Second, systems should be speaker independent; that is, the system should be able to recognize words spoken by any individual who might speak without first ''training" the system to recognize the words spoken by that individual. Third, systems for general medicine require large vocabularies; some domains and subdomains, for example, may require vocabularies in excess of 30,000 words or meaningful phrases. As vocabularies expand, both the costs and error rates generally become intolerable. Emerging voice-recognition technology is likely to ease the inputting of clinical data in future CPR systems, but the successful experiences discussed earlier with such systems as HELP, THERESA, and DIOGENE confirm the existence of currently available alternative approaches to capturing crucial clinical data (including text) in the CPR.

Assuming that text can be conveniently entered into the clinical system through voice-recognition technology or other means, the problem then becomes one of effectively analyzing in an automated way the content and meaning of the textual data. The raw material for epidemiological analysis and for effectiveness and outcomes studies is primarily text from patient records, which must be converted to coded data. For accurate comparisons, patient record data must be correctly transformed into precise, unambiguous codes that represent specific characteristic processes.

Text processing is generally considered to be a complex operation; its application to the data in the CPR, with its special and diverse vocabularies, further complicates the challenge of implementing it as a system capability. Often, the more experienced the practitioner, the more succinct or abbreviated the notes in the record. The notes thus may consist of abbreviations, acronyms, and mnemonics, which could be difficult to interpret, even by other health care professionals. Although text processors have improved markedly in recent years, they can approach but never exceed the quality of written or dictated information. Therefore, the quality of patient records can be improved only through more disciplined approaches to consistent vocabulary in the record. Although technology (voice-input or menu-driven input systems) can artificially impose more consistent terminology, practitioners should be encouraged as a rule to avoid idiosyncratic terminology and to use more formal, well-defined vocabularies. Additional technological research is needed in this area, as well as studies of incentives for behavioral change, before CPR systems can reach their full potential.

Among the important priorities for the 1990s is the further development