An Exploration of Regression Analysis Methods to Identify Significant Predictors of Visual Outcomes in Unilaterally and Bilaterally Injured Ocular Trauma Patients Público
Minihan, Adair (Spring 2019)
Published
Abstract
Background: Acknowledging the paired nature of ocular data is imperative to a sound statistical analysis, as it differs in unilaterally versus bilaterally injured patients and also in cross-sectional versus longitudinal analysis. The dataset in question contains unilaterally and bilaterally injured patients and is longitudinal. Previous literature does not address regression methods accounting for both unilaterally and bilaterally injured patients. It also does not attempt to identify significant clinical and demographic predictors for ocular trauma outcomes.
Objective: Identify regression methods for analysis of unilaterally and bilaterally injured ocular trauma patients in cross-sectional and longitudinal contexts.
Methods: The main predictor of interest was injury type and the outcome of interest was the logMAR score. LogMAR scores were divided into three categories: satisfactory (logMAR <= 0.3), moderately impaired (0.3 < logMAR <= 1.0), and severely impaired visual outcomes (logMAR > 1). Cross-sectional analyses consisted of selecting the injured eye for unilaterally injured patients and the most injured eye at immediate follow up for bilaterally injured patients. Three logistic regressions with a baseline logit link were created, one for each follow up time. In longitudinal analysis, both eyes were included for bilaterally injured patients. Two mixed models were created with a logit link and random effect for subject id, one comparing satisfactory to moderately impaired outcomes, the other comparing satisfactory to severely impaired outcomes.
Results: Injury type was significant in both the cross-sectional logistic regression for the immediate follow up and the longitudinal mixed model comparing satisfactory outcomes to severely impaired outcomes. For patients with blunt injuries (vs. penetrating), the odds of having severely impaired outcomes was 0.236 (95% CI 0.099, 0.563) times the odds of having satisfactory outcomes at the immediate follow up. For patients with blunt injuries (vs. penetrating), the odds of having severely impaired outcomes was 0.15 (95% CI 0.055, 0.408) times the odds of having satisfactory outcomes in the longitudinal analysis.
Conclusions: The mixed models better captured the complexity of the data. They account for the longitudinal and paired nature, as both eyes were included for bilaterally injured patients.
Table of Contents
Introduction………………………………………………………………………………….…….1
Background………………………………………………………………………………….…….2
Paired Ocular Data and Hypothesis Testing………………………………………………2
Regression Analysis Methods………………………………………………………..……3
Methods……………………………………………………………………………………………5
Ocular Trauma Data……………………………………………………………………….5
Data Cleaning and Categorization of Variables…………………………………………...5
Cross-Sectional Methods………………………………………………………………….6
Longitudinal Methods……………………………………………………………………..6
Results…………………………………………………………………………………………….7
Descriptive Analysis………………………………………………………………………7
Cross-Sectional Analysis………………………………………………………………….7
Longitudinal Analysis……………………………………………………………………..8
Discussion…………………………………………………………………………………………9
Conclusions………………………………………………………………………………..9
Limitations and Future Research……………………………………………………….....9
Reference……………………………………………………………………………………...…11
Appendix…………………………………………………………………………………………13
About this Master's Thesis
| School | |
|---|---|
| Department | |
| Subfield / Discipline | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Palabra Clave | |
| Committee Chair / Thesis Advisor |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
An Exploration of Regression Analysis Methods to Identify Significant Predictors of Visual Outcomes in Unilaterally and Bilaterally Injured Ocular Trauma Patients () | 2019-04-10 00:25:46 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|