Margriet Kwint

Prognostic value of volumetric changes during concurrent chemoradiation 91 5 with the lowest Bayesian Information Criterion (BIC) (where a difference of at least 10 points is regarded as a sufficient improvement), (ii) discriminatory power with the best mean posterior probabilities of class-membership (mean probability >0.80) and (iii) clinically relevant differences between the subgroups (where subgroups with <5% of patients were considered clinically irrelevant). The Grolts-checklist was followed as much as possible to report on the LCMM (30). Differences in baseline characteristics between the subgroups were tested using Pearson Chi Square tests (categorical variables), Kruskal-Wallis tests (non-parametric and >2 groups) or ANOVA (parametric and >2 groups). To test the potential benefit of adding tumor response to a baseline prediction model, we compared the baseline model and the baseline+LCMM tumor-volume change model using the Harrell’s c-statistic. Sensitivity analyses To validate current literature (13-15) on the prognostic potential of GTV changes during treatment, we performed supplementary analyses in which similar definitions, as used in this literature, of relative tumor volume change between first and last fraction were considered: (14). (In patients for whom no CBCT was available of the last fraction, the CBCT made in the last week was used.) And the relative tumor volume change of the GTV between fraction 11 and 21: (15). To replicate the analyses done in the literature, these relative changes were then dichotomized according to the median value of the group and the association with treatment outcome was subsequently tested using Cox regression analyses. Multivariate cox regression analyses were then performed to investigate the independent association of the GTV-change parameters with OS, PFS and LRC. Furthermore, univariate cox regression analyses were performed to associate the median cut-off values from literature with treatment outcome, namely; 39.3 % (rel_ GTV first-last ) and 22.8% (rel_GTV 11-21 ) (14, 15). Finally, the influence of pathology was also assessed by stratifying for; AC or SCC and NSCLC-not otherwise specified (NOS) (13).