Reevaluation of the Bayesian network model to support immunotherapy decision-making in recurrent/metastatic head and neck squamous cell carcinoma.

purposeRecurrent and metastatic head and neck squamous cell carcinoma (RM-HNSCC) is a challenging malignant disease due to limited treatment options and heterogeneous therapeutic responses. Despite Immunotherapy with PD-1 inhibitors such as Nivolumab and Pembrolizumab has become a standard of care, it is still complex and multifactorial to make a clinical decision in giving the patient its best individualized therapy. This study refines the use of Bayesian networks (BN) in application of supporting therapy selection in RM-HNSCC, integrating clinical, pathological, and molecular features.

methodsThe previously introduced immune-oncologic BN for head and neck cancer was rechallenged by clinical data from 82 patients with RM-HNSCC. Therapy decisions (Nivolumab vs. Pembrolizumab) were calculated from the primary patient data. Model performance was assessed by comparing predicted versus actual therapy using retrospective data.

resultsThe reevaluation of the immune-oncologic BN revealed an inconsistency in the conditional probabilities of the initial model leading to a correction. Consequently, the model achieved 94.7% predictive accuracy for Nivolumab administration (18/19 cases) and 95.2% (60/63) for Pembrolizumab, yielding an overall prediction accuracy of 95.1% with a Cohens Kappa of 0.859. The model provided transparent reasoning paths for therapy selection and enabled hypothetical simulations in clinical scenarios.

conclusionThe immune-oncologic BN offers a promising approach to enhance personalized immunotherapy selection in RM-HNSCC patients by integrating diverse clinical variables into an interpretable decision support framework. The results support further prospective validation and clinical integration. Given the increasing complexity of biomarker-driven therapy, probabilistic models may contribute to more consistent and evidence-aligned treatment decisions.