ESTRO 2021 Abstract Book

S327

ESTRO 2021

Figure 2: (a-b) Case 15.03-051, KBP suggested improvements in small and large bowel dose which resulted in improved plan delivered to the patient and (c-d) Case 15.03-71 showing improvements in the 12.5 Gy isodose line relative to bowel, resulting in improved robustness to bowel position variation at treatment Conclusion KBP for real-time treatment plan review is feasible and demonstrated ability to improve treatment plan quality in two cases. Challenges include interpretation of KBP results with respect to clinical trade-offs, and determination of appropriate plan quality improvement criteria. OC-0427 Statistical monitoring for a large multi-institutional clinical study in radiation oncology S. Ecker 1 , C. Kirisits 1 , Y. Seppenwoolde 2 , A. De Leeuw 3 , M. Schmid 1 , A. Sturdza 1 , J. Knoth 1 , R. Pötter 1 , K. Tanderup 4 , R. Nout 2 , N. Nesvacil 1 1 Medical University of Vienna, Department of Radiation Oncology- Comprehensive Cancer Center, Vienna, Austria; 2 Erasmus MC Cancer Institute, Department of Radiation Oncology, Rotterdam, The Netherlands; 3 University Medical Centre Utrecht, Department of Radiation Oncology, Utrecht, The Netherlands; 4 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark Purpose or Objective Monitoring clinical studies ensures patient safety, protocol compliance and data integrity. Major challenges are the volume and variety of collected data. The goal is to reach a high level of data quality and early identification of issues that may compromise patient safety or trial integrity. The ongoing multi-institutional EMBRACE II study for advanced radiotherapy treatment of cervical cancer collects over 1000 clinical, treatment and outcome related variables in >1000 patients from >50 centers. A monitoring tool has been developed to reach these goals in an efficient, automated, and accurate way. This tool is presented. Materials and Methods Software was developed with the programming language R, which produces a monitoring dashboard hosted on a server. It implements a monitoring plan that was designed by a multi-disciplinary team. The plan defines monitoring activities and methods. Two main methods are implemented within this strategy. First, we focus monitoring activities on areas that have a large risk to impact patient safety or trial integrity. This risk-based approach uses pre-defined rules and thresholds to monitor data quality, protocol compliance and outcome. Rules and thresholds were defined using evidence from previous studies and expertise from the research team (Tab. 1). Secondly, for monitoring data quality we use machine learning algorithms to detect outliers and data inconsistencies that can come in an unlimited variety and are therefore hard to capture by pre-defined rules. This approach scans the entire dataset and flags potential errors (Fig. 1). Participating centers are automatically informed of any findings via individualized reports. Results The monitoring tool produces interactive tables and graphs which facilitate the understanding of the comprehensive study data. Summary tables allow informed assessment of the current state of disease and morbidity outcome, specific protocol-related hypotheses, and data quality. This allows early identification of emerging risks and helps to determine whether certain aspects of the study should receive more attention. At present the data quality checks automatically detected 389 issues in 310/920 (34%) of patients. 355 (91%) were found by the pre-defined rules, and 34 (9%) by machine learning algorithms. Performance of the latter was assessed by human inspection of the flagged values. 3/34 (9%) were classified as false positives that