ESTRO 2021 Abstract Book

S416

ESTRO 2021

For AGE and OP, the model parameters are also much larger for LR than for the SCM (data not shown) and LR even predicts an unreasonable negative parameter for OP. As revealed by the SCM, all clinical covariates have a negligible effect on model performance, leaving EUD as sole covariate. The final NTCP model has a high discriminative power, AUC = 0.78 (0.69-0.86), see also the CEBL-free survival for EUD terciles in Fig. 1 (model predictions in dotted lines). Mean predicted NTCP of the LR and the SCM are 21 % and 39 %, respectively.

Table 1 P-values for EUD and the four clinical covariates Univar. LR Multivar. LR Multivar. SCM EUD < .001 .008 < .001 AGE .004 .25 .86 CHE .06 .43 .39 OP .01 .24 .82 TG .01 .44 .35

Conclusion We have developed an actuarial NTCP model for the analysis of CEBL incidence and compared the results to classical logistic regression. We found that LR not only strongly underestimates the NTCP due to data censoring, but also overestimates the importance and weight of clinical risk factors. We conclude that logistic regression is poorly suited for NTCP modelling for late complications with substantial loss of follow-up.

[1] Bahn E, Bauer J, Harrabi S, Herfarth K, Debus J, Alber M. Int J Radiat Oncol 2020;107:571–8.

OC-0530 Local TCP and NTCP after radioactive plaque therapy for uveal melanoma: a data pooled analysis L. Cella 1 , F. Buonanno 2 , M. Conson 3,1 , C. de Almeida Ribeiro 4 , C. Oliviero 5 , F. Itta 6 , R. Liuzzi 1 , R. Pacelli 3,7 , S. Clemente 5 1 National Research Council (CNR), Institute of Biostructures and Bioimaging, Napoli, Italy; 2 University Federico II, Post Graduate School in Medical Physics, Department of Advanced Biomedical Sciences, Napoli, Italy; 3 University Federico II, Department of Advanced Biomedical Sciences, Napoli, Italy; 4 Brazilian Institute of Cancer Control, Radiation Oncology, San Paolo, Brazil; 5 University Hospital Federico II, Unit of Medical Physics and Radioprotection, Napoli, Italy; 6 University Federico II, Post Graduate School in Medical Physics, Department of Advanced Biomedical Sciences, Napoli, Napoli, Italy; 7 National Research Council (CNR), Institute of Biostructures and Bioimaging, Napoli, Italy Purpose or Objective Eye plaque therapy (EPT) represents the most frequently used eye-sparing treatment option for small to medium sized uveal melanoma (UM). The excellent local tumor control (LTC) rate of over 90% achieved by EPT can however be associated with severe complications and adverse events with a range of radiation-induced ocular injuries which may affect visual outcome and patient quality of life. Here, we investigated the local tumor control probability (TCP) and normal tissue complication probability (NTCP) of UM treated with EPT. Materials and Methods PubMed-indexed articles from 2005 to 2020 on EPB for UM were considered eligible if reporting dosimetric details correlated with actuarial 5-year LTC and/or morbidities. Pooled clinical outcomes were fitted to logistic dose-response model using the weighted least chi-square method. Each study was weighted by the number of patients included. Results Fifty-two cohorts with 11653 (range [11-3703]) UM patients met the inclusion criteria. The most common radioisotopes used in the reviewed literature are the gamma emitter I-125 (46%) and the beta emitter Ru-106 (29%). Five-year LTC data were extracted from 33 cohorts (median 92%, range [59-100]%) and TCP as function of tumor apex dose (Figure a) resulted in an estimated D 50 = 63.1 Gy (95%CI 59.9–66.2 Gy), γ = 2.8 (95%CI 2.4– 3.3) [reduced κ 2 =1.28]. Data on radiation-induced cataract rate (median 23%, range [0-63]%) and lens doses were extracted from 22 cohorts while radiation–induced optic neuropathy (median 9%, range [0-54]%) and optic disc dose from 18 cohorts (Figure b and c). NTCP model parameters were reasonably consistent for cataract [ D 50 = 39.6 Gy (95%CI 27.3–51.9 Gy), γ = 0.39 (95%CI 0.20–0.57), reduced κ 2 =2.27] and optic neuropathy [ D 50 = 107.7 Gy (95%CI 70.6–144.7 Gy), γ = 0.98 (95%CI 0.64–1.33), reduced κ 2 =1.59]. Dose-response relationships