Margriet Kwint

Acute esophagus toxicity after IMRT and concurrent chemoradiation 119 6 predictive of AET. For this reason, in building the logistic model, we only selected the most significant parameter to predict grade ≥2 and ≥3 AET, as can be achieved with the forward conditional selection logistic regression method. For prediction of grade 3 AET, the V50 was shown to be the most significant (p=0.013) parameter with a β 0 of -2.486 and a β 1 of 0.032 (Table 3) . For grade 2 AET, the D max was the remaining parameter after forward selection. However, we expect that using D max in the plan optimization will not sufficiently influence the dose distributions. Since grade 3 AET is clinically more relevant, and also for practical reasons, we also estimated the binary logistic regression parameters for grade 2 AET using the V50, which was also shown to be highly significant (p=0.012) ( Table 3 ). Doing so, we establish one single parameter, which can be used to predict the probability of both grade ≥2 AET and grade ≥3 AET. The probability of developing acute esophagitis grade ≥3 can now be estimated using: and for grade 2 AET: The sigmoid shaped relationship between AET grade ≥2 and ≥3 and the V50 is plotted in Figure 2 , together with the actual incidences of AET. To illustrate the need to replace our current V35-model for grade ≥2 AET, we have plotted the actual V35 data of the current study with respect to the model ( Figure 3 ). The log-likelihood difference between the V35 data estimated with the old model and the V50 data estimated with the new model corresponded to a significant difference with a p-value of 0.011 (chi-square distribution with 3 degrees of freedom), indicating superiority of the V50 model. Using the Mann-Whitney test, patients experiencing grade 3 AET and receiving all planned doses of Cisplatin (26%) were compared to patients in whom Cisplatin was discontinued early (10%). This incidence was not significant different (p= 0.083).