Background: Re-use of patient data from an electronic patient record heavily relies on structured, coded data. Terminological systems (TSs) are meant to support this. TSs are more commonly implemented nowadays, especially those that support post-coordination. The effect of using a compositional TS on correctness and specificity of captured patient data is unknown.
Objective: To evaluate the agreement between free-text reasons for admission to intensive care captured in a patient data management system (PDMS) and reasons for admission that were recorded using a compositional TS embedded in the PDMS.
Methods: Observational study comparing pairs of free-text reasons for admission to intensive care with reasons for admission that were recorded using a compositional TS. Both reasons for admission were captured in the PDMS by clinicians during regular care practice. Each pair was judged as exact match, partial match or mismatch by two independent raters. Partial matches were further analyzed to investigate whether free-text or TS-based reasons for admissions included more detail and whether these differences could be explained by the content, the interface of the TS or by user or usability characteristics.
Results: Eleven percent of the pairs matched exactly, 79% of the pairs matched partially and 10% of the pairs did not match. Compared to free-text registration TS-based registration resulted in more detail for 21% of the partial matches, in less detail for 43% of the partial matches and in 36% of the partial matches some detail was added while at the same time other detail was lacking. In 65% of the cases in which the TS-based registration lacked some detail, this detail was available in the content of the TS. Physicians who used the TS occasionally had a significantly higher percentage of mismatches.
Conclusion: In practice, post-coordination leads to information with different detail but a level of detail comparable to free-text registration of reasons for admission. Details missing in the TS-based reasons for admission were most often available in the TS, indicating that user interaction with the system is more of an impediment than the contents of the TS.