Stata ic v 13.0

We calculated the proportion of patients not receiving thrombolysis for the entire period and for the time periods between 2003–2006 and 2007–2011. Descriptive statistics of non-thrombolysed and thrombolysed patients (baseline characteristics/mechanisms of stroke) and of the most common reasons for non-thrombolysis are presented. Patients arriving before and after 180 min and patients with admission during the first (2003–2006) versus the second (2007–2011) half of the observation period were compared using a logistic regression model that included potential predictor variables (age, sex, NIHSS score at admission, risk factors, stroke mechanism, time/severity/imaging/bleeding/other reasons for non-thrombolysis). The OR and its 95% CI or its associated p values were given to quantify and test the significance of the strength of the association. Predictors with p<10% in bivariate analysis were used to fit a multivariate logistic model. P Values of 0.683 (table 2) and 0.469 (table 3) in Hosmer-Lemeshow goodness-of-fit test suggested that each model fitted reasonably well. Analyses were conducted with STATA/IC (V.13.0; College Station, Texas, USA).

Regression analyses were performed for each body composition outcome as a dependent variable (lean mass, fat mass, body fat percentage, total mass). For each, six models were created. The initial model included only covariates. For the remaining five models, each independent variable (NUT_HC, NUT_FBR, NUR, NUR_CLS, PASE) was added independently to the covariate model. Likelihood-ratio tests were conducted to determine whether each independent variable explained significantly more variance than the covariate model. Further, because NUR_CLS was ordinal, effect sizes (d) were calculated between NUR_CLS (low NUR, moderate NUR, high NUR) for each outcome measure. Margins plots were created for statistically significant independent variables. Model diagnostics (collinearity, normality, heteroskedasticity) were assessed and visually inspected. Statistical analyses were performed in Stata/IC V.13.0 (StataCorp, College Station, Texas, USA).