The first version of the S2e guideline on rehabilitation of mobility following stroke (ReMoS) was published in 2015. Today, almost four years later, implementation in clinical practice continues to be sluggish. For practitioners, it appears to be difficult to extract relevant core statements for daily practice from around 150 pages of continuous text with more than 250 abstractly-formulated recommendations. Thus the question arises as to how such weakly-structured data can and must be processed in order to enterclinical practice.

The aim of the “ReMoS-Mining” project, which began in the middle of last year, was to process the existing ReMoS guideline as a text, in order to make the knowledge contained in it accessible to a wider public in condensed form.

Using a bundle of partly algorithm-based analysis methods, significance structures were initially extracted from the text data of the guideline and summarised into key messages and recommendations.
The document editing was followed by the actual exploitation of the content, which included methods of classification, segmentation and dependency analysis. The aim was to summarise the content according to the problem or objective, to group it and to analyse it in terms of informative value and level of recommendation. Relational structuring in a data matrix allowed for further quantitative and qualitative analyses, as well as the application of dedicated filter logic to select units of information.

The processing and analysis was performed using the statistical software R and current packages downloadable for it online. R, as a well-known and powerful tool for the statistical analysis and evaluation of structured data, proved to be equally suitable as a tool for processing weakly-structured textual data in this context.

One extraction from the ReMoS mining project, as of now, is a dynamic web application that presents the contents of the guideline in a concentrated form and also reflects the results of exploratory data analyses that even remain hidden from readers of the full text.

The objective, of obtaining clinically-relevant guideline information in less than ten seconds which succinctly summarises the body of research on rehabilitation of mobility after a stroke, looks to have been achieved. Now we wait to see if, and to what extent, the efforts bear fruit. The project is, however, still in its infancy. In the long-term, the application is intended to evolve into a tool that optimally supports clinical decision-making for evidence-based rehabilitation of post-stroke mobility and, in addition to guideline recommendations, also generates concrete exercise and therapy suggestions, based on specific algorithms.

Key definition: Text mining deals with the processing and analysis of text data in order to use linguistic and statistical methods to develop patterns and unknown information from documents or natural language sources and to prepare them for users. The automated processing of textual data is not trivial, as that data is unstructured and highly dimensional. As a result, the text data in the process must first be structured and the dimensional characteristics reduced. For this purpose, text normalisation and dimensional reduction methods are usually applied. In order to be able to analyse text data, it needs to be transformed into a document matrix, in a second stage. From this, vectors are generated with which similarities are calculated. This allows the data to be clustered and classified, in a third stage, to identify topics, groupings or patterns. In the final stage, data mining, the data is analysed and visually processed. The findings, data relationships or patterns derived from these can then be used for further purposes.