Abstract
This paper presents a case study comparing text analysis approaches used to classify the current status of a patient to inform scheduling. It aims to help one of the UKs largest healthcare providers systematically capture patient outcome information following a clinic attendance, ensuring records are closed when a patient is discharged and follow-up appointments can be scheduled to occur within the time-scale required for safe, effective care. Analysing patient letters allows systematic extraction of discharge or follow-up information to automatically update a patient record. This clarifies the demand placed on the system, and whether current capacity is a barrier to timely access. Three approaches for systematic information capture are compared: phrase identification (using lexicons), word frequency analysis and supervised text mining. Approaches are evaluated according to their precision and stakeholder acceptability. Methodological lessons are presented to encourage project objectives to be considered alongside text classification methods for decision support tools.
Acknowledgements
The authors thank Alex Poole and Leitchan Smith at Cardiff and Vale University Health Board for programming the in-house text search tool and provision of access and support to the dataset.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Textual data are text that is perceived as unstructured by a numerically driven database.
2 Note that the administrative outcome mainly consisted of “query follow-up”—hence the need for validation.