While multivariate statistical techniques have found wide-spread use in many areas of science and engineering, their application in the health care sector is not nearly as wide-spread and usually limited to certain areas. This work makes use of state-of-the-art multivariate analysis techniques by applying them to three life science problems where such an application has not previously been reported. Specific emphasis is placed on the fact that early clinical trials may include a smaller number of participants and a larger number of variables. These statistical techniques are able to identify variables that can lead to better treatments and diagnosis that may not have been considered otherwise, because these techniques can sort through multiple combinations of variables to find a combination with the highest classification potential. This can either aid in the diagnosis by showing which measurements are significantly different between a control group and an affected group, or these significantly different measurements can point to interventions that may be needed to normalize the variable profiles of the affected group. The first contribution of this work focuses on the use of Fisher Discriminant Analysis (FDA) applied to data of blood samples from pregnant mothers who currently have a child with Autism Spectrum Disorder (ASD) and pregnant mothers whose children are typically-developing (TD). Classification returned a 90% accuracy in predicting which group the mothers fall into by using blood metabolite measurements from the Folate-Dependent One Carbon Metabolism and Transsulfuration (FOCM/TS) pathways. The second contribution used logistic regression to find that non-pregnant mothers who have a had a child with ASD and non-pregnant mothers with typically-developing children were able to be classified through the use of general blood metabolite measurements from multiple pathways with 97% accuracy. The third contribution focuses on similar algorithms, but applied to a different problem in that FDA was used on sports-related concussion recovery measurements to find differences between recovered athletes and non-recovered athletes to identify important markers of recovery. This work included balance measurements that have not been widely used before as compared to the standard observational practice that is currently in wide-spread use.
This research highlights the importance of multivariate statistical approaches applied to a few selected life science problems. The approaches are able to find significant measurements in clinical data that may not have been detected through traditional univariate analysis. This work has the potential to change the direction of future research and provide insight into clinical data that may have not been discovered using traditional means.;
May2021; School of Engineering
Dept. of Biomedical Engineering;
Rensselaer Polytechnic Institute, Troy, NY
Rensselaer Theses and Dissertations Online Collection;
Restricted to current Rensselaer faculty, staff and students in accordance with the
Rensselaer Standard license. Access inquiries may be directed to the Rensselaer Libraries.;