Melanoma sentinel node biopsy and prediction models for relapse and overall survival.

BACKGROUND: To optimise predictive models for sentinal node biopsy (SNB) positivity, relapse and survival, using clinico-pathological characteristics and osteopontin gene expression in primary melanomas. METHODS: A comparison of the clinico-pathological characteristics of SNB positive and negative cases was carried out in 561 melanoma patients. In 199 patients, gene expression in formalin-fixed primary tumours was studied using Illumina's DASL assay. A cross validation approach was used to test prognostic predictive models and receiver operating characteristic curves were produced. RESULTS: Independent predictors of SNB positivity were Breslow thickness, mitotic count and tumour site. Osteopontin expression best predicted SNB positivity (P=2.4 × 10⁻7), remaining significant in multivariable analysis. Osteopontin expression, combined with thickness, mitotic count and site, gave the best area under the curve (AUC) to predict SNB positivity (72.6%). Independent predictors of relapse-free survival were SNB status, thickness, site, ulceration and vessel invasion, whereas only SNB status and thickness predicted overall survival. Using clinico-pathological features (thickness, mitotic count, ulceration, vessel invasion, site, age and sex) gave a better AUC to predict relapse (71.0%) and survival (70.0%) than SNB status alone (57.0, 55.0%). In patients with gene expression data, the SNB status combined with the clinico-pathological features produced the best prediction of relapse (72.7%) and survival (69.0%), which was not increased further with osteopontin expression (72.7, 68.0%). CONCLUSION: Use of these models should be tested in other data sets in order to improve predictive and prognostic data for patients.