One of the key challenges for clinical information systems developers is the development of interfaces for data entry. There is particular interest in capturing patient data in controlled form, directly from the clinician. This implies direct interaction between the user and the controlled vocabulary. When difficulties arise, it is easy to blame the user interface; however, problems may also be due to a dissonance between the cognitive model of the user and that underlying the vocabulary design.
We are using a cognitive science approach to the evaluation of user interaction to carry out controlled data entry in an electronic medical record system. Our controlled vocabulary, the Medical Entities Dictionary (MED) includes the usual dimensions of content (number of terms) and organization (hierarchical classification). The MED contains an additional dimension: knowledge - terms are linked through nonhierarchical semantic relationships. This knowledge is used primarily for vocabulary maintenance purposes. We hypothesis that this knowledge can also be brought to bear on the task of assisting the user in finding the desired term in the vocabulary. For example, if a user enters the term "Trypanasoma infection", the MED might contain such a term, with the term for the patient's actual diagnosis, Chagas' Disease, beneath it in the hierarchy. Alternatively, it may contain the term "Trypanasoma cruzi" (the organism) with a link to Chagas' disease. If the user fails to find the desired term, the question is: was the term missing from the vocabulary, was the term present but the knowledge needed to help the user find the term missing, or were the term and knowledge present, but the interface incapable of displaying them? If such knowledge is to be useful, is one KIND of knowledge (e.g., hierarchical or semantic) more useful?
We will obtain system usage data in several forms (audiotape "talk-aloud" protocols, videotape human-computer interactions, and computer session logs) and apply cognitive data analysis techniques to code the data (such as events of interest, use of system features, and time to completion of tasks) and use them to understand cognitive issues (such as the effects on reasoning of vocabulary organization and presentation). We will use the results of this analysis to identify and correct problems with the vocabulary content, knowledge and interface. Once corrections have been made, a second analysis will be performed to assess the impact of the changes on user performance.