This study used receiver operating characteristic curve (ROC) to analyse surveillance, epidemiology and end results (SEER) adenoid cystic carcinoma data to identify predictive models and potential disparity in outcome.

For the risk modelling, each factor was fitted by a generalised linear model to predict the cause-specific survival. The area under the ROC was computed. Similar strata were combined to construct the most parsimonious models. A random sampling algorithm was used to estimate the modelling errors. Risk of adenoid cystic carcinoma death was computed for the predictors for comparison.

There were 5,947 patients diagnosed from 1973 to 2009 included in this study. The mean follow-up time (SD) was 93·8 (90·6) months. Three out of five patients were women. The mean (SD) age was 58·55 (16·01) years. SEER stage was the most predictive factor of outcome (ROC area of 0·68). Sex, radiotherapy and surgery had ROC areas of about 0·57. None of the socio-economic disparities was found for treatment outcome. Radiotherapy was underused in localised and regional stages when the intent was curative, especially in older patients.

Anatomical staging was predictive and useful in treatment selection. Understaging and underuse of radiotherapy may have contributed to poor outcomes.