Abstract Background: Adenoid cystic carcinoma (ACC) of the head and neck is a rare malignancy marked by high recurrence, late metastasis, and unpredictable clinical behavior. Traditional prognostic tools often fall short in capturing complex outcome determinants—particularly those related to racial disparities. This study leverages machine learning (ML) to predict survival and examine race-associated risk patterns in ACC. Methods: We analyzed 1,771 patients with histologically confirmed ACC from the SEER database (2004–2015). Demographic, clinical, and treatment variables were included in both traditional survival analysis (univariate and multivariate Cox regression) and predictive modeling using eight supervised ML algorithms. SHAP (Shapley Additive Explanations) analysis was used to interpret model predictions and identify the most influential prognostic features. Results: Age, tumor TNM stage, and tumor site emerged as independent predictors of survival. Neural networks outperformed other models, achieving the highest AUC (0.899) and accuracy (0.774). SHAP analysis highlighted advanced age, advanced stage, and absence of radiation therapy as top contributors to poor prognosis. Higher mortality was observed in people with black race (HR: 1.36, P = 0.026). Older age was strongly correlated with worse survival (HR: 4.00, P 0.0001), female patients had better prognosis than males do (HR: 0.83, P = 0.032). Higher AJCC stages (III+) were significantly correlated with increased mortality (HR: 3.89, P = 1.36×10-27). Patients without distant metastasis (M0) had significantly lower risk than those with (HR:0.16, P 0.0001).Tumors located in the maxillary sinus were linked to worse outcomes. Among regression models, linear regression, ridge, and lasso models showed almost identical performance outputs with an average MSE of 1969, MAE of 32.93, and R2 of 0.222 for the testing dataset, indicating minimal overfitting with consistent performance. The Support Vector Regression model achieved the highest R2 value of 0.247 (test) with lower MSE and MAE compared with the other models, making it the model of choice for predicting patient outcomes. Conclusions: Black patients with head and neck ACC experienced significantly worse survival, highlighting a persistent racial disparity not fully explained by tumor biology alone. Machine learning models, particularly neural networks, offer powerful and interpretable tools for individualized prognosis. These findings underscore the need to integrate ML into clinical workflows and to address structural inequities that influence cancer outcomes in historically underserved populations. Citation Format: Fares Qtaishat, Mohammad-Amer Tamimi, Yousef Ateiwi, Ahmad AlKayyat, Adham Musa, Sarah Abdallah, Layan AlDaher, Jana Tarawneh, Mohammad Alghaniem, Sara Qutaishat, Ramez Odat. Racial disparities and machine learning-based survival prediction in head and neck adenoid cystic carcinoma: A SEER cohort analysis abstract. In: Proceedings of the 18th AACR Conference on the Science of Cancer Health Disparities; 2025 Sep 18-21; Baltimore, MD. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2025;34(9 Suppl):Abstract nr C006.
Qtaishat et al. (Thu,) studied this question.