The challenge has not yet been met. We do measure things but generally not outcomes. We combine and share information, but this information is generally about processes, algorithms, and "best practices" and not necessarily about outcomes. We are accountable to purchasers, communities, and individual patients but often in terms of costs, access, and satisfaction rather than clinical outcomes. The four articles in this section offer clues to the reasons Ellwood's vision of outcomes management has not yet become reality.

Spoeri and Ullman describe one particular area of performance measure effort-the 1994 HEDIS (Health Plan Employer Data and Information Set) pilot project. Many readers are likely familiar with HEDIS' basic content and utility. The issues about provider profiling and inclusion of outcome data raised in the second half of the Spoeri and Ullman paper are among the most interesting lessons learned from HEDIS because they force us to be more aware of, and perhaps seriously question, some of the basic assumptions that underlie many measurement and reporting initiatives. Who is really accountable for health outcomes in a population of health plan enrollees? What elements of case mix should be included in various adjustment models? If race, for example, is statistically associated with outcome in a particular patient population, should a plan or provider be "excused" for poor performance in the higher-risk group by including race in the model? If the answer is "yes," an incentive to improve care in a special population is removed; if it is "no," the plan or provider may avoid the higher-risk group.

Palmer provides an eloquent argument for using process rather than outcome measures. She identifies a key barrier to more extensive use of process measures of quality-the need for detailed clinical information that is typically not found in large, automated databases. However, careful use of some existing databases can yield useful process measures and developments in electronic medical records promise greater opportunities in the future.

Liang and Shadick shift the discussion to outcome measures and address the utility of including disease-specific measures of health status in large databases. They conclude that such an effort may not be worthwhile because of scale insensitivity, lack of evidence for improvement in patient care as a result of availability of this information to clinicians, and difficulty in deciding which of several possible disease-specific surveys should be administered to an individual patient with more than one relevant condition. These are reasonable concerns, but before we agree we should be sure that the issue has been clearly drawn. All administrative databases are not the same, nor do they serve the same purposes. Why focus on plans as the things being compared, and why focus on billing databases within plans? It seems much more likely that outcome data in a standard format would be useful in the databases of organized provider networks in which the data lie closer to where the care is delivered.

What information should we take away from studies that show no impact on patient care when data from functional health status surveys are provided to clinicians? The utility of the information for individual patient care does not seem to be the issue that the authors set out to address. If there has been a call for inclusion of outcome data in health plan databases, the purpose has typically been external accountability to purchases and comparisons of plans by purchasers. Improvement in patient care comes indirectly (if it is supposed to come at all) through pressures to show better results or lose customers. This is a very different activity from one in which individual clinicians collect and use health status data in individual patient encounters [3, 4].

McHorney paints a more optimistic picture of a future in which health status data are routinely collected and used. She focuses on the new opportunities that may come through a combination of computer-based survey administration techniques and a different conceptual approach to survey design. The new approach is tailored to patients by asking them only those questions relevant to their own health problems and functional levels. The interests of purchasers and some providers in this approach may ultimately determine how quickly (or if ever) this approach is adopted. Purchasers now feel relatively comfortable with well-established, validated generic measures of health, such as the SF-36. Despite some obvious theoretical advantages, the computerized adaptive testing and item response theory approach will have to show that the resulting scores have the same kind of intuitive appeal to users that the 0 to 100 scores of the SF-36 do.

We have made substantial progress in the past 10 years towards realizing Ellwood's vision. The path we ultimately follow will inevitably differ in some respects from the one he asked us to take. The relative emphasis on process versus outcome measures, for example, is still an unsettled issue whose resolution (if any is possible) will determine how close the two paths are. How far we travel on the outcome path will depend on technologic advances and on whether the basic assumptions about accountability for health outcomes are sound.

David R. Nerenz, PhD

Henry Ford Health System; Detroit, MI 48202