A probabilistic model of bilateral lymphatic spread in head and neck cancer

Current guidelines for elective nodal irradiation in oropharyngeal squamous cell carcinoma (OPSCC) recommend including large portions of the contralateral lymphatic system in the clinical target volume (CTV-N), even for lateralized tumors with no clinical lymph node involvement in the contralateral neck. This study introduces a probabilistic model of bilateral lymphatic tumor progression in OPSCC to estimate personalized risks of occult disease in specific lymph node levels (LNLs) based on clinical lymph node involvement, T-stage, and tumor lateralization. Building on a previously developed hidden Markov model for ipsilateral lymphatic spread, we extend the approach to contralateral neck involvement. The model represents LNLs I, II, III, IV, V, and VII on both sides of the neck as binary hidden variables (healthy or involved), connected via arcs representing spread probabilities. These probabilities are learned using Markov chain Monte Carlo (MCMC) sampling from a dataset of 833 OPSCC patients, enabling the model to reflect the underlying lymphatic progression dynamics. The model accurately and precisely describes observed patterns of lymph node involvement with a compact set of interpretable parameters. Midline extension of the primary tumor is identified as the primary risk factor for contralateral involvement, with advanced T-stage and extensive ipsilateral involvement further increasing risk. Occult disease in contralateral LNL III is highly unlikely if upstream LNL II is clinically negative, and in contralateral LNL IV, occult disease is exceedingly rare without LNL III involvement. This model offers an interpretable, probabilistic framework to inform personalized elective CTV-N volume reduction. For lateralized tumors that do not cross the midline, it suggests the contralateral neck may safely be excluded from elective irradiation. For tumors extending across the midline but with a clinically negative contralateral neck, elective irradiation could be limited to LNL II, reducing unnecessary exposure of normal tissue while maintaining regional tumor control.