A Comparison of Various Data Reduction Procedures in a Multiple Sclerosis Sleep Study

Abstract

Clinical studies often deal with datasets with numerous variables. As a result of the similarities between the variables, we frequently observe the presence of multicollinearity in the data. This study aimed to apply different data reduction strategies to sleep study variables in multiple sclerosis (MS) patients. The main objective was to use various data reduction strategies to explain a subjective measure of sleep quality (Pittsburgh Sleep Quality Index: PSQI) by the objective measures of sleep quality obtained during complete in-laboratory overnight polysomnography. Overall, we found that few objective measures of sleep quality were important in explaining the subjective PSQI, based on the results of various well-accepted statistical methods. Total sleep time was found to be the most important feature of objective sleep quality for explaining subjective sleep quality among all other investigated objective sleep quality variables in most of the approaches investigated in this study. The LASSO method for estimation worked best in terms of interpretability among all the approaches considered.