| Purpose. It is proposed that retrieval of clinical documents is facilitated and enhanced when these records conform to a semantic markup structure, specifically that proposed in HL7's Clinical Document Architecture (CDA) model. Hypothesis I: that tagging document elements will result in enhanced retrieval; Hypothesis II: that tagging document elements will result in a difference in relevance scores of two different question tasks.; Setting/subjects. Ten subjects with varying clinical experience, nine males and one female; 6 were fellows, 2 residents, 1 graduate of medical school beginning residency, and 1 an attending physician/faculty member. Three specialized in internal medicine, two in family medicine, and one each in infectious diseases, geriatrics, and medical informatics; two indicated they did not have a specialization. All were located at the University of Pittsburgh Medical Center.; Methodology. Crossover information retrieval study in which subjects served as their own controls; the document base consisted of 1000 deidentified clinical documents of 8 commonly occurring types in the MARS (Medical Archival Record) System in use at the University of Pittsburgh Medical Center.; Results. Hypothesis I was not supported. No statistically significant differences were found between the control (339.5 seconds ± 287.8) and treatment conditions (400 seconds ± 311.9) in mean time elapsed or the mean number of records in the final result set (control, 4.1 ± 7.8; treatment, 5.2 ± 10.5). However, tasks performed in the tagged treatment condition required a mean number of more steps in the search sequence to a degree that was statistically significant (control, 3.6 ± 2.9; treatment, 21.2 ± 13.2). Differences in precision were additionally investigated. The treatment condition tasks had a statistically significant lower rate of mean precision (control, .92 ± .231; treatment, .79 ± .383). Hypothesis II was also not supported. There was no statistically significant difference between the means of relevance of Type I and Type II questions tasks (Type I, .72 ± .388; Type II, .67 ± .345).; Discussion. The implications of this experiment are primarily for the development of the data model. Inadequacy of existing section headings—labels—for representation of clinical content may account for these negative results. Complex tag sets cannot ultimately resolve problems of unstandardized structure; the lack of structure is itself a significant limitation. |
