A case study on parsing chemotherapy related free-text data

When modelling and simulating healthcare related processes free-text data is often the only possible source of information. This data may contain vocabulary variations such as mistyped, misspelled and/or abbreviated words. This paper describes a semi-automated approach to free-text normalisation based on a combination of commonly used techniques and local expertise of oncological nurses. In our approach particular emphasis is given to the effectiveness of the vocabulary creation process through an interactive software application. When local knowledge is successfully captured, normalisation of large data sets can be done very rapidly with a high accuracy rate achieved. Furthermore, the techniques for localised normalisation can have significant benefits to free-text parsing accuracy when data is aggregated from multiple sites (hospitals). A vocabulary created at one site can be reused at multiple other sites to speed up the vocabulary creation but at the same time facilitate the capture of local idiosyncrasies that otherwise would be lost. We believe that this research may lead to better understanding the free-text data which in turn may impact patient treatment and outcomes.