THE manuscripts included in this special series present us with a significant body of research. The Veteran's Health Study, from which these manuscripts are derived, can legitimately be viewed as a next logical step beyond the Medical Outcomes Study and the earlier Rand Health Insurance Study. It features the use, and adaptation, of the SF-36 in the study of patient-based outcomes, such as health status and health-related quality of life, among a sample of patients from the ambulatory care system of the Veterans Health Administration, the nation's largest integrated health care system. The investigators go beyond that, however, in attempting to combine disease, or condition, specific measures for an important group of diseases with the SF-36 generic measure and in developing and utilizing instruments for the assessment of the severity of patient perceived illness. The resulting manuscripts make for interesting, and informative, reading on a variety of related topics.

The paper by Rogers et al. (in press), for example, on comparison of the health status of VA and non-VA ambulatory patients is useful for the information it provides on the differences found in the level of self-assessed health status between the two populations. Certainly the findings of significantly poorer health status in the VA ambulatory care population, a difference suggested by other studies, and the explanation of this difference by crude measures of disability and socioeconomic status have implication for the intensity of medical care required by that population, the resource allocation to the system, etc. Additionally, the important question of whether these differences are due to a difference in case mix in the populations or to a difference in treatment effectiveness (or to a combination of the two) is not ignored in this paper, as it often is. The authors argue effectively for the case mix interpretation in this case.

Despite the usefulness of these results, I found the most interesting part of the Rogers paper to be the innovative use of odds ratio comparisons in place of the usual method of adjustment for concomitant variables by means of a regression equation in making the comparison between the populations. Not only does this approach have the advantage of simplicity, but I find myself in agreement with the authors when they state that "the nature of statistical adjustment is usually to seriously underestimate how much of observed differences might be due to case-mix when important differences like these exist." This is a promising methodology that deserves much wider exploration and use in our field.

Many of the papers in this volume are concerned with the refinement and further understanding of patient-based measures of health outcomes in the VA. That is entirely appropriate. As we look to the future and to the next generation of studies, however, we should recall that, if we follow the Donabedian model, outcomes research must be regarded as having 2 major steps. The first of these is the definition of, and development of tools for the measurement of, appropriate health outcomes. The second is the explication of the relationship between the desired-and now, measured-health outcomes and the observable processes that lie within our control and are thought to affect, or perhaps even generate, those outcomes. Without this important second step, ie, the relating of processes to outcomes, we cannot seriously address such critical issues as the use of outcome measures as indicators of quality of health care because, of course, health care is itself a process.

Unfortunately, the inference problems associated with making this connection between processes and outcomes are very real. Once one gets beyond the use of experimental, eg, clinical trial, data-and one gets beyond that very fast in real life-there is very little to guide us in our decision making. Those of us who are methodologists need to turn more of our attention toward helping to provide assistance with the problems associated with the drawing of inferences from nonexperimental, eg, administrative, observational, etc, data sets because often these are the best data that will be available for the foreseeable future on our most important topics.

In the meantime, projects like the Veterans Health Study can help us take that next step, the step of examining how the health services processes, defined in the broadest sense, relate to the health outcomes and health-related quality of life that they have shown us that they can measure in the VA population. If the papers in this series are any indication, we can expect these investigators to contribute significantly to this next important series of questions. I look forward to seeing their results.

REFERENCE