Exploratory and prospective model of stent restenosis after Percutaneous Coronary Intervention using MultivariateGaussian Subspatial Regression

Abstract

A new statistical analysis methodology, "Multivariate Gaussian Subspatial Regression" (MGSR), has been applied to randomized clinical trial data collected from percutaneous coronary intervention (PCI) patients, which combines the descriptive quality of Factorial Techniques and the predictive power of Gaussian Processes. This model has been built from 3 different quantitative coronary angiographic core-lab measures of the same lesion from 2 separate angiograms (at baseline before PCI, at baseline immediately after PCI and at 12 months follow-up). Measurements of the pre-PCI variables of a patient are mapped to the factorial plane and predictions are visualized as regions of interest in this plane. MGSR makes it possible to detect patients at risk of coronary stent restenosis or patients in whom ruling out the disease, in a graphical way; avoiding unnecessary, costly and possibly risky treatments for patients with no complications predicted; and advising to closely follow patients at risk. In addition, the model recovers missing values regardless of the variables, and once fitted, it corrects itself when more dependent variables are included. MGSR software is freely available online at https://github.com/victorvicpal/MGSR.