R software version 3

Prehospital stroke scales were reconstructed with the NIHSS items assessed at baseline in the intervention centre. The scales were assessed as positive or negative, using the cut point proposed in the original publication. We calculated the sensitivity for the detection of LVO for each prehospital stroke scale, both stratified by occlusion location and for all occlusion locations combined. For each prehospital stroke scale, the sensitivities for different occlusion locations were compared using Chi-square tests. Additionally, we plotted the sensitivity for all possible cut points of the prehospital stroke scales, stratified by occlusion location. Potential differences in sensitivity across prehospital stroke scales may be caused by variation in the included NIHSS items. Therefore, we calculated the percentage of patients in our cohort who had an abnormal score on each NIHSS item. All analyses were performed using R software version 3.6.1 and Rstudio version 1.0.153.

For cytotoxicity assays significant differences among fractions of cytolysis obtained from unirradiated cells were compared to the ones of each irradiation dose at a given time point. Significance was determined via R software version 3.6.1 (RStudio, Boston, USA) using a two-tailed t test calculated with an R code created by the DKFZ Biostatistics Department. To correct for multiple comparison, we applied Holm-Bonferroni method.

The study population was described in terms of frequencies for qualitative variables, or medians, means and associated ranges for quantitative variables. Comparisons of the proportion of samples with a DCIS component before and after NAC were performed with McNemar tests.
Factors predictive of pCR were introduced into a univariate logistic regression model. A multivariate logistic model was then implemented. The covariates selected for multivariate analysis were those with a p-value for the likelihood ratio test below 0.05 in univariate analysis.
A significance threshold of 5% was used. Analyses were performed with R software, version 3.1.2 (RStudio Team (2018). Rstudio: Integrated Development for R. RStudio, Inc., Boston, MA, USA, URL http://www.rstudio.com).

Analyses were realized using the R software Version 3.1.2 and R Studio Version 0.98.1091 – © 2009–2014 RStudio, Inc. The tests were performed in bilateral formulation. p values < 0.05 were considered to be significant. For the univariate analysis of the relationship between bone involvement and vascular parameters, binary variables were analyzed by a comparison of means by the Student t test for normally distributed variables, and by the Wilcoxon test in the other cases. A univariate analysis of the association between cardiovascular risk factors and vascular parameters was realized by Fisher test. Variables with p < 0.2 were included in the multivariate analysis [50 (link)]. Multivariate analysis was performed by multiple linear regression. The selection of variables was based on the minimization of the Bayesian Information Criterion (BIC) by a bidirectional step-by-step method. The adjusted coefficient R² was the main benchmark for choosing the best model. Normality of residuals was verified by a graphical method (Quantile-Quantile graph) and controlled by the Shapiro and Wilk test. The quality of the models was assessed by monitoring the equivariance and independence of the residuals, and the research for influential or aberrant observations.

Pre- and post-NAC TIL levels were analyzed as continuous variables. All analyses were performed on the whole population and after stratification by BC subtype. To compare continuous variables among different groups, Wilcoxon-Mann-Whitney test was used for groups including less than 30 patients and for variables displaying multimodal distributions; otherwise, student t-test was used. Association between categorical variables was assessed with chi-square test, or with the Fisher’s exact test if at least one category included less than three patients. In boxplots, lower and upper bars represented the first and third quartile respectively, the medium bar was the median, and whiskers extended to 1.5 times the inter-quartile range. Factors predictive of pCR were introduced in a univariate logistic regression model. Covariates selected for multivariate analysis were those with a p-value no greater than 0.1 after univariate analysis. Survival probabilities were estimated by Kaplan-Meyer method, and survival curves were compared with log-rank tests. Hazard ratios (HR) and their 95% confidence intervals (CI) were calculated with the Cox proportional hazard model. Analyses were performed with R software version 3.1.2(RStudio Team (2018). RStudio Integrated Development for R.RStudio, Inc., Boston, MA URL)). The significance threshold was set at 5%.

The least absolute shrinkage and selection operator (LASSO) logistic regression algorithm was used for reducing the excessive dimensionality of data and selecting the most significant features in the training data set. Radiomic features with non-zero coefficients were selected from the training data. The analysis was performed using R™ software version 3.6.3, Vienna, Austria, and R Studio™ version 1.2.5033, Boston, USA using the “glmnet” package.