Methods for Analysis of Big Data.: A Combination of the Lasso Method and the Case-Crossover Design for Investigating Potential Drug Side Effects on Myocardial Infarction

Abstract

The current master thesis was written during the academic year 2013 − 2014 atthe Norwegian University of Science and Technology (NTNU). It concerns the im-plementation of the case-crossover design in a combination with the lasso method,for investigating potential effects of a set of drugs on myocardial infarction. Thedatasets that were used in the analysis were generated based on information aboutthe usage frequencies of the drugs in the period from 2008 to 2012. Furthermore,the thesis provides a brief explanation of how the lasso method can be used in thecase of generalized linear models, as well as the case-crossover design. The mainanalysis was based on two datasets such that probably weak aspects of the lassocould be discovered. Another aspect of the current thesis was the implementationof the relatively new inference method for the lasso, as well as the implementa-tion of two forms of the lasso method: simple lasso and bootstrap lasso. Finally,the current thesis shows, with numerical results, that the bootstrap form of lassois an effective variable selection method, and that the lasso inference is not yetsufficiently developed.