Margriet Kwint

Chapter 5 100 Discussion In this study, we identified 3 subgroups with distinct GTV-changes during treatment. The identified subgroups each showed an increasingly steep volume reduction during CCRT. However, no associations of membership of a particular subgroup with outcomes were observed. Recent literature (13-15) suggested associations between tumor volume change and OS. In our present study, we could not validate any of the published associations of tumor volume changes and OS, either by considering Latent Class Mixed Modelling or by replicating the original analyses. An explanation for this might be that these previous studies of Brink et al.(13) Jabbour et al.(14) and Wald et al.(15) were very small studies (99, 38 and 52 patients, respectively). As Wald et.al .(15, 32) already emphasized, those studies can only be considered as hypothesis generating rather than conclusive. Since our large study could not validate these results, the observed association could be clarified due to a power problem in those smaller studies (13- 15). In coherence with previous studies (4-11), a large baseline GTV of the primary tumor remained a significant predictor for worse OS and PFS. This remained not significant in the pathology subgroup analysis, probably due to a power problem, but the same tendency in the strength of association was observed. Although we identified 3 different subgroups with distinct volume changes, most patients (83%) were grouped into the same subgroup with limited tumor volume regression during treatment. When looking at the individual patterns of treatment response (Figure1), one might acknowledge that the assumption of heterogeneity in treatment response could not be confirmed. Moreover, Wald et al.(32) did not observe heterogeneity in treatment response either. Therefore, the addition of longitudinal data of tumor volume change during treatment might not improve the current risk models based on baseline information (4-11). Nevertheless, using longitudinal parameters to improve the predictive accuracy of outcome models are an important topic of ongoing research (33-35). Novel longitudinal parameters such as imaging features, circulating tumor DNA and molecular tumor profiling might be relevant in developing more accurate dynamic prediction models for a more personalized treatment and follow-up care approach.