Theses and Dissertations

Date of Award

5-2023

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Applied Statistics and Data Science

First Advisor

Dr. Tamer Oraby

Second Advisor

Dr. George Yanev

Third Advisor

Dr. Hansapani Rodrigo

Abstract

Chronic kidney disease (CKD) is a significant complication that contributes to diabetes-related mortality in the United States, and there is growing evidence that sodium-glucose cotransporter 2 inhibitors (SGLT2i) can slow its progression. However, observational studies may suffer from confounding by indication, where patient characteristics and disease severity influence the decision to prescribe SGLT2i. This study utilized electronic health records of individuals with diabetes (from TriNetX) to investigate the effectiveness of SGLT2i on CKD progression. The database provided detailed information on patients’ CKD status, demographics, diagnosis, procedures, and medications, along with corresponding dates of diagnosis and prescription. The study comprised of 38,776 patients aged 18 years and above with 1 (or >) year history of diabetes who initiated treatment with SGLT2i between May 9, 2013, and July 7, 2021. To address potential confounding by indication in observational studies, we utilized propensity score matching. We also addressed the issue of imbalanced classification in medical datasets by applying balanced bagging (with bootstrap aggregation) and Synthetic Minority Oversampling Technique (SMOTE). By overcoming the underrepresented minority class and reducing bias in the machine learning (ML) models, we ensured accurate causal identification of CKD prevalence by employing the following ML models: logistic regression, decision tree, random forest, extreme gradient boosting, support vector classifier and artificial neural network. Our results suggest that SGLT2i have a protective effect on CKD outcomes, providing valuable insights into the practical efficacy of this treatment and potentially serving as a clinical decision-making tool. Our findings have significant clinical implications for the healthcare sector and suggest that these techniques can improve the reliability of ML methods.

Comments

Copyright 2023 Solomon Eshun. All Rights Reserved.

https://go.openathens.net/redirector/utrgv.edu?url=https://www.proquest.com/dissertations-theses/machine-learning-approach-evaluate-effect-sodium/docview/2842776574/se-2?accountid=7119

Download

Share

COinS