Session Item

Temporal lobe radionecrosis (TRN) is a well-described side effect following high-dose proton beam therapy (PBT) to the base of skull (BoS) for radioresistant tumours, such as chordomas and chondrosarcomas. Due to their close proximity to the BoS, the temporal lobes are particularly at risk in these diagnoses, where incidence of symptomatic high-grade TRN is reported to be up to 10%. High grade TRN can have a major impact on quality of life and is sometimes life-threatening. Accurate estimation of risk is useful in the clinic for consent purposes and, whenever possible, plan optimisation. Schröder et al. developed a normal tissue complication probability (NTCP) model to quantify the risk of TRN on a structure (individual temporal lobe) or patient basis [1]. However, external validation of NTCP models is necessary to ensure their applicability in different settings. The aim of our study is the validation of this model, for its implementation into clinical practice.

Our data included patients (pts) treated via the proton overseas programme in Jacksonville, Florida between 2010 – 2018, before a National Health Service PBT service became available in the UK. Pts were included if they had received >60 Gy(RBE) PBT for BoS chordomas/chondrosarcomas and had ≥12 months follow-up data. Following the parameters identified by the Schröder model, data were collected on total dose received, age at treatment, hypertension, D1cc left and right temporal lobes and presence/absence of CTCAE grade (G)≥2 TRN. We calculated the patient-wise predicted NTCP using the formula in the Schröder paper.

A total of 93 pts were identified. Median age 48 years (range 7-77), median total dose received 73.8 Gy(RBE) (range 66 – 75.6). After a median follow up of 46 months (range 12 – 122), 8 pts (9%) developed G≥2 TRN. In 2 pts, the grade of TRN could not be ascertained from the notes and was described as unknown. The box plots of predicted patient-wise NTCP vs G≥2 TRN presence (figure 1) indicate good model discrimination. Patient-wise predicted NTCP values gave an AUC-ROC of 0.80, with a dichotomic threshold NTCP=0.2 optimising specificity (78%) and sensitivity (75%) (figure 2). Pts who went on to develop G≥2 TRN had a median predicted probability according to the Schröder model of developing toxicity of 29% (interquartile range 19-40%) versus 11% (interquartile range 6-18%) in those without G≥2 TRN.

Our results show a comparable high AUC-ROC value as in the Schröder paper, demonstrating that the model has a good predictive ability for TRN development. The predicted patient-wise NTCP of 0.2 could be suggested as a cut off, above which pts are at increased risk of developing TRN, however further work is required to develop a practical clinical tool.

Reference
[1] Schröder et al. NTCP Modeling for High-Grade Temporal Radionecroses in a Large Cohort of Patients Receiving Pencil Beam Scanning Proton Therapy for Skull Base and Head and Neck Tumors.Int J Radiat Oncol Biol Phys 2022:1–8.