Here is a tragic case where the diagnosis was missed on three separate occasions over a 19-month period. Both terminal conditions are treatable and potentially separable if discovered and aggressively managed. It is not the intent of this presentation to criticize the evaluation of this patient, but rather to point out the need and, also, the potential benefits of an alternate approach, which could have significantly changed the course of this particular patient's evaluation. We have conducted similar analyses with other CPC cases with almost equivalent specificity. These same evaluations have been performed in a real clinical setting with live patient data. What we have learned so far we think will be extremely helpful in extending the potential application of this technology: A text-based decision support tool is only as good as its practitioner. It takes practice and training to learn to use this system effectively. There are many traps in logic, and the use of words and terms within the text must be understood to effectively utilize this tool. There are shortcuts in logical analysis which we mentally use all the time but which cannot be accepted using this system (e.g., you must not rely on your memory or any specific associations to circumvent the system). The system will only become clinically relevant when the entire field of medicine is included in the data base. This is one of our current limitations with only two-thirds of the Merck medical text available for reference. This makes it difficult to apply to a general medical problem since we are not sure which direction the case might take, and often these are multisystem diseases or problems that put us at a severe disadvantage if we don't have the necessary data base. The structure and integrity of the data base are critical to the success of the system. Since numbers are ubiquitous, these cannot be used for key word elements. Techniques must be introduced to create word-oriented numbers that can be uniquely identified (e.g., "AGE14-20"). The publisher of the data must supply a continuous flow of up-to-date material that can be incorporated within the framework of the working system. It is possible to train medical technicians to use the system if they are familiar with medical terminology. The speed and, perhaps, the precision of their analysis could not be expected to rival that of a medical specialist.(ABSTRACT TRUNCATED AT 400 WORDS)