Support Vector Feature Selection for Early Detection of Anastomosis Leakage From Bag-of-Words in Electronic Health Records

摘要: The free text in electronic health records (EHRs) conveys a huge amount of clinical information about state and patient history. Despite rapidly growing literature on the use machine learning techniques for extracting this information, little effort has been invested toward feature selection features’ corresponding medical interpretation. In study, we focus task early detection anastomosis leakage (AL), severe complication after elective surgery colorectal cancer (CRC) surgery, using extracted from EHRs. We bag-of-words model to investigate potential strategies. purpose is earlier AL prediction with data generated EHR before actual occur. Due high dimensionality data, derive strategies robust support vector linear maximum margin classifier, by investigating: 1) simple statistical criterion (leave-one-out-based test); 2) an intensive-computation (Bootstrap resampling); 3) advanced (kernel entropy). Results reveal discriminatory power complications CRC (sensitivity 100%; specificity 72%). These results can be used develop models, based that surgeons patients preoperative decision making phase.