ESTRO 2025 - Abstract Book

Brachytherapy - Gastro-intestinal, paediatric brachytherapy, miscellaneous

S183

ESTRO 2025

Purpose/Objective: Locally advanced rectal cancer (LARC) is a major contributor to rectal cancer morbidity and mortality worldwide (1). In most countries, radical surgery remains the standard treatment for the majority of patients. However, many patients, especially some elderly individuals, have high surgical risk and may wish to consider alternative treatment options (2). Consequently, organ preservation techniques involving (chemo)radiotherapy, sometimes with a Contact X-ray Brachytherapy (CXB) boost have gained popularity (3). This study aims to develop a novel clinical prediction model for CXB in rectal cancer, aiming to provide clinicians with an algorithm to predict overall survival (OS) and recurrence free survival (RFS) in LARC patients treated with (chemo)radiotherapy and a CXB boost. Material/Methods: This study developed and validated a multivariable prediction model using data from 575 patients treated with CXB at Clatterbridge Cancer Center from 2008 to 2019. Model development adhered to transparent reporting of a multivariable prediction model for individual prediction model prognosis or diagnosis (TRIPOD) guidelines (4), and internal validation was performed using 1,000 bootstrapped iterations. Discriminative ability was assessed using Area Under the Curve (AUC), and calibration was assessed through the Hosmer-Lemeshow test and brier score and displayed on calibration curves.

Results:

Use of concurrent chemotherapy, patients’ age at diagnosis, tumour size and fitness for surgery emerged as key predictors for OS at 3-year and 5-years. The OS models demonstrated strong discriminative ability [AUC of 0.76 (0.728 - 0.816) Figure 1 and 0.77 (0.728 - 0.807) respectively] with the calibration plots suggesting potential overfitting, corroborated by the brier score and Hosmer-lemehsow test as shown in Figure 2 . The model for 3-year OS achieved positive predictive value (PPV) of 90.1% and negative predictive value (NPV) of 40.26%, while the 5-year OS had a PPV of 82.81% and NPV of 58.97%. The RFS models were moderately accurate, with AUCs of 0.62 (95% CI: 0.5730-0.6953) and 0.63 (95% CI: 0.5730 0.6953) for 2-year and 3-year RFS, respectively. The 2-year RFS PPV was 78.7% with an NPV of 36.0%, while the 3 year RFS PPV was 83.8% with an NPV of 30.3%.