Statistical Analysis
Descriptive statistics were reported using medians and interquartile range (IQR) for continuous data. Categorical variables were presented as frequency (percentage). Variables were tested for normality using the Shapiro-Wilks test. To establish associations between dependent and independent variables, Spearman’s rho correlations were conducted. To determine the trajectory of dysphagia and dysphonia from initial SLT assessment to SLT discharge, medians of ordinal dependent variables at both time points were compared using two-tailed Wilcoxon signed rank tests.
To determine independent predictors of oral intake status at time of initial SLT assessment, a binary logistic regression was used. The seven-point ordinal FOIS rating scale was divided into feeding tube reliant (FOIS Levels 1-3) and not tube feeding reliant (FOIS Level 4-7) categories as the binary dependent variable. To prevent over-fitting the model, six independent variables were selected for oral intake status (age, duration of intubation, proning, neurological manifestations, maximum cuff pressure, history of respiratory disease).
For voice quality, an ordinal logistic regression was completed with the overall (G) four-point ordinal GRBAS rating as the dependent variable. Six independent variables were selected for the voice regression model (intubation injury, proning, maximum cuff pressure, duration of intubation, number of comorbidities, and history of respiratory disease).
For both regression models, independent variables were selected based on evidence from previous research and a visual review of the data. Where a significant association was identified between independent variables (e.g., duration of intubation and presence of tracheostomy), only one was selected for a model. Mean imputation was made for one independent variable (maximum cuff pressure) as it was missing 24/100 cases. Model fits were confirmed using likelihood ratio chi- squared tests. A two-sided α of less than 0.05 was considered statistically significant. Statistical analyses were completed using the SPSS (v26) software.