DETECTION OF HEALTHCARE ASSOCIATED INFECTIONS USING TEXT AND DATA IN SWEDISH PATIENT RECORDS

Speaker: Hercules Dalianist Seminar Date: Tuesday March 7 2017 12.00pm Brief abstract: Over 10% of all in-patients today obtain a Healthcare Associated infection (HAI), this causes a lot of suffering for the patients and immense costs for the society. Healthcare managers cannot easily obtain statistics of the level of HAI in each clinic so they can put actions to prevent HAI. In this talk we explain what a HAI is and how one can detect it using both textual information and structured data from the patient record. We are using a combination of pre-classified Swedish text from the patient records as well as structured data as drug use specifically antibiotics, body temperature and microbiological values as training data. Both Support Vector Machine and Gradient Tree Boosting machine learning methods were used in addition to a number of pre-processing steps for the classification. Our best results in detecting HAI with respect to high recall is with the Gradient Tree Boosting, where we obtained 93.7% recall and 79.0 % precision.