Development and validation of a prediction-score model for distant metastases in major salivary gland carcinoma.

Abstract

6085 Background: We developed and validated a prediction-score for distant metastases (DM) in major salivary gland carcinoma (SGC). Methods: Patients with SGC treated with curative-intent surgery +/- postoperative radiation therapy (PORT) at 4 tertiary cancer centers were divided into discovery (institution A&B) and validation (institution C&D) cohorts. Multivariable analysis using competing risk regression was used to identify predictors of DM in the discovery cohort and create a prediction score. The optimal score cut-off for high vs low-DM risk was determined using a minimal p-value approach. The results were subsequently evaluated in the validation cohort. The cumulative incidence and Kaplan-Meier methods were used to analyze DM and overall survival (OS), respectively. Results: Overall, 1035 patients were included (Table). In the discovery cohort, DM predictors (risk score coefficient) were: positive margin (0.6), pT3-4 (0.7), pN+ (0.7), lymphovascular invasion (LVI; 0.8), and high risk histology* (1.2). High DM-risk SGC was defined by sum of coefficients greater than 2. In the discovery cohort, the 5-year cumulative incidence of DM for high vs low risk SGC was 50% vs 8%; p < 0.01; these results were similar in the validation cohort (44% vs 4% at 5 years; p < 0.01). In the combined cohorts, this model predicted distant-only failure (40% vs 6%, p < 0.01) and late ( > 2yr post surgery) DM (22% vs 4%; p < 0.01). Patients with high DM-risk SGC had an increased incidence of DM in the subgroup receiving PORT (46% vs 8%; p < 0.01) or concurrent chemotherapy (71% vs 34%; p < 0.01). The 5-yr OS for high vs low risk SGC was 48% vs 92% (p < 0.01). Conclusions: This validated prediction score model may be used to identify SGC patients at increased risk for DM and select those who may benefit from prospective evaluation of treatment intensification and/or surveillance strategies. Baseline characteristics. [Table: see text]