ESTRO 2024 - Abstract Book

S1305

Clinical - Head & neck

ESTRO 2024

Results:

The model was generated on 937 patients, with a median age of 54 years (range 21 - 85 years). Pathological T3 and T4 patients constituted 339(36.2%) patients. The median tumor size and depth were 2.7cm and 1mm respectively. . 115(12.3%) had well differentiated, 602 (64.2%) had moderately differentiated and 220 (23.5%) had poorly differentiated tumours. Margins were involved for 22 (2.3%) patients and close (<0.5cm) for 117 (12.5%) patients. Lymphovascular invasion was present in 331 (35.3%) and perineural invasion was present in 411 (43.9%). There was cortical bone invasion in 192 (20.4%). 397(42.4%) had nodal metastases with the total number of positive nodes varied between 0 and 58. Extranodal extension was present in 204 (21.8%) patients. The median follow up was 36 months. The 2-year and 5-year DFS was 74% and 63% respectively The global likelihood ratio χ2 test (LR) for prediction of DFS was 243.17 with 20 degrees of freedom and Pr(>χ2 ) of <0.0001. The Akaike information criterion on χ2 was 203.17. The optimism corrected R2 was 0.193 and c-index was 0.74. The model showed excellent calibration for prediction of 2-yr DFS as shown in Figure 1.

Figure 1: Calibration plot

Conclusion:

A multivariable individual prediction model for DFS was built and internally validated with robust discrimination and calibration. We plan to extend this approach to a multi-institutional model with independent external validation for general use.

Keywords: oral cancer, prognostic model, disease outcomes