Data Analysis
Summary statistics were reported as median (IQR) or frequencies (n, %)
as appropriate. Logistic regression was used to perform a multivariate
analyses of outcome measures and data from the multivariate analyses
were reported as odds (Exp (B)) with the 95% confidence interval (CI).
Demographic, clinical, technology-related and follow-up variables were
initially included. Diagnostic category was included in the analysis to
account for heterogeneity of the sample. Subjects were classified into
five diagnostic categories according to their underlying condition
(upper airway; central nervous system; musculoskeletal; cardiopulmonary;
other category for those ones with characteristics of more than one
category). Individual underlying disease conditions previously shown to
influence long-term NIV outcomes15 were also included
in the initial multivariable analysis. We also included the time period
in which NIV was initiated (Epoch 1, Jan 2005-Apr 2008; Epoch 2, May
2008-Aug 2011; Epoch 3, Sept 2011-Dec 2014).
Predictor variables that had a p<0.20 on univariate analysis
for each outcome were tested in the respective logistic regression
models. Co-variates considered clinically relevant, such as age, sex,
diagnostic category, number of co-morbidities, number of additional
therapies, and NIV type, were forced into multivariable analyses,
regardless of their univariate significance. A stepwise purposeful
addition and removal of covariates was performed until a best-fit model
(using the Nagelkerke R-square) was produced. Analysis was performed
using IBM SPSS Statistics version 26.0 (SPSS, Inc., Chicago,
IL)20. Summary statistics were reported as median
(IQR) or frequencies (n, %) as appropriate. A p-value of
<0.05 on analysis was determined to be statistically
significant.