Data Mining for Description and Prediction of Antibiotic Treated Healthcare-Associated Infections
Abstract: Healthcare-associated infections is the most common healthcare related injury and affect almost every tenth patient. With the purpose of reducing these infections Infektionsverktyget, The Anti-Infection Tool, was developed for registration and feedback of infection data. The tool is now used in all Swedish county councils resulting in a wealth of data. The purpose of this thesis was thus to investigate how data mining can be applied to describe patterns in this data and predict patient outcomes regarding healthcare-associated infections that need to be treated with antibiotics. Data mining was performed with Microsoft SQL Server 2008 in which models based on six different data mining algorithms with different parameter settings were developed. They used the attributes gender, age and previous diagnoses and medical actions as inputs and antibiotic treated healthcare-associated infection outcome as output. The predictive performance of the models was evaluated using 5-fold cross validation and macro averaged measures of recall, precision and F-measure. Patterns generated by selected models were extracted. Models based on the Naive Bayes algorithm showed the highest predictive capabilities with respect to recall and models based on the Decision Trees algorithm with low pruning had the highest precision. Although, none were considered to perform sufficiently well and several areas of improvement were identified. The most important factor in the inadequate performance is believed to be the relatively rare occurrences of infections in the dataset. Extracted patterns based on the Association Rules algorithm were considered the easiest to interpret. Patterns included clinically valid and invalid as well as trivial relationships. Future studies should be focused on further model improvements and gathering of more patient data. The idea is that data mining in Infektionsverktyget in the future could be used both to provide ideas for further medical research and to identify risk patients and prevent healthcare-associated infections in daily clinical work.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)