ESTRO 2020 Abstract book

S884 ESTRO 2020

50, 60, 70, 80, 90, 100%). In addition, proportions of patients experiencing the outcome event were adapted (60% versus 80%). Results The sample size needed for a study increased dramatically, when the biomarker performance decreased (Figure 1). For example, if the hazard ratio for an experimental treatment was expected to be 0.46, the sample size needed would be 89 patients. However, if the biomarker that was used to select the study population had a PPV of 80%, 70% or 60%, this number increased to 131, 166 and 218 respectively. Moreover, if sample size is not adapted accordingly, this decrease of the PPV would result in underestimating the experimental treatment effect (HR of 0.53, 0.57 and 0.61 instead of 0.46).

calculated using the relative seriality model, including simulation of organ movement. The complication-free TCP (P + = TCP(1-NTCP)) is used to interpret the balance between TCP and NTCP. Results The results show that the SIP concept decreases the NTCP while maintaining a stable P + for all patients. The ideal dose prescription for the two target volumes PTV SIP and PTV dom is dependent on the individual patient anatomy (e.g. location of GTV relative to PTV SIP ). However, the general results are similar for all the patients. The figure shows the calculated probabilities (TCP, NTCP, P + ) for different treatment plans of one patient. Based on this patient-specific plot a decision for the optimal dose prescription can be made by either aiming for the highest P + , a threshold NTCP or a combination of both.

Figure 1. Sample size calculations for different biomarker performances for a time-to-event analysis. The colored lines show the sample size calculation for the different positive predictive values (PPV). A two-sided alpha of 0.05 and power of 90% is used. The overall probability of experiencing an event during the study is 80%. Conclusion These simulations show that studies with an enrichment design can heavily underestimate the needed sample size, due to less optimal biomarker performance. As underpowered studies could lead to erroneously abandoning of promising new treatments, it is of utmost importance to use a reliable and realistic estimation for biomarker performance. PO-1543 Public radiomics data collections in an open access Semantic Web (SPARQL) endpoint P. Kalendralis 1 , Z. Shi 1 , C. Zhang 1 , A. Choudhury 1 , A. Traverso 1 , M. Sloep 1 , J. Van Soest 1 , R. Fijten 1 , A. Dekker 1 , L. Wee 1 1 GROW School for Oncology and Developmental Biology- Maastricht University Medical Centre+, Department of Radiation Oncology MAASTRO, Maastricht, The Netherlands Purpose or Objective In a groundbreaking investigation, Aerts et al. (1) showed that quantitative imaging features (radiomics) could potentially be used to decode information about tumour phenotype that is relevant to disease prognosis, but not directly interpretable via an unaided human eye. This publication has been the subject of intense interest ever since, and there have been numerous requests for more information about the datasets – RIDER, Interobserver, Lung1 and Head-Neck1. To support research into repeatability, reproducibility, generalizability and explainability in radiomics, we have now made the clinical follow-up, extracted pyRadiomics (2) features and DICOM metadata of these 4 datasets more readily findable,

Conclusion The radiobiologically based plan evaluation allows for an individual optimisation of the dose prescription for patients treated with the SIP concept. While increasing the clinical workload, the presented method works as a decision-making tool and can help to increase plan quality in terms of a lower predictive NTCP for a stable P + . PO-1542 Integrating biomarker performance in sample size calculations for therapeutic trials C. Oberije 1 , R. Lieverse 1 , Y. Van Wijk 1 , L. Dubois 1 , A. Van der Wiel 1 , D. Marcus 1 , S. Sanduleanu 1 , P. Lambin 1 1 Maastricht University, Precision Medicine- D-Lab and M- Lab- GROW - School for Oncology & Developmental Biology, Maastricht, The Netherlands Purpose or Objective Precision medicine relies heavily on biomarkers to deliver the right treatment to the right patient. However, the performance of biomarkers is often overestimated and based on small patient populations. In addition, quality control and laboratory standard operating procedures can differ, resulting in differences in sensitivity and specificity. Moreover, confidence intervals for the estimated performance further increase the uncertainty. The aim of this simulation study was to investigate the effect of using realistic biomarker performance (e.g. markers for hypoxia, biallelic or monoallelic DNA repair defects, regulatory T (Treg) cells) on sample size calculations for a randomized clinical trial, that uses an enrichment design. In such a clinical trial the study population, selected based on biomarkers, is more likely to benefit from the treatment than in an unselected population. Material and Methods We carried out sample size calculations for a time-to- event study, based on a total number of 288 scenarios. We used hazard ratio’s (HR) ranging from 0.1 to 0.6, a power of 80% and 90%, a two-sided alpha of 0.05, and a range of positive predictive values (PPV) for the biomarker (30, 40,