outcomes given a particular course of treatment. By virtue of this neglect, physicians have not had the probability estimates at hand to inform patients about theft likely outcomes. A fundamental dimension in the IOM committee's definition of quality (IOM, 1990), namely the likelihood of an outcome, is not now understood in any systematic way. Because of the influence of the new agency, in the next few years we will come to understand much better what works in medicine; I will return to this point below.

The IOM committee's definition also includes the idea of a desired health outcome. I think we are now in a position to begin to learn what it is that patients actually want in medicine as opposed to what they have used. Most economists and policy analysts have in the past confused utilization with demand, as if the rate of service use expressed the wants of patients. Now we know from the small-area variation studies and from all we have learned about the scientific weaknesses in medicine that utilization does not necessarily indicate what patients want.

In Paul Griner's paper (Griner, 1991), I am struck by the use of the word "appropriate." This word has always been a problem for me. When I was on the house staff at Johns Hopkins, my chief of medicine and other senior clinicians were always telling the house staff what was appropriate, which usually meant what they wanted us to do. Although their theories were very plausible, many were not tested; they were simply part of the conventional wisdom of the day. At that time, for example, it was appropriate, particularly at Hopkins where Dr. Halstead had invented the idea, to do a radical mastectomy for women diagnosed with breast cancer. We now know that this theory was hardly appropriate, if by the word appropriate we mean something that patients want based on knowledge about outcomes.

In our search for definitions, therefore, modification is needed. As a house staff officer, we thought the words appropriate care meant necessary care. If it is inappropriate, then it is unnecessary care. Now we know that medicine is much more complex, that there are multiple morbidities, multiple treatments, and multiple outcomes. Care is appropriate only if two conditions exist: (1) it works in some important dimensions of the multiple-morbidity, multiple-outcomes complex; and (2) patients prefer this treatment and these outcome probabilities among those offered by the alternatives. When these conditions are met, it seems fair to say that care is appropriate. Care that patients do not want is not appropriate, even though it may be effective.

I want to elaborate a bit on this notion of choice between valid treatments. As we completed our assessment of watchful waiting and surgery as