Exploring nature of the structured data in GP electronic patient records

In this research project, two novel approaches have been developed for understanding and interpreting data from electronic patient records from general practitioners captured in the form of Read codes. The first method has examined medical records as language; the second has developed new visualisations of medical data to support interpretation and hypothesis formation.Using methods borrowed from computational linguistics it has been possible to show that most of the data in medical datasets coded using Read codes show behaviour similar to that seen in language, in particular the power law behaviour known as Zipf's law. However, Read codes encoding the medication showed different behaviour. Computational linguistic can therefore have a role in assessing the quality of medical data. In a related study we show that the methods developed are applicable outside of medicine, providing insights into the quality of gene function annotation captured using the Gene Ontology.In a second approach, a novel method was developed for mapping high dimensional patient records (at a population level) into a lower dimensional vector space. This mapping is crucial for exploring data and applying various data mining methods on such data. Most importantly, such mapping makes it possible to visualize the patient records for a given period of time in terms of all patient data at a whole population level. It provides us with a tool for exploring potential and weaknesses of the coded data originating from electronic patient records. This visualisation has been applied to data regarding irritable bowel syndrome (IBS) patients in Salford. Using the methodology it was possible to show firstly, that it will not be possible to develop a diagnosis for IBS based solely on gastrointestinal symptoms, and secondly that at the diagnosis level, the IBS are scattered in different regions of diagnosis space, suggesting that patients with diagnosis of the IBS were also recorded to have different diagnosis data other than IBS diagnosis.EThOS - Electronic Theses Online ServiceIranian Ministry of HealthGBUnited Kingdo