VALIDATE VALIdation of Decision support models for rAdioThErapy

Validation projects

Four areas where published models are being put to the test across member institutions. Each project follows the seven commitments: the model travels, the data stay put, the analysis plan is locked first, and both discrimination and calibration are reported. The working programme was last updated at the VALIDATE meeting at ESTRO 2026, 17 May 2026.

Local tumour control

Diffusion-weighted MRI (ADC) as a predictor of local control in head-and-neck cancer

Leads: Warren Floyd and Sara Thrower (MD Anderson Cancer Center); René M. Winter (NTNU, Trondheim)

Scientific question

Does the pre-treatment apparent diffusion coefficient (ADC) of the primary tumour, and of involved lymph nodes, predict local failure after definitive radiotherapy for head-and-neck cancer?

Approach

  • Co-register diffusion-weighted images to T2-weighted images and exclude noisy diffusion voxels from the regions of interest.
  • Evaluate an expanded ADC band and a wider range of b-values.
  • Delineate tumours and nodes with AI auto-segmentation developed at Memorial Sloan Kettering, excluding cystic regions, and harmonise the imaging between participating sites.

Long-term xerostomia

Dose to the parotid stem-cell region compared with mean parotid dose

Leads: Maria Thor (Memorial Sloan Kettering Cancer Center); Peter van Luijk (UMCG); Per Munck af Rosenschöld (Lund University)

Scientific question

Is the dose to the stem-cell-rich central-duct region of the parotid gland a stronger predictor of long-term xerostomia than the mean parotid dose, as reported in the model's development cohorts?1

Approach

  • A containerised pyCERR pipeline delineates the parotid glands, the masseter and the van Luijk central-duct point of interest at every site.
  • Xerostomia endpoint definitions are harmonised across saliva-based, patient-reported and physician-rated scales.
  • A DICOM and outcome-data quality checklist is applied before any model is fitted.

Lymphopenia and survival in thoracic radiotherapy

Published lymphopenia and overall-survival models across independent centres

Lead: Laura Cella (CNR, Naples)

Scientific question

Do published models of radiotherapy-related lymphopenia, and of overall survival, generalise across independent centres treating thoracic cancers?

Models under validation

  • The Manchester lymphopenia model, with predictors including mean lung dose, age, baseline lymphocyte count, concurrent chemotherapy and the volume of the thoracic vertebrae receiving 20 Gy, and the companion overall-survival model with lymphopenia, concurrent chemotherapy, radiotherapy duration and prescribed dose.2
  • The MD Anderson clinical nomogram developed in oesophageal cancer, with age, baseline lymphocyte count, body-mass index, treatment technique and planning-target volume.3

Approach

  • A common data-collection template and dose–volume analysis across the participating centres.
  • An open methods question is whether to use AI auto-contouring or each centre's clinical contours: robustness to contouring variability against the bias that automated contours may introduce.

Radiation pneumonitis

Mean lung dose combined with pre-treatment FDG-PET uptake of normal lung

Lead: Maria Thor (Memorial Sloan Kettering Cancer Center); parallel interstitial-lung-disease track: Andrew Hope (Princess Margaret Cancer Centre)

Scientific question

Does mean lung dose combined with the 90th percentile of FDG uptake (SUV-P90) in the normal lungs on pre-treatment PET predict early radiation pneumonitis, including immunotherapy-associated cases, across institutions?

Models under validation

  • The Thor et al. model, developed in stage III non-small cell lung cancer treated with chemoradiation and immunotherapy and prospectively validated at the developing institution.4,5

Approach

  • Apply the model to standard dose–volume data plus the pre-treatment FDG-PET SUV-P90 of the normalised lungs, metrics that existing dose–volume and PET pipelines already extract.
  • Compare calibration in immunotherapy-treated and non-immunotherapy sub-cohorts.
  • A deep-learning pneumonitis model with peritumoral attention maps is a candidate secondary arm; an interstitial-lung-disease scoring approach from Princess Margaret runs as a parallel track.

Investigators with a published model they would like to see validated, and institutions that would like to take part, are welcome to contact the Chairs.

References

  1. van Luijk P, Pringle S, Deasy JO, et al. Sparing the region of the salivary gland containing stem cells preserves saliva production after radiotherapy for head and neck cancer. Sci Transl Med 2015;7(305):305ra147. doi:10.1126/scitranslmed.aac4441
  2. Abravan A, Faivre-Finn C, Kennedy J, McWilliam A, van Herk M. Radiotherapy-related lymphopenia affects overall survival in patients with lung cancer. J Thorac Oncol 2020;15(10):1624–1635. doi:10.1016/j.jtho.2020.06.008
  3. van Rossum PSN, Deng W, Routman DM, et al. Prediction of severe lymphopenia during chemoradiation therapy for esophageal cancer: development and validation of a pretreatment nomogram. Pract Radiat Oncol 2020;10(1):e16–e26. doi:10.1016/j.prro.2019.07.010
  4. Thor M, Lee C, Sun L, et al. An 18F-FDG PET/CT and mean lung dose model to predict early radiation pneumonitis in stage III non-small cell lung cancer patients treated with chemoradiation and immunotherapy. J Nucl Med 2024;65(4):520–526. doi:10.2967/jnumed.123.266965
  5. Thor M, Apte A, Grkovski M, et al. Prospective validation of a pretreatment 18F-FDG PET/CT and mean lung dose model for early radiation pneumonitis. Phys Imaging Radiat Oncol 2025;36:100844. doi:10.1016/j.phro.2025.100844