Background In 2007, HAS (Haute Autorité de Santé, France) started experimenting the use of computerised guidelines analysis with the G-DEE software as part of the guidelines production process. G-DEE uses shallow natural language processing techniques to recognise the expression of recommendations in a guidelines’ text, and structures the document accordingly using on-screen text highlighting and colouring. Project managers were offered, on an entirely voluntary basis, the possibility of using this tool during the final phases of guidelines production to i) analyse overall guidelines structure and ii) review the formulation of individual recommendations. To date, this experiment has concerned 38% of guidelines that have been officially released.

Objectives Improving the performance and usability of the deployed G-DEE system.

Methods We have collected feedback from guidelines’ project managers who have identified clear benefits in terms of readability of individual recommendations, overall text balance, fine-tuning and testing of alternative wording. Main limitations identified were processing time, which required additional planning and the absence of a tuning facility that would allow different levels of grammar coverage.

Results Notable enhancements include faster document processing (x10) and incremental document analysis allowing direct text input in the system.

Discussion These will make its use possible throughout the production process, e.g. directly as part of consensus meetings once the first draft is available. In addition, the possibility to define additional grammar files for text processing supports the processing of new document types (shorter guidelines, summaries) which may adopt different writing styles.