a reasonably accurate practice assessment, would permit better use of implicit review methods. Perhaps more importantly, it would be useful to support more sophisticated studies on integrating expert judgments effectively into techniques that also rely in part on adherence to criteria, observed outcomes, or other less costly methods. As a number of investigators have demonstrated, combining features of the different approaches to assessing care can be a very effective way to minimize the weaknesses of individual approaches. A key goal of such studies should be to develop optimal methods in the assessment process for involving physicians, a limited and costly resource.

Another important area for further investigation is determining what relationships exist between the quality of a physician's care for one condition and the quality of the physician's care for other conditions-the issue of generalizability. The few studies that exist provide a sense that each condition is different, but whether assessments can focus on a limited number of diagnoses or must measure the quality of care across the entire spectrum of a physician's practice is obviously a crucial logistical question. Although measures do not appear generalizable at present, more sophisticated analyses might detect underlying patterns or correlations in physician treatment behaviors.

Another key area for further work is the development of better techniques for extracting relevant information from medical records. This is essentially an issue of data quality. Evaluations of care using patients' medical records can assess only items that should be present reliably in the charts, and ambulatory records have much more uneven quality than hospital information systems (479). Increasing computerization of patient data bases is a positive development in this regard. Some larger health care organizations and group practices are relying on such systems, and some quality assessments within hospitals involve manipulation of computerized patient data (446,547). The claims that physicians and hospitals submit to third-party payers could also provide computerized information, especially if entries concerning patients' diagnoses and clinical status were improved. Although a major segment of ambulatory practitioners has not yet adopted computerized office data systems, the creation of some kind

of national standards for computerized patient records could be an effective approach to improving reliable access to relevant information on the care process. More generally, uniform standards for data collection and reporting could be developed for all ambulatory records. Such measures would have to consider balancing increased time and cost of more detailed records with the benefits to quality assessment and other activities possible through more reliable or complete data.

Even with such improvements, many critical aspects of medical practice will remain difficult or impossible to capture in a provider's written record. Thus, increasing sophistication in measuring interpersonal aspects of care and physician influence on patient compliance with a therapeutic regimen could result in substantial improvements in the validity of process measures. These deficiencies can be addressed at least in part through patients' assessments of care (see ch. 11), and a physician assessment system featuring medical record reviews complemented by patient surveys could be a powerful approach to developing information on both the technical and interpersonal aspects of care provided. Borgiel and his colleagues currently use this combination in their practice assessments (84,86). Other creative approaches to measuring interpersonal aspects of care, as well as the other physician services not well reflected in the medical record, might also be useful.

Much research has already been devoted to setting standards for evaluating physicians' performance, but the development, evaluation, use, and timely revision of criteria and standards remain a central issue in any assessment that involves explicit criteria. In part, the development of criteria and standards requires clinical studies: much uncertainty remains about what clinical practices and procedures are most strongly associated with medical effectiveness. Ideally, only effective processes should form the basis for criteria developed for evaluations of care (411). In the care of hypertension and some other conditions, the processes that are effective have been relatively well established, and many useful criteria sets have been developed over the last 15 years. Some type of national clearinghouse, perhaps administered

through professional medical organizations, might both promote the effective use of these criteria sets and coordinate their refinement with guidelines for the content of medical records.

Improving the quality of ambulatory care assessments will also require further attention to more practical matters related to feasibility. Some of these concerns-such as promoting efficient use of medical experts—have already been mentioned. Many other approaches could also lead to lower assessment costs; examples include improved training methods, improved coordination with other quality-related projects and with organizations and activities designed to promote medical quality, and innovative approaches such as selfaudits (417). Another key area is the adaptation of computer technologies to assist in the collection of assessment information. For example, office audits can be expedited using software programs to enter data on adherence to criteria (419). Conceivably, these methods could be coordinated with computerized data base record systems to make assessments more fully automated.

Two other crucial considerations related to the feasibility of using evaluations of physicians' management of specific conditions to evaluate quality deserve final mention. One is the need for further deliberation on whether attention to all of the research items detailed above is worthwhile, or whether less ideal or entirely different approaches would be better alternatives for provid

ing consumer information or for increasing the likelihood that patients will receive high-quality care for specific conditions, such as hypertension. Although considerable experience with assessment methods in both research and practical settings indicates that these methods-especially combined approaches-have considerable promise, the discussion in this chapter suggests that serious technical, organizational, and economic obstacles remain before a functional system could be implemented nationally to provide useful information to consumers about individual physician performance for certain conditions. In this regard, it is important to recognize that almost no research has been directed specifically toward the question of providing information to consumers about the quality of the processes of care they receive for the treatment of hypertension or any other condition.

The other crucial consideration, running through out this chapter, is that evaluations of the process of care clearly require the leadership and assistance of the medical profession. Historically, professional medical associations have played the paramount role in evaluating physicians' performance; at present, they are continuing to expand their activities in promoting highquality care. Independently of its own assessment activities, or in coordination with them, the Federal Government can support the medical profession's efforts.

INTRODUCTION

There is a common notion that "practice makes perfect." In the medical care setting, this adage is often interpreted as "high-volume hospitals and physicians achieve better outcomes." The word "volume" in this context refers to the number of procedures or number of patients with the same diagnosis treated in a specific hospital or by a particular physician. For some procedures and diagnoses, better patient outcomes and lower inhospital mortality have been associated with higher volumes.

In its simplest form, the hypothesized relationship between volume and outcome may be displayed as a graph with volume (e.g., number of patients undergoing a specific procedure per year in a hospital) on the horizontal axis and outcomes (e.g., mortality rate) on the vertical axis. The graph in figure 8-1 shows high mortality in hospitals with low volumes and low mortality in hospitals with high volumes. The flattening of the curve at high-volume levels indicates that there is little additional reduction in mortality above a certain volume threshold.

It is important to limit the conclusions drawn from this graph. Even if a relationship is found between volume and outcome, it is inappropriate to conclude that increasing the volume in a hospital will improve outcomes or that reducing the volume will worsen outcomes. Conclusions cannot be drawn about how changes in volumes affect changes in outcomes, because most analyses use data from a cross section of hospitals observed at a point in time rather than data from the history of mortality and volume over time. Instead of causality from volume to outcome, there may be causality from outcome to volume; that is, medical providers with low mortality rates may attract higher volumes of patients. Another possibility is that some unmeasured factor may account for an observed relationship between volume and outcome. For example, high-volume hos

pitals or physicians may have relaxed admission criteria; these relaxed criteria, in turn, may mean. that some of their patients are healthier and less likely to suffer adverse outcomes. In this case, both the higher volume and the better outcomes may be caused by the relaxed admission criteria.

For OTA's review of the literature on the volume-outcome relationship, the abstracts of approximately 100 papers were read. Of the 50 articles that were thoroughly reviewed, 26 presented reportable findings.' Studies were included if they examined a sufficient number of hospitals (over 20) and cases to offer statistically valid volumeoutcome results or if the study purported to ex

'This chapter is based on a paper prepared for OTA by Harold S. Luft, Deborah W. Garnick, David Mark, Stephen J. McPhee, and Janice Tetreault (395).

'Additional technical information on the studies included in the literature review is available in the paper on volume prepared by Luft and colleagues (395).