care are advantageous and which are not? Future research should attempt to deconstruct the oversimplified notion of “provider type” and examine which aspects of the care process predict better outcomes. The goal of such research is not to declare one provider group “the winner” but rather to synthesize a system that borrows the best aspects of both generalist and specialist care.

The relationship between generalists and specialists has profound implications for policy, education, training, care, and reimbursement. Payers and consumers hope to spend their money wisely, policy-makers wish to shape the health care system appropriately, and patients should expect the best possible care. Hence, we must devote resources to developing a body of literature that will allow us to make decisions on the basis of scientific data rather than intuition, administrative edict, or interprovider politics.