Analyzing EHR Data with NDL Methods Sample Clauses

Analyzing EHR Data with NDL Methods. When new problems first appear, the most natural reaction from researchers is to try existing methods and examine if they work on new problems. EHR data is not a new problem, but in the most recent several years, it gradually attracts much more atten- tion from researchers than before. One reason is that the development of technology enables researchers to analyze the whole EHR data from one or multiple data sites. Another reason is that the rich information contained in EHR may provide great support to detect existing treatment patterns and to develop personalized treatment regime. In order to analyze EHR data, a few NDL methods have been proposed. We briefly explain what techniques those study utilize and how the proposed methods perform. The work by Wu et al. (2010) reviews the challenges and strategies of analyzing EHR data. They examine three common machine learning techniques: logistic regres- sion with BIC as selection criterion, support vector machine (SVM), and boosting. The EHR records for a total of 536 subjects with heart failure are extracted and used as the experiment dataset. SVM is a classification technique, which transform the original data space into a higher dimensional space. To reason to do such transformation is that some classifi- cation decision making can be easier in higher-dimensional space. Assume the data is consisted of (xi, yi), i = 1, . . . , N for N patients and each xi is assumed to have p inputs. By defining the M feature function as h(xi) = (h1(xi), h2(xi), . . . , hm(xi)), the decision boundary can be found by Σ − || || N min [1 yif (xi)]+ + x x 0, (0.00) x0,x