Arthritis and musculoskeletal diseases comprise more than 100 conditions [1]. The importance of the effect of these disorders on function and quality of life has been recognized since 1949. Since the early 1980s, questionnaires have been developed to measure physical function and health status [2-12]. Both general and arthritis-specific measures have been applied to systemic rheumatic diseases and musculoskeletal disorders.

New disease-specific measures are available for rheumatoid arthritis, osteoarthritis, spondylitis, fibromyalgia, lower back and neck conditions, shoulder conditions, lower-extremity conditions, upper-extremity conditions, the carpal tunnel syndrome, and spinal stenosis. In general, the questionnaires have excellent psychometric properties, measure patient-centered outcomes, are sensitive to changes in clinical status, are more sensitive than generic measures, and serve as proxies for disease severity and activity. They are firmly entrenched in the clinical literature and accurately predict outcomes, mortality, and utilization of health services. Outcome measurement for several conditions have been incorporated in some practices, and progressive professional groups, such as the American Academy of Orthopedic Surgeons, the Pediatric Orthopedic Society of North America, the Hand Society, the Foot and Ankle Society, and the North American Spine Society, have invested substantially in the production, dissemination, and reporting of information on minimal essential outcomes. These measures are sufficiently mature that we can consider how and whether they might be collected as part of routinely obtained administrative data to answer questions in health services research and administrative research.

Feasibility of Incorporating Disease-Specific Outcome Measures into Administrative Databases

Administrative databases contain a wealth of demographic, diagnostic, and utilization information but usually lack clinical information. A sparse body of literature suggests that although these databases have tremendous power and potential, the precision of diagnostic coding varies and is related to the underlying diagnosis. Little work has focused on validating the information on utilization. Yet, the biggest problem with most large databases is lack of clinical detail and critical information on patient outcomes. Adding patient-derived information to administrative databases would probably require a fiat as a condition of payment in traditional indemnity systems or a major administrative decision in other systems of care. For practical reasons, it makes sense to obtain information directly from the patient rather than through physicians' offices, which are already overwhelmed with paperwork.

Generic forms could be administered routinely. Although they are less sensitive to change in persons with specific disorders, generic instruments are useful for making comparisons across different disorders. Some are simpler and shorter than disease-specific measures. With disease-specific instruments, the challenge is daunting. How would one decide which disease-specific instrument should be sent to the patient? In a person with many conditions, which condition should be assessed? Diagnostic codes for presenting patients' problems may not bear on the real diagnosis. It is difficult to imagine that patients would tolerate completing a questionnaire for each condition. Functional status in patients with one chronic disorder may be compromised by the existence of other illnesses, and it may be difficult to determine which condition is causing the functional problem and which physician is treating that condition. No single standard of measurement exists: The available instruments for arthritis and musculoskeletal diseases overlap considerably in the domains that they cover, and these in turn overlap with the domains covered by generic instruments. Customizing the number and type of forms or generating nonoverlapping combinations of scales by using an algorithm or an automated system may be a solution, but the comparison of patients by use of different outcome measures presents considerable methodologic difficulties. We think that this quandary is inherent to the current paradigm of outcome measurement, in which standard closed-ended questionnaires are the one size that is assumed to fit all. Finally, regardless of which and how many measures should be collected, it is essential that providers of care receive feedback in a timely manner, no matter how the information might ultimately be used. If health care providers are to be accountable, they must know what is being counted.

Should Generic or Disease-Specific Outcomes Data Be Included in the Administrative Databases?

The question of whether outcomes data should be included at all in administrative databases is as important as the practical considerations of implementing and administering extensive data collection. We discuss this with regard to the potential utility of these data in improving patient care by better monitoring, problem identification, and assessing the quality of care.

Four randomized, controlled clinical evaluations [13-16] of the effect of providing functional status information to clinicians and the effect of this information on decision making and patient outcomes showed disappointing results. These trials indicate that much work needs to be done before the technology that has been so useful in clinical research is transferred uncritically to the larger community.

McVey and colleagues [13] did a randomized, controlled trial of multiple health status instruments in conjunction with a geriatric consultation team that performed a comprehensive functional evaluation with specific recommendations. They were unable to show significant differences in functional status between case-patients and controls. Rubenstein and coworkers [14] performed a randomized, controlled trial of the Functional Status Questionnaire (a generic measure) and an educational intervention directed at internists. Although most physicians considered the functional status evaluation useful, the functional status in the case-patients and controls did not change over 1 year. Kazis and associates [15] studied the effects of providing rheumatologists with the Arthritis Impact Measurement Scale (AIMS) and the Health Assessment Questionnaire (HAQ), two widely used health status and physical function questionnaires on rheumatoid arthritis, for use by patients with stable rheumatoid arthritis during routine office visits. Health status information had no effect on either the process or the outcome of care. Most recently, Rubenstein and colleagues [16] evaluated a computer-generated functional status questionnaire, patients' self-reported chief symptom, and problem-specific resources and management suggestions on all new patients, augmented by two brief educational sessions for primary care housestaff. At 6 months of follow-up, physical function was not improved; however, emotional well-being and social activities that were limited because of health improved in the experimental group (309 patients and 40 physicians) but not in the control group. The lack of improvement in physical functional status observed in these studies may have been caused by the stability of disease in the participants, the timing of the provision of information, insensitive or inappropriate outcome measures, the lack of an effective intervention, the ignorance of physicians on how to use the information, or the possibly mistaken premise that the information is useful at all.

Physical function may not be the most appropriate end point. In the musculoskeletal and rheumatic diseases, which share features with other chronic diseases, the model of disablement is a causal chain that begins with impairment at the cellular and organ level. If left untreated and uncontrolled, these derangements lead to diminished function that is perceived in varying degrees by individual persons depending on their needs, aspirations, and expectations. Functional disability instruments may also handicap one's social role. Quality-of-life instruments that measure pain, function, or social role are downstream on the causal chain, and although these issues are of tremendous relevance and importance for patients, their immediate concern is usually joint pain and associated symptoms (such as stiffness and fatigue). In rheumatoid arthritis, for example, the basic pathologic lesion is rheumatoid pannus; this causes local inflammation that leads to structural damage and pain and sets up a cascade of events that ultimately results in crippling and disability. Both the patient and the physician focus on controlling the synovitis and the symptoms; function, although important, is only one aspect of the problem and is not directly treatable. In osteoarthritis, stiffness and gradual destruction of cartilage eventually result in mechanical pain and disability. Studies show that most patients give pain the most weight in their assessment of overall function and quality of life [17], and this is reflected in the primary therapeutic strategies used by physicians. Outcome measures that can be useful to improve management tend to be simple (numeric rating scale of pain); have a physiologic basis (for example, blood pressure); are modifiable over a short period (such as temperature); serve a mnemonic function (for example, the Apgar score); and, most important, measure what the patient perceives as his or her problems [18]. Measuring function and quality of life meets fewer of these criteria than does measuring swollen, painful joints.

If disease-specific measures are to be used and useful, they must measure something relevant, important, and mutable. Functional measures and multidimensional instruments or subscales of instruments used in arthritis and musculoskeletal disease have floor or ceiling effects in their ability to detect change. Their responsiveness is a function of the patient's baseline function, the number of items on the questionnaire, or a gradation of change that addresses one end or the other of the continuum of observed function.

Psychometrically sound instruments assume that function can be measured in all patients by using the same instrument. However, a patient's reported function is determined not only by physiology but also by age, sex, motivation, social support, priorities, goals, and needs. Because function is relative, a small change may make a big difference. A small change in function may be adequate for the person's needs yet not be detectable or meet statistical norms. Measures exist to better capture patient priorities, but whether the greater resources required to administer them result in sufficiently improved validity or responsiveness remains unexplored. Moreover, these measures tend to be impractical for the large-scale data collection that would be required if this type of test were to be used. Finally, is it fair to hold physicians solely accountable for deteriorating function in a patient? Function is determined by multiple factors and is often influenced more by social and psychological factors than by health care. Even with the best care, a patient's functional status can decline.

Functional outcome measures have been proposed as quality indicators of health systems for groups of patients or populations, such as those in the Health Plan Employer Data and Information Set (HEDIS). This is an attractive application but is largely unstudied with respect to disease-specific measures. Generic and disease-specific measures are responsive to differences between groups; this means that both could be used as indicators to highlight potential areas for detailed study and quality improvement initiatives. We suspect that the information derived from the measures would be too crude to identify actionable items alone. However, systematic assessment of function to monitor patients and identify problems is a critical component of patient care, and any of the instruments discussed are useful in this regard. The doubt lies in the utility of including such measures of function in administrative databases.

The rheumatic and musculoskeletal diseases have probably seen more progress than any other disorder in the development of disease-specific, patient-derived, and patient-centered outcome measures. All of the instruments currently used have excellent psychometric properties and have been useful in clinical and health services research. The question of how and whether these measures should be incorporated into administrative databases is answered here from a logistical and conceptual point of view. We think that logistical barriers might be overcome with work, but conceptual issues remain. The evidence from randomized, controlled trials shows that these measures add little or no value to patient care. The questions of whether routine use of such outcome measures in patient care would improve outcome over longer periods, why they do not improve outcome, and how they would become more useful to clinicians are unresolved.