A comparison of logistic regression and artificial neural networks in predicting central lymph node metastases in papillary thyroid microcarcinoma.

Objective: Prophylactic central lymph node dissection(CLND) is a controversial issue in papillary thyroid microcarcinoma( PTMC) patients without lymphatic metastasis. Artificial neural network(ANN) has been proposed as an alternative statistical technique for predicting complex biologic phenomena. Our aim is to develop an ANN model in predicting central lymph node metastases(CLNM) in patients with PTMC, in comparison to traditional logistic regression(LR) analysis.

Study Design: Eighty patients who underwent total thyroidectomy plus CLND for PTMC were included in the study. The factors associated with CLNM were determined by using both ANN model and LR analysis. The predictive performances of these two statistical models were compared.

Results: Twenty (25%) patients had CLNM. In univariate analysis, age >45 years, tumor diameter >7 mm, and multifocality were the associated parameters with CLNM. These parameters were used to create LR and ANN models. LR test revealed tumor diameter >7 mm and multifocality as independent factors for CLNM. ANN (AUC: 0.786) had a higher predictive value for CLNM, in comparison to LR model (AUC: 0.750).

Conclusions: Tumor diameter >7 mm and multifocality are the independent prognostic indicators of CLNM in patients with PTMC. ANN model has higher predictive value for determining central metastasis, in comparison to LR analysis.

Key Words: Artificial neural networks, Central lymph node metastasis, Logistic regression, Papillary thyroid microcarcinoma.