ESTRO 2024 - Abstract Book

S190

Brachytherapy - GI, paediatric, miscellaneous

ESTRO 2024

Rectal cancers are the fourth most prevalent type of cancer in men and women in the United Kingdom (1). Contact X-ray brachytherapy (CXB), also known as Papillon, has the potential to avoid invasive surgery and permanent stomas in a subset of rectal cancer patients (2). The UK National Insititute for Health and Care Excellence (NICE) has recommended this organ-sparing approach for patients not suitable for surgery (3). A clinical prediction model can prognosticate an individual's treatment outcome, identifying patients who are more likely to have a favourable outcome under a wait-and-watch regimen for rectal cancer (4). However, there are currently no standardised, clinically predictive models for rectal cancer patients who undergo treatment with CXB. We aim to identify, summarize, and critically appraise published prediction models for radiotherapy in rectal cancer patients to aid in developing a contact therapy model in the future. This review was prospectively registered (CRD42022277704) and was performed in accordance with the Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) guidelines (5,6). PubMed, Embase, Scopus, Web of Science and Cochrane CENTRAL, and Medical Literature Analysis and Retrieval System Online (MEDLINE) databases were interrogated using search terms "Rectal Cancer", "Contact Brachytherapy", "Radiotherapy", "Clinical Prediction Models", and their synonym terminologies. The literature review was performed by two independent reviewers, and disagreements were resolved by discussion with a third reviewer. Data Extraction was done using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist, and the Prediction Model Risk of Bias Assessment Tool (PROBAST) was utilized for evaluating the risk of bias in these studies (7). The extracted data included study setting, patient recruitment, demographic information, selection of predictors, modelling techniques, validation methods, and performance measures. The primary outcome was to assess the predictive accuracy of various models in evaluating the length of survival and the histopathological response to radiotherapy. This was achieved through subgroup analyses, which combined models with similar outcomes and summarized calibration statistics and discrimination statistics. Additionally, this review assesses the robustness of analysis and applicability in real-world clinical practice of selected models. Five thousand nine hundred and thirty-three studies were reviewed, and eighteen models published before April 2023 fulfilled our predefined inclusion and exclusion criteria. Models predicting rates of histopathological response based on Tumor Regression Grade (TRG) demonstrated an Area Under Curve (AUC) of 0.81 with a 95% Confidence Interval (CI) of 0.71 to 0.93. Similarly, models predicting pathological complete response (pCR) had an AUC of 0.76 with a 95% CI between 0.69 and 0.83 (Figure 1). These values were based on apparent performance on the training cohorts and both subsets had 3% heterogeneity on the I^2 test. Although these results are encouraging, existing models have drawbacks related to either retrospective data collection or single-centre prospective sampling. This may result in a high risk of bias, as these methods might fail to consider all available predictors, compromising the study's overall robustness and the model's predictive power. All but two studies lacked true external validation, and all papers failed to comment on rates of missing data and the use of imputation, to account for missing raw data, in their respective samples (Figure 2). Heterogeneity and inconsistency in reported calibration measures prevented us from performing a formal analysis of models reporting on the length of survival (overall survival or disease-free survival). Furthermore, this review excluded all papers that had evaluated the role of radiomics. The Cochrane review group has yet to publish Material/Methods: Results: