LYMPH NODES STATUS IN T1 ESOPHAGEAL CANCER: A PREDICTIVE MODEL BASED ONLY ON CLINICAL CHARACTERISTICS
Giovanni Capovilla*, Elisa Sefora Pierobon, Lucia Moletta, Luca Provenzano, Giorgia Cornaviera, Renato Salvador, Mario Costantini, Stefano Merigliano, Michele Valmasoni
Padova University Hospital - Center for Esophageal Diseases, Padova, Italy
BACKGROUND
T1 esophageal cancer could be successfully treated with endoscopic mucosal resection or submucosal dissection. The radicality of the procedure depends on the T stage but also on the N stage. At present we don’t have sufficiently accurate diagnostic methods to assess the lymph nodes status and hence to make a definitive treatment decision, in particular for patients with low risk for esophagectomy. Some predictive scores have been proposed to asses the nodal status to guide the treatment choice, but they all contemplate one or more independent variables derived from surgical specimen. The aim of this study was to build a predictive model using only independent variables derived from clinical practice, before any endoscopic or surgical decision are made.
METHODS
From our prospectively maintained esophageal cancer patients database, we selected all patients who underwent esophagectomy, without any prior chemoradiation treatment, that resulted to have a T1 disease and N status defined on pathology. The dependent variable of our model is the dichotomous lymph nodes status (positive or negative) and so we used a binomial logistic regression model. We selected as independent variables all that satisfied statistical significance criteria or clinical background knowledge criteria, and then we simplified our model with backward elimination method, guided by the Akaike Information Criterion (AIC) that permits to identify models with high goodness-of-fit and to penalise overly complex models.
RESULTS
The dataset used to define the model counted 89 patients with pT1 esophageal cancer (M/F=66/23 (74/26%), mean age 60.06 yrs range 34-79 yrs). Tumor location was upper esophagus in 17 (19%), middle in 30 (34%) and lower in 42 (47%) patients. Median endoscopic tumor length was 20 mm (SD±19.47). Pathology: 53 patients had SCC and 36 adenocarcinoma; 36 tumors were pT1a and 53 pT1b, lymph nodes status was negative in 77 (87%) cases and positive in 12 (13%, 2 pT1a and 10 pT1b); grading was G1 in 26 (32%), G2 in 47 (54%) and G3 in 15 (14%) cases. After recursive model optimisation, in the final model the predictors resulted to be patient age, tumor location, tumor grading, tumor histology and tumor length. Our model shows statistical significance (p=0.04), and the ROC curve have a good area under the curve (AUC) of 0.82.
CONCLUSION
Our predictive model is based on simple and easily obtainable clinical variables. This model has to be validated, and we are now collecting data from centers of our national society for the diseases of the esophagus to perform an external validation, in order to verify and refine the model as a useful tool to complement the choice of the optimal treatment for T1 esophageal cancer.
Back to 2020 Abstracts