A method for processing multivariate data in medical studies

Abstract

Traditional displays of principal component analyses lack readability to discriminate between putative clusters of variables or cases. Here the author proposes a method that clusterizes and visualizes variables or cases through principal component analyses thus facilitating their analysis. The method displays pre-determined clusters of variables or cases as urchins that each has a soma (the average point) and spines (the individual variables or cases). Through three examples in the field of neuropsychology, the author illustrates how urchins help examine the modularity of cognitive tasks on the one hand, and identify groups of healthy vs. brain-damaged participants on the other hand. Some of the data used in this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. The urchin method was implemented in MATLAB, and the source code is available in the online Supporting information. Urchins can be useful in biomedical studies to identify distinct phenomena at first glance, each having several measures (clusters of variables) or distinct groups of participants (clusters of cases).

Diagnosis of malignant pleural mesothelioma from pleural fluid by Fourier transform-infrared spectroscopy coupled with chemometrics

“…The high-distance values >3, as obtained in the model distance plot of the MPM, LC, and BPE groups, indicate a very robust and clear differentiation of both groups from the BPE fluids-group and from each other. [76][77][78] Moreover, Cooman's plot obtained from SIMCA analysis was also used to show the differentiation between two classes and the distance of test samples to their model. Cooman's plot of MPM versus LC and MPM versus BPE models demonstrated the correct classification of tested samples from each group with an average of 96.2% accuracy.…”

Effect of characteristics on the clinical course at the initiation of treatment for human immunodeficiency virus infection using dimensionality reduction

“…In particular, if the number of variables to be corrected increases, the dimension increases and, consequently, the distance between coordinate values increases, which is not suitable for the multi-model under the linear assumption. In this case, the problem can be resolved through dimensionality reduction, which is a method of multivariate analysis [8][9][10] . In a typical multivariable model, if the correlation of the main independent variable with the dependent variable can be presented after adjusting for some covariates, the multivariate analysis can identify the combination of factors that exhibits the greatest correlation with the dependent variable 11 .…”