Probabilistic modelling of bilateral lymphatic spread in oral cavity squamous cell carcinoma for personalised elective nodal treatment

OBJECTIVE: Current guidelines for management of the neck in oral cavity squamous cell carcinoma (OCSCC) recommend elective nodal treatment of a large part of the regional lymphatic system. We present a model of bilateral lymphatic metastatic progression that estimates the risk of occult nodal involvement based on a patient's clinically diagnosed lymph node level (LNL) involvement, T-stage and tumour extension, enabling a more personalised risk assessment. APPROACH: We extended a previously published hidden Markov model (HMM) of lymphatic metastatic progression in oropharyngeal squamous cell carcinoma to the oral cavity. The lymphatic drainage system is represented as a directed acyclic graph (DAG). The DAG's nodes correspond to the primary tumour and LNLs I-V. The DAG's edges are associated with probabilities of tumour spread to and between LNLs, and encode correlations between LNLs. These probabilities were inferred using Markov chain Monte Carlo (MCMC) sampling on the basis of a dataset of 407 OCSCC patients. Different DAGs were evaluated using model evidence computed via thermodynamic integration to identify the structure that best reflects lymphatic spread patterns in OCSCC. MAIN RESULTS: Assuming an acceptable risk threshold of approximately 5% for occult metastases per LNL, the model suggests that ipsilateral LNL III should be included in elective treatment only if LNL II is clinically positive, and to treat LNLs IV and V only if involvement in LNL III is confirmed. For lateralised tumours not crossing the midline and a clinically negative contralateral neck, the contralateral neck may be excluded from elective treatment. In tumours crossing the midline, elective treatment of contralateral levels could be limited to LNLs I and II. SIGNIFICANCE: The presented model provides an interpretable probabilistic framework for estimating patient-specific risk of occult disease and may inform future guidelines for personalised elective nodal treatment.