ESTRO 2020 Abstract Book

S71 ESTRO 2020

SP-0151 Statistical Process Control for the analysis of IGRT data R. Louwe 1 1 Wellington Blood & Cancer Center, Radiation Oncology, Wellington, New Zealand Abstract text Statistical Process Control (SPC) is a method that was developed to monitor the stability of manufacturing processes, and provides a means to decide whether or not to investigate outliers in a consistent and time/cost- efficient way. SPC has increasingly been used in radiation oncology during the last decade, predominantly to monitor the results of treatment machine quality control (QC) and patient-specific treatment plan QC. However, only a few studies describing applications of SPC in image-guided radiotherapy (IGRT) are available in literature. Current SPC applications in IGRT include the verification of image registration consistency, monitoring patient deformation, and assessing process changes and the accuracy of patient positioning over time. These studies all employed univariate SPC charts both for individual patient applications and for larger patient cohorts. However, when large data sets are analysed that include many patients and multiple quality metrics for each patient, the data may be correlated. Multivariate SPC charts, such as the multivariate generalisation of the exponentially weighted moving average (EWMA) chart, take these correlations into account and may be well suited for the analysis of ‘big data’ in IGRT, but this is currently still uncharted territory. This presentation will explore examples of multivariate SPC charts including a re-analysis of previously published results (Figure 1), and will discuss potential benefits and additional requirements of this approach. In addition, important aspects such as the sensitivity and specificity of control charts, as well as the impact and management of measurement uncertainty will be discussed.

SP-0152 “Rapid learning”: Using real world data to improve clinical practice G. Price 1 1 Manchester Cancer Research Centre, Radiation Related Research Department 58 The Christie NHS Foundation Trust, Manchester, United Kingdom Abstract text Real world data – the information routinely collected about patients over their care pathway – offers an opportunity to provide evidence where Randomized Controlled Trials (RCTs) are not practical. There is an unmet need for such approaches in radiotherapy where many changes to practice are not suited to RCTs meaning there is often only limited assessment of their impact on clinical outcomes. The quantity and quality of data collected in modern radiotherapy mean it is ideally suited to such analyses. Furthermore, if real world data can be used to evaluate the effect of changes to radiotherapy practice, it opens the door to the use of iterative quality improvement techniques to optimize treatments. In this approach, often called rapid learning, a change to practice is made, its effect evaluated, and this information used to refine the next change before testing its effect again. It has the potential to transform the way in which new technologies and protocols are introduced into the radiotherapy clinic. It is not yet, however, in widespread use. This lecture will explore the promise of rapid learning and consider some of the challenges to its routine implementation. It will discuss the advantages and disadvantages of working with real world data in different ways, comparing the use of selective ‘simple trials’ and Trials within Cohorts (TwiCs) to before-after and time- series analyses. As well as examining the trade-offs in the evidence produced by different methodologies we will discuss their practicalities, including consideration of different patient consent models. Finally we will use a case study of heart sparing in lung radiotherapy to discuss the steps that need to be taken to move rapid learning into the clinic.