Abstract 162P
Background
High-quality evidence to provide specific recommendations on the nature of clinical and imaging surveillance after radiotherapy for NSCLC is lacking. Risk-stratification models are needed to determine optimal follow-up. We present a radiomics model to predict overall survival (OS) 2 years post-treatment, using radiotherapy planning CTs and the gross tumour volume (GTV), contoured for radiotherapy purposes as the region of interest for feature extraction.
Methods
A retrospective multi-centre study of patients receiving SABR or conventional (chemo)radiotherapy for stage I-III NSCLC was undertaken. Cases with a GTV encompassing the primary tumour were included from 5 hospitals, divided into training, validation, and external test sets. For classification purposes, survival status (alive/dead) at 2 years from the start of radiotherapy was recorded. We explored a combination of 9 feature reduction techniques with 11 machine learning classifiers. The combination resulting in the highest validation set AUC was selected and deployed on the external test set. and the model was bench-marked against TNM stage. Youden index, calculated from the validation set ROC curve, was used to define high and low risk groups.
Results
509 patients were included, training=302, validation=75, test=132. Median follow-up was 762 days. Train-validation and test set mean age was 74 and 71 respectively. Death rates at 2 years were 33.7% vs 32.6%. Respective validation and test set AUCs and 95% confidence intervals are shown in the table. Our model had superior AUC to the TNM model in both validation and test sets. Kaplan Meier curves showed marked separation with significant log-rank tests. Table: 162P
| OS Model | Validation Set Results | External Test Set Result | ||||
| AUC | 95% CI | AUC p-value | AUC | 95% CI | AUC p-value | |
| Radiomic | 0.712 | 0.592-0.832 | 0.013 | 0.685 | 0.585-0.784 | 0.621 |
| TNM | 0.573 | 0.442-0.704 | - | 0.663 | 0.579-0.77 | - |
Conclusions
We present a validated and externally tested survival-stratification model that uses routinely available radiotherapy data. This model could be integrated into the radiotherapy workflow to inform personalised surveillance at the point of treatment for patients with NSCLC.
Clinical trial identification
NCT04721444.
Legal entity responsible for the study
The Royal Marsden NHS Foundation Trust.
Funding
UKRI AI for Healthcare Centre for Doctoral Training, Imperial College London and the ICR and Royal Marsden NIHR BRC.
Disclosure
All authors have declared no conflicts of interest.