Session Item

To fit radiation pneumonitis (RP) models to two patient cohorts treated at the same clinic with different RT treatments and compare the risk factors related to the incidence of RP. To compare the predictability of each model in the other cohort.

We included patients consecutively treated with 50-66Gy in 25-33 fractions (433 pts 2010-2017). In April 2013, the daily online positioning strategy was changed from match on the vertebrae to match on the primary tumour. Combined with an adaptive radiotherapy (ART) strategy, this reduced PTV volumes significantly. The patients were split in two groups: Treated with ART (253 pts, ART group) or before ART was introduced (180 pts, preART group). Basic characteristics (age, gender, smoking status, histology, stage (AJCC, 7thEd), performance status (PS)), chronic obstructive pulmonary disease (COPD), chemotherapy, previous surgery and RT treatment (GTV volume, mean dose to lungs (MLD) and heart (MHD)). RP was evaluated retrospectively (CTCAE v3.0), dated at maximum observed RP grade. Separate multi-variable Cox proportional hazard models were fitted to each group. The performance of each model in both groups was investigated graphically by calibration plots (six subgroups, equal size, based on predicted probabilities 180 days after RT start).

Patient characteristics in the two groups were not significantly different. However, treatment changed significantly. The use of concomitant chemotherapy (64 % vs 78 %, p<0.001) and IMRT (86 % vs 100 %, p<0.001) increased, median PTV volume (473 cm3 vs 272 cm3, p<0.001), median MLD (16.4 Gy vs 12.1 Gy, p<0.001) and MHD (8.0 Gy vs 4.5 Gy, p<0.001) decreased. Fig1 shows the HRs for the preART and ART model. The effect size of MLD was equal in both models, but non-significant in ART (both HR=1.53, p=0.009 preART, p=0.094 ART), while a nonsignificant dose effect for MHD was observed only in the ART model (HR= 1.25, p=0.072). Patients treated post-operatively (HR=3.13, p=0.041) or for recurrence after surgery (HR=2.94 (p=0.006)) had a significantly higher risk in preART, but both were non-significant in ART. Patients treated with sequential chemotherapy had a significantly higher risk (HR=5.45, p=0.017) in ART, but were non-significant in pre-ART. The predicted probabilities in both models fit the data set they originated from (Fig2). However, the preART model overestimated the probability in the ART group, and the ART model underestimated the probability in the preART group, except for a group of patients with very high predicted probabilities.

The effect of lung dose response was constant across the two groups. Other variables, such as surgery and chemotherapy presented larger variations. While the models fit the cohort they originate from well, they over/underestimated the probability of RP in the respective other cohort. This was possibly caused by the underlying changes in treatment strategies, or by patient selection, and highlights the need for continuous model adaption.